public Hibrarj) This Volume is for REFERENCE USE ONLY ^(y^[^l^(^iy^tr8vl[^t^t^(y8?l^^t^t^ • • ••••• •• • • • ? • • •. , • •• •• • • • • • • • • • • • • • • • • • • • • • • HE BEL L S Y S T E M / mcai i VOTED TO THE SCIENTIFIC ^^r>^ AND ENGINEERING PECTS OF ELECTRICAL COMMUNICATION LUME XXXII JANUARY 1953 ,?NfUMBER 1 Surface Properties of Germanium WALTEB H. BRATTAESr AND JOHN BARDEEN 1 Frequency Economy in Mobile Radio Bands KENNETH BULLINGTON 42 Intermodulation Interference in Radio Systems WALLACE C. BABCOCK 63 Magnetic Resonance: Part I— Nuclear Magnetic Resonance KARL K. DARROW 74 Delay Curves for Calls Served at Random John riordan 100 The Evaluation of Wood Preservatives — Part I REGINALD H. COLLEY 120 Motion of Gaseous Ions in Strong Electric Fields GREGORY H. WANNIER 170 Abstracts of Bell System Papers Not Published in this Journal ^55 Contributors to this Issue ^^^ COPYRIGHT 1953 AMERICAN TELEPHONE AND TELEGRAPH COMPANY TRB B^lELJ.:|nf^?FSM, l^CHNI^ JOURNAL ADVISORY BOARD S. BRACKEN, President, Western Electric Company F. R. KAPPEL, Vice President, American Telephone and Telegraph Company M. J. KELLY, President, Bell Telephone Laboratories EDITORIAL COMMITTEE E. I. GREEN, Chairman A. J. B U S C H F. R. L A C K W. H. DOHERTY J. W. MCRAE G.D.EDWARDS W. H. N U N N J. B. FISK H. I. ROMNES R. K. HONAMAN H. V. SCHMI D T EDITORIAL STAFF M. E. STRIEBY, Managing Editor R. L. SHEPHERD, Production Editor THE BELL SYSTEM TECHNICAL JOURNAL is published six times a year by the American Telephone and Telegraph Company, 195 Broadway, New York 7, N. Y. Cleo F. Craig, President; S. Whitney Landon, Secretary; Alexaii ott, Treasurer. Subscriptions are accepted at $3.00 per year. Single cupK-a are 75 cents each. The foreign postage is 65 cents per year or 11 cents per copy. Printed in U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XXXII JANUARY number 1 Copyright, 195S, American Telephone and Telegraph Company Surface Properties of Germanium By WALTER H. BRATTAIN* and JOHN BARDEENt (Manuscript received September 3, 1952) The contact potential (c.p.) and the change of contact potential with il- lumination {Ac.p.)l of several germanium surfaces have been measured. The reference electrode used was platinum. It was found that the c.p. could be cycled between two extremes about 0.5 volts apart by changing the gaseous ambient. Ozone or peroxide vapors gave the c.p. extreme corresponding to the largest dipole at the Ge surface. Vapors with OH radicals produced the other extreme. There is a one to one correlation between c.p. and (Acp.)/, . For 12-ohm cm n-type Ge (Ac.p.)L was large and positive when the surface dipole was largest, decreased to zero and became slightly negative as the surface dipole decreased to its smallest value. The variation for 12-ohm cm p-type Ge was just opposite as regards both sign and dependence on surface dipole. The surface recombination velocity was found to be inde- pendent of c.p. For a chemically prepared surface it was 50-70 cm/sec and 180-200 cm/ sec for n and p-type surfaces respectively. A theory is given that explains the results in terms of surface traps, Na per cm donor- type traps near the conduction band and Nb per cm acceptor-type traps near the filled band. A quantitative fit with experiment is obtained with only one free parameter. The results are direct evidence for the existence of a space charge layer at the free surface of a semiconductor. INTRODUCTION Every one is familiar with the fact that it is necessary to expend energy to remove an electron from a conducting solid. This energy is * Bell Telephone Laboratories. t University of Illinois. The contributions of the second author to this work started while he was a Member of the Technical Staff of Bell Telephone Labora- tories and continued at the University of Illinois. 2 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 called the work function. The work function is caused in part by a charged double layer or dipole at the solid surface. In metals this dipole extends over a distance of the order of 10~ cm. In semiconductors how- ever part of the dipole extends into the semiconductor to a distance of the order of 10~® to 10"^ cm depending on the properties of the semi- conductor. This part of the surface dipole is called the space-charge layer. The rest of the surface dipole has approximately the same extent as in metals. In the space between any two conducting solids there is a contact potential caused by the difference between the work functions of the two surfaces. One can measure this contact potential by several meth- ods. We have used an adaptation of the well known method of Kelvin. If one has a reference electrode whose work function remains constant, then by measuring the c.p. between this electrode and another surface one can measure any changes in work function or total dipole of this second surface. The above method has been used to study the properties of the germanium surface in a gaseous ambient at atmospheric pressure. It has been found that the total dipole at the germanium surface can be changed by changing the ambient and further that by proper control of the ambient the surface can be cycled back and forth between two extremes of small or large dipole corresponding to a c.p. change or work function difference of the order of one-half volt. If one upsets the thermal equilibrium in the germanium by creating excess electron-hole pairs near the surface, the potential of the surface will change until a steady state is reached. When the extra electron- hole pairs are introduced by illuminating the surface with light, the potential change shows up as a measurable change in contact potential between the reference electrode and the Ge surface.^ It has been found that this contact potential change on illumination (Ac.p.)l is large and positive on n-type Ge when the surface dipole is large, then decreases to zero and becomes slightly negative as the surface dipole decreases to the smaller extreme. For p-type Ge the (Ac.p.)^ is large and negative when the surface dipole is small and go6s through zero and becomes slightly positive as the surface dipole increases to the larger extreme. One can describe qualitatively what is going on as follows. The extra hole and electron pairs created by the light diffuse either to the interior or to the surface to recombine. The recombination in the interior is governed by the body life time r. The surface recombination is charac- terized by a recombination velocity v, . When the surface is illuminated its potential, with respect to the interior, changes until the combined Surface properties of germanium 3 flow of holes and electrons to the surface and interior is just equal to the rate they are being created by the light. The sign and magnitude of the potential change for a given illumination depends on the body properties of the germanium and on the size of the space charge layer. These experimental results are direct evidence for the existence of a space charge layer at the free surface of a semiconductor. They not only confirm the results obtained for silicon surfaces but go much further in that they enable one to determine how the layer is changed by the gaseous ambient used. It* is kno\\Ti that the surface recombination velocity, Vs , can be changed, by large factors, by surface treatment.^ For mechanically treated surfaces Vs approaches thermal velocities. Every hole or electron striking the surface recombines. For such a surface it is found that (Ac.p.)l is too small to be measured. On the other hand Vs can be as low as 100 cm/sec for chemically polished or etched surfaces such as those used in the experiments where (Ac.p.)l was measured. In this case one wishes to know how Vs depends on the gaseous ambient. This was meas- ured for the same surface used in measuring (Ac.p.)z, and it was fou^d that, for the ambients used, Vs is approximately a constant and there- fore independent of the other surface changes. A quantitative theory, some details of which are in the Appendix, has been formulated to explain the results. It is proposed that there are two types of recombination traps at the surface: donor type, Na per cm^, with energies, Ea , near the conduction band and acceptor type, Nb per cm^, with energies, Eb , near the filled band. Surface recombination takes place by electrons and holes successively going into one of the two types of traps. To account for the fact that Vg is unchanged by changes in ambient, it is assumed that the concentrations of these traps are independent of ambient. Changes in c.p. with ambient are assumed to result from adsorption and desorption of fixed ions which are at an effective distance ^^2X 10"^ cm outward from the surface traps. A sche- matic energy level diagram is given in Fig. 13, to be discussed later. The charge of the ions is compensated mainly by charges in the surface traps w^hich, together with the ions, form a double layer, A large part of the change in c.p. with ambient results from changes in this double layer. There is also a change in barrier height, —eVsy associated with the redistribution of electrons in the traps. An increase in negative ions on the surface requires a decrease in number of electrons in traps, and thus a higher barrier. Part of the change with light, (Ac.p.)l , occurs in the body of the semiconductor and part occurs across the barrier layer. Changes in Vb 4 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 and occupancy of the traps compensate in their effect on surface re- combination, so that Va is unchanged by ambient. This means that changes in concentrations of electrons and holes in the interior with illumination and thus the body contribution to (Ac.p.)l are independent of ambient. Changes in (Ac.p.)l with ambient result soley from changes in Vb and are in the directions we have described earlier. The concentrations of carriers, n and p, and their change with light are obtained as follows. Drift mobilities, Hn and jip , and the equilibrium product, np, are knowTi from earlier experiments of G. L. Pearson, J. R. Haynes and W. Shockley.^ From these, and the resistivity of the sample, which was measured, n and p can be determined. The light source is calibrated in terms of hole-electron pairs created per cm per sec. The recombination rate is determined from the body life time r and the surface recombination velocity. From these latter three measurements, one can calculate the steady state density of electrons and holes near the surface when it is illuminated. Mention should be made here of the fact that oxygen has been found to play a definite role on the Ge surface. The large extreme in dipole is obtained when active oxygen (ozone) is introduced into the gaseous ambient. Peroxide vapors have the same effect. The other extreme is produced by vapors having an OH radical, water vapor, alcohol etc. A number of vapors not falling in either of the above classes have little or no effect on the surface dipole. Another result is that the difference in work function or dipole between n- and p-type Ge is small. This is to be expected from previous work. We shall first discuss the experimental technique and then give the experimental results. The main conclusions of the theory will then be outlined and compared with these results. EXPERIMENTAL METHOD The Ge surface to be measured and the reference electrode are mounted under a bell jar. Oxygen or nitrogen as desired is allowed to flow through the bell jar at a rate of approximately 2 liters per min. The volume of the bell jar is 16 liters. The gas used flows over a drying column of silica-gel and then calcium chloride. Means are provided for bubbling this gas through any desired liquid before it enters the bell jar. A spark discharge can be run in the gaa flow line. The reference electrode is placed parallel to the Ge surface about 1 mm away. It is mounted on a vibrating reed which is driven, electromagnetically, at its resonant frequency of about 90 cycles per second and at an amplitude of the order of 0.1 mm. This varies the capacity sinusoidally giving rise to an electrical signal when any potential, SURFACE PROPERTIES OF GERMANIUM 5 contact or other, exists between the surfaces. When the proper dc potential is applied between the surfaces this signal goes to zero. If no other poten- tials are present this dc potential is equal and opposite to the contact potential. A phase reference method* is employed to determine this balance point with a relative accuracy of d=o X lO"'^ volts. A diagram of the Ge- reference electrode circuit is shown in Fig. 1 . Care must be taken to shield this circuit. Stray capacity reduces sensitivity and should be minimized. Charged insulators inside the shield will produce an apparent c.p. All con- ducting surfaces other than the Ge should be relatively far from the moving reference electrode. The surface is illuminated through a compound lens system by focusing the filament image, of a suitable projection bulb, on the germanium surface. This light passes through the grid of the reference electrode which removes about 10 per cent of the light. The light can be modulated by a square wave chopper, so that (Acp.)/, can be measured on an ac basis. Both Ta and Pt reference electrodes have been used. The Pt electrode appears to be somewhat more constant. If the Ge surface is replaced by a gold electrode the contact potential difference is practically independent of the changes in the gas ambient. The arrangement is such that two samples can be mounted in the bell jar and the reference electrode moved from one to the other without opening the bell jar. In this way two surfaces can be compared, without any question arising of long time drifts in the reference electrode. EXPERIMENTAL RESULTS /. Change of c.p. between Ge and Pt reference electrodes as a function of the gaseous ambient WTien this work was started the object was to find some means of varying the c.p. It was thought that the actual values of the c.p. would Ge SAMPLE reference ^ electrode' POTENTIOMETER C^ TO AMPLIFIER Fig, 1 — Schematic of experimental circuit. H.R. Moore designed and made the electronic equipment used to do this. 6 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 be highly dependent on past history of the Ge sample. If however one could measure one or more other properties, such as (Ac.p.)l and Vs , on the same surface at the same time, then one could look for correla- tions between these properties. In this manner one might be able to eliminate the past history as a factor. From previous experiments in the highly variable ambient of room air we knew the order of magnitude of the variation to be expected. At first it was impossible to produce this range of variation under the bell jar. The contact potential always drifted in the direction of a positive extreme, i.e., small total dipole at the Ge surface. The only way found to get a really large change in the opposite direction was to lift the bell jar and expose the sample to room air. These phenomena were finally traced to the presence of negative ions, possible salt ions, in the room air. These were not present in the oxygen and nitrogen supplies we used for creating the ambient in the bell jar. It was found that the opposite c.p. extreme could be produced, under the bell jar, by running a spark discharge in dry oxygen as it was flowing into the system. The next step was to cycle the c.p. from one extreme to the other and back again. The procedure was to start with the spark discharge in dry oxygen, change to either wet O2 or wet N2 and to end with dry O2 . The development of this dependable and reproducible cycle was a great aid to the proposed study. Fig. 2 is a plot of contact potential versus time for a single crystal slice D of Ge cut from a melt that was p-type. The surface was prepared by removing some of the Ge with a silicon carbide (180 mesh) blast of approximately ten pounds air pressure. The Ge was then mounted in the bell jar within one-half minute of the "sandblast," and the dry O2 flow started. The c.p. was followed for a few minutes to be sure ever5rthing was working properly. 0.6 20 24 26 32 TIME IN MINUTES Fig. 2 — Contact potential cycles for sandblasted sample D. SURFACE PROPERTIES OF GERMANIUM 0.8 0.7 0.6 ? 0.5 _i < Z 0.4 0.3 0.2 ^ ^— N ->— N . •-— * 11 ^ J / — V \ / ^ r »-..,^_^^^ • A-. ___^^ L/ r / /^ — -•"^ 1 ^ y .--•' \ 8 12 16 20 24 28 32 36 40 44 48 52 TIME IN MINUTES Fig. 3 — Contact potential cycles for etched sample A. In Fig. 2 zero time is taken after the spark discharge was run in the O2 flow line for 2 minutes. This started the first cycle. After approximately 17 minutes the O2 was made to bubble through H2O. Fifteen minutes or so later the O2 was changed back to dry. At this time, 32 minutes, the flow rate was increased by a factor of three. The c.p. was followed for about 17 minutes, then the process was repeated. The results and the reason for the choice of time intervals, are all evident from a study of Fig. 2. The spark discharge in O2 decreases the c.p. After this treatment the c.p. increases with time, most of the change occurring in the first 15 minutes. The wet O2 then increases the c.p. to a maximum value which is reached in about 15 minutes. Finafly the dry O2 reduces the c.p. It is evident that there is a quasi-equflibrium value of c.p. in dry O2, to which the c.p. returns after either extreme treatment. At first there is quite a large shift from cycle to cycle but this shift gradually disappears as the cycling is continued. In Fig. 2 cycles 2, 4, 6 and 10 are shoA\Ti. Very little change takes place after cycle 10. Such results have been obtained many times over a period of two years. When al- lowance is made for shifts in work function of the reference electrode and for the fact that the experimental technique improved as the work progressed it is found that all the results for a given sample of Ge are very consistent. In Fig. 3 are sho^vn similar results for an n-type slice A. In this case the surface was first ground or sandblasted to remove any films, and THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 0.8 20 24 28 32 TIME IN MINUTES Fig. 4 — Contact potential cycles for etched sample D. then given a polishing etch (CP-4). After the etch the surface was washed in running distilled water of reasonable quality. The surface was then dried with filter paper and kept covered until placed in the bell jar. From here on the procedure was the same as before. In this case cycles 1, 3 and 11 are shown. The surprising result is the similarity be- tween Fig. 2 and Fig. 3. To a first approximation one is tempted to say that the dipole of a Ge surface, in the bell jar atmosphere is independent of past history. Within certain limits this is approximately true. There are differences between Fig. 2 and Fig. 3 but they are small and probably due to differences in surface treatment. Fig. 4 shows the results for p-type slice, D, when this surface is etched as above. This slice and the slice A were placed in the bell jar at the same time. Cycle 1, Fig. 3, was taken on A, cycle 2, Fig. 4, on D and so on. Any differences between the results in these two figures are to be attributed to the differences in samples. They cannot very well be ascribed to the reference electrode and the initial surface treatments were as nearly the same as they could be made with reasonable care. By making runs of this type two samples at a time, different samples and different surface treatments can be intercompared. This method eliminates the shifts in the work function of the reference electrode that sometimes occur from run to run. Such resulte can \)e illustrated by plotting the data for different samples and different surface treatments for cycles 10 or greater where the resulte for successive cycles are the same. This has been done in Fig. SURFACE PROPERTIES OF GERMANIUM 9 5 where one cycle each is plotted for both sample A and D, for each of the two surface treatments, sandblasting and etching. The curves are approxi- mately all of the same shape so that the differences between them can be described by giving the shift in contact potential necessary to superimpose the curves. This treatment works very well except for the first part of the cycle after the spark coil where the shift necessary is sometimes more and sometimes less than that needed to make all of two curves superimpose well. Note in Fig. 5 that the contact potential for sample A etched is always greater than for sample D etched. In the case of the sandblasted surface just the reverse is true. Also the contact potential for the etched surface of either sample is always larger than for the sandblasted. Using these methods comparable results were obtained on two other n-type samples, C and E, of increasingly lower specific resistance. Taking advantage of the relation between carrier concentration and the position of the Fermi level we have plotted in Fig. 6, the contact potential for both the etched and the sandblasted surfaces versus the position of the Fermi level (Ep — Ei) in electron volts. The contact potential values used were taken from the saturation values in wet O2 but the shapes of the curves w^ould be much the same for any other point in the cycle. The contact potentials are thought to be accurate to about 0.01 volts. Solid lines have been drawn through the points. Note that the contact potential for the etched surface is always greater than for the sandblasted surface, i.e., the work function is always less and that this difference is greater when Ep = Ei or the germanium is nearly intrinsic. 20 24 28 32 TIME IN MINUTES Fig. 5 — Comparison of cycles for samples A and D with sandblasted and etched surfaces. 10 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 0.80 < a 0.76 cr wO.72 0.68 z ^0.64 I- z UJ ^:o.60 0.52 /^ -^ / / \ ETCHED ^ y \ V ^ A \ ^ANDBLASTED V / -ai2 -0.8 -0.4 0 0.4 0.8 0.12 0.16 0.20 0.24 0.28 Ef-Et, IN ELECTRON VOLTS Fig. 6 — Contact potential dependence on position of the fermi level. For variety Fig. 7 shows results on c.p. for a sandblasted polycrystal- line sample of silicon. It is apparent that similar phenomena are taking place on the silicon surface, however the behavior is different. These preliminary results are shown to illustrate the generality of this method for investigating semiconductor surfaces in a controlled gaseous ambient. The time between cycles was somewhat variable. The first and second cycles were always taken immediately after the specimen was placed in the bell jar. After this, successive cycles were taken one or two a day over a period of a week or more. The bell jar was not opened during a run and a small flow of dry gas was maintained between cycles. Little if anything occurred during these idle periods, showing that the changes that did take place were due to the cycling. //. Change of contact potential with illumination When the Ge surface is sandblasted the change of contact potential with illumination (Ac.p.)i, is too small to be measured. If however the surface has been prepared by the polishing etch, the (Ac.p.)z, is easily observed. This change if not too small can be measured by finding the balance on the potentiometer for light off and light on. When the con- tact potential is changing with time, or if the change is small this is difficult to do. A better procedure is to chop the light at a definite frequency and meiusure the amplified output on an ac meter. Tliis L>ives a continuous residing that can be read easily at any given lime. If a filter is used to pass only those frequencies near the fundamental of SURFACE PROPERTIES OF GERMANIUM 11 the chopping frequency, the improvement in signal to noise enables one to read very small changes. A plot of (Ac.p.)l in volts versus the contact potential for samples A n-type and D p-type is shown in Fig. 8. For the n-type sample the signal is large and positive when the contact potential is small and becomes small and negative as the contact potential increases. Except for the shift in the contact potential where (Ac.p.)/, goes through zero, the re- sults for the p-type sample are practically the opposite of those for the n-type sample. Similar curves have been obtained cycle after cycle in many complete runs. Within experimental error the curves have the same shape for a given sample for all cycles in all runs. Sometimes the curve for the first cycle in a run will differ in shape from the rest. How- ever there are shifts such that the c.p. for which the light goes through zero (c.p.)o does vary from cycle to cycle throughout a run. Fig. 9 is a plot of (c.p.)o versus cycle number for p-type sample D and n-type sample A. The data are plotted for two distinct runs. Both of these units were in the bell jar when the measurements were taken. Consistent results of this kind give one confidence that the Pt reference electrode is staying constant. These experimental results can be summarized as follows. All data for a given sample can be superimposed by shifts in contact potential scales for the different cycles. A plot of (Ac.p.)l versus c.p. — (c.p.)o can be represented by a single curve for each unit. Moreover all the curves for all units both n- and p-type have quite similar shapes provided in the latter we plot [ — (Ac.p.)i,] versus (c.p.)o — c.p. The shape of this curve is shown in Fig. 8 for a given sample and Fig. 9 shows how (c.p.)o varies from cycle to cycle. These plots adequately describe the results. 20 24 28 32 TIME IN MINUTES Fig. 7 — Contact potential cycles for silicon. 12 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 0.024 0.020 0.016 0.012 -0.006 0.004 q: (J -0.004 < -0.008 0.012 - 0.016 ■0.020 -0.024 \ \ \ , UNIT A Vn-TYPE \ \ a \^ N ^v^ ■N \ ^■*^ y— \ s. \ UNIT D ^ p-TYPE \ \ \ 0.35 0.40 0.45 0.50 0.55 0.60 0.65 0.70 0.75 0.80 C.R IN VOLTS Fig. 8 — Change of contact potential with illumination versus contact po- tential, samples A and D etched. Not shown in Fig. 8 because of the scale used is the result that (Ac.p.)l for n-type material does not increase indefinitely as c.p. decreases but approaches a maximum value. Likewise — (Ac.p.)l for p-type approaches a maximum as c.p. increases. Figures illustrating these results will be discussed after the theory is presented. Some of the experimental details require discussion. It is necessary to calibrate the ac response in terms of absolute potential change. This can be done by comparing the ac reading with the dc reading when the light signal is large. One can also do this by introducing a known square- wave signal across the potentiometer in Fig. 1, and reading the ac signal out. Both methods agree when allowance is made for variation of the light signal with frequency. The latter response is almost flat from 25 to 300 cycles, but there is evidence for some very low frequency com- ponents in the dc measurements of (Ac.p.)^ . When the light signal goes through zero the signal is small and the dc value is difficult to read. SURFACE PROPERTIES OF GERMANIUM 13 At times evidence has been obtained to indicate that the dc value changed sign. At other times the dc (Ac.p.)l behaved as if the place where it went through zero was shifted in c.p. from the point where the ac signal goes through zero. In view of this it was necessary to prove that the ac signal was changing phase at the zero point and not just going down in the noise and then increasing again without phase change. This was done by comparing the phase of the signal with a signal from a photocell placed in the same chopped light beam. By this means it was proved conclusively that the ac light signal was actually going through zero. Some data were obtained on change of contact potential with light on n-type samples C and E having progressively smaller specific re- sistances. It was found that (Ac.p.)l decreased with specific resistance. It also decreased into the noise as the contact potential was increased, so that it could not be determined if it changed sign as for samples A and D. Because of the smaller signal (Ac.p.)z, could not be measured easily except by the ac method and so far the signal has not been cali- brated properly. Some preliminary data on p-type silicon indicate that (Ac.p.)i, for this sample was negative and that it decreased as c.p. was decreased. The magnitude of (Ac.p.)l for the same light intensity was much larger than for germanium and so far it has not been found to go through zero and change sign in the experimental range. III. Other methods of varying the c.p. In some cases N2 was used in place of O2. A spark discharge in the N2 had very little effect on the c.p. On the other hand wet N2 produced much the same effect as wet O2. The positive extreme in c.p. was about 0.1 volt greater in the case of wet N2. After the wet treatment dry N2 was not nearly as effective as dry O2 in reducing the c.p. to its inter- mediate value. This can hardly be due to a difference in dryness of the two gases since the same drying column was used in both cases. The results indicate that dry N2 tended to leave the surface in whatever condition obtained before the dry N2 flow was started, and that O2 counteracts the effect of H2O. With dry N2 as a carrier, other vapors were tried. A. N. Holden sug- gested trying a peroxide and picked out ditertiary butyl peroxide as being reasonably safe. Use of this vapor was found to produce the same changes as the spark coil in the O2 flow. Other vapors having OH radicals such as methyl alcohol and acetic acid were found to act the same way 14 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 0.8 0.7 [3 0.6 O > ^0.5 q: U 0.3 =^s^^ -^ A €3- ^ ^"a UNIT A n-TYPE / •r / UNIT D o A ^ 0 - A i 1 p-T YPt r f A ^^o ^-^"o 0 2 4 6 8 10 12 14 1 CYCLE NUMBER 3 18 20 Fig. 9 — Contact potential for zero light effect, (c.p.)o , versus cycle number, two runs for each sample. as water vapor. To prove that this was not caused by traces of water in the alcohol or acid the N2 w^as bubbled through water solutions of H2SO4. These results indicated that one needs appreciable amounts of water vapor to produce the effect, much more than could be present in the alcohol or acid. Other vapors, such as carbon tetrachloride, methyl- chloride, nitrobenzene and ether, were found to have no effect on either the contact potential or (Ac.p.)l . Acetone has a small effect in the same direction as water. This is to be expected because this compound exists in part in a tautomeric form having an OH group. Vapor from 30 per cent H2O2, 70 per cent H2O acted at first like a peroxide vapor and with a longer time of exposure behaved like water vapor. Small amounts of CI2 gas in N2 produced the same change as the spark discharge in O2 and after 14 minutes of flushing the bell jar with N2 produced an ad- ditional effect when water vapor was introduced. On n-type samples before the usual increase in c.p. and decrease in (Ac.p.)l there was a large increase in (Ac.p.)^, . This was attributed to the reaction between the water vapor and the CI left on the Ge surface, producing oxygen. The nature of the change was a rapid shift in (Ac.p.)o and thus a mo- mentary increase in (Ac.p.)l . This effect is only large in the first cycle. It indicates that the shifts in (Ac.p.)o plotted in Fig. 9 are probably due to an oxidation of the Ge surface as the cycling progresses. In all these experiments the relation between the c.p. and (Ac.p.)^ was essentially the same as that obtained in the standard cycle. One can detect the presence of thin surface films by electron diffrac- tion techniques. R. D. Heidenreich took electron diffraction pictures of a germanium surface immediately after the polishing etch and water SURFACE PROPERTIES OF GERMANIUM 15 wash. He found the surface to be quite clean, with a film thickness less than 10 A. He also took pictures after the germanium surface had been cycled about fifteen times and found in this case definite evidence for a thin surface film. The film was either amorphorus or composed of very small crystals. He estimated the thickness to be between 20 and 50 A. In the discussion of the experimental work it has been assumed that all the changes in c.p. are to be attributed to the Ge surface and not to the Pt. It is not easy to give a definite proof that this is true. The fact that it is a reasonable assumption is suggested by the nature of the re- sults themselves. Almost identical results were obtained using a Ta electrode. In Fig. 10 we show the results of tw^o cycles when the Ge was replaced wdth gold. While there are some changes they are almost an order of magnitude smaller than the changes when Ge is present. It follows that one would expect much the same results with either Ta, Pt or Au reference electrodes and that most of the changes are due to the Ge. No change of c.p. with illumination is observed when both electrodes are metals. In one case a small amount of H2 was added to the N2 flow. Here the c.p. between Pt and Ge changed rapidly but the (Ac.p.)i, did not change at all. Subsequent runs using the regular cycle indicated that the c.p. scale had been shifted corresponding to a reduc- tion in work function of the Pt. Except for this shift the results were the same and the shift disappeared in about one day. The conclusion was that H2 had little effect on Ge but reacted w^ith the Pt decreasing its work function. This is a good illustration of the powder of this method of measuring more than one property of a semiconductor surface at the same time. If only c.p. had been measured the conclusions would not have been so clear cut. If one knew the work function of the Pt electrode then one would know the work function of the Ge. Work functions are all measured in high vacuum. We have not been able to think of a method of determin- ing the work function of any electrode in a gaseous ambient unam- 0.2 en 0.1 CL L)-0.1 -0.2 £i4°Ju^°.4o^ H±I4^ O f> o 24 28 32 36 40 44 48 52 10 4 8 12 16 20 TIME IN MINUTES Fig. 10 — Contact potential cycles when germanium is replaced by gold. 16 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 600 500 400 300 200 100 60 60 50 40 30 20 O 1ST CYCLE A 2ND CYCLE D 3RD CYCLE K °A o B A > o a UNIT A D D p- o TYPE Sa D D 6 [J o - ■ A o -^ °n 0^ 8^ O Da -ZS" UNIT A n-TYPE o D 0 D [Jo 12 20 24 28 32 TIME IN MINUTES 36 40 46 52 Fig. 11 — Dependence of the velocity of recombination, y« , on the contact potential cycles. biguously. It may well be impossible to do this in Bridgman's operational sense. Most physicists would agree that the Pt reference electrode probably has a work function of around 5 to 7 volts under these con- ditions, but this of course does not help much. IV. Measurement of surface recomhination velocity Va At first we tried various methods of measuring Va on the Ge surface at the same time that c.p. and (Ac.p.)l were measured. No method was found that could be trusted. As the study progressed we realized that the c.p. of a Ge surface could be cycled in a reproducible way. Since the proper geometry for measuring Vg was a rod or filament,^ rods were prepared of samples A and D with the same chemical surface treatment. The decay of hole electron pairs, created by a point light source, was measured as a function of distance from the light. If the body life time t and the dimensions of the filament are known one can then determine v, . These measurements were made while the gaseous ambient was cycled in the same manner as before. The results for fila- ments cut from samples A and D are plotted against time in Fig. 11. The ambient atmosphere was changed as a function of time just as it was in Figs. 2 to 4. The main conclusion is that within the accuracy of this experiment there is no evident dependence of v, on the changes in SURFACE PROPERTIES OF GERMANIUM 17 gas ambient and therefore no dependence on the corresponding changes in surface dipole. This result was somewhat unexpected and the first time the experiment was performed it was hard to believe that the previously measured changes in c.p. actually were taking place. In this case the gas ambient was experimented with to try to change Vs and it was found that Vs could be changed from the order of 10^ cm/sec to greater than 10^ cm/sec and back again by exposure of the filament to (NH)40H fumes and then HCl fumes respectively.^ While interesting, this has no direct bearing on the other experiments. The experiment was performed again with freshly prepared rods and this time the cycling used in the c.p. experiments was rigidly adhered to, giving Vg equal to 70 cm/sec and 200 cm/sec approximately for samples A and D respectively. The experiments were then repeated again using new filaments cut from the samples as close as possible to the surfaces used in the c.p. experiments leading to the results shown in Fig. 11, namely, Va equal to 50 cm/sec and 170 cm/sec respectively. From these experi- ments it was concluded that Va is approximately constant in the range involved and is determined by the nature of the sample and the surface treatment used. It was noted in some of these experiments that Va for the first cycle was somewhat larger than for the subsequent cycles. This change when it occurred is probably to be correlated with the changes in the early cycles in the c.p. measurements. Similar measure- ments on sample C gave Vg equal to 1500 cm/sec. No measurements of Va were made on samples B and E. V. Other experimental measurements The specific resistance of each sample was measured near the surface used in the experiments. It w^as approximately constant across the surface but did vary slowly with depth in some of the samples. The body life times were measured on each sample. The thickness of the slices used was intentionally made large compared to their cor- responding diffusion lengths, about 0.5 cm for A, B, C and E and about 2.0 cm for D. The mobilities were taken from J. R. Haynes^ measure- ments: Mn = 3600 and Mp = 1700 cmVvolt sec. There is some uncer- tainty as regards. the exact value of the equilibrium product of holes and electrons, np, at 300°K. We have used the value 6.3 X 10^® obtained from some unpublished data of G. L. Pearson. The light source was calibrated by replacing the germanium sample with one of F. S. Goucher's n-p junctions.^ The bell jar, with every- thing else including the reference electrode, was left in their normal 18 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 positions. An average was taken over the filament image, and the ef- fective area of the p-n junction was determined and allowed for. This was done for all light intensities used. Most of the experiments were performed with a fixed intensity and the averaged result for this in- tensity was 6.0 X 10^^, hole electron pairs per cm^ sec. The rate of pair production was found to be proportional to the light intensity. Since practically all the light is absorbed in a depth (10""^ cm or less) that is small compared to the diffusion length, it is a simple matter to calculate the steady state increase dp in hole electron pairs due to the light. The relation is 8p = N/{vs + Vd) (1) where N is the rate of pair production, Va is the velocity of recombination at the surface and va is the diffusion velocity for the minority carrier. Since N is proportional to light intensity it follows that bp is too. The magnitude of (Ac.p.)l should depend on the light intensity. One might at first expect it to be proportional to light intensity. That this is not the case is shown by curve 1 in Fig. 12. Curve 1 is a plot of (Ac.p.)l versus bp for unit D on a log-log scale. A smooth curve has been drawn through the experimental points. As we shall see later, theory predicts that if (Ac.p.)^ is large, it should be proportional to ln{\ 4- bp/a) where a is the equilibrium density of the minority carrier, In {-%) 10"' 2 3 4 5 681 2 34568 10 2 34 « 5 4 3 ^ 2 s • U 6 < 5 4 3 2 ../ ^=^ 1/ 3^ [/ ^ X yt ^{ / / / Ov^ ^ / ■ - / /^ - / y /' / > jr / / / , J JL 10' 2 3 4 5 6 6 )q12 2 3 4 5 6 6 ^q'^ 2 3 4 tfp NO. PER CM 3 Fig. 12 — Dependence of contact potential change with illumination (Ac.p.)^, on light intensity. SURFACE PROPERTIES OF GERMANIUM 19 n in p-type material and p in n-type. That this prediction is borne out is shown by how well the experimental points fit the straight line curve 2 in Fig. 12 where (Ac.p.)i, is plotted versus this quantity. The scale for 8p is shown along the bottom of Fig. 12 and that for tn(l -{- 8p/a) along the top. Similar results were obtained for unit A. In Table I the parameters, specific resistance p in ohm cm, life time r in microseconds and the surface recombination velocity Vs in cm/sec are given for each unit used. Also given are some pertinent quantities derived therefrom, namely the equilibrium densities of electrons and holes, n and p in number per cm and the increase in density at the surface 8p in number per cm^ when the rate of pair production due to the illumination was 6.0 X 10^^ per cm^ sec. THEORY The constancy of Vs throughout the range of surface dipole investi- gated puts rather stringent requirements on any theoretical model to be constructed. W. Shockley and W. T. Read^ have investigated the theory of recombination via traps. It is evident from their work that if one assumes a trap density peaked near a single energy and very small elsewhere, then Vs will be constant over a range of surface dipole values provided that the peak energy is either near the conduction band or near the filled band. The experimental results make it appear very unlikely that the trapping mechanisms on the n and p-type surfaces are essentially different. It is assumed that both types of traps are present on the surface and that the traps are approximately the same for both n- and p-type samples. Further, it is assumed that the traps Na of energy Ea near the conduction band are donor type, i.e., neutral when filled and positively charged when empty. Likewise the traps Nb of energy Eb near the filled band are assumed to be acceptor-type traps, i.e., neutral when empty and negatively charged when filled. The absolute charge on the traps is not important, however, because we are Table I "a n n V n n p MO i>0 T V, 5p C A B D C E 12.5 15 12.0 2.5 0.008 1.38 X 1014 1.14 X 1014 2.1 X 1012 7.0 X 1014 4.4 X 10" 4.56 X 1012 5.6 X 1012 3.0 X 1014 0.91 X 1012 1.45 X 109 900 600 4000 48 50-70 (100) 170-200 1.5 X 103 2.2 X 10" 1.9 X 10" 1.75 X 10" 1.67 X 10" 4.4 X 10-" 6.0 X 10-" 20.0 X 10-" 20 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 concerned only with differences between the occupied and unoccupied states. The fraction of the traps which are occupied depends on the trap energy and on the position of the Fermi level at the surface. The latter in turn depends on the condition that the surface as a whole be neutral. A very obvious mechanism for changing the total surface dipole is the adsorption or desorption of ions on the surface. If this happens the position of the Fermi level at the surface shifts until the total charge on the surface, adsorbed ions, charge in the traps and charge in space- charge layer, adds up to zero. The consequences of this model have been carried through. The basis for the theory is illustrated in the energy level diagram of Fig. 13. At the semiconductor surface there is a space-charge layer of thickness (b which gives a change in electrostatic potential of Vb , corresponding to a potential energy of an electron of —cVb . Outside of the surface of the germanium proper, there is a surface film of thick- ness Id . A double layer giving a potential change Vd , is formed from a charge of ions, cti , on the outer surface of the film and charges in the surface traps of types a and h. Changes in c.p. with ambient result from changes in o-/ and consequent changes in Vb and Vd . It is as- sumed that the remainder of the work function is independent of ambi- DOUBLE LAYER SPACE CHARGE LAYER Iq^IO-^CM X=0 INTERIOR Fif?. 13 — Schematic of energy level diagram at germanium surface. SURFACE PROPERTIES OF GERMANIUM 21 ent. When light shines on the surface, Vb is changed to Vb + ^Vb and there is an additional potential drop, 8Vi , in the body of the germanium resulting from the recombination current which flows to the interior. The change in contact potential with light is equal to 8Vb + dVi . The film thickness ^d is shown on an exaggerated scale. We expect //> ^ 10"^ cm and ^b '^ 10~^ cm, so that Ib^ ^d . TABLE OF SYMBOLS A. Energies: Ea = Ea (true) + kT In {oiunoc/fj^o^, is the effective energy of the a-traps for F^ = 0. Here oiunoc and Woc are the statistical weights of the unoccupied and occupied states, respectively. Eh = effective energy of the 6-traps for Vb = 0. Ec = energy of lowest state of conduction band in interior of semi- conductor just beyond the space-charge layer. Ev = energy of highest state of valence band at the same position. Ef = Fermi energy. Ei = Ef when material is intrinsic. Vd = potential drop across surface film. Vb = potential drop across space-charge layer. Vbo = value of Vb for which Ua = Pb , see below. Vq = value of Vbo for an intrinsic sample. B. Concentrations: n = Nc exp [{Ef — Ec)/kT] = equilibrium concentration (no./cm^) of conduction electrons in interior of semicon- ductor just beyond the space-charge layer. p = Nv exp [{E^ - EF)/kT] = corresponding hole concentra- tion. ni = intrinsic concentration. ns , Ps = equilibrium concentrations of electrons and holes, re- spectively, at the surface. Na,Nb = concentration (no./cm^) of a- and 5-traps, respectively. Ua = equilibrium concentration (no./cm^) of occupied a-traps. Pj, = iVft — rib = equilibrium concentration (no./cm ) of unoc- cupied 6-traps. UaOyPbo = values of na and pb for an intrinsic sample with Vb = 0. ni = n + 8n,pi = p + 8p, and n,i , psi , riai , Pbi = concentra- tions in presence of light. Electrical neutrality requires that 8n = 8p. 22 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 The theory is based on the following postulates: I. Changes in c.p. with ambient result from changes in ui and the consequent changes in Vb and Vd • c.p. = 7^ + 7^ + const. (2) The charge o-/ is largely compensated by charges in the surface traps. The barrier height, Vb , is determined by the requirement of electrical neutrality: (Ti = e{na — Vb) + const. (3) Since Vb and Vd are of the same order, the net charge per unit area in the space-charge layer will be smaller than ai in the approxim.ate ratio (d/Ib and may be neglected. II. Traps of type a have energies above Ep and of type h below Ep for all values of Vb attained in the different ambients used. More ex- actly Ea - eVB - Ep > kT, (4) Ep-E,+ bVb > kT, (5) for all Vb . One may then use the Boltzmann approximations for Ua and pb : _ Na '''^ - 1+ exp [{Ea - eVs - Ep)/kT] (6) - Na exp [{Ep - Ea + eVB)/kn __ Nb P* - 1-1- exp [{Er - Eb + eVB)/kT\ (7) -- Nb exp [{Eb - bVb - Ep)/kT]. It is not necessary for our arguments to assume that all traps of each type have the .same energy. The only requirement is that the distributions of trap energies are such that the a-traps are always above and the 6-traps always below the Fermi level for any ambient. III. Creation of electron-hole pairs by absorption of light occurs near the surface in a distance that is small compared with the diffusion length. Optical constants of germanium indicate that practically all of the light with energy sufficient to create electron-hole pairs is absorbed within a distance of 10~^ cm of the surface. The diffusion length is of the order of 0.2 cm. IV. In the presence of light, the concentration of electrons in a-traps SURFACE PROPERTIES OF GERMANIUM 23 is in equilibrium with the concentration of electrons in the conduction band and holes in 6-traps are in equilibrium with holes in the valence band. The barrier height is adjusted so that the total charge in both types of traps is unchanged by illumination. We shall show in the ap- pendix that the resistance to flow of electrons from the conduction band across the space-charge layer and into a-traps is small compared with the resistance to flow of electrons from the valence band to the traps. Similar considerations apply to flow of holes to 6-traps from the valence band as compared with flow from the conduction band. V. Recombination is limited by holes going into traps of type a and electrons going into traps of type h. The two types of traps act in parallel for recombination. The contributions to the surface re- combination velocity are proportional to psUa and UsPb , respectively. These products are independent of Vb and thus of ambient if postulates II and IV are satisfied. If other types of traps were important in re- combination, one would expect Vs to depend on ambient, contrary to what is observed. Postulates I and II are used to relate Vb and c.p. with changes in ambient. Postulates III, IV, and V are used to relate (Ac.p.)l with Vb and the trap densities, and also to obtain an expression for the surface recombination velocity. As Vb is made more positive, corresponding to a decrease in barrier height for electrons, na increases and pb decreases. It will be convenient for the theoretical discussion to introduce the particular barrier po- tential Vbo , for which Ua = Pb . With use of the Boltzmann approxi- mations in Equations (6) and (7), this gives exp 12^Vbo\ = (Nb/Na) exp [{Ea + ^6 - 2E,)/kT)] = {Pbo/nao){p/n), where /3 = e/kT. The last form follows from the definitions of pbo and Uao and by noting that exp [2(Ei - Ef)/^!] = p/n. We then have Ua/Pb = exp [2/3(7b - Vbo)1 (9) As so defined, Vbo depends on the Fermi level and thus on the con- ductivity of the specimen. We shall let Fo be the value of Vbo for an intrinsic specimen. From (8) Uao/Pbo = exp [-2/3Fo]. (10) If riao = Pbo , then Vo will be zero. Postulate II sets limits on Vo , but Vo is otherwise undetermined in our experiments. 24 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 We shall first relate Vd and Vb . Since the electric field in the film is ^Trai/Ko , Vn = 4.TtDcri/Kn , (11) where Kd is the dielectric constant of the film. Substituting for ai from Equation (3) and making use of (8) and (9), we find Vd = 2H sinh ^{Vb - Vbo) + const, (12) where H = ^ {NaN, exp [(E, - Ea)/kT])^ = (47reVi^x>)(n„oP6o)^ (13) Kd If Vd is expressed in volts and tn in cm. i/ = 1.8 X 10-\£n/Kn)(naoPbQf. (13a) The contact potential Equation (2) may be expressed in the form c.p. = Vb - Vbo + 2H sinh ^(Vb - Vbo) + const. (14) The change in contact potential with illumination results from a potential drop 8Vi in the interior and a drop 8Vb across the space- charge layer. The former comes from the recombination current of holes and electrons diffusing from the surface to the interior. Since electrical neutrality requires that 8n = 8p, the concentration gradients are equal. However, the mobility of electrons is greater than that of holes, so that the diffusion current of electrons is larger than that of holes by the mobility ratio "6." Since there can be no net current flow to the interior, an electric field is established which is in such a direction as to enhance the flow of holes and retard the flow of electrons. The net potential drop associated with this electric field is /3(6 4-1) i bn+ p ) where 8p = bn is the change in concentration in the interior just beyond the space-charge layer.* This potential is positive for both n- and p-type material, and is independent of ambient, since 8p is independent of ambient. The ob- served (Ac.p.)i, is generally much larger than given by (15) and is of opposite sign for n- and p-type material. To account for the observa- tions, it is necessary to assume that the major part of the effect is associated with a space-charge layer at the free surface. We believe that the present experiments give the most convincing evidence obtained so far for the existence of' such a space-charge layer. ^ SURFACE PROPERTIES OF GERMANIUM 25 The value of 8Vb , the change in potential across the space-charge layer due to light, is determined by the requirement that there be no net change in charge in the surface traps, or that 5na = dpb . (16) The changes 8na and 8pb come both from changes in 8p and 8n and from 8Vb . According to postulate IV, riai = ^a + 8na is in equilibrium with the conduction band and pti = Pb -\- 8pb with the valence band. We have then riai/ria = Usi/ris = (ni/n) exp [138V b], (17) Pbi/Pb = Psi/ps = (pi/p) exp [-I38Vb]. (18) from which it follows that dna/ria = (ni/n) exp [^8Vb] - 1, (19) WPb = ipi/p) exp [-^8Vb] - 1, (20) ^ = exp [2/3(7. - Vbo)] = y^\^"P ^-;ff^ -/ . (21) Pb (ni/n) exp [138V b] — 1 Equation (21) is a quadratic equation in exp [(38V b] which may be solved to give an explicit expression for 8Vb • The role of electrons and holes may be interchanged by changing the sign of 8Vb and of Vb — Vbo . This accounts for the difference in behavior of n- and p-type samples. The total change with light is the sum of 8Vi and 8Vb : (Ac.p.)^ = 8Vl = 8Vb + 8Vi . (22) In the analysis of the data, 8Vi is calculated theoretically from (15) with 8p = 8n determined from (1) and 8Vb is obtained from the ob- served (Ac.p.)l and 8Vi using (22). Equation (21) is then used to find Vb — Vbo - A plot of c.p. — {Vb — Vbo) versus sinh ^{Vb — Vbo) should be a straight line with slope 2H. An analysis of the observed data in this manner is given in the following section. We turn finally to a discussion of the surface recombination velocity, Vs . According to postulate V, recombination is limited by flow of holes to a-traps and of conduction electrons to 6-traps. The flow of holes from the valence band to a-traps (really electrons from a-traps drop into the vacant levels in the valence band corresponding to the holes) is proportional to the product of the hole concentration at the surface Psi , and the concentration of electrons in a-traps, Uai . The reverse flow is that of thermal generation of holes: electrons from the valence 26 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 band go into unoccupied a-traps. Since, according to postulate II, the number of unoccupied traps is nearly equal to the total number of traps, and is thus independent of 8Vb and 8p, the reverse flow will be practically equal to the thermal equilibrium value. If Spa is the re- combination cross-section for holes, the net flow of holes to a-traps is Upa = SpaVpipsiriai — Psna)(holes/cm^ scc). (23) Here, Vp is the velocity factor which when multiplied by the concen- tration gives the number of holes crossing a unit area from one direction: Vp = (kT/2Trmp)K (24) With use of the relations, Uaj/ria = risi/ns , Pbi/pb = Psi/ps , from equations (17) and (18), which follow from the fact that the a-traps are in thermal equilibrium with the conduction band and the 6-traps with the valence band, (23) becomes Upa = SpaVp{na/ns){psinsi — PsUs). (25) This expression may be simplified further. The ratio ria ^ Ng exp [{E, - Ea - eVs)/kT] ^ Na ^ (E, - Ea\ . n. Nc exp [{E, - Ec - eVs)/kT] Nc ^""^ \ kT J ^ ^ is independent of Ef and Vb . The ratio may be evaluated for an intrinsic specimen with Vb = 0, in which case ria = Uao and ris = Ui . Thus Ua/ris = Uao/Ui . (27) We also have from postulate IV, PsiUsi = piTii . (28) The equilibrium products are PsUs = pn = n]. (29) With use of (27), (28) and (29), (25) becomes Upa = SpaVp{nao/ni){pini — pn). (30) Similarly, it is found that net flow of electrons to 6-traps is equal to: Unb = Sr^n{pbQ/nx){pini - pu) . (31) The total rate of recombination is given by the sum of Unb and Upa and is given by an expression of the form : U --U^-^Upa^ Cipiui - pn) = C(n + p)8p. (32) SURFACE PROPERTIES OF GERMANIUM 27 It should be noted that the coefficient C is independent of the Fermi level and thus of the conductivity of the specimen, whereas Vs is not. The relation between them is: For n-type C = Vs/n, (33) For p-type C = v^/v- (34) Values of C may be determined empirically from observed values of Vs . Referring to Table I, for sample A, Vs = 60 cm/sec and n = 1.4 X lO'Vcm', so that C = 4.3 X 10~'' cm'/sec. For sample D, Vs = 180 cm/sec and p = 3.0 X lO'Vcm', so that C = 6 X 10~'' cmVsec. The values of C are approximately the same, indicating that the traps are not much different for the two specimens. The theoretical value of C involves Uao and pto • It is the product naPb = najoPbo , which is related to the parameter H and which can be estimated from empirical data. To obtain the concentrations them- selves, a value must be assumed for Vq . We have riao = {UaaPbof exp [ — ^Vq], (35) Pbo = (riaoPbo)' exp [l3Vo]. (36) As we have mentioned previously, there is no way to determine Vo from our experiments, although postulate II sets limits on its value. Let us for simplicity assume that SnbVn = SpaVp = StV. Then, using (35) and (36) in (30) and (32), we have C = 2Stv{naoPbo/n]f cosh ^7o . (37) This equation may be used to estimate the trapping cross-section S, for an assumed Fo • COMPARISON BETWEEN THEORY AND EXPERIMENT In Fig. 14 we have plotted c.p. - (c.p.)o - {Vb - Vbo) versus 2 sinh 0{Vb - Vbo) for p-type sample D, (see Table I). Each symbol represents experimental points for one cycle. Results are sho^^^l for five different cycles not all in the same run. These results are typical of all the data obtained for this sample. It is seen that these data can be fitted over most of the range by a straight line as dra\vn, giving a value oiH = 0.02 e.v. In Fig. 15 we have used the same experimental results to plot 8Vl versus c.p. - (c.p.)o . Since the experimental values of 8Vl cover a range of a factor of 20 or more, a logarithmic scale was used to plot 28 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 1 0 -1 -2 -3 ^~0 A. / A • k • o / X X / /■ a X a!/ / 1 ^-« (\» -7 -8 -9 -10 -11 -12 y / X °> k J^ r / : n/ /^ •/ X • >/ /XA • / / -0.04 0.04 C.R- 0.08 •(C.R)c 0.24 0.28 0.12 0.16 0.20 ;0-Vb+Vbo in volts Fig. 14 — Plot of 2 sinh /3(7b — Ybo) versus c.p. —(7b — Ybo) for sample D as suggested by theory. hY L ' Both positive and negative branches of the curve are plotted on the same figure. The symbols used for the experimental points are consistent with Fig. 14. When c.p. — (c.p.)o is greater than zero, hV l is negative. As c.p. increases it approaches a maximum negative value, this is the negative branch. When c.p. — (c.p.)o is less than zero, hY l is positive and as c.p. decreases hY l approaches a positive maximum value that is less in magnitude than the negative maximum. The solid curve represents the prediction of theory for B. — 0.02 e.v. The agree- ment between theory and experiment is good. It should be emphasized that this fit is obtained with only one adjustable parameter. The data for the other samples were analyzed in the same way. The results are shown by plotting hYj^ versus c.p. — (c.p.)o as in Fig. 15. Fig. 16 is for n-type sample A and Fig. 17 for n-type sample B. The values obtained for H were 0.015 and 0.022 e.v. respectively. The fits obtained are about equally good in all cases with some deviation between theory and experiment near the extremes of contact potential. The values of H obtained are all of the same order as they should be if the surface trap structure is approximately the same from sample to sample. When Yb — Ybo is large and positive and thus c.p., Eq. (14), is large, 6Vl approaches the negative maximum SURFACE PROPERTIES OF GERMANIUM 29 [G'-?)— ]. equation 21, and as Vb — Vbo becomes small and negative, c.p. decreases and 8Vl goes through zero and approaches the positive maximum + [(?'»?)+-■]■ For p-type germanium Ui/n is greater than pi/p so that the negative maximum for 8Vl is greater than the positive maximum. Just the opposite is true for n-type germanium. In this comparison of theory and experiment a tacit assumption has been made that the germanium surface is uniform in its properties. It might well be that this is not the case. The surface might be "patchy." To estimate what effect patches or non-uniformities might have the 10 ^ k o • • -2 A V D X < 8 6 / X 5 4 3 a/d X ,• ° -3 ^ A A / n 8 o/ 5 -iV- S— 4 1 t— \ f -4 ■ -0.15 -0.10 -0.05 0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 C.P. -(c.p. Jo IN VOLTS Fig. 15 — Change of contact potential with illumination, SVl , versus c.p.; experiment and theory sample D, p-type. 30 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 z 2 >-^ 10-3 8 6 5 4 3 10" X , o X \ a • • \ u ' > o > 3 c \ - n •\ - 1 D t \ o\x \ 0 ) • X - - / - A* x/ 7 1 f \ /a ] 1 -0.35 -aaO -0.25 -0.20 -0.15 -0.10 -0.05 0 0.05 0.10 0.15 C.R-(C.R)n IN VOLTS Fig. 16 — Same as Fig. 15 for sample A, n-type. following was done. It was assumed that the surface was made up of two parts each having the theoretical dependence of 8Vl on c.p. but with the contact potential where the light effect goes through zero differing by 0.05 volts. The values of 5Fl at a given contact potential were averaged and plotted against c.p. The difference between this curve and the original one was almost entirely a simple shift in (c.p.)o . It is therefore unlikely that non-uniformities in c.p., over the surface, of the order of 0.1 volts or less would change the relation between 8Vl and c.p. sufficiently to be detectable. One can use equations (8), (10) and (21) to calculate values of (Vb — Vq) for each experimental value of 8Vl and then plot these values against the corresponding values of contact potential. These results are shown in Fig. 18 for n-type sample A and p-type sample D. (Vb — Vo) changes with c.p. as one would expect and in an approxi- SURFACE PROPERTIES OF GERMANIUM 31 mately linear fashion throughout most of the experimental range. The change in Vb is approximately one-fifth the change in contact potential. {Vb — Vo) is positive for the p-type sample and negative for the n-type sample throughout most of the range. If the trap distribution on the surface were symmetrical about the intrinsic position of the Fermi level, then pbo would be equal to Uao and Vo would be zero. In this case the space charge layers on n- and p-type germanium would be about equally developed. All the experimental evidence on germanium indi- cates that this is not the case but rather that — Vbti is much larger than Vbp . This then means that Fo is negative and that the trap distribution is unsymmetrical either in number or energy or both in such a way that riao > PbO . In Table II values are given for the differences {Vb — Vbo), etc., -0.35 -0.30 -0.25 -0.20 -0.15 -0.10 -0.05 0 0.05 0.10 0.15 C.R-(C.R)o IN VOLTS Fig. 17 — Same as Fig. 15 for sample B, n-type. 32 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 0.08 -0.06 0.36 0.40 0.44 0.48 0.52 0.56 0.60 0.64 0.68 0.72 0.76 0.80 CONTACT POTENTIAL IN VOLTS Fig. 18 — Potential across space charge layer, Vb , versus contact potential. also 8V% for certain special cases for samples A and D as calculated from theory. From equations (8) and (10) (V. BO e 2/3 ln(p/n). (38) It is a constant for any given sample as is dVi (see equation 15). From equations (21) and (38) it follows that when 8Vb = 0 then (Vb - Vq) = 0 for all samples. To say this another way, if the traps are symmetrically distributed about Ei and Vb is zero, then illuminating the germanium would produce no potential difference across the surface. One would not suspect offhand that (Vb - Vq) for 8Vl = 0 was also approximately independent of the material in the sample but it is. For the case where 5p and the minority carrier are both small compared to the majority Table II A («-type) D (^-type) (Vbo - V.) -0.044 +0.064 iVi +0.0019 +0.0015 {Vb - Vo)6Vb = 0 0 0 ^\^^--lBo)6ys =0 +0.044 -0.064 V* " l'\^^ " ^ +0.010 +0.010 (Vb - VBo)iVL - 0 +0.054 -0.054 iVL , {Vb - Fbo) - 0 +0.024 -0.026 SURFACE PROPERTIES OF GERMANIUM 33 carrier one can easily show that when 8Vl = 0, (Vb- Vo) = (1/2^) In 6. This follows from equations (15), (21), (22) and (38). The rest of the results hardly need comment. It is interesting to compare differences in contact potential, etc., for units A and D. In equation (2) the Fermi energy has been included in the constant. This equation predicts that if the trap distributions on both surfaces are the same, then when the contact potentials of both surfaces are equal the values of Vd should also be equal and the dif- ference between the Fb's should be equal and opposite to the differences between the Fermi energies in electron volts. This equation can be written c.p. = (Vb - Vo) - (Vbo - Vo) + Vd+Vo + const . (40) In comparing units we shall use A*s to denote differences and these differences are always taken A-D. For the case where Ac.p. = 0, A(Vb — Vo) can be read from Fig. 18 and A(Vbo — Vo) from Table 11. We have then A(Fz> + Fo + const) = -0.05 . This indicates that our simple picture is not quite right. If AVd were zero for this case it should follow from equation (12) that both A2^ sinh ^{Vb — Vbo) and the difference in the const in this equation should be zero. It turns out, however, that A2^ sinh ^(Vb — Vbo) is not zero but +0.07. Equation (12) can be substituted into equation (40) giving c.p. = {Vb - Vo) - {Vbo - Vo) + 2H sinh KVb - Vbo) (41) + Fo + const2 . Where the constant now includes the constant part of Vd and is labeled const2 to distinguish it from the constant in equation (40). We have then for Ac.p. = 0 A(Fo 4- consta) = -0.12. This difference A(7o + consta) can be calculated three other ways, using the experiment results and theoretical values where necessary. These ways are 34 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 1) at the same time in the cycle A(Fo + consta) = -0.10 2) when 6 7l = 0 A(Fo + const2) = -0.13 3) when 8Vb = 0 A(Fo + consta) = -0.12. All four results are consistent within the probable experimental accuracy and give an average result of A(Fo + const2) = —0.12 instead of zero as one might expect from the simple picture. This indicates that thfe trap distributions are different on the two surfaces. The constant part of Vd is proportional to — (Na — Nb) and from equation (10) one would expect Fo to decrease as Na increases with respect to Nb , thus AFo and A const2 should be of the same sign and additive so that A const2 is less than 0.12 volts; assuming that ^d/Kd is the order of 2 X 10~^ cm, this indicates that A(Na — Nb) is the order of 3 X lO^Vper cm^ which is small compared with the probable trap density, Na , Nd , of the order of 10^* per cm^ as we shall see in the next paragraph. Assuming that to/Ko in equation (13a) is the order of 2 X 10~^ cm one can calculate (uaoPboY obtaining 4.1 X 10^° and 5.5 X 10^^ for samples A and D respectively. Using these values one can solve for St in equation (37) and obtain for the case of Fo = 0 the value 5.0 X 10"^^ cm for the average capture cross section of the surface traps. The values of St , riao and pbo depend on what one takes for Fo . The relations are „ ^ 5 X 10-^^ ' cosh iSFo ' n„o = 5 X 10'' exp [-^Vo] , Pw = 5 X 10'° exp [iSFo] . This dependence is shown in the graph in Fig. 19. As already mentioned there are reasons for thinking that Vo is less than zero. If one takes —0.06 volts as a reasonable value then one gets St = 10~'^ cm^ Uao = 5 X 10 /cm and Pao = 5 X lOVcm^ respectively. One can push these calculations still further to estimate Na , Nb , Ea and Eb . One knows that Ea and -Eb must be greater than l/kT. Also Na and A^6 should be less than the number of germanium atoms per cm^ of surface which is 1.4 X 10"/cm'. Values of Na and Nb of the order of 1 X lO'Vcm' for the number of traps per cm* with energies Ea and —Eb of the order of 0.2 SURFACE PROPERTIES OF GERMANIUM 35 e.v. measured from the midband energy are reasonable and not incon- sistent with the original assumptions. As mentioned in the experimental section, some data on (Ac.p.)l for samples C and E have been obtained. While no complete analysis of these results has been made, one can see that they are of the right order of magnitude. As the specific resistance of the sample decreases, the body life time r decreases. This empirical result is to be expected.^ Consequently 8p for the same light intensity decreases and one would expect (Ac.p.)l to decrease as it does. For sample E of course one could not neglect the charge in the space layer so that the theory would be more involved. The comparison between the contact potentials of samples A, C, D and E shown in Fig. 6 can be understood in part at least. Consider first the over-all result that at the same time in the cycle the c.p. for a sand- blasted surface is less than for an etched surface, i.e., the work function for the sand-blasted surface is greater. It is known that the surface re- combination increases enormously when the surface is sand-blasted. This means that either the surface trap density has increased or that the distribution has changed or both in such a way as to increase surface lO'^i 1 10 \ 1 / (If \ \ '""'] / / \ < \ / 2 / \ / \ \ \ \c\' 10^' / \ /' \ O e 7^ / \ / \ (S \ C 4 I" / / \ >v ? Pk \n< JO 10^° / r \ 8 / \, fi / / s \ ^ 4 2 ~-r \ 2 -0.08 -0.06 -0.04 -0.02 Vq in volts Fig. 19 — Dependence of carrier densities in surface traps, n«o and Fm , for Vb = 0, and cross section of trapping St on Fo . 36 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 recombination. The results of Fig. 6 indicate that this change in trap distribution also has been in a direction to increase ( — Fb). This is what one would expect if Nb has increased with respect to A^o , ie., sand-blasting increases the effective number of acceptor traps. About all that can be said about the rest of the results in Fig. 6 is that if one knew the surface trap structure in number and energy distribution for the etched surface then one could deduce from the results in Fig. 6 the distribution of traps for the sand-blasted surface over part of the energy range at least. The theory developed here for the germanium surface may not apply at all for a silicon surface. In a previous discussion of the silicon surface it was assumed that the resistance to surface trapping occurred mainly in the flow across the space charge layer rather than from the surface to the traps. This may well be the case in silicon. An investigation of the silicon surface using these same techniques should clarify this point. It should be emphasized that the ambient used for this study and the variations in it are special in that they do not necessarily correspond to the atmosphere of ordinary room air. It is very probable that constituents in room air other than oxygen ions and water vapor are important, such as salt ions for instance. It is quite probable that Vs for such a surface exposed to room air would increase considerably with time instead of staying relatively constant as it does under the bell jar. CONCLUSIONS A method has been developed for studying the surface properties of Ge in a gaseous ambient at atmospheric pressure. It has been found that the Ge surface interacts with this ambient. Two atmospheric constituents that are important in this interaction, oxygen and vapors with OH radicals, have been isolated. With the controlled use of these, the surface dipole of Ge can be cycled between two extremes. Thus the dependence of other properties of the Ge surface on surface dipole, such as change of contact potential with illumination and surface recombina- tion, can be determined. It is evident from the results that the method is very powerful. In order to complete the present study, it was necessary to stop trying new experiments since almost all directions of variation open up new and interesting phenomena. These results on Ge are really just a beginning, and the preliminary data on silicon indicate that the same method would also be fruitful on other semiconductors. The technique is of course not limited to atmospheric pressure. A tentative theory of the Ge surface has been developed that is suflficient to explain the experimental results on a semi-quantitative basis. Theory and the experiment together predict approximately the SUKFAGE PROPERTIES OF GERMANIUM 37 number, type and distribution in energy of the surface traps. One has therefore a tentative model of the Ge surface that should be very useful in any further investigation of its properties. ACKNOWLEDGMENTS We wish to acknowledge the help and assistance of all our colleagues who have contributed in many ways to make this investigation a success. We wish to mention in particular E. G. Dreher and R. E. Enz who took most of the experimental data, H. R. Moore who designed and made the electronic equipment used in making the measurements, and Conyers Herring for suggestions regarding the theory of large amplitude signals. Appendix We have assumed (Postulate IV) that traps of type a are in equi- librium with electrons in the conduction band and that traps of type h are in equilibrium with holes in the valence band. We wish to show that this is not really a separate assumption but follows as a consequence of Postulate II if the density of traps is not too high. For simplicity, we shall restrict the discussion in the appendix to the limiting case of small departures from equilibrium so that the equations are linear. The problem may then be discussed most conveniently by means of quasi- Fermi levels* or imrefs, n and p , for the conduction electrons and holes, respectively. Departures 50n and 8ns)l (A.5) Pel = Ps + 8ps = p exp [-KVb + 8Vb - 8ps)]. (A.6) * Reference 2, pages 302-308. 38 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Changes in imrefs of the traps are defined by: n„ = n, + &n, = i + gxp [(£, - E, -e(V, + SVs) + eS^,)/m ' ^^'^^ p„ = p, + Sp, = 1 ^. exp [(g, _ £;, + ,(V, + sv,) - e«0,)/fcr] • ^^-^^ In these last two equations, t may refer to either type of trap (a or 6) and 6pt = — 5nt . For small signals, the recombination current via a given set of surface traps may be considered as a flow produced by differences in the imrefs, 50n and 8p , through four effective resistances in series, as shown in Fig. 20. Here RnB is an effective resistance for flow of electrons across the barrier layer, Rnt for flow of electrons from the conduction band at the surface to the traps, Rpt for flow of holes from the filled band to the traps, and RpB for flow of holes across the barrier layer. Under steady state conditions, the net flow of conduction electrons to the surface is balanced by an equal flow of holes. The recombination current may be thought of as a flow of electrons from the conduction band via the traps to the valence band. The recombination current per unit area is I = -eU = (50n - 5p)/Rt , (A.9) where Rt is the sum of the four resistances in series. Here U is the particle current and / is the corresponding electric current. If dn and 6p are expressed in volts, Rt is in ohms/cm^. We may define a re- combination constant Ct by an equation corresponding to (34) : U = Ctiviui - n]). (A. 10) o ^^n 6¥>^ Fig. 20 — Circuit analogy of surface recombination. SURFACE PROPERTIES OF GERMANIUM 39 The relation between Ct and Rt is obtained by use of (A.l) and (A.2) in (A. 10). Since (A.9) is valid only to terms of the first order in 50n and bp , or in other words that a-traps are in equilibrium with the conduction band and 6-traps with the valence band. First consider the flow across the space-charge layer. The resistances Rnt and Rh depend on the sign of Vb . When Vb is negative, the net electron current across the space-charge layer is: / = evniusi — ni) exp [^{Vb + 8Vb)] (A.12) ^ e^Vnns(8(f)n — 50ns), where Vn is defined by an equation similar to (24). The second form is the linear approximation. Thus we have 1/RnB = e^VnUs . (A. 13) If Vb is positive, Us is replaced by n. Correspondingly, for Vb negative, the hole current is: / = evj^ipi - psi) exp 1^(Vb H- 8Vb)] ^ e^Vpp{8(l)ps — 84>p), and l/RpB = e^Vpp. (A. 14) For Vb positive, p is replaced by ps . The maximum value of Rns is obtained for the ambient which makes Vb most negative and RpB is a maximum when Vb is most positive. The current from the conduction band to the traps is calculated as in (23): / = eiVnSntrisiPn - gtUa) (^-15) ^ e^VnSntnsPt{84>ns - 8t). (A.16) 40 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Here gt is the rate of thermal generation of electrons from occupied traps to the conduction band. Since, for equilibrium conditions, / = 0, we have: gt = VnSntrisPt/nt . (A. 17) It may be verified easily that gt is independent of Ef and Vb . The ratio, net = nspt/nt = Nc exp [{Et - Ec)/kT], (A.18) is the equilibrium concentration of electrons in the conduction band when the Fermi level is at the level of the traps. The expression (A. 16) is again the linear approximation. Thus 1/Rnt = e^VnSntrisPt . (A. 19) Similarly, we have 1/Rpt = e^VpSptPsUt . (A.20) We shall show that the ratio Rnt/Rpt = {vnSpt/vpSnt)(Psnt/nsPt)y (A.21) may be expected to be small compared with unity for a-traps and large compared with unity for 6-traps. First consider a-traps. If anything, 'Spa < Sna , bccausc holes must give up a larger energy than conduction electrons in going to a-traps. The second factor is small compared with unity if Ps « net , (A.22) with net defined by (A.18). This will be the case if the a-traps are closer to the conduction band than the top of the valence band at the surface is to the Fermi level. According to Postulate II, this should always be the case. Similarly, for 6-traps Rpt/Rnt « 1 (A.23) if n. « pw , (A.24) where Pvi = p.nt/pt = N, exp [-(Et - Ev)/kT]. (A.25) is the concentration of holes in the valence band when the Fermi level is at the level of the traps. SUKPACE PKOPERTIFS OP GERMANIUM 41 The values of RnB and Rps relative to Rnt and Rpt are: Rns/Rnt = SntVt (A.26) Rps/Rpt = Sj,tnt. (A.27) In our experiments these are always small compared with unity, since the trapping cross-sections are of the order of 10"^^ cm^ and p< and n« are always less than Nt , which is of the order of lO^Vcm^ Barrier resistances may be important for surfaces with a large number of surface traps. Analysis of earlier data on the change of the contact potential of silicon with light was made on the assumption that the probability that an electron or hole reaching the surface be trapped is relatively large, and that consequently the barrier resistances are large compared with the trapping resistances. The present experiments on germanium throw doubt on this interpretation, but further experiments are required to clarify the situation. REFERENCES 1. Bardeen, J., Phys. Rev., 71, p. 717, 1947. Brattain, W. H., Phys. Rev., 72, p. 345, 1947 and Semi-Conducting Materials Butterworths Scientific Publica- tions Ltd., pp. 37-46, 1951. 2. Shockley, W., Electrons and Holes in Semiconductors. D. Van Nostrand, pp. 318-325, 1950. 3. Pearson, G. L., unpublished data. J. R. Haynes and W. Shockley, Phys. Rev. 81, p. 835, 1951. 4. The CP-4 etch is due to R. D. Heidenreich. The formula is given in Phys. Rev., 81, p. 838, 1951. This method of treating a Ge surface is due to C. S. Fuller. 5. C. S. Fuller suggested the use of these fumes. 6. Goucher, F. S., G. L. Pearson, M. Sparks, G. K. Teal and W. Shockley, Phys. Rev. 81, p. 637, 1951. 7. Shockley, W. and W. T. Read, Phys. Rev., 87, p. 835, 1952. 8. van Roosbroeck, W., Bell Sys. Tech. Jl., 29, p. 560, 1950. Frequency Economy in Mobile Radio Bands By KENNETH BULLINGTON (Manuscript received August 20, 1952) The various factors affecting the usability of mobile radio channels are discussed, and estimates are obtained for the number of usable channels per megacycle for several present and proposed methods of operation. The lack of radio-frequency selectivity is the principal barrier to maximum frequency economy, but this difficulty can be avoided by sufficient geographical and operational coordination. The increasing demand for all types of radio services emphasizes the need for efficient use of the radio frequency spectrum. In mobile radio operation the number of usable channels that can be obtained in the VHF and UHF mobile bands depends not only on the width of the in- dividual channels, but also on how and where each channel is to be used. Activity on the same frequency at neighboring locations, and on neigh- boring frequencies at the same location both affect the usefulness of a channel. Halving the channel spacing doubles the number of potential assignments, but it does not double, and in some cases it does not ap- preciably increase the number of usable channels. The usefulness of a single isolated channel is determined by the in- tensity of its signal above the noise level. Because of the very wide varia- tion in received signal strength caused by distance, terrain, building shadowing, etc., the coverage area of a channel can be discussed only in statistical terms. There are likely to be islands of poor signal-to-noise ratio even close to the transmitter, and the coverage gradually fades out into more spotty conditions at greater distance. If the same frequency is used at a neighboring location, the familiar problem of co-channel interference arises. There will now be locations where the desired signal is above noise, but the undesired signal is still stronger. Thus, the coverage area of a channel is reduced by the existence of the co-channel transmitter; again, it is possible to discuss this reduc- tion only in statistical terms. When two channels are being operated on different frequencies in the same general area, the coverage area of each is limited by signal-to- 42 FREQUENCY ECONOMY IN MOBILE RADIO BAND^ 43 noise considerations. In addition, each channel may affect the other because of spurious radiation from transmitters, insufficient receiver se- lectivity, receiver oscillator radiation, etc. The recent trend toward re- ceivers with greatly improved IF selectivity is worthwhile, but even infinite IF selectivity cannot solve many of the present interference problems. When three or more channels are operating in the same general area, another type of interference occurs because of intermodulation in trans- mitters or receivers. If it were technically feasible to build into the equip- ment sufficient radio frequency selectivity to separate the working chan- nels, this interference could be removed. In fact, this is not feasible, and it is necessary to consider possible modulation products from channels falling within a frequency band several percent wide. The number of possible interference conditions that result from intermodulation (third order) rises from 9 for 3 working channels to 50 for 5 channels, to 450 for 10 channels, and to 495,000 for 100 working channels. Some of these interference combinations overlap and fall on the same channel; but even considering all possible duplication, intermodulation inter- ference rapidly becomes controlling as the number of closely spaced chan- nels working in the same area is increased. It is not technically feasible to achieve enough radio frequency selec- tivity to permit unrestricted and uncoordinated use of many channels in a given area, unless the channels are, on the average, separated by about 1 per cent of the operating frequency. For any kind of efficiency of frequency utilization, it is necessary to have some coordination in the location of fixed transmitters and in the use of channels. The maximum efficiency of utilization requires the maximum coordination. The technical factors that determine channel width, channel spacing, and the number of usable channels are described and tabulated below. The first section discusses the principal factors that affect the useful- ness of channels equipped with transmitters and receivers with perfect filtering. This is followed by a consideration of the limitations imposed by insufficient total filtering and by insufficient radio frequency filtering. The next section shows the reduced requirements that are possible by coordination between systems. Finally, the quantitative data are used to illustrate the capabilities and efl^iciencies of various present and pro- posed methods of mobile system operation. CHANNELS WITH PERFECT FILTERING It has been found by experiment that the radio path loss between an- tennas in a mobile radio system can be ascribed to three principal fac- 44 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 tors: distance, shadowing and standing wave patterns. The variation \vith distance from the base station follows the theoretical free space loss up to 500 feet or more, as long as the points are within line of sight. Typical values of the free-space loss are shown in Table I. Beyond about one-half mile the median path loss over plane earth increases about 12 db each time the distance is doubled out to distances of 20-30 miles.^' ^ In addition to the increase in path loss with distance, which is ac- counted for reasonably well by the theory of radio propagation over plane earth, bold features of geography such as mountains and large buildings cause shadow losses that result in irregular coverage patterns. For example the median loss at street level for random locations in New York City is about 25 db greater than the plane earth values com- puted for the distance and antenna heights involved ; the corresponding 10 per cent and 90 per cent losses are about 15 and 35 db respectively.^ Superimposed on the above effects which vary relatively slowly with location are standing wave patterns whose effect on path loss can change substantially within a foot or so. The standing waves are the result of random additions of multiple reflections from nearby buildings or ter- rain, and the variation in path loss follows the Rayleigh distribution for small changes in distance in urban areas. In other words, there is no theoretical limit on the deviation from the median but in 1 per cent of the possible locations the signal is likely to be more than 8 db above the median value and in 99 per cent of the possible locations the signal level is not expected to be more than 18 db below the median value. The motion of the mobile unit through the standing wave patterns causes signal fluctuations or flutter in the received signal. The flutter Table I — Free Space Loss Between Dipoles Separation Between Transmitting and Free-Space Loss Receiving Antennas 150 mc 450 mc 6 ft. 50 ft. 500 ft. ^ mile 16 db 36 66 70 26 db 46 66 80 > Young, W. R., Jr., Comparison of Mobile Radio Transmission at 150, 450, 900 and 3700 Mc. Bell Sys. Tech. Jl., 30, i)|). 1068-1086, Nov., 1952. * Aikens, A. J., and L. Y. Lacy, A Tost of 450 -Megacycle. Urban Transmission to a Mobile Receiver. I.R.E., Proc, pp. 1317-1319, Nov., 1950. » Bullington, K., Radio Propagation Variations at VHF and UHF. I.R.E., Proc., pp. 27-32, Jan., 1960. FREQUENCY ECONOMY IN MOBILE RADIO BANDS 45 rate at 150 mc may be as much as 15 cycles per second for a speed of 30 mph and increases as either the radio frequency or the speed of the mobile unit is increased. The fast acting gain control needed to minimize the flutter effects is obtained automatically with frequency modulation but is more difficult to obtain with amphtude modulation. This factor is one of the principal advantages of the use of FM instead of AM for mobile radio systems. The co-channel interference to be expected between stations having equal transmitter powers depends on the path loss statistics for both the desired and undesired signals. At the edge of the desired coverage area there must be a high probability that the desired signal will be strong enough to be useful and only a small probability that the undesired signal will be strong enough to be troublesome. The geographical separa- tion needed between co-channel stations varies from about four to six times the desired coverage radius when FM is used and from six to eight times when AM is used.^ If the needs for mobile channels were uniformly distributed geographically only a small part of the potential channel assignments would ever be used in a given area. However, the needs for mobile channels are usually concentrated in areas of high population density so that a large percentage of the channel assignments may be needed in the same area. The above estimates on co-channel spacing depend somewhat on the antenna heights and the type of terrain, and assume that the same fre- quency is used in both directions of transmission. When the two-fre- quency method is used with adequate separation between the trans- mitting and receiving frequencies, the co-channel spacings can be reduced to about three to five times the coverage radius for FM and to about four to six times for AM. This reduction of approximately 30 per cent is possible because the most troublesome interfering path in the single frequency method (from base transmitter to base receiver) can be elim- inated in the two-frequency method by sufficient selectivity. The principal reason for using the single frequency method is to pro- vide communication between tw^o mobile units when they are relatively near each other but are beyond the range of the base station. When transmission of all messages through the base station is desirable, or at least not objectionable, the two-frequency method is preferable. It is shown in a later section that close geographical and operational co- ordination is needed to achieve maximum efficiency in the use of fre- quency space and that this coordination can be obtained only with the two-frequency method. * See reference in Footnote 3. / 46 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 The bandwidth needed to pass the desired signal depends on the fre- quency stabiHty that can be maintained as well as on the type of modula- tion. The allowance for frequency drift includes the variations in both transmitters and receivers. The importance of these figures is indicated in Table II which shows the tolerances needed for frequency instability. For example, with an overall frequency stability of d=0.002 per cent the channel width at 450 mc needs to be 18 kc wider than the minimum band- width required to pass the modulated signals. The use of frequency modulation has several important advantages that cannot be readily obtained with AM. The instantaneous gain con- trol and the closer co-channel spacings have already been mentioned. In addition, for the same radiated power, FM with a frequency swing greater than about ±3 kc has the well known advantage of providing a higher output signal-to-noise ratio throughout most of the coverage area than is possible with double sideband amplitude modulation; this FM advantage is substantially reduced when the IF bandwidth is large compared with the bandwidth required to pass the desired sidebands. The bandwidth required for frequency modulation of a 3 kc voice band must be at least ±3 kc. For reasonable FM signal-to-noise ad- vantage, particularly in the presence of impulse noise, the frequency swing should be at least ±5 kc which requires a bandwidth of d=8 kc for good quality. The corresponding bandwidth for amplitude modula- tion is ±3 kc; the use of single sideband AM transmission does not seem feasible, at least not for single channel operation. LIMITATIONS IMPOSED BY INSUFFICIENT (tOTAL) FILTERING The frequency separation between carrier frequencies must be greater than the bandwidth required to pass the desired signal because addi- tional frequency space or guard bands are needed to build up receiver selectivity against undesired signals and to avoid the extra band radia- tion from transmitters. The power of a 100 watt transmitter is about Table II — Tolerance Needed for Overall Frequency Drift Frequency SUbility Allowance for Frequency Drift 150 mc Band 450 mc Band ±0.001% ±0.002 ±0.005 ±1.5 kc ±3 ±7.5 ± 4.5 kc ± 9 ±22.5 FREQUENCY ECONOMY IN MOBILE RADIO BANDS 47 Table III — Required Suppression versus Distance Between Antennas Distance Between Transmitting and Total Selectivity or Filtering Required Receiving Antennas 150 mc 450 mc 0 ft. 50 ft. 500ft. ]4 mile 160 db 124 104 90 160 db 114 94 80 160 db greater than the minimum signal that is useful in the receiver (140 db below one watt), so ideally no appreciable interference would result if the overall selectivity of the receiver and the suppression of extra band radiation in the transmitter could be in excess of 160 db. This amount of isolation is difficult to obtain by filtering. The inter- action between transmitter and receiver of the same system is frequently avoided by the use of ''push-to-talk" operation, but the potential inter- ference between different systems requires the full 160 db (based on 100 watt transmitters). Fortunately, a substantial part of the desired isola- tion can be obtained by modest geographical separation. The net re- quirements for either receiver selectivity or transmitter filtering are less than 160 db by the losses shown in Table I and are summarized in Table III. Receiver selectivities of 90-100 db or more are feasible except on nearby channels and possibly on certain image channels. Typical values of the guard bands that are required between the edge of the desired pass band and the frequency at which the desired attenuation to inter- fering signals can be obtained are estimated in Table IV. Even if the guard band, shown in Table IV, required to provide ade- quate selectivity in the receiver could be reduced to zero by providing infinitely steep sides on the IF selectivity curve, there would still re- main the guard band needed to avoid the extra band radiation from the transmitter. The amount of suppression of extra band radiation needed Table IV — Guard Band versus IF Selectivity Desired IF Selectivity 40 db 60 80 100 120 Required Guard Band 12 kc 15 20 25 30 48 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 for unrestricted operation is equal to the required receiver selectivity- given in Table III and can be translated into frequency space in the following manner. Both AM and FM transmitters radiate some noise and distortion products outside of the ideal modulation bandwidth. In addition, some of the sideband energy in FM falls outside the desired modulation band- width. The magnitude of the undesired FM sideband radiation is higher than the noise immediately outside of the desired band, but it decreases more rapidly with the result that the noise is usually controlling in the region where the extra band radiation is more than 60 to 70 db down. The guard bands that are required between the edge of the desired transmitted band and the frequency at which the necessary suppression of extra band radiation can be obtained are estimated in Table V. These values depend on the width of the voice band and are relatively independent of the radio frequency since ri. selectivity is not possible. Measurements on present day transmitters correspond to the above estimates for values of suppression less than 80 db, but a frequency separation of nearly one megacycle or more is needed for suppressions of 100 and 120 db. This limitation is not expected to be inherent so more optimistic estimates are indicated in Table V. If the present charac- teristics cannot be improved, that is, if suppressions greater than about 80 db cannot be obtained. Table III indicates that some interference may be expected within about one-half mile of an unwanted transmitter. A comparison of the information given in Tables III, IV and V in- dicates that the guard bands required for unrestricted operation are approximately 100, 50 and 25 kc for minimum separations between transmitter and receiver of 50 feet, 500 feet and one-half mile, respec- tively. These values together with the bandwidth needed for modula- tion and for frequency instability determine the frequency separation required between channels operating in the same area and are sum- marized in Table VI. Table V — Guard Bands Required to Avoid Extra Band Radiation Suppression of Extra Band Radiation Guard Bands Required AM FM 40 db 60 80 100 120 3kc 10 25 60 100 9kc 15 25 50 100 FREQUENCY ECONOMY IN MOBILE RADIO BANDS 49 Table VI — Channel Spacing Required for Unrestricted Operation of Two FM Channels in Same Area VERSUS Antenna Separation Channel Spacing, Neglecting Intermodulation Minimum Separation Between Transmitting and Receiving Antenna 150 mc 450 mc ±0.002% ±0.005% ±0.002% ±0.005% 50 ft. 500 ft. }4 mile 112-122 kc 62- 72 37- 47 121-131 kc 71- 81 46- 56 124-134 kc 74- 84 49- 59 151-161 kc 101-111 76- 86 The above table shows that if interference of the types so far con- sidered is to be kept below the minimum usable signal at all distances greater than about 500 feet from undesired transmitters, the channel spacing needs to be at least 62 to 75 kc in the 150 mc band and 74 to 105 kc in the 450 mc band. The channel spacings for AM are equal to the minimum shown above, while the higher figure is for FM with ±5 kc swing (a modulation bandwidth of =b8 kc). Since the above channel spacings are considerably greater than the necessary IF bandwidth, it should be possible to use intermediate chan- nels in adjacent non-overlapping areas. This geographical limitation does not appreciably decrease the overall efficiency in the use of fre- quency space as long as the needs for mobile channels are more or less uniformly distributed within a large region, but it becomes important where a large percentage of the available channels are needed in the same metropolitan area. In a later section it is shown that channel spacings less than the values given in Table VI are feasible in the same area providing sufficient co- ordination is achieved in both geographical spacings and operating methods. The estimated channel spacings shown in Table VI do not take into account the effect of intermodulation interference which is discussed in the following section. Intermodulation interference may limit the num- ber of usable one-way channels to only 1 or 2 per megacycle instead of the above 6 to 20 per megacycle, unless further restrictions are placed on the selection of frequencies and on the method of operation. limitations imposed by insufficient rf filtering When a strong unwanted signal on a frequency within the RF bandwidth is present at the input to a receiver, overloading occurs and the receiver 50 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table VII — Required RF Receiver Selectivity versus Antenna Separation Minimum Separation Between Transmitter RF Selectivity and Receiver 150 mc 450 mc 0 ft. (common antenna) 50 ft. 500 ft. ]^i mile 95 db 59 39 25 95 db 49 29 15 is said to be desensitized. When two or more strong unwanted signals are present desensitization also occurs, but in addition, extraneous fre- quencies are generated by intermodulation in the receiver itself. As the levels of the unwanted signals become greater than about 75 db below one watt (1 or 2 millivolts across a typical receiver) the intensity of the modulation products rises rapidly above the set noise. The resulting interference can be 60 db or more above set noise and the number of the modulation products increases by at least the cube power of the number of operating channels. Ideally, the intermodulation interference in the receiver caused by 100-watt transmitters (20 db above one watt) can be eliminated by 20 + 75 = 95 db RF selectivity even when the receiver and the unwanted transmitters are connected to the same antenna. In practice, the effect of geographical separation assuming the free space loss given in Table I reduces the RF selectivity requirement to the values given in Table VII. The RF selectivity requirements given in Table VII cannot be ob- tained on nearby channels. The approximate RF bandwidths associated with various amounts of RF selectivity in mobile receivers is shown in Table VIII. For example, in mobile receivers it seems feasible to provide 40 db of RF selectivity at frequencies removed from the desired chan- nel by about 3 mc in the 150-mc band and by about 10 mc in the 450-mc band. At fixed stations the RF bandwidth required for a given selec- Table VIII — Frequency Spacing from Midband versus RF Selectivity Desired RF Selectivity Frequency Spacing from Midband 150 mc 450 mc 20 db 40 60 ±1.6 mc ±6 ± 5 mc rblO ±20 FREQUENCY ECONOMY IN MOBILE RADIO BANDS 51 Table IX — ^Significant RF Band versus Antenna Spacing Minimum Separation Between Receiver and Unwanted Transmitters RF Band 150 mc 450 mc 50 ft 500 ft. \i mile ±6 mc ±3 ±2 ±14 mc ± 7 zh 4 tivity can be reduced to one-third and possibly to one-fourth of the above values by the use of bulky and expensive filters. The critical frequency band that needs to be considered in deter- mining the usefulness of any given channel can be obtained by combining the information given in the two preceding tables with the results shown in Table IX. For example, if it be desired to work mobile receivers un- restricted to within 500 feet of two or more unwanted transmitters, all frequency assignments within ±3 mc in the 150-mc band (or within ±7 mc in the 450-mc band) must be carefully chosen if intermodulation interference is to be avoided. When the zb3-mc band is divided into 100 potential channel assign- ments of 60 kc each and when the channels assigned to a given area are chosen at random, 7 channels working 50 per cent of the time (or 37 channels working 10 per cent of the time) mil, on the average, cause third order intermodulation interference about 10 per cent of the time on each channel within the band. The interference is expected to be above the minimum usable signal level in all receivers located less than about a mile from the unwanted transmitters. Even if the operating fre- quencies are selected carefully instead of at random, no more than 11 channels out of 100 can be found that are free of third order intermodula- tion when used simultaneously in the same general area. These results are discussed more completely in a companion paper. ^ When the num- ber of potential channel assignments is greater or less than 100, the corresponding number of usable channels limited by third order modula- tion alone is shoAvn in Table X. The numbers of usable channels shoA\ii above are further reduced when fifth and higher order intermodulation products are considered. A reduction in the nominal channel spacing from 60 kc to 20 kc means a three-fold increase in the potential channel assignments, but Table X shows that the number of usable channels increases much more slowly. 5 Babcock, W. C, Intermodulation Interference in Radio Systems. Page 63 of this issue. 52 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table X — Number of Usable Channels versus Number of Potential Channels ' No. of Usable Channels No. of Potential Channel Assignments in RF Band Shown in Table IX Careful Selection No Interference Random Selection 10% Chance of Interference % of Time Transmitter Is On 50% 25% 10% 20 50 100 200 500 I 11 12 14 5 6 7 9 12 10 12 15 18 24 25 30 37 45 60 Thus far, only the intermodulation interference generated in the re- ceivers has been considered. Intermodulation also occurs at the same frequencies in the transmitters, but it usually can be made less impor- tant than the corresponding interference in the receivers. Ideally, the intermodulation products generated in the transmitters should not be stronger than 140 db below one watt (about 1 microvolt at the input to the receiver) which requires about 75 db RF filtering in each transmitter output. This ideal requirement is based on 100 watt transmitters with both the transmitters and receiver working on the same antenna. In prac- tice, the RF filtering requirement is less than 75 db because of physical separation between transmitters and receivers, and typical values based on free space transmission are shown in Table XI. A comparison of the filter requirements on 100 watt transmitters with the corresponding receiver selectivity requirements given in Table VII shows that the receiver requirements are greater as long as the effective Table XI — RF Transmitter Filtering versus Antenna Separation RF Filtering Needed in Each Transmitter in db DUUnce Between Receiver and Unwanted Transmitters ISO mc Distance Between Transmitters 450 mc Distance Between Transmitters Oft 10 ft so ft SOOft Oft 10 ft so ft 500 ft Oft.* 60 ft. 600 ft. Mmile 76 57 47 40 46 36 29 39 29 22 19 12 75 52 42 35 36 26 19 29 19 12 9 2 * Common antenna. FREQUENCY ECONOMY IN MOBILE RADIO BANDS 53 separation between transmitters is greater than about 50 feet. For example, with a 500-foot separation between the transmitting and re- ceiving antennas, Table VII shows that the 150 mc requirement on r.f. selectivity is 39 db. The bandwidth between the 39 db points on the receiver selectivity characteristic determines the number of potential channel assignments to be used in Table X. INCREASED EFFICIENCY OBTAINED BY COORDINATION The preceding selectivity and filtering requirements are severe and in some cases virtually unattainable except at considerable sacrifice in frequency space. The principal reason for these exacting requirements is that the assumed unrestricted and independent operation results in large differences in field intensities among closely spaced frequencies. In order to pick out the weak signals from among the strong, sufficient selec- tivity must be provided to suppress the potential interference to below the minimum usable signal. An alternative is to reduce the level differences and hence the filter- ing requirements by geographical and operational coordination. This means that the level of the potential interference can be permitted to be many db above set noise as long as it is always at least 10-20 db below the desired signal at all possible locations. By proper coordination the troublesome RF filtering problems can be eHminated within the co- ordinated system and the remaining IF selectivity problems can be minimized. The first step is to use the two frequency method of operation with adequate separation between the frequencies used for the opposite directions of transmission. In this way substantial RF filtering can be obtained to eliminate the interference between one or more base trans- mitters and a base receiver. This type of interference is particularly troublesome between single frequency systems because of the relatively high base transmitter power and because the high antennas at both locations reduce the radio path loss to a minimum. The corresponding possible interference between transmitters and receivers on different mobile units is also reduced by the two frequency method but inter- ference between mobile units is much less important because of the lower power and much lower antenna heights. The potential interference between base transmitters and mobile re- ceivers caused by insufficient total filtering can be reduced by locating all base transmitters at or near a common point so the level differ- ences between the desired and undesired signals will never be exces- sive. When all transmitters radiate from a common antenna, a selec- 54 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 tivity or filtering requirement of about 40 db (instead of the values shown in Table III) is sufficient for a reasonable signal to interference ratio plus an allowance for differential path losses resulting from stand- ing wave effects. The RF selectivity or intermodulation problem in the mobile receiver can be eliminated by reducing the power level at the first converter to about 75 db below one watt. This can be done by providing a simple automatic gain control in the RF stage of the mobile receiver. In regions where the desired and undesired signals are weak the receiver has full sensitivity, while at locations near the transmitters both the desired and undesired signals are reduced in level before reaching the first con- verter. The result is that the intermodulation products generated in the receiver are reduced about 3 db for every db that the desired signal is lowered and the distortion becomes negligible before the output signal- to-noise ratio is reduced appreciably. In order that the a.g.c. circuit can be fully effective it is necessary that the transmitters be grouped to- gether and that the desired carrier be transmitted to control the gain of the receiver. Grouping the base transmitters at or near a common point together with the associated measures of transmitting the carriers and using a.g.c. greatly reduces the requirements on the mobile receiver, but these measures complicate the design of the base transmitter. The intermodu- lation products generated in the closely associated transmitters result in potential interference both within and outside of the desired trans- mitting band. The intermodulation that falls on the mobile receiver fre- quencies needs to be suppressed by at least 25 db below^ the carrier on any channel to prevent mutual interference within the coordinated system. The intermodulation that appears as extra band radiation out- side the frequency range of the coordinated system must be suppressed by RF filters. The guard band needed to prevent mutual interference between the coordinated system and its neighbors is small compared with the frequency space that is saved by the close spacing of the chan- nels within the coordinated system. In the direction of transmission from the mobile transmitters to the base receivers, the above coordinating methods cannot be used but equally effective ones are available. The RF selectivity requirements shown in Table VII can be reduced 20 db by using 20 db less power in the mobile transmitter than in the base transmitter. This measure is somewhat analogous to the use of a.g.c. in the opposite direction of transmission; a further step would be automatic control of the radiated power but this complication does not appear to be necessary. In order to regain the full coverage area, multiple base receivers at FREQUENCY ECONOMY IN MOBILE RADIO BANDS 55 different locations are needed and this use of space diversity techniques provides an opportunity to pick the receiver having the best signal-to- noise ratio. Moreover, the low power in the mobile transmitter together with the better RF filters that are possible in fixed locations reduces the critical bandwidth within which intermodulation interference can arise to about =b0.4 mc at 150 mc and to about d=0.6 mc in the 450 mc range. In these bandwidths approximately 20-25 channels can be obtained which with random location of the mobile units would be divided more or less uniformly among five or more base receiving stations. Since no more than 4 or 5 channels would be operating within the critical RF bandwidth at any one receiving location, the possibility of intermodula- tion interference is almost negligible. Finally, an off-channel squelch circuit is provided which disables the base receiver at a location where serious adjacent channel interference is most hkely to occur and forces the choice of another base receiver in a different location. Another ef- fect of the off- channel squelch circuit is that it keeps the base receiver quiet during idle times, and in this respect it is analagous to the advan- tage gained in the mobile receiver by continuous transmission of the desired carrier at the base transmitter. Most of the above coordinating methods tend to emphasize and to increase the characteristic differences between the two directions of transmission. The net effects are that greater frequency economy is ob- tained and that the electrical requirements are reduced on the mobile equipment where size, weight and power are critical and where cost savings are important because of the large number involved. An increase in complexity occurs at the multi-channel base station but this seems economically justified because the cost can be divided among many working channels. When the above methods of coordination are fully utilized, the RF requirements are eliminated in the mobile equipment and can be met in the base station equipment. In addition, the IF selectivity require- ment on nearby channels is reduced to about 40 db in the mobile re- ceiver and to about 60 db in the base receiver. The extra band radiation requirement on nearby channels is reduced to about 25-40 db in the base transmitter, depending on whether one or more than one antenna is used ; and to about 60 db in the mobile transmitter. These requirements coupled with the data given in Tables II, IV and V lead to the frequency separation between coordinated channels operating in the same area as given in Table XII. The channel spacings are shown for AM and for FM with a frequency swing of ±5 kc (which requires a bandwidth of ±8 kc for good quaUty). The spacings shown in Table XII assume that each channel is trans- 56 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table XII — Channel Spacing versus System Stability- Coordinated Systems in Same Areas Channel Spacing Stability 150 mc 450 mc Mobile Receiver Base Receiver Mobile Receiver Bass Receiver AM fm am FM AM FM AM FM dzO.001% ±0.002% ±0.005% 21 24 33 31 34 43 25 28 37 35 38 47 27 36 63 37 46 73 31 40 67 41 50 77 mitted on an individual carrier. Single-channel operation seems to be the only practical arrangement for transmission from the mobile trans- mitter to the base receiver. In the other direction of transmission, from base transmitter to mobile receiver, the question naturally arises whether additional frequency economy could be achieved by multichannel meth- ods. In this case individual carrier operation is also indicated for trans- mission and economic reasons. The multiple echoes that exist at street level in urban areas Umit the number of usable channels that can be transmitted on a single carrier.^ While the exact number is somewhat indefinite, it appears to be less than about 20 and perhaps less than 10 channels. In addition the selectivity and linearity requirements on multi-channel receivers (even for two channels) are much more severe than for single channel equipment. From these considerations it appears that the use of more expensive receivers and channel separation equip- ment in each mobile unit is not economically feasible. frequency economy in present and proposed mobile SYSTEMS The technical factors given above provide a basis for estimating the number of usable mobile channels that can be obtained in a given band- width. This bandwidth must be sufficiently large to be isolated by RF filtering if the results are to be well defined. The following examples assume two different geographical distribu- tions: (1) the number of usable channels with overlapping coverage areas that can be obtained within a city or metropolitan area, and (2) the number of usable channels that can be obtained when the channels are distributed more or less uniformly over a state or other large area. The examples are based on the use of frequency modulation with a • Young, W. R., Jr., and L. Y. Lacy, Echoes in Transmission at 450 Megacycles from Land-to-Car Radio Units. I.R.E., Proc, pp. 255-258, March, 1950. FREQUENCY ECONOMY IN MOBILE RADIO BANDS 57 modulation bandwidth of ±8 kc and a frequency stability of ±.002 per cent; with these assumptions, the IF passband should be at least 22 kc in the 150 mc band and 34 kc in the 450 mc band. Narrower band- widths could be used but this would result in a substantial sacrifice in coverage under impulse noise conditions. Five cases are considered: (1) Single Frequency Semi-Coordinated — In this case, substantially no interference is expected from third order modulation problems, which are avoided by careful selection of operating frequencies, but higher order modulation products may be important. Base station locations are unrelated geographically to other systems in same general area, except that a minimum spacing of 500 feet between receiver and in- terfering transmitter is assumed. (2) Single Frequency with Interference — In this case, the choice of frequencies is unrestricted, but a 10 per cent chance of third order inter- modulation interference is accepted within 500 feet of unwanted trans- mitters, when transmitters are in operation 25 per cent of time. (3) Two Frequency Semi-coordinated — This is the same as (1), ex- cept with two-frequency operation. (4) Two Frequency with Interference — Same as (2) except with two frequency operation. (5) Fully Coordinated Broad-hand — This case assumes: (a) two fre- quency operation with the land transmitters coordinated in location, power, antenna height and emission of protective carriers ; (b) low power mobile transmitters; (c) multiple land receivers; (d) no interference from third or higher order intermodulation ; and (e) guard bands to protect mobile and neighboring services from mutual interference. The number of usable channels that can be obtained in the same area is estimated in Table XIII for frequencies near 150 mc. The minimum channel spacing shown in the first column of Table XIII is calculated as follows: in cases (1), (2), (3) and (4), the extra band radiation from the base transmitter is controlUng. As shown in Tables III and V, to avoid interference for distances greater than 500 feet from the interfering transmitter requires a guard band of about 50 kc. This is added to the 22 kc required IF pass-band of which ±8 kc is allowed for the FM signal, and ±3 kc for 0.002 per cent system instability. In (5), the adjacent channel receiver selectivity is controlHng: Table IV shows the required 60 db can be obtained in 15 kc, which added to the required 22 kc IF band gives approximately 40 kc. It will be noted from Table VII that the assumption of a separation of 500 feet between the receiver and the interfering transmitter requires 58 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table XIII — Usable Channels in City at 150 mc Method of Operation Minimum (Not Average) Channel Spacing in Same Area Number of Usable Channels in 6 mc (\^ Sincrlp frpoupncv semi-coordinated 75 kc 75 75 75 40 10 (2) Sinerle freauencv with interference 14 (3) Two frequency semi-coordinated 5 (4) Two frequency with interference 7 (5) Fully coordinated broad-band* 45 * Includes three guard bands of 0.8 mc each to protect mobile and neighbor- ing services from mutual interference. about 40 db RF selectivity, and from Table VIII that the 40 db selec- tivity requires that all frequencies within d=3 mc need to be considered. With 75 kc channel spacing, there are 80 potential assignments in 6 mc. Table X indicates that 10 one-way channels can be found that are free of mutual third order intermodulation interference. If the available bandwidth were 12 mc the number of interference-free channels would be doubled. By the same process from Table X, we derive the number of usable channels shown for case (2). For cases (3) and (4), the methods are the same, but the number of usable channels is reduced to one-half that shown for the single fre- quency cases. In the fully coordinated broad-band system (case 5) a usable one-way channel can be obtained every 40 kc. However, three guard bands total- ing 2.4 mc are provided to protect both the mobile and neighboring systems from mutual interference. If the available bandwidth were 12 mc the number of interference-free channels would be increased from 45 to 120 since no additional guard bands would be required. The comparison between various methods of operation given in Table XIII applies to 150-mc channels operating in the same city. When the channels are distributed more or less uniformly over a large area, the number of usable channels is increased by several factors. The separation between carrier frequencies in non-overlapping areas needs to be only slightly greater than the IF pass-band of the receiver, say, 30 kc at 150 mc. The guard bands needed in one location can be used in other areas at geographical separations less than co-channel spacings. Finally the required geographical separation between co-channel stations is less for the two frequencjy method than for the single frequency method and is less for FM than it would be for AM. An estimate of the maximum number of usable channels within a large FREQUENCY ECONOMY IN MOBILE RADIO BANDS 59 Table XIV — ^ Usable Channels in State or Large Area AT 150 MC Method of Operation Minimum Channel Spacing* Number of Usable Channels in 6 mc (1) Single frequency semi-coordinated (2) Single frequency with interference. . . . 25 kc 25 25 25 25 108 171 (3) Two frequency semi-coordinated 108 (4) Two frequency with interference (5) Fully coordinated broad-band 171 240 * Assumes adjacent channels are not assigned in same area. area can be obtained by considering an area whose radius is about six times the coverage radius of the individual transmitter. A larger area is unnecessary because single frequency FM channel assignments can be repeated at this distance, while a smaller area would tend to approach the common area concept used above. The large area can be divided into 9 subareas, each of which can be treated in the manner used in Table XIIL The results are shown in Table XIV, which again assumes an FM modulation bandwidth of ±8 kc and d=0.002 per cent overall sys- tem frequency stability. The entries in Table XIV are calculated as follows: Once again, the smallest band to be considered is limited by the RF selectivity in mobile receivers to 6 mc ; with 25 kc as the minimum channel spacing, there are 6000/25 = 240 potential assignments. From Table X, only 12 can be found to be free of third order intermodulation . With 12 channels in each of 9 subareas, there is a grand total of 108 channels usable in the state or large area. With more frequency space, the usable number is increased in proportion. By the same process, from Table X we derive the number shown for case (2). In the two frequency cases, the co-channel separation can be made smaller than in the single frequency cases, since the most troublesome case of interference (that between base transmitters and base receivers) is eased by RF selectivity. Thus, the co-channel separation needs to be only about 0.7 that for single frequency operation, which means that there are now effectively 18 instead of 9 subareas. It follows that the grand total of usable channels is the same in cases (1) and (3) and cases (2) and (4). In considering case (5), we note from Table XIII that 40 kc is the minimum channel spacing usable in a single subarea. However, the larg- est grand total of channels is found by using 50 kc spacing in the sub- areas, and assigning the adjacent 25 kc channels to other subareas. 60 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table XV — Usable Channels iin [ City at 450 mc Method of Operation Minimum (Not Average) Channel Spacing Number of Usable Channels in 14 mc (1) Single frequency semi-coordinated 85 kc 85 85 85 50 12 (2) Single frequency with interference 17 ^3^ Two freouencv semi-coordinated 6 (4) Two frequency with interference 9 (5) Fully coordinated broad-band* 68 * Includes three guard bands of 2,4 mc each to protect mobile and neighboring services from mutual interference. Similarly, the guard bands of one subarea can be used for channels else- where so all of the available 240 channels can be used. The examples given in Tables XIII and XIV represent the two extreme conditions and the practical situation lies in between the two. By similar reasoning it is possible to estimate the number of usable channels that can be obtained at frequencies around 450 mc. The num- ber of usable channels shown in Table XV is for overlapping coverage areas in a city or metropolitan area and the estimates given in Table XVI are based on a uniform distribution over a state or other large sec- tion of the country. Again, FM modulation with a bandwidth of ±8 kc and a system frequency stability of 0.002 per cent are assumed. For a band^\'idth of 28 mc instead of 14 mc the number of usable channels is doubled for the first four cases and is increased from 68 to 208 for the fifth case. The corresponding estimates for bandwidths less than 14 mc are indefinite because of insufficient r.f. selectivity. conclusions The principal conclusions that result from Tables XIII, XIV, XV and XVI, and from the preceding discussion can be summarized as fol- Table XVI — Usable Channels in State or Large Area AT 450 MC Method of Operation Minimum Channel Spacing* Number of Usable Channels in 14 mc (I) Single frequency semi-coordinated 35 kc 35 35 35 35 117 i2) Single frequency with interference 198 (3) Two frequency semi -coordinated 117 (4) Two frequency with interference 198 (6) Fully coordinated broad-band 400 * Aflflumes adjacent channels are not assigned in same area. FREQUENCY ECONOMY IN MOBILE RADIO BANDS 61 lows : 1. A fully coordinated system requires a band of several megacycles that can be treated as a unit, but it offers substantial overall frequency economy and freedom from interference that can be obtained in no other way. This is particularly true in large metropolitan areas where the demand is greatest. With the same equipment and the same standards of quality and reliability, coordinated channels can always be spaced much closer in frequency than uncoordinated systems. 2. The advantages of coordination increase rapidly as the number of channels per unit area is increased. However, in areas where only three or four channels are required, the advantages of complete coordination are sufficiently small that only the semi-coordination of careful frequency allocation is required to preserve overall frequency economy. 3. For maximum economy, where full coordination is not used, the channels should be assigned as in FM and TV broadcasting first to areas and then to users within areas. The allocation of a block of channels to a particular service with a minimum of operational and geographical restriction frequently results in an ever-increasing interference problem as each additional station is placed in operation. 4. Single-frequency operation is most suitable where the operational need for single channel communication between mobile units (as con- trasted with fixed-to mobile) is more important than frequency econ- omy. 5. A frequency separation between potential channel assignments of 25 kc in the 150 mc range, and 35 kc in the 450-mc range seems tech- nically feasible; but adjacent channels with these minimum spacings cannot be assigned in the same area. These values may be reduced to about 20 and 30 kc, respectively, at the sacrifice of an appreciable reduc- tion in coverage under impulse noise conditions. A further reduction in channel spacing would not appreciably increase the total number of usable channels, since the controlling factors are RF selectivity and extra band radiation, rather than IF selectivity or the total number of potential channel assignments. 6. The average spacing needed between channels operating in the same area varies from about 40 to 500 kc or more, depending on the method of operation and the criterion of usability. 7. The need is for a certain small number of channels in all areas, plus a peaked demand in centers of population. In the semi-coordinated cases, the maximum number of channels that can be allocated to the peak area is a small fraction of the total number of channels available. In the fully coordinated, broad-band case, there is much more fiexibil- 02 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 ity and the peak area can be allocated a large fraction of the total avail- able. 8. FM- is preferable to AM for land mobile service because its instan- taneous gain control feature minimizes the flutter caused by the motion of the mobile unit through standing wave patterns. This advantage in- creases in importance as the carrier frequency increases. In addition, FM with an adequate frequency swing provides an increased signal-to- noise advantage over most of the coverage area. The somewhat greater channel width required by FM is more than offset on an area coverage basis by the closer co-channel spacing. Intermodulation Interference in Radio Systems Frequency of Occurrence and Control by Channel Selection By WALLACE C. BABCOCK (Manuscript received August 25, 1952) Intermodulation interference becomes a serious factor in frequency usage when a block of consecutive channels is provided for a given type of radio service in a confined area. Formulas are presented which show the number of potentially interfering 3rd and 5th order intermodulation products that can be formed in a band of n consecutive channels. The probability of en- countering interference when a number of operating channels are picked at random from this band of n channels is developed and the number of in- terference free operating channels that can be obtained by careful selection in this same band is also derived. When a block of consecutive radio channels is used in a confined area to provide a given type of service, interference becomes a serious prob- lem. The situation is aggravated by the fact that whenever energy at two or more radio frequencies combines in a nonlinear circuit, as in transmitter output stages or in receiver input stages, products at other than the original frequencies are created. These are called intermodula- tion products, and they are capable of causing serious interference within the block of channels assigned to a given type of service as well as in other bands assigned to other types of service. It is important in engineering a service to know something about the nature of these products in order to evaluate their interference poten- tialities and to study means of controlling or minimizing that interference. The numbers and locations of various types of intermodulation products are susceptible to mathematical computation. Whether or not all of these products would produce actual interference depends on the geo- graphical locations of transmitters and receivers, and on their specific 63 64 THE BELL SYSTEM TECHNICAL JOUKNAL, JANUARY 1953 electrical characteristics. This paper discusses the number of potential interferences, and in effect, envisages a situation where potential inter- ferences are strong enough to be actual interferences.* GENERAL Intermodulation products are commonly referred to as 2nd, 3rd, 4th ••• nth order products depending on the order of nonlinearity which gives rise to the products. Interference within a system is not generally experienced from the even order products because the frequency sepa- ration between the channels involved and the product formed by them is so great that the selectivity of the transmitter and receiver radio fre- quency circuits is sufficient to reduce it to a negligible amount. Some of the odd order products can be discounted also for the same reason. There are odd-order products, however, involving both sums and dif- ferences of operating frequencies in such fashion that the frequencies of the products formed are very close to those which generated them. These products are those referred to throughout the remainder of this paper since they are the most likely to cause interference. The most general form of 3rd order interference occurs when three frequencies, A, B and C, intermodulate in such fashion as to produce interference on a channel operating at frequency D. In this case A+ B - C = D Another form of 3rd order interference occurs when the second har- monic of A intermodulates with B to produce interference on a channel operating at frequency C. In this case 2A - B = C In like fashion the following forms of 5th order interference may occur. A + B + C-D-E^F 2A+B-C-D = E A+B+C-2D = E 2A + B -2C = D 3A - B - C = D 3A -2B = C * Bullington, K., Frequency Economy in Mobile Radio Bands. Page 42 of this issue. INTERMODULATION INTERFERENCE IN RADIO SYSTEMS 65 NUMBER OF INTERMODULATION PRODUCTS AND FREQUENCY BAND AF- FECTED It is of interest to know how many intermodulation products can be produced by a block of n uniformly spaced channels and where they will fall ^vith respect to the frequency band occupied by the n channels. Third Order Products When all possible combinations of n uniformly spaced operating channels are activated three at a time, n\n — l)/2 products mil be formed and they vnll lie between A — {n — 1) and A -\- 2{n — 1) where the operating band lies between channels A and A -\- {n — 1). This means that the bandwidth of the intermodulation products is very nearly three times that of the operating channels. The products are sj^mmetrically distributed with respect to the midpoint of the operating band. There will be 2n^ + 3n' - 2n - 6 24 third order products that will fall in the n — 1 channels immediately below the operating band and a like number of products that will fall in the n — 1 channels immediately above the operating band. The remaining products, 4n^ - 9n' + 2n + 6 12 in number, will fall in the n channels that constitute the operating band. Not all of these products, however, are capable of producing interference since some products of the A + 5 — C type can fall on the very transmit- ting channels that combine to produce them. These products are harmless since they do not fall on receiving channels and are generally of much lower level than the carrier on the channels in which they do fall. If such products are not counted, there remain I (n - l)(n - 2) products which fall within the operating band. The formulas presented here and in Table I are empirical and were derived for values of n up to 10. However, it is believed that they are reasonably accurate for much larger values of n. 66 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Fifth Order Products When all possible combinations of n operating channels are activated five at a time n\n - l)(n' - n + 4) 12 fifth order products will be formed and they will lie between A — 2{n — 1) and A + 3(n — 1) where the operating band lies between channels A and A + (n — 1). This means that the bandwidth of the intermodulation products is very nearly five times that of the operating channels. The products are sjonmetrically distributed with respect to the midpoint of the operating band. It has been found empirically that if all products known to fall on transmitting channels are excluded, there can still be formed 3^[6n4 _ 37^3 _^ 575^2 _ 2214n + 3922] potentially interfering products which fall within the operating band. This is strictly true only for values of n that exceed 8 and are not mul- tiples of 3. There will remain K2[^' - 14n4 + I79n3 - 1154n2 + 4428n - 7844] products that will fall either outside the band or on transmitting chan- nels within the band. Since relatively few of these products fall on trans- mitting channels within the band the above expression gives to a high degree of approximation the number of products that will fall outside the operating band. Numbers of Potentially Interfering 3rd Order and 5th Order Products The formulas given in the two preceding sections were developed by considering individually the various types of 3rd order and 5th order products. Table I shows the general formulas which apply to these individual types of 3rd and 5th order products from which the summa- tion formulas given in the preceding sections were obtained. The number of potentially interfering 3rd and 5th order products is shown in Table II for specific transmission bands containing 10 and 20 consecutive channels. PROBABILITY OP INTERMODULATION INTERFERENCE WHEN OPERATING CHANNELS ARE PICKED AT RANDOM In an uncoordinated radio communication system it may be assumed that p operating channels are assigned on a random basis in a band con- iXTERMODrLATIOX INTERFERENCE IN RADIO SYSTEMS 67 Table I — Number of Potentially Interfering Intermodulation Products n = Number of consecutive channels in available band Tjrpe of Product 2A - B A + B - C SA - 2B SA - B - C 2A + B - 2C 2A + B - C - D A + B + C - 2D A + B + C-D-E Number of Products (n - 1)=^ 2 if n is odd (n - l)(n - 3)(2n -1 ) 6 if n is odd 5 (« - 3) if n is a multiple of 3 3(n2 - lln + 32) }iin^ - 9n2 + 34/1 - 56) 3(n - 3) (71 - 5)2 f orn > 6 4(n - 3)(n - 5)^ {n -o){n- 6)(n2 - lln + 37) for n 5^ 8 njn - 2) 6 if n is even njn - 2)(2n - 5) 6 if n is even (n - l)(n - 2) 3 other n's taining ^ consecutive channels. Let us suppose further that the average busy time of the channels is such that T represents the portion of time that an average channel is activated by a transmitter and R represents the portion of time that an average channel is connected to a receiver. The probability of interference, /, may be defined as the probability that one or more of the intermodulation products that are formed when pT channels are transmitting will fall on a specific channel m the operating band. The method used to determine / is based on the assmnption that the distribution of intermodulation products is uniform over the operating band. It is further assumed that the magnitudes of the intennodulation products as encountered at the receiver input, are always strong enough to cause interference. Table III shows formulas for / that have been developed for each t>T)e of third order and fifth order product. Fig. 1 shows plots of p versus / when only 3rd order products are considered and Fig. 2 shows plots of p versus / when both 3rd and oth order products are considered with n and T as independent variables. 1. Fig. 2a shows that a band in excess of 500 adjacent channels is required to limit the probability of 3rd and 5th order interference to 10 per cent (/ = 0.1) when 10 operating channels are picked at random from that band if traffic is such as to fully use these 10 channels, 5 for 68 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table II — Type and Number OF Intermodulation Products Number of Products Type of Product 10 Channels 20 Channels 2A - B 40 180 A + B - C 200 2,100 3A - 2B 24 114 SA - B - C 66 636 2A -\- B - 2C 192 2,512 2A + B - C - D 525 11,475 A + B + C - 2D 700 15,300 A + B + C - D - E 540 45,570 Table III — Formulas for /, Probability of Interference n = Number of consecutive channels in available band, p = Number of operating channels picked at random from those in band. T = Portion of time average channel is activated by transmitter. m = Number of intermodulation products of a specific type expected to fall within band of n channels when pT channels are activated. Type of Intermodulation 2A - B A + B - C ZA - 2B ZA - B - C 2A+ B - 2C 2A-\- B - C - D A + B + C -2D A+B+C-D-E Formula for m mi nh Tlli ^ pT{pT - l)(n - 2) 2(n - 1) ^ pTjpT - l){pT - 2)(2n - 5) 6(n ~ 1) ^ pTjpT - l)(n - 2) _ SpTipT - - IXpT - 2)(n2 - lln + 32) W4 nin - l)(n - 2) _ pT(pT - l)ipT - 2)(n3 - 9n2 + 34/i - 56) 2n(w - l)(n - 2) _ ZpTipT ■ - \){pT - 2){pT - 3)(n - 5)2 7«J n{n - l)(n - 2) _ ApTipT ■ - DipT - 2){pT - 3)(n - 5)2 n{n - l)(n - 2) mg pTipT - l){pT - 2){pT - dXpT - 4)(n - 5) (n - 6)(n« - lln + 37) n{n - l)(n - 2)(n - 3)(n - 4) I — -] when only third order products are considered. /"I— (1 j when both third and fifth order products are conaidered. INTERMODULATION INTERFERENCE IN RADIO SYSTEMS 69 transmitting and 5 for receiving. (7" = 0.50 = R) In this case it is the 2A -{- B — 2C type of product that requires the use of such a wide band. 2. Fig. 2a shows that there is practical certainty of interference in a band of 500 adjacent channels when 30 operating channels, picked at random from that band, are fully used. In this case it is the A + B + C — D — E type of product that requires the use of such a wide band. 3. If the same total traffic as was assumed in (1) is handled by a greater number of operating channels, the number of available consecutive channels required for the same probability of interference remains the same. Thus, Fig. 2b shows that a band in excess of 500 consecutive channels is still required to limit the chance of interference to ten per cent when the traffic is distributed among 20 randomly selected operating channels; {T = 0.25 = R) similarly Fig. 2c shows that the required number of available consecutive channels remains the same when the traffic is dis- tributed among 40 operating channels. {T = 0.125 = R). CHANNEL SELECTION FOR THE ELIMINATION OF INTERMODULATION IN- TERFERENCE Discounting the effect of selectivity in the radio equipment, it was shown in the preceding section that only a very limited number of channels can operate together without some degree of mutual inter- ference when these channels are picked at random from a very consid- erable number of available channels. This is of course extravagant of frequency space. In this section, it is proposed to determine whether frequency space can be conserved by carefully selecting the operating channels in such fashion that the various tjrpes of intermodulation products that are formed ^vill all fall on other than operating channels. This is readily accomplished in the case of 3rd order products by select- ing the operating channels in such fashion that the frequency difference between any pair of these channels is unlike that between any other pair of channels. Many other rules inherently more complicated and more cumbersome to apply than the one stated above must be obeyed if 5th order as well as 3rd order products are to be controlled in this way. Table IV presents p operating channels selected from a band of n adjacent channels (numbered sequentially in order of ascending fre- quency) in such fashion as to avoid 3rd order interference within the system. Considerable effort has been spent in selecting these channels to insure that the number n associated with each value of p is the lowest 1.0 0.6 0.4 a2 ao6 ao4 0.02 - y^ y^y^ '^ y T - / /// / TRANSMITTER IS ACTIVATED R = PORTION OF TIME AVERAGE RECEIVER IS CONNECTED n = NUMBER OF CONSECUTIVE CHANNELS IN BAND / / / / / / / / / / / / / / A ^1 / / c f m / - / . 7 - / 1 1 1 / / i / i 1 / / * 1 n f 1 (a) T = 0.5 = R 1 1 ..L. 1 1 - ' ,^ •^ y^ - y ^/ A / / / / / ^ J f f / / / / 0 2 // j / 0 1 fj (i / - r>i it ' / / 0 06 - f / / y i I f 0X}4 ' 1 1 — 1 1 — OiOZ 1 1 f OOI (C) T= 0.125 = R 1 1 / n 1 1 6 8 10 20 30 40 p = NUMBER OF OPERATING CHANNELS 60 80 100 Fie. 1— Probability of intermodulation interference I versus number of operat- ing channelH p when only 3r(i order products are considered, (a) T = Portion of time average transmitter is activated « 0.50; li = Portion of time average receiver is connected - 0.60. (b) Same as above except T - 0.25 - R. (c) Same as above except T - 0.126 - R. 70 0.6 0.4 0.2 0.1 0.06 0.04 0.02 0.01 1.0 0.6 0.4 - '/^"y^y^^ ^ 1 - / / / T = PORTION OF TIME AVERAGE TRANSMITTER IS ACTIVATED R = PORTION OF TIME AVERAGE RECEIVER IS CONNECTED n= NUMBER OF CONSECUTIVE CHANNELS IN BAND w / / / / //. / / / /// / /// J §/o ^ y / - / - / / t } i r / / 1 (a) T = 0.5 = R 7 / ._!_ J-. 1 1 0.06 0.04 - /^ /y^^ - / / // / / / A ' / / // / / / / / //// / /w / - III - III II / / 1 (b) T = 0.25 = R 1 V 1 1 — 1_ 5 6 8 )0 20 = NUMBER OF OPERATING 30 40 50 60 CHANNELS 200 Fig. 2— Probability of intermodulation interference I versus number of operat- ing channels p when both 3rd and 5th order products are considered, (a) T = Portion of time average transmitter is activated = 0.50; R = Portion of time av- erage receiver is connected = 0.50. (b) Same as above except T = 0.25 = R. (c) Same as above except T = 0.125 = R. 71 72 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table IV — Specific Operating Channels Having No Third Order Intermodulation Interference p is the number of interference free operating channels which can be obtamed from n consecutive channels. p n Operating Channels Having No 3rd Order Interference 3 4 1,2,4 4 7 1, 2, 5, 7 5 12 1, 2, 5, 10, 12 6 18 1, 2, 5, 11, 13, 18 7 26 1, 2, 5, 11, 19, 24, 26 8 35 1, 2, 5, 10, 16, 23, 33, 35 9 46 1, 2, 5, 14, 25, 31, 39, 41, 46 10 62 1, 2, 8, 12, 27, 40, 48, 57, 60, 62 8* 137 1, 2, 8, 12, 27, 50, 78, 137 * Neither 3rd nor 5th order interference exists with this selection of eight op- erating channels. possible number from which p channels having no 3rd order interference can be selected. A plot of p versus n based on the above table is shown in Fig. 3. The curve which includes both 3rd and 5th order intermodulation prod- ucts shows that 300 consecutive channels must be made available to provide for the careful selection of 10 operating channels which are in- terference-free at all times, regardless of the traffic load. For comparison, it was shown earlier (Fig. 2) that more than 500 consecutive channels must be available to permit picking at random 10 operating channels UJ u ■^ ui UJ UJ UI >■ HO, - 10 u 3 Z II 20 10 .»' ^^^ - 3R0 ORDER INTERMODULATION ONLY^^ ^ ^ ^^ - ^ ^ -^ ^ ^ ^^^^^ 3RD AND 5TH ORDER INTERMODULATION y^ ^ ^ "i^ ^ ^ ^ / -1_ 1 1 200 400 2 3 4 5 6 8 10 20 30 40 60 80 100 n = NUMBER OF CONSECUTIVE CHANNELS Fig. 3 — Number of consecutive channels n required to provide a number of interference free operating channels p. INTERMODULATION INTERFERENCE IN RADIO SYSTEMS 73 which are subject to interference 10 per cent of the time when fully- loaded with traffic. Careful channel selection is therefore a step toward better frequency- usage. It is, however, only a small step, and a more effective solution must be found if efficient frequency usage is to be achieved in areas where sizeable numbers of operating channels are required. Such a solu- tion for the mobile radio service is a "coordinated system," proposed in a companion paper previously referred to, wherein additional measures are described for reducing the probability of interference by proper geographical system layout and the use of certain practicable operational features. Magnetic Resonance PART I — NUCLEAR MAGNETIC RESONANCE By KARL K. DARROW (Manuscript received September 3, 1952) Magnetic resonance is the name of a phenomenon discovered less than sixteen years agOj which from the start has had a high theoretical importance and is now attaining a notable practical value. Nuclear magnetic resonance occurs when a substance containing magnetic nuclei is exposed to crossed magnetic fields, one being steady and the other oscillating, and the strength of the former field and the frequency of the latter are matched in a particular way. When these are properly matched, the nuclei are turned over in the steady field, and energy is absorbed from the oscillating field. Another way of describing the effect is to say that resonance occurs when the applied frequency is equal to the frequency of precession oj the nuclei in the steady field. This phenomenon illustrates very clearly some of the fundamental laws of Nature. For the purposes oj nuclear physics it is used to determine the magnetic moments of nuclei and their relaxation-times in the substance that contains them. It is also used for chemical analysis, fbr measurement of magnetic fields, for analysis of crystal structure and for locating changes of phase of the substance containing the nuclei. Magnetic resonance of electrons is similar, but for a fundamental reason is confined almost ex- clusively to free atoms of certain kinds, to ferromagnetic substances and to certain strongly paramagnetic salts. For these last it serves to throw light on the fields prevailing within the crystals. "Magnetic" is an ancient word in physics and so is "resonance," but "magnetic resonance" is something new. It is the name of a phenomenon which is sharp and clearcut and easy to evoke, which springs directly from the ultimate magnetic particles of matter, which illustrates the fundamental laws of these, and which has found and still is finding uses of importance. There are two types of it, the nuclear and the electronic. Nuclear magnetic resonance is the theme of the first part of this article: it will recur from time to time in the second part (to appear in a later issue of this Journal) but the main topic of that second part will be 74 NUCLEAR MAGNETIC RESONANCE 75 electronic resonance. It is fitting that they should be treated in this order, for the nuclear type of resonance is less distorted by complexities than is the other. Perhaps it is not premature to say that while nuclear magnetic resonance always goes by that name, the electronic type is usually called "paramagnetic resonance" or "ferromagnetic resonance." Nuclear magnetic resonance was realized in 1937, in molecular-beam experiments. The war distracted physicists, and the next great step was not made until after — but very soon after — the armistices. In the winter months of 1945-46 the phenomenon was produced in liquids and in solids. The news burst upon the world from the pages of The Physical Review in the early weeks of 1946, causing among physicists an immediate and an immense sensation. Of some discoveries one wonders how they came to be made at all, of others one wonders after- ward why they were not made earlier. Nuclear magnetic resonance is of the latter class. But this is a discovery that could not have been made much sooner than it was, for it required the apparatus and techniques of short-wave radio and microwaves, and these are recent. The work of 1945 was done by two independent groups three thousand miles apart, using somewhat different experimental methods and ex- pounding the theory in somewhat different ways. The differences are really superficial, and in the course of time will probably be minimized; but the two streams of later work that rose from those two sources are still distinguishable. The methods are called the nuclear resonance absorption method and the nuclear induction method: I treat them in this order. A sketchy account of the molecular-beam method will follow upon these, and then several of the applications — which of these are major and which are minor must be left for history to decide. On the first few pages, and on many thereafter, the talk will be of protons. Protons are the commonest material particles in Nature, electrons excepted (neutrons are also an exception but not an important one here, as they are seldom found free). Protons also have the happy attribute called "spin J^" soon to be explained, which simplifies the exposition greatly. This is one of the rare fields of physics in which the simplest case, the commonest case, and the most useful case, are all three of them one and the same. PROTON RESONANCE ABSORPTION To begin with, there must be a sample containing hydrogen, protons being the nuclei of ordinary (as distinguished from heavy) hydrogen atoms. It may be pure hydrogen in gaseous, liquid or solid form, or any 76 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 one of countless compounds of hydrogen. I first take water for the sample, and enumerate the other particles in water. There are the nuclei of the common isotope of oxygen, oxygen 16: they are non- magnetic and produce no resonance. There are the nuclei of rarer isotopes of oxygen and hydrogen: they will be mentioned later. There are the electrons: they are reserved for Part II of this article. We are now left with the protons. The sample is placed between the poles of a magnet, Fig. 1, so that it is in a magnetic field which should be homogeneous and is usually strong. The strength of the field is denoted by H, and its direction is always that of the 2-axis (and usually vertical). I should hke to call it "the steady field," but usually it is modulated during the experiments, so I shall call it "the big field". Actually it can be very small, but nearly always it is between 8,000 and 15,000 gauss, and 10,000 gauss is a good figure to keep in mind. The big field must not be the sole magnetic field applied to the sample. There must also be an alternating or oscillating field — stationary electromagnetic waves, formed in a solenoid (or sometimes in a resonant cavity). Such waves comprise, as Maxwell taught us long ago, an alternating electric field and an alternating magnetic field. In most of the uses of electromagnetic waves it is the electric field that counts, and the magnetic field is remembered only as something demanded by Maxwell's equations to keep the electric field going. In this application the electric field takes a back seat, and it is the magnetic field that counts. This oscillating magnetic field must be at right angles to the big field; we lay the x-direction along it. Its amplitude, to be denoted I t i i t t i t t t ( t i i t ^ SAMPLE ■*-X Figr. 1 — Scheme of the apparatus for observing nuclear magnetic resonance. The detecting circuits are omitted. The nuclei indicated by the arrows are of "spin ^/' protons for example. NUCLEAR MAGNETIC RESONANCE 77 Fig. 2 — Peak of nuclear resonance absorption. This is the first peak to be published other than those obtained with molecular beams. It pertains to protons in water. (Courtesy of E. M. Purcell). by Hi , is of the order of a small fraction of one gauss up to several gauss. Its frequency must be of the order of tens of megacycles. To be more specific, the effect that is sought with protons is located at 42.6 mc when H is 10,000 gauss. Finally there must be circuits and detectors for measuring the ab- sorption of the electromagnetic wave-energy in the sample. These are well kno^vn to those proficient in the art: we pass them over. Now of the two quantities H and v either is to be varied while the other is to be kept constant, and the absorption is to be measured. Usually H is varied while v is kept constant, and the data consist of a plot of absorption against H for a set value of v. When such a curve is plotted it proves to be, in the main, a smoothly- sloping curve, of no interest in the present connection. What is of interest is that it is interrupted by a magnificent peak of extraordinary sharpness, deserving to be called a needle. Probably there is nothing that can please an experimenter more than a curve with a fine sharp peak: here he has it. Fig. 2 exhibits the first such peak on record. But neither Fig. 2 nor any other picture can convey an adequate idea of the sharp- ness of the peak, for the distance from this imposing feature to the axis of the ordinates at field-strength zero may be, and often is, tens of thousands of times as great as the breadth of the peak. (With the induction-method, peaks have been distinguished from each other that are separate by only a millionth of the value of H at which they are found). This needle has the narrowness that is characteristic of fine lines in optical spectra; and this is as it should be, for a spectrum-line 78 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 is just what it is, even though it lies in the radiofrequency range and for technical reasons appears on a scope instead of on a photographic film. The peak is the phenomenon of magnetic resonance. We shall now interpret it in terms of theory somewhat oversimplified, for it is not the office of these opening paragraphs to introduce all of the complexities of quantum mechanics. In Fig. 1, inside the rectangle which represents the sample, appear a number of arrows. These are symbols of the protons. In other circum- stances we might imagine the protons as solid balls, in still others we might imagine them as centres of force; but for the present purpose we are regarding them as tiny bar-magnets, and the arrows symbolize the directions in which they are pointing. It is necessary to label these directions with perfect clearness. The figure has been drawn with the south pole-piece of the big magnet (the one responsible for the big field H) above. The point of each arrow represents the north pole of the protonic magnet. Thus the arrows pointing upward represent protons in the orientation into which the big field would like to turn the protonic magnets, and would indeed succeed in turning them if they were literal compass- needles in literal compasses. The arrows pointing downward represent protons in the opposite orientation. I will call these, for shortness, the "up" orientation and the "down" orientation. Evidently if the physicist could reach into the sample with fingers or with forceps and turn a proton from the up orientation into the down one, he would be doing work upon the proton at the expense of energy from his muscles. Well, he cannot reach into the sample with fingers or with forceps and grasp and turn a proton. But he can reach into the sample with the oscillating field and turn the protons, and this is the experiment we are considering. Magnetic resonance is the turning of protons from the up orientation into the down one, from the orientation or ''level" of lesser energy into the orientation or level of greater energy. But why does the effect occur at one frequency only? And what determines that frequency? To cope with this problem we shall have to introduce symbols, equations, and quantitative reasoning. The first step is to evaluate the work required to turn the proton, or, in other words, the energy-difference between the two orientations or levels. It shall be denoted by Wy and the magnetic moment of the proton by /Xp . We proceed by strictly classical reasoning. The torque exerted on the proton by the magnetic field H is -fipH sin 0. Here 6 stands for the angle between the direction of the steady field and the direction in which the magnetic moment of the proton is pointing. We have NUCLEAR MAGNETIC RESONANCE 79 admitted the existence of only two values of 6, viz. the values 0** and 180°; more will be said about this later; but for the duration of this particular argument we shall have to admit all values of 6 from 0° to 180°. The value of W which we are seeking is the integral of iipH sin 0 from 0° to 180°, from the up orientation to the down one. It is easily obtained : /.ISO" TT = - / iXj,H sin BdB = 2^j,H (1) Having arrived at equation (1) by strictly classical reasoning, we must now approach equation (2) by a starkly quantal argument. Immense amounts of evidence have sho\\Ti that when energy is absorbed from electromagnetic waves of frequencies v in the optical range of the spectrum and in the X-ray range, not to speak of other ranges, it is invariably absorbed in parcels or quanta equal to hv, h standing as always for Planck's constant. If this doctrine is sometimes difficult to assimilate when applied to the optical spectrum, how much more difficult it is to accept w^hen applied to waves of radio frequencies! Yet here also it is to be accepted, so we put: W = hv (2) Now we transfer the value of W from equation (1) to equation (2), and arrive at the destination: H = y2hv/f.^ (3) In this equation h is kno^^^l with very great accuracy, and /ip had also been measured when the first experiments upon magnetic resonance were made, though not with nearly the accuracy that physicists now claim for it. It remains only for the experimenter to insert for v the value of the frequency in his experiment and for H the value of the fieldstrength at which the peak appears. The test is whether the two sides of the equation agree. Needless to say, the test has been brilliantly passed. Quantum-theory has entered into this argument in more ways than the one which led to equation (2). I return now to the fact that we have arrived at equation (3) by postulating two, and only two, ''per- mitted" orientations of the protonic magnets in the steady field. This is illustrated by the presence, in Fig. 1, of arrows pointing up and arrows pointing do\Mi but no arrows pointing slantwise. We might have assumed that there are protons, and therefore arrows, pointing in every direction. We might have assumed that there is a proton pointing, say, at angle 80 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 76°13' to the vertical, and that it can absorb a quantum hv of just the right energy to turn it to the angle 118°36'. This would have led to the inference that instead of absorption confined to the fieldstrength hv/2yLp corresponding to the actual peak, there would be absorption at every fieldstrength from hv/2yLp on upwards toward infinity. The experiment frustrates this inference, and so declares for the two and the only two permitted orientations. One did not have to wait for this experiment to learn this fact: it has been known for thirty years, both as a consequence of quantum mechanics and as a fact of experience. However this is a very pretty proof of it. We now must generalize equation (3) so as to make it take care of all nuclei and not the proton only; and in the course of this process we shall meet the actor behind the scenes who determines the permitted orientations. His names are spin and angular momentum. THE GENERAL EQUATION FOR NUCLEAR MAGNETIC RESONANCE We are now en route to the general equation of which (3) is the special case appropriate to the proton. Our first step takes us to the deuteron or nucleus of heavy hydrogen. Its magnetic moment differs from that of the proton, so we must write Md instead of jxp . More sig- nificant is the fact that the deuteron has three permitted orientations in the big field instead of two. The orientations of proton and deuteron are shown in the first and third columns of Fig. 3; beside them are horizontal lines depicting their energy-values, energy being measured vertically upward from an arbitrary zero. One guesses from the aspect of Fig. 3 that the deuteron will show three peaks of magnetic resonance ; for it seems possible for the deuteron to be turned from orientation a to orientation 6, from 6 to 2 and from a all the way to z. But of these three conceivable "transitions" the third a PROTON i DEUTERON .N z a U^pH ►b b a JMdH 2 't : Fig. 3 — Orientations and energy -levels of protons and deuterons in a mag- netic field, according to the "old" quantum-theory. NUCLEAR MAGNETIC RESONANCE 81 does not occur at all: the theorists know the reason why, and call it a "forbidden" transition. As for the other two, the mere symmetry of the picture shows that they involve equal absorptions of energy and there- fore contribute coincident peaks. There is therefore only one dis- tinguishable peak, and we have to find the value of H at which it appears. This is easily done. Going back to equation (1), we integrate the inte- grand which there appears from 0° to 90° (or from 90° to 180°) ; and so we come to the analogue of equation (3) which applies to the deuteron: H = hv/na (4) In the course of this argument we have met with an example of two general rules : no matter how many permitted orientations there are, transi- tions occur only between consecutive ones, and these permitted transitions always agree in energy-absorption, so that there is never more than one peak. Yet equation (4) differs from equation (3), because in the right- hand member hv/fi — and now I am using ju as the general symbol for magnetic moment — is multiplied by J^ for the proton and by one for the deuteron. Now, J^ is the value of the spin of the proton and one is the value of the spin of the deuteron. We generalize from these two instances: we use / as the general symbol for the spin; and we arrive at the following: H = (I/f.)hv (5) The generalization is sound; and equation (5) is the fundamental equation of nuclear magnetic resonance. I now have to interpret the word ''spin." Spin is a particular measure of the angular momentum of the nucleus. That a magnetic nucleus has angular momentum is surely not surprising. We are trained to ascribe magnetism to the motion of charged bodies: an electric current flowing in a loop has the same magnetic field as a bar-magnet. When a nucleus is observed to have a magnetic moment and an angular momentum, it is natural to correlate one property with the other: one does not quite know how far the analogy may safely be pressed, but at least it is helpful. But what sort of a measure of the nuclear angular momentum is the quantity 7? The answer to this question is confused by the fact that in our times there have been two forms of quantum theory: the "new" quantum mechanics which is undoubtedly more competent in general, and the "old" quantum theory of the nineteen-twenties which is certainly more simple in the present case. Desire to be clear has led me to employ 82 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 the older theory up to now, but conscience obliges me to introduce the new one. In the old theory, / is the nuclear angular momentum in terms of the unit V2ir. Two of the permitted orientations, which I will call the "extreme" ones, are straight along and straight against the field- direction. For these, the projections of the angular momentum upon the field-direction are -\-Ih/2Tr and —Ih/2Tr. For 'the proton / = M, and the two extreme orientations are the only ones. In the new theory, the nuclear angular momentum in terms of the unit h/2ir is \//(/ + !)• For the two extreme orientations, the pro- jections of the angular momentum on the field-direction are +//i/27r and —Ih/2ir, just as they were in the old theory. But now these orienta- tions are no longer straight along and straight against the field-direction. They must be inclined, one to the up direction and the other to the down direction, at the angle of which the cosine is I/^s/I/{I -\- 1). Thus I has partly changed its meaning: it is still the maximum permissible projection, upon the field-direction, of the nuclear angular momentum in terms of the unit V^tt, and this is what it was before; but it is no longer the magnitude of the nuclear angular momentum. So also has /i changed a part of its meaning. It is the maximum per- missible projection, upon the field-direction, of the magnetic moment of the nucleus, and this it was before ; but it is not the magnitude of the nuclear magnetic moment. The language of this subject has not been well adjusted to this change. Fortunately I is called the "spin," which does not necessarily convey the impression that it is quite the same thing as angular momentum; but ^ is still called the "magnetic moment," and in the new quantum mechanics this is a mistake. Fig. 4 is Fig. 3 redrawn in the spirit of quantum mechanics. The arrows now represent angular momentum and magnetic moment jointly. PROTON 2//pH t 3EUTER0N f — a If M [/ / b b |/^dH ^< l\ K UdH h 277= z Fig. 4 — Oriontutions and energy-levels of protons and dcuterons in a mag- netic field, according to quantum mechanics. NUCLEAR MAGNETIC RESONANCE 83 but the numbers affixed to them are the values of angular momentum. The energy-levels in the second and fourth columns are often known as ''Zeeman levels." I take this occasion to complete the statement about the allowed orientations, which in recent paragraphs has been made for the extreme orientations only. The projections of the nuclear angular momentum upon the field-direction are +//i/27r, +(/ - l)h/2T, {I - l)V2x, -Ih/2Tr. From this principle combined with the fact that transitions occur only between consecutive levels, follows rigorously equation (5), which I de- rived in a looser way. Spins are ascertained in various ways, usually from their influence on the electrons surrounding the nuclei, which manifests itself in details of optical spectra and in cleverly-designed molecular-beam experiments. They are always integer multiples of }4- Important instances of nuclei of spin 3^ are the proton and the nucleus F^^ The neutron and the electron also belong in this category, as we shall see later on. The deuteron has already provided us with an important instance of a nucleus of spin one. Spins as high as % are certainly known, and this is probably not the limit. Nuclei of spin zero are common: I have already mentioned one of them, oxygen 16. Such nuclei do not produce magnetic resonance; we shall have nothing to do with them. A brief table shall conclude this section. To what has already been stated it adds the number of permitted orientations corresponding to each value of spin. Spin Number of orientations H for peak 3^ 1 V2 2 3 4 (^^)Wm hp/fi (%)Wm I 2/ -\- 1 THE LARMOR PRECESSION AND NUCLEAR INDUCTION Now we go back to first principles, make a fresh start, and arrive by a different route at the equation for magnetic resonance. On this route we meet with a vivid justification of the use of the name "resonance." Resonance implies a tuning or a matching between an applied fre- quency and a frequency either actually or potentially present in the substance in question. A piano-wire, a membrane, the air in an organ- pipe, an electrical circuit comprising capacity and inductance, all resonate to the frequency which is that of their own natural vibrations. No mention has yet been made of a frequency peculiar to the nucleus which is matched by the applied frequency when magnetic resonance 84 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 occurs. There is indeed such a natural frequency, not however a fre- quency of vibration; it is a frequency of precession. Precession is a concept well known to astronomers and to such physicists as have to do with gyroscopes, perhaps not so well known as it should be to others. In Fig. 5, the vertical is again the direction of the big magnetic field. The arrow represents the angular momentum of the nucleus, which I now denote by p. The magnetic field H exerts a torque on the nucleus. I have already given an expression for this torque, but I gave it in the language of the ''old" quantum-theory. To employ this ex- pression with as little apparent change as possible, I introduce the symbol fio for the magnetic moment of the nucleus, and reserve /x for PSIN0 Fig. 5 — Illustrating the Larmor precession. the maximum permissible projection of juo on the field-direction. The torque now appears on the right-hand side of the following, purely classical, equation: dp/di = fjLoH sin 6 (6) This is a vectorial equation, but I will endeavor to express its vectorial content by words instead of symbols. Fix the attention on the tip of the arrow. The torque makes it describe a circle of radius p sin d in the horizontal plane, with a frequency which I denote by v. Its peripheral speed in this circle is v multiplied by the circumference of the circle, therefore u'2irp sin d. This speed is dp/dt. Putting its value into (6), we observe with pleasure that 0 vanishes from the scene: the result is going to be the same for all orientations of the magnet: this is it: H = 2tpv/ho (7) NUCLEAR MAGNETIC RESONANCE 85 Making the substitutions that have already been describe, we get: H = {I/n)hv (8) and this is none other than formula (5), the fundamental equation of magnetic resonance. The precession-frequency is the resonance-fre- quency. This precession is often called the "Larmor precession," and the frequency given by (5) or (8) is called the "Larmor frequency." The name is a posthumous honor; Larmor died before magnetic resonance was discovered; his theory was applied to the Zeeman effect, the effect of magnetic fields upon optical spectra. It is not hard to believe that when the applied frequency coincides w4th the Larmor frequency, something drastic must happen to the precession. The theory has been worked out on a classical basis. I will not pursue it into its details; but at least the first step should be taken. I have said that the alternating field is perpendicular to the big field. We take the x-direction as its direction. The magnetic field, or magnetic vector as I will henceforth call it, has then Hi cos 2Tro)t for its a;-component (I use co for the frequency so as to distinguish it from the Larmor frequency) and zero for its ^/-component. Now imagine a vector, of constant magnitude (3^)^i , lying in the a;?/-plane, pointing away from the 2-axis and revolving around this axis clockwise with frequency oj. Its x-component will be (K)^i cos 27rco^, its ^/-component will be —Q/2)H\ sin 2iroit. Imagine another such vector revolving counterclockwise. Its x-component will be Q/QHi cos 2iro)t, its y-Qovn- ponent will be (}^)Hi sin 27rco^. (It is evident that we have chosen their phases so as to bring about this result). The sum of these two vectors has Hi cos 2T(at for its a;-component and zero for its ^/-component. But this is the vector that we started out w^ith. In the language of optics, we have resolved a plane-polarized wave into two circularly-polarized ones. The foregoing is pure mathematics. Now comes the physics. Of these two revolving vectors, one is whirling in exact unison with the precessing magnet when co is exactly equal to the Larmor frequency, the other is rushing round and round in the opposite direction. Our intuition tells us that the former may be expected to produce a great effect on the precession, the latter a small one. The latter is not always negligible, but may be neglected here. Thus in this artful way w^e have substituted a circularly-polarized field for the actual plane-polarized one. The theory further leads to the prediction that when resonance exists, the precession will be exaggerated in such a way as to produce 86 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 an alternating magnetic flux across the xz plane. Now I describe an actual experiment, the first of its type. The sample is water (or something else) in a spherical container. Around the container are wrapped two coils at right angles to one another. The coil of which the axis is parallel to the x-axis produces the alternating field. The coil of which the axis is parallel to the y-axis is connected with a rectifier and a detector. At resonance there is an alternating magnetic flux through the latter coil, and by the operation of the rectifier this is converted into a signal on the scope. The signal locates the resonance-frequency as accurately as does the peak in the absorption-method. This is the phenomenon called ''nuclear induction." I terminate this section by mentioning a paradox resulting from precession. Everyone has seen a compass-needle turning to point to the north : it is natural to infer that when a magnetic field is applied to a piece of matter, the elementary magnetic particles of which the nuclei (and also the electrons) are examples will automatically turn to point along the field. Yet the analogy fails and the inference is false: the nuclei do not turn to point along the field, but each of them maintains a constant angle with the field while it precesses. It seems to follow that matter cannot be magnetized by a magnetic field, but again the inference is false. Animistically speaking, the field makes the nuclei want to turn into its direction, but they cannot fulfill their desire without assistance from something other than the field. This something-other is not absent, and in the section on "relaxation" we shall meet with it. THE MOLECULAR-BEAM EXPERIMENT There are three methods for detecting and locating nuclear magnetic resonance, and we have now considered two of them. In one of these, the resonating nucleus makes itself manifest by absorbing energy; in the other, that of nuclear induction, by radiating energy; in the one which is to come, by simply failing to turn up at the scene of the measure- ment. This singular attribute is that of the molecular-beam experiment, which (I repeat) was done before the others and so receives the credit of revealing nuclear magnetic resonance. Molecular-beam experiments are so remarkable that it is hard to speak of them without yielding to temptation to say more than is essential to the purpose, but here the temptation must be withstood. Conceive a narrow stream of hydrogen-containing molecules coming along the (horizontal) axis of y, and cutting across a big magnetic field parallel to the (vertical) axis of z. This big field differs from that of NUCLEAR MAGNETIC RESONANCE 87 Fig. 1 in one important way: it is non-uniform, increasing in strength from (say) the bottom to the top. In one respect the protons behave just as they do in a sample in a uniform field: roughly half of them are pointing up and the other half pointing down. But in the non-uniform field the ''up" protons experience a net force pushing them upward and the "down" protons a net force pushing them down. (Visualizing each of the protons as a tiny bar-magnet, one sees that the fieldstrength is bigger where the upper pole of the magnet is than where the lower pole is). The beam is parted into two diverging pencils, the one con- taining the "up" protons only and the other the "down" protons only; I call the first the "up" pencil and disregard the second. The "up" pencil now passes through a region just like that implied in Fig. 1 : a magnetic field which is big and vertical and uniform — H will stand for its strength — and an oscillating field with the magnetic vector parallel to the x-direction. If in this second region some of the protons are turned by the oscillating field into the "down" orientations, that will make no difference to their course across the remainder of the second region where H is uniform. But beyond the second region lies a third where again there is a big field that is non-imiform. In this third region the "up" protons go one way and the "down" protons go another. The detector lies athwart the first way; the "down" protons will miss it. The detector-reading is plotted against H for a set value of v. One might think that two curves would be plotted, one with the alternating field off and the other with it on, and that the latter would be sys- tematically lower than the former. But the latter will be lower than the ^95 z HI cc LLI z ^ 80 to z f^ 75 2 70 < 65 60 1^ • . • i • -• ■■ • •._« • 112 113 114 115 116 117 118 119 120 MAGNET CURRENT IN AMPERES Fig. 6 — Negative peak or valley of nuclear resonance absorption obtained by the molecular-beam method. It pertains to lithium nuclei in lithium chloride molecules. This was the first experimental evidence of nuclear magnetic resonance. (I. I. Rabi, J. R. Zacharias, S. Millman and P. Kusch). 88 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 former only in the immediate vicinity of the value of H which conforms to equation (3) ; for the protons are turned over only when the Larmor frequency agrees very nearly with the applied frequency. Accordingly one keeps the alternating field on all the time and plots a single curve; and this is marked by a fine sharp peak, but this time a peak that points downward, Fig. 6, for it testifies to the absence of the overturned protons that have missed the detector. The first experiment of this kind was done on molecules of lithium chloride. The reader may have been puzzled that I spoke of a beam of molecules and then of the deflection of protons: the protons, or whatever other magnetic nuclei are being studied, carry the molecules with them. In the experiments on LiCl, the peaks of lithium and of chlorine were found in different parts of the curve. Later the proton-resonance was discovered by using molecules of KOH and NaOH, and confirmed with molecules of H2 and HD (the latter being a hydrogen molecule of which one nucleus is a proton and the other a deuteron). It is from this molecule of HD that the proton-resonance, and for that matter the deuteron- resonance also, stand out most clearly and sharply. In H2 and in D2 the resonances are perturbed and multiplied, but for reasons which are well understood so that the theory is strengthened instead of being weakened; but to describe these pretty things would be confusing unless they were explained, and to explain them would take us far afield. SOME APPLICATIONS OF NUCLEAR MAGNETIC RESONANCE The first of the uses of nuclear magnetic resonance is of interest mainly to the nuclear theorist. He wants to know (//m) for as many nuclei as possible; and this knowledge may be found by locating the resonance-peaks, and applying to their values of H and p the equation (3) or (5) which I repeat: H = il/fi)hp (9) Anyone who is going to burrow into the literature of this subject must be apprised beforehand, or else find out the hard way, that this simple statement is variously expressed. Here is a sad case of the ruination of a beautiful terminology by carelessness. The terms which have been mined are "gyromagnetic ratio" and "magneto-mechanical ratio." The former ought to mean, as originally it did mean, the ratio of angular momentum to magnetic moment. The latter ought to mean the ratio of magnetic moment to angular momentum. Both have by NUCLEAR MAGNETIC RESONANCE 89 now been used in both these senses, and there are variants within each sense, depending on the unit that is preferred by the user. The ap- pearance of either ''gyromagnetic ratio" or "magneto-mechanical ratio" in a paper is a red light warning the reader to make sure just what the author means by it. In this paper both of these terms are discarded with regret. An experimenter may give his value of (ju//) directly, or may give his value of g, which is (fi/I) expressed in terms of a peculiar unit. The peculiar unit is eh/4:TrmpC, in which nip stands for the rest-mass of the proton and the other symbols have their normal meanings. This unit got into the picture because there was a doctrine that (fi/I) for the proton ought to be just two of it. This was based on an analogy with the electron which, to the consternation of theorists and the complica- tion of Nature, proved to be fallacious. Reported values of g range from nearly 6 to 0.143; the proton has one of the two highest values, the triton or nucleus of hydrogen 3 has the other. Since all these values may be described without much extravagance as being "of the order of 2," the use of g remains convenient.* Many people say that they have measured fx. Formally this is all wrong, but practically it is usually all right, for in most if not all cases / is known from experiments of other kinds. Most of these people give the value of n in "nuclear magnetons." This means that they are giving the value of gl, as is seen from the following equation which resumes in notation what I just said in words, and provides the definition of g: II = gl(eh/^irmpc) (10) The quantity in brackets is called "nuclear magneton." Now that this tiresome but necessary passage is behind us, we can review the results. Values of gl — or of some other of the quantities catalogued above — have been published for about forty nuclei. The values of gl available toward the end of 1950 were gathered together and published in an article to which I give the reference in a footnote, f The largest is about twenty-five times the smallest: this is a wide range of variation, yet not so wide as that of the nuclear charges or the nuclear masses. Isotopes of one another may have values nearly the same or considerably dif- ferent; the same is true of isobars. Most of the values are positive: this * It is perhaps not premature to mention that in optical spectroscopy and m electronic magnetic resonance, the symbol g is used with a similar but not an identical meaning. _ t Pake, G. E., American Journal of Physics, 18, pp. 438^52, pp. 473-86, 1950. The table is on p. 440 of the October issue. 90 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 means that the angular momentum and the magnetic moment are parallel. A few are negative: for these (one of which is the neutron) the angular momentum and the magnetic moment are anti-parallel. These values of /x (for, I repeat, gl is ju expressed in terms of a par- ticular unit) are useful as challenges and as aids to the nuclear theorists. They are challenges, because the /x-value of a given nucleus is something to be explained ; they may be aids, because a theory may be fortified by giving the right value of /z or confuted by leading to a wrong one. Now, nuclear theory is difficult, and by and large it is not so far advanced that it can demand experimental values accurate to let us say, the legendary "sixth place of decimals." This is a piece of temporary good fortune, for two reasons. First, the strength of the big field H may not be known with adequate accuracy at the place where the nuclei are. It can however be ignored if one is concerned only to measure the ratio of the (m//) values of two nuclei. The experimenter has then to put into his apparatus successively samples containing the two kinds of nuclei, or a single sample containing them both: the ratio of the frequencies at which the resonance-peaks appear is the ratio of the {ixjl) values, and H vanishes in the division. Often the comparison-nucleus is the proton, so that many published values of gl come ultimately from ratios in which gl for the proton stands in the denominator. Such ratios are frequently adequate for the testing of theories, and their accuracies may be very good indeed, even attaining the sixth significant figure. (The basic determination of /x for the proton itself will be mentioned in Part XL) Second, the true field which the nuclei experience may be slightly different from the big field H, because of local fields within the sub- stance. This is of course an admission that our fundamental equation, (5) or (9) in this article, can be wrong. So it can be, and this is a de- velopment that may be thought distressing. But such developments are almost the rule in physics, whenever the art of measurement is bettered ; and in the present case the errors in equation (9) must be regarded as felicitous, for they lead to some of the most fascinating applications of nuclear resonance. Thus when ammonium nitrate, NH4NO3, is put into the apparatus, there are two peaks of nitrogen instead of one. They are not far apart — if for the frequency in use one is at i/ = 10,000 gauss the other is at 9,997. The formula NH4NO3 suggests, and the diagram of the molecule would confirm if we had it here, that the two nitrogen nuclei are dif- ferently placed in the molecule: one may say that they have different atomic surroundings. Thus the position of either of the peaks is dis- NUCLEAR MAGNETIC RESONANCE 91 tinctive not of nitrogen alone, but of nitrogen in its particular sur- roundings. These same surroundings might recur in several different types of molecule, or might be confined to one. The formula of ethyl alcohol may be written as CH3-CH2-OH. This compound presents three proton-peaks, Fig. 7, separated by a few per cent of one gauss when the big field is of the order of 10,000: they have been ascribed to protons in the three "groups" CH3 and CH2 and OH. To identify a group is to perform a process of chemical analysis, and this is a nascent application of nuclear magnetic resonance. This is a good place to speak of the efficacy of nuclear resonance in revealing the presence of chemical elements or of individual isotopes. The proton is one of the easiest nuclei to discern in this way, largely owing to its relatively high magnetic moment. It has been calculated that 2-10^^ protons suffice to give a detectable "signal" by the induc- Fig. 7 — Breakup of the proton resonance peak of ethyl alcohol into three peaks, each believed to arise from protons in distinctive 'groups within the molecule. (Courtesy of M. E. Packard). 92 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 tion-method ; a small capsule of gaseous hydrogen at a pressure of only one atmosphere will show the resonance of protons. The second isotope of hydrogen is normally present in that substance in an abundance of only 1.5 parts in ten thousand, the second isotope of oxygen exists only in an abundance of four parts in ten thousand; neither was discovered for more than a decade after the search for isotopes was well under way; but both of them have been detected by nuclear resonance. Another application is to crystallography. In the crystal called gyp- sum, each proton is exposed to a magnetic field of the order of ten gauss from its neighboring protons. The resonance-peak is split into two or three or even four, depending on the inclination of the big field to the crystal axes. It would take many pages to describe this effect in detail, but it is so intelligible that one may deduce from it the positions of the protons in the crystal lattice. Nuclear resonance in fact seemed called to play a great role in crystallography, since the principal tool of the crystallographer has been the diffraction of X-rays, and this will not disclose the presence nor a fortiori the locations of protons in a crystal lattice. However this promising child of resonance has apparently been throttled in its cradle, for the still newer art of neutron-diffraction has proved itself adequate for finding the protons in a lattice. Another application is to the measurement of magnetic field strengths. One sees that if proton-resonance is produced at a measured frequency in a steady field of which the magnitude H is unknown, H may be determined by equation (3) with an accuracy contingent on the accuracy with which fjLp is kno\vn, and this is pretty high. This has become a common method of measuring magnetic field strengths. RELAXATION If anyone were asked to guess the most important use of nuclear magnetic resonance, he would have two good reasons for choosing the study of relaxation. More pages of the scientific journals have been devoted to it than to any other application. Moreover, the discoverers spoke of it almost as soon as they spoke of the discovery; one has the feeling that they were so confident of the discovery, that as soon as it was made they considered it much less important for its own sake than as a tool. "Relaxation" is a word that entered long ago into physics. Its general meaning is the gradual self-adjustment of a system to a sudden change in conditions. In the immediate instance the system is our sample in the big magnetic field; the sudden change in conditions is the starting or NUCLEAR MAGNETIC RESONANCE 93 the stopping of the oscillating field; and what gradually adjusts itself is the distribution of the protons between the up and the down orienta- tions. Now I will describe an experiment such as has been performed on protons in water. Let the sample be placed in the big field some time — several hours will always be ample — before the experiment is to begin. The experi- menter should know in advance the frequency of the Larmor precession, so that he can apply the oscillating field of proper frequency as soon as he and the sample are ready. The sample then enters into what I will call "the state of resonance." The experimenter is to measure the height of the peak as soon as the oscillating field is switched on: I call this initial stature Ao . The big field and the alternating field are now both to be kept on. The height of the peak, A, is to be measured from time to time, say once every tenth of a second (this has been done with movie techniques). It is found that A is a declining function of time; the peak is shrinking. After a while, let the alternating field be switched off while the big field continues to be on. The state of resonance is now suspended. Again the height of the peak is to be recorded every tenth of a second. Need- less to say, the alternating field must be on while the record is being made, but it shall be off all of the rest of the time, which is most of the time. It will be found that the peak is growing again. It is, in fact, trending back to its initial stature Ao , and the law of its rise is the exponential law: A = Ao[l -exp(-t/TO]* (11) The constant Ti , which this experiment determines, is called the "spin- lattice relaxation-time." "Lattice" will be recognized as a term ap- propriate to crystals: in the literature of this subject it is however applied to all solids and liquids. Its meaning in this field may be put as follows: the "lattice" is all of the sample except the nuclear spins. The actual experiment is not usually done quite as I just described it. The alternating field does not have to be switched on or off, because if its frequency is far from the Larmor frequency it is practically in- effectual. If the observer wants to end the state of resonance, he dis- places H or V away from the resonance-value; if he wants to restore it he brings H or v back to the resonance-value. By modulating the big field with say a 60-cycle frequency, he may pass the system briefly * This formula implies that A = 0 when i = 0; the reader can recast it to cover the general case in which 0. The striking feature of Table I is the increase in delay time for ran- dom service, which becomes more pronounced with decreasing F(u) and increasing occupancy (or traffic) level, a. The increase throughout the table is an effect of the limitation to small values of F{u). For given * Thanks are due to George W. Abrams for directing this work, to Dr. Richard W. Hamming for transforming the equations into forms suitable for the differen- tial analyser and for supervising its operation, and to Miss Catherine Lennon for a great deal of calculation. 102 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 1.0 0.6 0.4 0.2 0.1 V \, \ s 0.06 v\ N,^ \^ \ \^ FN ao2 \\ \ \ \. ^^ ^ A \ \ \ 50 60 70 80 120 Fig. 1 — Delay curves for random service. F{u) = conditional probability of delay at least u; u = ct/h, c = no. trunks, h = av. holding time, a = call input in time h, a = a/c. a, the delay curves for order of arrival and random service include the same area, which is in fact equal to the mean delay (of calls delayed), (1 — a)~ . Since F{0) = 1 for both, and the random service curve decreases more slowly for large u, the curves must intersect at some point, say for u = iio ;ior u < Uo , the o.a. curve must be above the ran- Table I — Delay-Time and Random Service Delay Times, w, for given F(u) and a and for order of arrival (o.a.) and ran- dom service. a P - •0.1 F = 0.01 F = 0.001 o.a. Random o.a. Random o.a. Random 0.1 2.56 2.58 5.12 5.47 7.68 8.60 0.2 2.88 2.91 5.76 6.57 8.63 10.68 0.3 3.29 3.34 6.58 8.05 9.87 13.35 0.4 3.84 3.91 7.68 10.04 11.52 16.95 0.5 4.61 4.68 9.21 12.89 13.82 22.09 0.6 6.76 5.82 11.51 17.25 17.27 29.97 . 0.7 7.68 8.28 15.36 23.14 23.03 43.33 0.8 11.51 12.57 23.03 36.80 34.54 70.29 0.9 23.03 25.99 46.05 77.24 69.08 156.63 DELAY CURVES FOR CALLS SERVED AT RANDOM 103 dom curve. This is shown in Fig. 2 for a = 0.9, but the logarithmic scale for F(u) obscures the equaUty of area. The character of the comparison may be clearer if the picture is changed. Consider a department store counter with c clerks (correspond- ing to c trunks) in attendance. The time for a sale corresponds to the trunk holding time, and the rate of arrival of customers is like that of call input. For service in order of arrival customers are given serially numbered tickets on arrival; for random service, these tickets may be supposed drawn from a hat, or numbered from a series of random num- bers, or since aggressiveness and the clerks' attention are subject to devious rule, it may be that no attention at all to order of service is equivalent to random service. The fact that the average delay is independent of the order of service may be explained roughly by saying that the average rate at which wait- ing lines are removed depends only on the average rate of arrival of customers and the rate at which they are served. Notice however that service at random causes more variable delays (the second and all higher moments are larger than for order of arrival service). Thus with random service the proportion of waiting customers receiving quick 1.0 0.6 0.4 0.06 — 0.04 LL 0.02 0.01 0.008 0.004 0.002 t. \\ \. \ \ s Vs \\ \ \, s^ \ \ \ \ N S. RANDOM V \ \ \ \ N \ v^ ^ \ \ ^ ^^ ORDER OF \ ARRIVAL ^ K ^^ ^ ^ 1 ^ — '^ 0.001 . . . . . , 0 to 20 30 40 50 60 70 80 90 100 110 120 130 140 U Fig, 2 — Comparison of delay curves for order of arrival and random service; a = 0.9. 104 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 service is increased (over order of arrival) but this is achieved at the cost of making other customers wait much longer. Service in order of arrival has the advantage to the customer that his delay is independent of all who come after him, and this is particularly appreciated in times of heavy crowding when long delays are possible for random service. In Table I, these crowded conditions correspond to small values of F{u) or large values of a, or both. In this picture it seems intuitively clear that much longer delays are possible for random service, for those unlucky customers who keep missing their turn. (Of course, a more realistic model would also include the effects of cus- tomers leaving before service, a factor of considerable telephone interest also.) As noted at the start of this section, F(u) is a conditional probability, the probability of delay at least it of a call that is surely delayed. To obtain unconditional probabilities of delay, F(u) is multiplied by the probability that all trunks are busy, which is the probability that a call is delayed. This probability is given by a well-known formula due to Erlang and customarily written as c r 2 c-i C(o.a) =,-_,^^^_^ !_! + - + 2,+ ••• + (c - 1) ! (c - a) L 1! 2! ■ ' (c - 1)! + (c - 1) ! (c a)J Tables of this function are available*. Finally it may be noticed here that for random service and light traf- fic (roughly, a less than 0.7), with sufficient approximation with 7/1 = 1- VW^y 2/2=1 + VoA * But there seems to be no extensive tabulation. However, the table for the Erlang B function made by Conny Palm (Stockholm, 1947) may be used with the relations 1 1 1 C(c, a) B(c, a) B(c - 1, a) ^ a , 1 - (g/c) " c B(c, a) Notice that C(c, a) also has the recurrence relation 1 -1 . (c - a){c - 1) C{c, a) c - I - a a(c - 1 - a)C(c - 1, o) DELAY CURVES FOR CALLS SERVED AT RANDOM 105 3. BASIC FORMULATION As noted above, the following notation is used: c is the number of trunks, h is the average holding time (the distribution of holding times is exponential) and a is the average number of calls arriving in time in- terval h. Then, if Fn(t) is the probability of delay at least t of a call arriving Avhen n other calls are waiting, the differential recurrence relation given by Vaulot is ^ = ^ ^ F„_,(0 - ^ FM + ^ F„+.(0 (1) dt n+l/i h h This may be derived as follows. Consider the interval dt after the epoch of arrival of the call in question. In this interval three events may occur: (i) a call may arrive, (ii) a trunk may be released, or (iii) neither of these. The probability of a call arrival is {a/h)dt and if a call arrives the delay function is Fn+i{t — dt). The probability of a trunk release, because of the assumption of exponential holding time, is {c/h)dt, and if a trunk is released the number of waiting calls is reduced by one; the probability that the call seizing the waiting trunk will not be the call in question is n/{n + 1). Finally the probability of the third event is 1 — (c + a)dt/h. All this is summarized in the differential relation FM) = ^dt Fn+i{t - dt) + -^ I dt Fn-iit - dt) h n -\- 1 h + (i - ^ dty^ - dt) Passing to the limit gives equation (1). Using new variables : tt = ct/h, a = a/c, equation (1) may be written more simply as dFniu) du n + 1 Fn-l{u) - (1 + a)Fn(u) + aFn+lW (l^) This equation is a mixed differential-difference equation of the first order as a differential equation and of the second order as a difference equation; hence three boundary relations are required. For the differen- tial part, it is clear that Fn{0), which is the probability of some delay of the test call, is unity for all n in question, that is, for all integral non- negative n. Also Fniu) = 0 for all negative n, is an obvious necessity, and, since Fn is a distribution function Fn{^) = 1- Finally the third 106 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 condition may be stated as lim Fn{u) = 1, all 2^ n=oo The probability of delay at least u of an arbitrary call is the sum on n of the product of the probability that n calls are waiting when the call arrives and the probability, Fn(u), that for this condition the call is delayed at least u. The first probability (for statistical equilibrium) is known to be (1 - a) dc, a)a" where C{c, a), as stated above, is the probability that all trunks are busy; (1 — a)C(Cj a) is the probability that all trunks are busy and no calls are waiting. Hence the probability in question, say /(it), is given by or by /(«) = (1 -«)C(c,a)2:«"^n(«) /(«) = C(e, a)F(u) F(.u) = (1 - a) Z a"F„(M) (2) 0 F{u), like Fn{u), is then a conditional probability, the probability at least w of a delayed call. Notice that, consistent with this, F(0) = 1. It is interesting to notice that Mellor's basic equation, which in pres- ent notation may be written as du n + 1 follows from (1) if first it is supposed that Fn-\{u) = Fn{u) = Fn+i(u) and then, for clarity, Gn replaces Fn . Hence, as indicated by the third boundary condition, it may be expected to be useful for large values of n. Its solution is Gn{u) = e-"^^"-^^^ (4) A somewhat better approximation may be determined by the Mac- Laurin series obtained by repeated differentiation of (la) and evaluation DELAY CURVES FOR CALLS SERVED AT RANDOM 107 Sit u = 0; this is as follows J, . . ^ , u , a (uY a{2a - 1) {uf (5) n + 1 2 (n + 1) 3! (n + 1) ai2oL - l)(3ce - 2) {uY _ ■^ 4! (n + 1) * * * But this is the same* as: Fn(u) ;^ [1 - (1 - a)u/(n + 1)]^'^^-^ (5a) As a approaches unity, (5a) approaches (4). Equation (5a) has been used, for large values of a, in the direct computations mentioned above. It may also be noted that for a = 0, equation (la) has the solution (now writing Fn{u, a) for Fn{u)) Fniu, 0) = {u, n) - — ^ (u, n-1) (6) where 0, Jo the last by integration by parts. * G. W. Abrams is due credit for noticing this. (10) 108 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Follo\Wng (2), this may also be written as 00 Mk = a - a) "Z cc^'mn.k , (9) 0 wifh rrin.k = / wl-F^(w)] du, Jo = k f u^~'Fn{u) du, k> 0. Jo First, notice that mn.o = -f F'niu) du = FM = 1; hence Mo = (1 - «) f: a" = 1, 0 showing that F(u) is properly normalized. Next, by integrating both sides of (la) with respect to u from 0 to « , and using the second form of (10) (with k = 1) -(n + 1) = nmn-1,1 - (n + 1)(1 + a)mn,i + (n + l)amn+i,i (11) In the same way, after first multiplying (la) throughout by u^~^, it is found that — k(n+ l)mn,k-i (12) = nmn-i,k — (n-\- 1)(1 + a)mn,k + (n + l)amn+i,k Unfortunately, neither (11) nor any other instances of (12) have simple solutions; nevertheless they may be used to determine Mk . Consider first the simplest case, Mi . If (11) is multiplied throughout by a" and summed on n, the result may be written — Lio = aLn — (1 + a)Ln + Ln — Loi = — Loi where for convenience in writing and of later notation Loi = S Q!"mn.i = (1 — oc)~^ Ml Lii = ^{n -\- l)a"m„.i Lio = E (^ +i)«" = ^ Z «"'"' = (1 - «)" (13) DELAY CURVES FOR CALLS SERVED AT RANDOM 109 and D = d/da. Hence • Ml = (1 - «)-' This is the mean delay of calls delayed and as mentioned above is the same as for service in order of arrival. In the general case*, the following notation is convenient I/O* = Y^ oL'mn,k = (1 — a)~^Mk Ljic = E (^ + 1)(^ + 2) • • . (n + j)a''mn,k Using the relations n(n + 2) • • • (n + j) = n(n + 1) . • • (n +i - 1) + (j - l)n(n + 1) • • • (n +i - 2) + • • • + (i - l)^Mn + 1) '" (n + j - i - 1) + "' + (i - l)In, n(n + 1) • • • (n + i - 1) ' = (n + l)(n + 2) . • • (n + i) - j(n + l)(n + 2)(n + j - 1) with (j- 1)^= (i- 1)0' -2) --'(j-i), the summing of (12) is found to result in kLj,k-i = [j - {j - l)a]Lj-i,k - a[(j - l)2Ly_2.fc (14) + (i - l)3Ly_3.. +"•-{- (j - l)iLj-i,k + ■" + U - l)!^iJ But this may be simplified by multiplying through by j and sub- tracting from the same equation with j replaced hy j + 1 ; the result is (i + 1 - Jot)Ljk - fLj-i,k = kLj+i,k-i - jkLj,k-i (15) Notice that for j = 0, k = 1, Ui = Lw , as in (13). Notice also that so that Lyo = jLj-1,0 + «2)Ly_i.o = i!(l - «)"^' *This procedure is the development of a suggestion made by S. O. Rice. 110 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Then the ratio Ui^iLmT' = (1 - af^'U,/k\ (16) = (1 - afMu/k\ = R, is the ratio of these moments to those for order of arrival service; the last relation is a definition. In the same way the ratio might be considered, but to avoid fractions the following somewhat odd change of variables seems convenient: {'t'')gkLjk = VjkLj+k,o (17) where go = gi = 1 and g^k = (2 - aY'-\3 - 2af'-' . . . (k + 1 - ka) gu+i = (2 - afiS - 2af'-' . . . (/c + 1 - kaf Notice that g2k+i(g2k)~^ = g2k{g2k-i)~^ = (2 - a)(3 - 2a) • • • (/c + 1 - /ca) = Dk+i the last being a definition, again. Since (1 — a)Lkfi — kLk-1,0 it follows from (15) that 0' + 1 - JodPjk - i(l - (x)pj-i,k , ^ (18) = (gk/gk-i)[(j + l)pj+i,k-i - i(l - oi)pj,k-i] By taking differences of this equation and writing Qok = Pok qik = pik — pok = Apok qu = P2k — 2pik + pok = A^Pok Qjk = Aqj-i.k = A%k a somewhat simpler recurrence relation is found to be as follows 0' 4- 1 - ja)qjk = (gk/Qk-i) (19) L/agy_i.jfc_i + 0' H- 1 + joc)qj,k-i + 0" + ^)qj+i,k-i] DELAY CURVES FOR CALLS SERVED AT RANDOM 111 Since pyo = 1, ally, goo = 1, and qjQ = 0, j 9^ 0. From these boundary conditions, it follows at once from (19) that Qjk = 0, i > k. By comparison of (17) and (16) QkRk = Pok = Qok A short table of the g's is as follows: j/k 0 1 2 3 0 1 2 3 1 I 0 1 «(2 - a)-i 0 0 2 4a (2 - a)-l 2a2(3 - 2a)-i 0 2(2 + a) 2a (18 - 5a - 4a2)Z)Ti 2a2(18 - 7« - 2a2)(3 - 2a)-2 Continuation of this leads to the values of Rk listed in Table II. Notice that for a = 1, by (18) Vjk (1) = (j + l)py+i..-i (1) = (i+l)(i+2)p,+2..-2(l) = (j + 1) (i + 2) . . . (j + k) since Qk = 1 for a = 1 and pjo = 1, all j. From this PokH) = QkiDRka) = ^;fc(l) = k\ On the other hand, for a = 0, gyt = 0, j > 0 and, by (19) ^0^^(0)^.(0) = go..-i(0)/^.-i(0) so that R,{0) = Rk-i{0) = RM = 1 5. MELLOR APPROXIMATION It is useful to have the moments of the distribution corresponding to the Mellpr approximation, since they serve as a guide. Here, following equation (4) Fiu) = (1 - a) D a"e n -«(n+l)-l (20) 112 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 apd 00 exp - xM = E Mk{-xY/k\ 0 = {I- a) \ du g-^" Z (n + i)-i^-e-"("+i)-^ (21) JQ 0 = (1 -a) Z all + a:(n + 1)]"' Hence ikT, = fc!(l - a) Z (n + 1)V (22) These moments are expressible in terms of polynomials associated with the distribution of permutations into classes according to the number of readings left to right necessary to find the elements in standard order. ^ Indeed the ratio n{a) =Mu{l- af/h\ has the recurrence relation n+i(a) = {ka + l)n{a) + a(l - a)r\{a) (23) and the first few values are as follows ri = 1 n = I -\- ^a -\- a r2 = 1 + a r4 = 1 + lla + lla' + a r5 = 1 + 26a + 66a' + 26a' + a Notice that rk{0) = \, rk{l) = /c!, just as for the precise results. 6. EXPONENTIAL SUMS The shape of the delay curves, from direct calculation, and also from Mellor's results, suggests representation in exponential sums. If Fiu) = Ai e-^'-"^""'' + A2 e-<^-"^«/^« + . . . (24) then M, ^^ ~ "^' = yl,xj + A^t+... (25) by a simple calculation. For k exponentials, 2k moments (including DELAY CURVES FOR CALLS SERVED AT RANDOM 113 Mo) may be fitted exactly by solution of 2k equations of form (25), as will be shown. The first approximation (k = 1) is the order of arrival curve, say F,{u) = 6-^'-"^" which has Ai = Xi = 1, Ak = Xk = 0, k > 1, and matches Mo and Mi . The next approximation (/c = 2) is determined from equations Ai + A2 =1 AiXi + ^-20:2 = 1 Aixl + A2X2 = R2 Aixl + A2XI = R^ Eliminating A2 from successive pairs, Ai (xi — 0:2) = 1 — X2 AiXi (xi — X2) = R2 — X2 Aixl{xi — X2) = R-i — R2X2 Eliminating Ai from these, 2:1 + 0:2 — 0:1X2 = R2 (26) {xi + X2)R2 — 0:10:2 = Ri or, writing ai = 0:1 + 0:2 , 02 = 0:1X2 , so that x^ — aiX -\- a2 = {x — Xi) {x — X2) ai — a2 = R2 (26a) aiR2 — a2 = Rz From the first of the second set of equations, and from symmetry (or from Ai + A2 = I) 1—0:2 Ai = A2 = Xi — X2 1 - xi X2 — Xi (27) f Taking R2 and R^ from Table II, it turns out that Xi"' = 1 - VW^ = 2Ai X2~' = 1 + vW2 = 2A2 (28) 114 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table II - Moment Ratios, Calls Served at Random RM = Mk(i - ccy/k\ Rx' 1 i22 = 2 2 - a /2, = 2(2 4- a) (2 - «)« R* = 4(6 + 5a - 4a« - a3) (2 - «)'(3 - 2a) R.= 4(36 + 60« - 59a2 - 24«3 4- I5a* + 2a6) (2 - a)*(3 - 2a)2 i2.= SMa) (2 - a)H3 - 2a)='(4 - 3a) R7^ 8/7(a) (2 - a)«(3 - 2a)*(4 - 3a)2 /6(a) = 432 + 972a - 2016a2 - 437a' + 1790a^ - 528a'^ - 196a6 + 67a7 + 6a8 /7(a) = 10368 + 34560a - 89208a2 - 32772a' + 177926a* - 104287a5 - 29260a« + 43876a7 - 9158a8 - 2039a9 + 588aW + 36a» and the second approximation is 2F2(u) = (1 - vW2) e-"^^-«^^^-"^) + (1 + VW2) e-^''-'''^^^' ^^^^ which turns out to be a good fit for a roughly less than 0.7. Curiously the corresponding Mellor approximation has a more complicated ex- pression. Following the same procedure for three exponentials, it turns out that the correspondent to the set of equations (26a) is 0'iR2 — Oi + as = Rz aiRz - a2R2 + a^ = R, (30) (i\Ra — (hRi 4" dzRi = Ri with ai = Xi -\- X2 -\- X3 , a2 = XiX2 + X\Xz + 2:2X3 , az — 0:1X2X3 , that is, the symmetric functions. DELAY CURVES FOR CALLS SERVED AT RANDOM 115 Using Table II for values of the R's, it is found that • ai = (18 -7a- 2a') (2 - a)-\3 - 2a)-' ^2 =18 (2 - a)"'(3 - 2a)-' (31) as = 6 (2 - a)-\3 - 2a)-' a;i , a;2 and 0^3 are then the roots of the cubic equation x^ — aix' + a2X — aa = 0 The coefficients Ai , i = 1, 2, 3 are determined from equations like A = ^2 — {xi + x^ + 3:2X3 /„ s (:ci — 2:2) (xi — 0:3) For the fourth approximation, matching 8 moments, the equations for the symmetric functions are CLiRz — (I2R2 ~h ^3 — cii = R\ aiRi — aiRz -\- a^H^ — a^ = Rt, cliRq — cLiRh ~\~ clzRa ~ CI4R3 = Ri and 0:1 , a:2 , Xz and x^ are roots of the quartic equation x^ — aix^ + a2X^ — azx -\- a4 = 0 Coefficients Ai are determined from equations like . _ Rz — (x2 -{• X3 + Xi)R2 + (x2a;3 + 3:23:4 + x^x^) — ^2X3X4 ,^4) (xi — 0:2) (xi — a:3)(a:i — 0:4) It may be noted that 3:2 + xs + X4 = ai — xi X2XZ + 0:23:4 + 3:33:4 = a2 — xi(ai — xi) 3:23:33:4 = az — Xi[a2 — Xi{a — Xi)] = a^i which gives the general structure. It is worth noting that equations (33) may be used to determine the 7^'s if the a's may be determined otherwise. As a matter of fact, they have led to the determination of Rt and Ri in the following way. The (33) 116 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 results for A; = 2 and 3 suggest that a4 = 4!(2 - a)-\S - 2a)-\4 - 3a)"' as = 4a4 Then by the first two of equations (33) flii^s ~ (I2R2 = Ri — 0,3 ~h 0^4 dlRi — Ci2Rz — Rb — ^3^2 4~ O'i the solutions of which are ai = 4(24 - 23a + 3a') [(2 - a) (3 - 2a) (4 - 3a)r' 02 = 2(72 - 23a - 10a' - 3a')[(2 - a) (3 - 2a) (4 - 3a)]-' By the last two of equations (33), Rq and i^7 are determined to be the values given in Table II, which have been verified independently. Note that for a = 0, both R^ and Rj are 1, and for a = 1, R^ = 6!, i?? = 7! Table III tabulates, for /c = 2 to 5, for convenience in avoiding frac- tions the symmetric functions hkj related to those above by hkj = Dkaj with, as before, Dk = (2 - a)(3 - 2a) • • • [/c - (/c - l)a] and oo = 1. The functions for /c = 5 were obtained by a process like Table III — Symmetric Functions for Exponential Sums of Calls Served at Random ik = 2 A; = 4 k = 5 6jo = 120 - 326a + 329a2 _ U6a^ + 24a< 651 = 600 - 978a + 329a2 + 146a3 - 72a< 652 = 1200 - 978a - 172a2 + 78a3 + 72a* 66J = 1200 - 326a - 172a2 - 78a3 - 24a< 654 = 600 655 = 120 620 = 2 - a k = 3 630 = 6 - 7a + 2a2 621 = 4 631 = 18 - 7a - 2a2 622 = 2 632 = 18 633 = 6 640 = 24 - 46a + 29a2 - 6a3 641 = 96 - 92a + 12a3 642 = 144 - 46a - 20a2 _ 6a3 643 = 96 644 = 24 DELAY CURVES FOR CALLS SERVED AT RANDOM 117 Table IV — Symmetric Functions for Exponential Sums, Mellor Approximation k = 2 ai = 3+a A; = 3 ai = 6 + 3« a2 = 2 a2 = 11 + 5a + 2a^ a3 = 6 A- = 4 ai = 10 + 6a: a2 = 35 + 26a + 11«' as = 50 + 26a + 14a2 + Ga^ a4 = 24 A - 5 ai = 15 + 10a a2 = 85 + 80a + 35a2 as = 225 + 200a + 125a2 + 50a3 ai = 274 + 154a + 94a2 + 54a3 + 24a* a, = 120 A: = 6 ai = 21 + 15a a2 = 175 + 190a + 85a2 as = 735 + 855a + 585a2 + 225a3 a4 = 1624 + 1604a + 1194a2 + 704a3 + 274a4 as = 1764 + 1044a + 684a2 + 444a3 + 264a* + 120a5 as = 720 that sketched above, and without determining Rs and Rq . Notice that which may be proved independently. All values in Table III satisfy the recurrence relation hj = [k - (k - l)a]h-ij +[k+ (k- l)a]h-ij-i (35) — {k — 1) ahk-2,j-2 which also satisfies the boundary relations for a = 0 and 1 given above for all values of k. The corresponding symmetric functions for the Mellor approximation are given in Table IV. These have the recurrence relation akj = ak-i,j + [/c + (/c - l)a]ak-i,j-i - (k - 1) W-2.i-2 (36) For a = 0, the values are the signless Stirling numbers of the first kind, that is, the numbers given by the expansion of (1 + x)(l + 2x) ••• (1 + kx). For a = 1, the results are the same as for the exact case, as given above. 118 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table V — Approximations to Delay Function F{u) for Random Service Two Exponentials 0.1 .3590 .1351 .0220 .0041 .0008 .0002 0.2 .3490 .1344 .0256 .0058 .0015 .0004 .0001 0.3 .3392 .1332 .0291 .0079 .0023 .0007 .0002 .0001 0.4 .3292 .1315 .0325 .0101 .0033 .0011 .0004 .0002 0.5 .3190 .1293 .0357 .0125 .0046 .0017 .0006 .0003 0.6 .3085 .1265 .0386 .0151 .0061 .0025 .0010 .0004 0.7 .2978 .1232 .0412 .0177 .0078 .0035 .0015 .0007 0.8 .2868 .1193 .0434 .0203 .0097 .0047 .0022 .0011 0.9 .2756 .1148 .0451 .0229 .0118 .0061 .0031 .0016 Three Exponentials 0.1 .3586 .1354 .0219 .0040 .0008 .0002 0.2 .3491 .1356 .0254 .0057 .0014 .0004 .0001 0.3 .3393 .1358 .0288 .0074 .0022 .0007 .0002 .0001 0.4 .3291 .1360 .0322 .0092 .0030 .0011 .0004 .0002 0.5 .3186 .1363 .0358 .0112 .0040 .0016 .0007 .0003 0.6 .3071 .1359 .0392 .0133 .0050 .0022 .0010 .0005 0.7 .2951 .1354 .0428 .0156 .0063 .0028 .0014 .0007 0.8 .2822 .1344 .0466 .0181 .0077 .0036 .0018 .0010 0.9 .2683 .1325 .0504 .0210 .0094 .0045 .0023 .0013 7. numerical results Table V gives both two-exponential and three-exponential 4 decimal approximations to the delay function F(u) for and for a = 0.1(0.1)0.9(0.1 to 0.9 in steps of 0.1) u{l - a) = 1(1)2(2)14, in the same abbreviated notation.* The variable v = u(l — a) is in- troduced to reduce the spread of these tables. It will be noticed that, as expected, the two orders of approximation agree closely for small values of a; indeed, only for the three largest values of a are the dif- ferences appreciable from the engineering standpoint. ♦ The results for two exponentials, some of those for three-exponentials, and all special results given below, have been obtained by Miss Marian Darville, whom I also thank for her careful drawing of the curves. The entire three-exponen- tial table has been computed independently by Miss Lennon. DELAY CURVES FOR CALLS SERVED AT RANDOM 119 For a = 0.9, results for four exponentials have also been obtained and compare with those of Table V as follows {k = number of ex- ponentials) : k V 1 2 4 6 8 10 12 14 2 3 4 .2756 .2683 .2748 .1148 .1325 .1402 .0451 .0504 .0483 .0229 .0210 .0195 .0118 .0094 .0091 .0061 .0045 .0047 .0037 .0023 .0026 .0016 .0013 .0015 It is somewhat surprising that two exponentials should do as well as they do for large values of v (in fact for 2; = 12 and 14 better than three) ; a similar behavior appears in the following comparison of approxima- tions on the Mellor basis, again for a = 0.9 k V 1 2 4 6 8 10 12 14 2 4 6 .2725 .2671 .2777 .1115 .1379 .1408 .0446 .0502 .0477 .0237 .0207 .0205 .0129 .0097 .0102 .0070 .0051 .0054 .0038 .0029 .0031 .0021 .0018 .0018 From these comparisons, it appears a relatively small number of exponentials is sufficient for engineering purposes. The curves of Fig. 1 are those for three exponentials, for uniformity. BIBLIOGRAPHY 1. Erlang, A. K., L0sning af nogle Problemer fra Sandsynlighedsregningen af Betydning for de automatiske Telefoncentraler. Elektroteknikeren 13, p. 5, 1917; The Life and Works of A. K. Erlang. Copenhagen, pp. 138-155, 1948. 2. Molina, E. C, Application of the Theory of Probabilities to Telephone Trunk- ing Problems, Bell System Tech. Jl., 6, pp. 461-494, 1927. 3. Mellor, J. W., Delayed Call Formulae when Calls Are Served in a Random Order. P.O.E.E.J. 25, pp. 53-56, 1942. 4. Vaulot, E., Delais d'attente des appels tdl^phoniques trait^s an hazard. Comp- tes Rend. Acad. Sci. Paris 222, pp. 268-269, 1946. 5. Pollaczek, F., La loi d'attente des appels t^l^phoniques, Comptes Rend. Acad. Sci. Paris 222, pp. 353-355, 1946. 6 Riordan, J., Triangular permutation numbers. Am. Math. Soc, Proc, 2, pp. 429-432, 1951. The Evaluation of Wood Preservatives Part I Interpretation and Correlation of the Results of Laboratory Soil-Block Tests and Outdoor Test Plot Experience, with Special Reference to Oil -Type Materials By REGINALD H. GOLLEY (Manuscript received September 22, 1952) This paper offers a review and interpretation of laboratory and field experi- ments aimed at determining the necessary protective threshold Quantities of wood preservatives. It details the procedure followed in the soil-block tests at the Bell Telephone Laboratories^ Incorporated. Discussion of specific criti- cisms of the techniques involved and replies to these criticisms are included. The paper also presents for the first time a correlation of the results obtained from soil-block culture tests, outdoor exposure tests on stakes and on pole- diameter posts as well as pole line experience. It demonstrates that the same levels for toxicity -permanence requirements (thresholds) are obtained from the three different types of accelerated experimental evaluations. There is every reason to believe that the same limits apply for the outer inch of sap- wood in pine poles in line. TABLE OF CONTENTS Introduction 121 A Short History of the Development of Laboratory Evaluation Procedures 124 Evaluation by Soil-Block Tests. 132 General Procedures 132 Inoculation and Incubation Rooms 134 Soil Characteristics and Moisture Content 134 Even-Aged Cultures 138 Standard Test Organisms 138 The Scope of the Soil-Block Evaluation Test 139 Preparation of the Test Blocks — Manufacture 140 Test Block Selection for Density 141 Average Block Volume 141 Treatment of the Test Blocks 142 Retention Gradients 143 The Amount of Preservative in the Blocks 144 120 EVALUATION OF WOOD PRESERVATIVES 121 Block Density and Preservative Absorption 146 Weathering 150 Conditioning 162 Sterilization 152 Flow Chart for the Bioassay Test 152 Some Madison Test Results 154 Check Tests at the Murray Hill Laboratories 159 Across the Threshold 160 The Significance of the Results of Laboratory Soil-Block Tests on Oil- Type Preservatives 160 Bibliography 163 Subjects to be Covered in Part II Evaluation by Treated % Inch Southern Pine Sap wood Stakes in Test Plots Rating the Condition of the Stakes Depreciation Curves for % Inch Stakes Estimating Threshold Retentions and Average Life Evaluation b}^ Treated Pole-Diameter Posts in Test Plots Evaluation by Pole Test Lines and by Line Experience; Service Tests Discussion Density and Growth Rate Size and Shape of the Test Blocks Toluene as a Diluent for Creosote Treating Solutions The Distribution of the Preservative in the Block Heat Sterilization of the Treated Blocks The Weathering of Creosote and Creosoted Wood General Considerations; Creosote Fractions Creosote Losses Creosote Losses from Treated Blocks Creosote Losses from Impregnated Filter Paper An Interpretation of Creosote Losses The Gross Characteristics of the Residual Creosotes in Soil-Block Tests of Weathered Blocks The Evaluation of Greensalt The Evaluation of Pentachlorophenol Swedish Creosote Evaluation Tests Shortening the Bioassay Test Toughness or Impact Tests for Determining Preservative Effectiveness Other Accelerated Bioassay Tests Other Observations Conclusions Acknowledgments INTRODUCTION In discussing the problems involved in the evaluation of wood pre- servatives over the years, it has generally been found necessary to onent the audience — in this case the readers of this Journal — in the field of biology, and particularly m the field of biological tests involving wood- 122 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 destroying fungi. It is impractical to expect from such tests the degree of accuracy in results that one would look for as a matter of course in certain types of well conducted physical or chemical experiments. One can, however, look for high reliability in the biological sense. In the half science, half art of wood preservation there is as yet no generally accept- able laboratory technique for measuring the preservative value of a given material. Although much development work has been done, both here and abroad, in an effort to promote standard laboratory procedures, their proponents have had very little success in bringing into line the techniques used in the various areas. The interest of the Bell System in establishing a standard bioassay test will become convincingly evident as this story unfolds. When the first American telephone lines were built there was an ade- quate supply of naturally durable pole timber in northern cedar and chestnut forests. The chestnut trees have been killed by a fungus disease, the chestnut blight; and the chestnut supply failed completely about twenty years ago. Northern cedar trees are not straight enough nor large enough nor plentiful enough to meet the demands of the power and communication utilities, but they are still used to some extent in the Lake States area. Usually they are incised at the ground line by toothed machines ; and they are then given a preservative treatment with creosote or with pentachlorophenol in petroleum to prolong the life of the butt and ground Une section. In the northern and western states the increasing demands for poles 35 feet and longer brought in western red cedar, a straight and nearly perfectly shaped pole tree. The present Bell System use of the species is relatively small, about 4 per cent of the total annual production. Butt treatment of western red as well as northern cedar began in earnest about thirty years ago. This procedure protects the ground section. Many western cedars are now full length treated because, although the species is durable, the tops and sapwood layers are subject to infection and decay, sometimes after a relatively short service life. In the South and Southeast the great favorite is naturally the southern pine pole, full length pressure-treated with creosote. Such poles made their way in the Bell System as far north as Memphis and Washington by the turn of the century. Their use increased rapidly after World War I, and they moved into virtually all parts of the country. They now make up about 73 per cent of the telephone pole plant. New treatment procedures for southern pine employing pentachlorophenol in petroleum applied by pressure processes are now under way at a number of plants. Pressure-treated Douglas fir and butt-treated western red cedar dom- EVALUATION OF WOOD PRESERVATIVES 123 inate other species on the West Coast and in the Pacific Northwest, while pressure and non-pressure treated lodgepole pine poles are favored in the Mountain States area. Pressure-treated jack pine and ponderosa pine move into telephone plant in small quantities in the Lake States and in the California areas, respectively. To render telephone service the Bell System has some 20,000,000 wood poles carrying its wires and cables. Many of these poles are used jointly with the power companies. Since poles of the joint use sizes are not available in sufficient quantities in the southern pine forest to meet all the demands of the utilities all of the time it is inevitable that western cedar, Douglas fir, lodgepole pine, red pine and western larch should move into various parts of the System, either for the direct and sole use of the Operating Companies or for joint use. The pole plant is continually changing. Pole species from the North- west vary greatly in their treatability and they are generally harder to treat than southern pine. It is not possible to use traditional creosote pressure treatments for some of these species without running the risk of objectionable exudation, or bleeding, of the creosote. The development of practical specifications for the application of new preservatives such as pentachlorophenol and greensalt, as well as the various types of creosote, to all of the pole species now used in Bell System plant calls for setting as exactly as possible necessary protective quantities of the various preservative materials. This is particularly true in view of the fact that for normal telephone use as well as for joint use it is absolutely essential to deHver to the Operating Companies poles that are clean and satisfactory for use in all types of telephone lines, without compromising on the question of adequate physical life for the treated units. This purpose is back of the Laboratories' efforts to develop bioassay tests that come as close as practicable to measuring the neces- sary protective amount of any given preservative, and to predicting its relative permanence in poles and crossarms in plant. It has been pointed out in earlier papers^"' ^^' ^^ that Bell Laboratories' concept of preservative evaluation involves (a) laboratory evaluation tests, (b) test plot experiments with small stakes, (c) similar tests of pole size specimens, and (d) test fines selected for long time observation. The latter are chosen with the cooperation of the Operating Companies. Lumsden^^ has recently presented a summary of a quarter century of experience with pole-diameter posts in one test plot located at Gulf port, Mississippi. The principal aims of the present paper are to interpret the results of various laboratory methods of preservative evaluation, and to 124 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 indicate how these results may possibly be correlated with test plot and field experience. A SHORT HISTORY OF THE DEVELOPMENT OF LABORATORY EVALUATION PROCEDURES The practice of laboratory evaluation of wood preservatives developed along different lines in Europe and in the United States. Here the Petri dish method was the early favorite. ^^' ^°° The basic scheme of this test is to use agar culture media containing gradient concentrations of the preservative material to be tested, and to employ various easily grown test fungi as indicators of inhibiting or lethal doses. The same scheme is employed mth stoppered Erlenmeyer flasks. The fungus now known as Madison 517, formerly referred to as Fomes annosus, has been used most frequently as the standard test organism although other fungi were also used.^°^ In the culture phase of the European standard agar-block method^^ the test fungi are grown in Kolle flasks on a malt agar medium. The impregnated wood test blocks are supported on glass ''benches" juSt above the surface of the agar and the growing test fungus. Wood pulp or paper boards saturated with malt extract are used by some investi- gators^' ^ ^^^ in place of the agar medium alone. Generally an untreated block and a treated block are placed together in the same flask. The concept of a test for wood preservatives that motivated the proponents of the German agar-block method was broad enough to include selection of the test blocks and test fungi, treatment and hand- ling procedures except weathering tests, culture technique, determination of the protection boundary, and directions for reporting the results. Differences in the behavior of water solutions of single chemical com- pounds such as sodium fluoride, and of volatile oily preservatives such as the creosotes were recognized; and provision was made for dealing with both types of materials. The formahzing of both the Petri dish agar method in the United States and of the agar-block method in Europe developed as a result of conferences called by Dr. Hermann von Schrenk, the first in St. Louis in 1929, and the second in BerUn in 1930. The Laboratories' representa- tives at the St. Louis conference were the writer and R. E. Waterman. The action taken at St. Louis was published by Schmitz in 1930.^°° In a previous paper'' about a year earlier, Schmitz had discussed various laboratory test procedures, and had offered an ''improvement" in the Petri dish technique based on the idea of preventing evaporation of volatile materials. Some of his statements at that time now seem by EVALUATION OF WOOD PRESERVATIVES 125 hindsight to have something of the character of a judgment before trial; but their bearing on the questions under discussion and their possible effect in retarding the development of more reaUstic methods appear to be important enough to warrant quoting at this time. For example, with special reference to Petri dish agar tests he says: 'The determination of the toxicity of relatively volatile substances, such as coal tar creosote, is particularly difficult, owing to the control of the loss of preservative during the sterihzation process. In order to pre- vent this loss, it is proposed to place the preservative in small sealed glass ampules, which are later broken to Hberate the preservative to form preservative-agar mixtures of any desired concentration." He considers laboratory tests of toxicity of preservatives to have httle or no application in commercial practice, and his opinions are definitely stated as follows : ' 'Toxicity studies deal only with the poisonous properties of a wood preservative, and therefore they do not give a complete picture of the value of any particular substance as a wood preservative. . . . "For commercial work, however, it is of interest to know the amount of material that must he initially injected into the wood to maintain the desired amount of preservative for a definite period of time. (Author's itafics). Lab- oratory studies of the toxicity of wood preservatives do not give this information. Attempts to calculate the amount of material which must be injected into the wood from laboratory studies of toxicity are, there- fore, based upon an erroneous conception of the value of such studies." Writing about the laboratory use of impregnated blocks of wood in testing wood preservatives, which was already well under way in Europe, he says that by using wood one may obtain conditions more or less closely resembling but not identical with conditions in actual service; but one would not only have to use a solvent in treating to low retentions, but there would be difficulties in obtaining an even distribution of the preservative in the wood. Furthermore: "Getting rid of the solvent would require considerable time, during which a considerable loss of creosote would occur. . . . "The composition of the creosote in the impregnated wood after the solvent has evaporated may be quite different from that of the original sample. More important still, the movement of the solvent in the wood during drying would cause an uneven distribution of the creosote." The reader will bear in mind that these opinions were expressed in advance of the St. Louis and Berlin conferences on laboratory evaluation methods. Schmitz repeated them essentially in his 1930 paper, saying, for instance: "Toximetric values are not in themselves an index of the wood preserv- 126 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 ing value of the substance tested. Other factors, such as leaching, vola- tility, chemical stability, penetrability, cost, cleanliness, etc., must all be considered in the final evaluation of a wood preservative." With respect to the European wood block test, he felt that " . . . until more confidence can be placed in the even distribution of the preservative in the test block (s) their use will be greatly limited." He has maintained his arguments with a high degree of consistency in later papers, and they have unquestionably influenced American thought on laboratory procedures and their practical application. The Petri dish method adopted as a possible American standard pro- cedure at the 1929 St. Louis meeting followed closely the techniques that had been developed and published by Humphrey et al.,^^ Batemen^ and Richards.^ Bell Telephone Laboratories made an intensive study of the Petri dish method during this period. The data obtained were never organized for publication since it was felt that the required evaluation of toxicity and permanence of toxicity of preservatives could not be ob- tained by the Petri dish test. European workers would accept neither the Petri dish test method nor Madison 517 as the test fungus. In 1931, about a year after the Berlin conference, and four years before Liese et af ^ reported on the task force development of the agar-block method, A. Rabanus of the I. G. Farben- industrie Aktiengesellschaft, Germany, published his ''Die toximetrische Priifung von Holzkonservierungsmitteln" (Toximetric testing of wood preservatives).^^ A somewhat expurgated and amended translation of this paper was presented to the American Wood-Preservers' Association in 1933. In the writer's opinion much of the force of the Rabanus argument was lost in the translation. The emphasis on the relative merits of the agar toximetric test and of the agar-block test was considerably diluted; and the cautiously guarded but nonetheless positive philosophy on the possibilities of using the results of agar-block tests in actual wood pre- serving practice was made water thin. Apparently there was an understanding that subsequent to the 1930 conference in Berhn^^ tests by the agar-block method would be run in the United States. For this project samples of creosotes as well as Scotch pine wood blocks were sent to a number of workers; but to the writer's knowledge no treated blocks were ever tested, or if they were no results were ever published. At Bell Telephone Laboratories some of the un- treated Scotch pine blocks were put through preliminary trials with the Kolle flask technique,^^^ and also a considerable number of plate and flask agar toxicity tests were run with the two sample creosotes. The inconsistency of the results — as far as translation to practical wood EVALUATION OF WOOD PRESERVATIVES 127 preservation was concerned — was a strong stimulant toward the Lab- oratories' development of a block test, referred to later. It was more or less general information at the time that agar toximetric tests with native and European strains of test fungi were being run in other laboratories in this country; but again — as far as the writer knows — the results were not published. In the meantime, and, as in the case of the Rabanus article cited above, before the publication of the Liese^^ report, Flerov and Popov^* published in 1933 in German the basic general principles of a soil-block test. The significance of the article by these two Russian investigators was apparently completely lost on American workers until the publica- tion in England in 1946 of Cartwright and Findlay's ''Decay of Timber and its Prevention. "^^ Findlay had been a member of the Berlin con- ference. Flerov and Popov were familiar with the discussions and result of the conference, and decided in favor of the soil base for their cultures after a critical review of the various methods then in use. Their proposals to all intents and purposes were unknown here. Van den Berge's comprehensive thesis"^ on "Testing the Suitability of Fungicides for Wood Preservation" appeared in Dutch in 1934. A mimeographed English translation was made available soon after for limited distribution. European workers were about ready to confine the use of the agar toximetric test to determining relative toxicities only of various preservatives in an agar medium. Liese and his colleagues sum- marized the arguments and experiments on the agar-block method in 1935, and launched it into a status of general acceptance in Europe and Great Britain. The British^^' ^' and German^^ editions of the standard were issued in 1939. The Rumanian version"" — closely following the German — came out in 1950. Jacquiot,^^ Lutz^^ and AUiot^ worked out proposals for standard procedures that would be more, comprehensive and in their opinion better applicable to wood preservation research in France. The Petri dish — and later the stoppered Erlenmeyer flask — agar methods continued to be used by many American investigators for test- ing wood preservatives, and there is no denying a certain utility in these methods for developing information about fungus poisons. The persist- ency of the agar techniques can be traced through publications by Richards,"" Schmitz,'"' '°' Snell and Shipley,'"' Schmitz, ^ Buckman and von Schrenk,'"^ Schmitz, von Schrenk and Kammerer,'"' Bland'" and Hatfield. ^^ Baechler still uses the closed flask-agar method and the fungus called Madison 517 for determining basic toximetric values;*' ^ and Fin- holt'^ has recently been bold enough to state that 'Tungitoxic materials 128 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 can be evaluated as wood preservatives by mixing the toxic substances with a malt extract agar solution and then testing the mix against standard fungi." Flerov and Popov had used sand in their preliminary experiments, but they were by no means the first to do so (see Falck^^). Rabanus^^ had reported his experiments with sand-block cultures two years earlier. He placed a pair of wood blocks — one treated and one untreated — on glass rods on wet sand in Erlenmeyer flasks; and after sterilization he inocu- lated the blocks directly with his test fungi. He points out that in this procedure the conditions were less favorable for the fungi than when the treated wood is placed above or on a vigorously growing culture, as in the agar-block test. Since the papers by Rabanus and by Flerov and Popov appeared in the same journal, one can assume that the latter knew of Rabanus' work. How much any of them knew of still earlier work by Breazzano is un- certain. His work in Italy,^^' ^^' ^^' ^^ begun in the first decade of the century, is evidence of the intense interest of the management of the Italian railroads in some practical laboratory means for testing wood preservatives that would provide results sooner and with more definite- ness than the traditional service tests. Parts of Breazzano's report of Oct. 9, 1913 are worth quoting in full from the EngHsh translation as historical background information. He reviews the situation as he sees it, and says: ''New systems and various substances for injection into wood are constantly being put on the market by industrial concerns, so that the Railway Administration finds itself confronted by an ever increasing number of processes to be examined and tested for efficiency." By 1910 the Railway Experimental Institute "... is well on the way toward testing the efficacy of a system of wood preservation by a method which gives dependable results even after a few months of observation. "... after making use also of the advice on the subject received directly from Prof. Tubeuf and from Netzsch's laboratory . . . positive results were obtained with the following technique: "On the bottom of an Erlenmeyer flask of 200 ml capacity was placed a thin layer of sand." After sterilization in dry heat at 180°C "sterihzed water was poured on the sand to moisten it well. Then there was placed on the sand the sample of wood, of dimensions about 9x2x1 cm., with one end resting on the damp sand and the other on the inside wall of the flask." The whole setup was sterilized in an autoclave at 120°C for about 20 EVALUATION OF WOOD PRESERVATIVES 129 minutes. The wood was inoculated by placing a piece of a culture of Coniophora cerebella, grown on agar medium, directly on the wood. Bre- azzano states that the wood was kept moist enough because of the water in the sand, that the fungus grew luxuriantly, and that ''the development of the fungus was evidently at the expense of the wood, since no other nutritive substance was at its disposal." He used blocks cut from treated beech ties. The fungus grew readily and he concludes that the treatment was not effective. He ends this early report with the statement: "... If the experiment is carried on under carefully defined conditions the various methods proposed for immunizing woods can be judged all by the same standard." Breazzano presented his method at Pisa in 1919, and m 1922^^ the principles of the sand-block culture were proposed as standard procedure (for Italy) for evaluating wood preservatives. Precise directions were given for the whole test technique, Avith important modification of the cultures, as indicated in the steps outlined below: 1. SteriUze by dry heat, at 180°C, "soyka" boxes 8 cm in diameter and 4 cm in height "in which is first placed a layer of sand 1 cm deep". 2. Prepare blocks of wood — treated and untreated — 4x4x2 cm, cutting them so that the broader faces will be transverse sections; and place these test blocks broad face down on the sand. 3. Sterilize at 100°C for one hour. 4. After steriHzing and cooling add sterile water in an amount that will be slightly in excess of what the sand can absorb. 0. After the wood blocks become moist plant Coniophora cerebella — without carrying over any agar medium with the transplant. 6. Incubate the "soyka" box cultures in a covered crystalUzing dish in a dark place for one month at 20-25°C; and "Take care that in this time the water which the sand absorbs does not evaporate completely, and add sterile water when necessary." At the end of the test the wood blocks were to be examined for decay; and if there was any doubt the wood was to be sectioned and examined microscopically for the presence of wood-destroying fungus hyphae (threads). In retrospect the subsequent changes involving the use of soil instead of sand, and in the testing of blocks specially treated for the experiment, seem hke refinements of Breazzano's methods. He later shifted to the use of very thin pieces of treated wood for his test specimens,^ ^^ severely criticizing the agar-block method that grew out of the BerUn conference as time consuming and inaccurate (loc. cit.). 130 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 In Bell Telephone Laboratories, R. E. Waterman and his colleagues started work on a wood-over-water block method for testing wood pre- servatives soon after the St. Louis conference, and they published their early results in 1937 and 1938.'^' ''^' ''^ Their block was a %-inch cube with a hole drilled through it in the approximate center of a transverse face. The J^-inch cubes simply represented sections of the J^-inch square stakes that had been substituted for round saplings^^ in the small speci- men test plot experiments. The hole served a double purpose — it facili- tated handling the blocks during drying and sorting operations'^ ^ and it served as a point of entrance of moisture, which was purposely provided for the block by means of a wood wick. Leutritz^^ formalized a soil-block test completely independently of Flerov and Popov, and published his method in this Journal (Vol. 25) in 1946, following an earUer short article in 1939^^ suggesting soil as a culture medium. Beginning in the summer of 1944 and continuing until June 30, 1951, Bell Telephone Laboratories subsidized in part a series of studies by the Madison Branch of the Division of Forest Pathology, of the United States Department of Agriculture, Bureau of Plant Industry, in coopera- tion with the Forest Products Laboratory at Madison, Wisconsin. The results of these studies and of parallel investigations have appeared in eight papers"' '''''•''•''• ''''"•'' from 1947 to date. The differences between the agar-block and the soil-block techniques, and the results obtained in comparable test series by the two methods are of funda- mental importance. They are presented and discussed at length in a paper by Duncan."*' Already some 40,000 blocks have been tested by the soil-block method at Madison, with 75 oil-type preservatives. Both at Madison and at Bell Telephone Laboratories, Murray Hill, additional work aimed at further refining of the soil-block technique is under way. Subsequent to discussions of the new soil-block techniques between representatives of Bell Telephone Laboratories and of the Forest Prod- ucts Laboratories of Canada, Sedziak'"^ has developed a soil-block test involving burying the block in the soil all but one corner; and instead of placing it on a fungus culture growing on feeder blocks, he inoculates a corner of the test block directly. For a general review of laboratory and test plot methods for evaluating wood preservatives the interested reader should have available, in addi- tion to Cartwright and Findlay's book,^* at least two more recent books, namely "Wood Preservation During the Last 50 Years" by van Groenou, Rischen and van den Berge,"* and the third edition of Holzkonservierung by Mahlke-Troschel-Liese/* Hunt and Garratt*^ survey wood preservation EVALUATION OF WOOD PRESERVATIVES 131 with particular reference to the American scene. The works of Boyce,^^ Baxter^" and Hubert ^^ should be consulted for general information on wood-destroying fungi and the pathology of timber products. Kaufert" prepared a concise bibHography of pertinent articles in 1949. For a fuller coverage the book by van Groenou, Rischen and van den Berge will be found most stimulating. Much of the European work on the testing and application of wood preservatives has been summarized in challenging form by the investi- gators at the Berlin-Dahlem testing station. ^^ In this memorial volume, the first paper, by Schulze, Theden and Starfinger, is a compilation of the results of comparative laboratory tests of wood preservatives by the agar-block method. So much work has been done that the ingenious graphical summary table is about 12 feet long; and even then the authors have omitted many results because the conditions of the standard test^^ were not observed. Becker^^^^ brings up to date the results of testing insecticides in the second article; Becker,^^^^^ in the next paper, summarizes tests for termite control; and Becker and Schulze^^"^^ in the fourth article cover laboratory tests of preservative materials for the control of marine borers. Six additional articles on subjects directly related to wood pre- servation complete an excellent supplement to the Mahlke-Troschel- Liese book already cited. The emphasis is, somewhat naturally, centered on the work of the Berlin station. Rennerfelt and his colleagues^' ^^' ^^ are conducting a series of labora- tory, decay chamber and test plot experiments in Sweden, aimed at evaluating wood preservatives for use in that country, and at possible correlation of experimental results with actual experience. Bienfait and Hof^^ are working in Holland on what appear to be the broadest test post experiments in Europe at the present time, under both land and water exposure conditions. Their tests of 10 preservatives and some 3350 posts of Douglas fir, Scotch pine, European larch, Sitka spruce, poplar and willow rival Bell Telephone Laboratories' installa- tions in four test plots at Gulf port. Miss., Orange Park, Fla., Chester, N. J., and Limon, Colo.'^' ^'' ^^ and the Forest Products Laboratory installations in Mississippi.^^ Bienfait and Hof, like Rennerfelt, have been using the standard European agar-block test in their plan for correlation of laboratory and field results. No report on the Holland tests has appeared since 1948. Narayanamurti and his associates'^ in their first interim report on laboratory and field tests of creosotes of Indian origin present the results of some fifteen years work at the Forest Research Institute at Dehra Dun, indicating from still another quarter the compelling force that is 132 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 leading to the development of preservative evaluation methods to sup- plement or partly displace long and uncertain service tests. The authors present a mass of information on six different creosotes, on four creosote fractions, and on mixtures of the creosote with fuel oil of Persian origin. Sal (Shorea robusta) railway ties were used for the field trials. Many of the data are condensed into graphs that are small and difficult to read. The findings in general are favorable to the creosote-petroleum blends. The writer, on the basis of personal experience, is dubious about either the theoretical or practical significance, in experiments of the type re- ported, of the values given for standard deviations and standard errors. The scope of the work entitles it to more complete review than is prac- ticable at this particular time and place. Bell Telephone Laboratories are represented in a group carrying on comprehensive cooperative investigations of pedigreed creosotes on which four papers have already been published.^' ^^' ^^' ^^ The results of outdoor tests of small stakes and fence posts are issued periodically by the Forest Products Laboratory at Madison, Wis.^^* ^^' ^^' ^^ In this connection, the Proceedings of the American Wood-Pre- servers' Association are in the class of required reading. Additional ref- erences will be cited at appropriate points in the succeeding paragraphs. Data will first be presented on some of the experience of Bell Tele- phone Laboratories and others with laboratory soil-block tests, with outdoor tests of small stakes, of pole-diameter posts, and with pole test lines in evaluating wood preservatives. Through analysis and discussion an attempt will be made to interpret the significance of the results ob- tained by the various evaluation procedures and to correlate the evi- dence. Emphasis will be placed naturally on creosote and pentachloro- phenol because of their great importance to the Bell System pole plant. The writer intends to support his interpretations with experimental data wherever possible, reserving the privilege in some cases to make sugges- tions ais to possible significance, even though complete technical proof may be lacking at present. EVALUATION BY SOIL-BLOCK TESTS General Procedures Soil-block cultures have been described in a number of papers^^' ^^' ^^' ^' ^* since Leutritz presented his method in this Journal in 1946.^° Some of the following statements, therefore, will be repetition; but the intent is to outhne the technique employed at Bell Telephone Labora- tories as a base for later discussion. EVALUATION OF WOOD PRESERVATIVES 133 The culture jars are wide mouth cyhndrical 8-ounce bottles, provided with screw caps. The moisture content of the soil is predetermined on a representative sample, and enough distilled water is placed in the bottle so that when the soil is added its moisture content will be somewhat above 40 per cent by weight. The bottles are filled approximately half- full of screened field top soil — which means about 140 grams of an oven-dried sandy loam. The soil handles better if it is reasonably dry so that it can be poured through a suitable funnel; and putting the water in bottles before one puts in the soil results in a practically clean glass surface on the inside of the bottle above the soil level. Two southern pine sapwood feeder blocks, measuring 1 J^ inches in the direction of the grain by % inch by approximately ^^2 inch (35 x 20 X 4.0-4.5 mm) are placed carefully on the flattened soil surface, as shown in Fig. 1(a). The soil and feeder block setups are then steriHzed for one- half hour at a pressure of 15 pounds per square inch, after which they are allowed to cool in the autoclave. Inoculation is accomplished by carefully placing a piece of inoculum, cut from a fresh Petri dish culture of what may be called a standard test organism, at or near the middle of the feeder block surfaces. Under Fig 1— Four eight-ounce cylindrical bottles illuslrating the soil-block cultures: (a) Bottle half-full of top soil, containing 40 per cent moisture on an oven-dry soil basis, with two fiat feeder blocks of southern pine sapwood on top. (b) A sixteen-day old culture ready to receive the impregnated southern pine sapwood test blocks, (c) Two laboratory weathered test blocks from the same series treated to a below threshold concentration of pentachlorophenol 0.051 lb dry penta/cu ft, attacked by the test fungus, Lenzites trabea. (d) Two test blocks laboratory weathered, treated to a retention of 0.194 lb dry penta/cu ft, near the threshold retention, showing resistance to fungus attack. 134 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 the temperature and humidity conditions of the incubation room, the growth of the fungus mycelium covers the feeder blocks in about two weeks and the fungus threads are then well started downward into the soil. Treated test blocks are weathered and then conditioned under con- trolled temperature and humidity to approximate constant weight. They are then sterilized, along with untreated control blocks, in an autoclave for 15 minutes at 100°C, atmospheric pressure. As a rule two treated blocks having approximately the same retention of preservative are placed together in a single test bottle. The incubation period is three months, in an incubation room held at a temperature of 80 zb 2°F and at a relative humidity of 70 dz 2 per cent. At the end of this period the cultures are taken down. This means that the blocks are removed from the bottles, brushed free of fungus mycelium, and weighed immediately. They are given a preliminary examination for decay evi- dence, and then reconditioned, under the same temperature and humid- ity conditions as before sterilization, to approximate constant weight. Fungus attack is determined by observation and by weight losses. The general setup of the cultures is illustrated in Fig. 1 (a-d). Inoculation and Incubation Rooms To faciUtate handUng the soil-block cultures, an inoculation room and an incubation room have been built (Fig. 2) at the Murray Hill Labora- tories. Both are held at approximately the same temperature and rela- tive humidity, that is, 80°F and 70 per cent. The inoculation room senses as a lock chamber, and passage from it to the incubation room has a negligible effect on the humidity and temperature of the incuba- tion room. The latter is provided with an illuminated double plate glass window (Fig. 3), so that the interior can be exhibited without the neces- sity of entering the room. This window is fitted with a heavy roller shade, and the room ordinarily is kept dark. Soil Characteristics and Moisture Content The question that is asked most often about the cultures is whether a standard soil is used. European and American criticism has been defi- nitely directed^^' ^^ at the fact that the use of different soils might have so much effect upon the growth and the reaction of the test fungi in the cultures that quite different results would be obtained by investigators in different laboratories. This possibihty is recognized; but the evidence to date seems to point to the general conclusion that perhaps the prin- EVALUATION OF WOOD PRESERVATIVES 135 m iMMiii iiiiiniiii mtmmmmW Fig. 2— Soil -block cultures on the shelves in the incubation room The un- painted wood shelves come to equilibrium with the temperature and relative humidity and thus are a factor in keeping the conditions stable. Ihe back ea^es of the shelves are set away from the wall to provide spaces for air circulation. iHe front edges of the shelves are provided with metal labeling strips. 136 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 EVALUATION OF WOOD PRESERVATIVES 137 cipal and most important factor in the soil-block culture is the moisture holding capacity and content of the soil, rather than its nutrient func- tion. If continued experimentation supports this conclusion it would not be necessary to limit the type of soils used except within rather broad limits. It also appears that the size and thickness of the feeder block now employed introduces enough wood into the culture bottle to mask any minor variations in the soil itself. The all important thing is to have enough water in the soil throughout the test period to keep the air above the soil essentially at 100 per cent humidity and the blocks at about fiber saturation — say about 27 per cent, oven-dry weight basis. The soil in use at Bell Laboratories at present is obtained from a plot that has been set aside at the Chester (N. J.) Test Station. This plot has been fallow for twenty-five years. It supports a general grassy flora. The soil is a sandy loam with the following general description: pH 4.9-5.0 Available magnesium 37.5 lb/acre Available phosphorus 4.5 lb/acre Available potassium 70.0 lb/acre Organic matter 3.0 per cent The cultures at the Forest Products Laboratory^^ have been made with a silt loam having a pH between 5.5 and 6.0. Bell Telephone Labora- tories' tests have indicated the desirabihty of avoiding soils of either very sandy or very heavy clay types. The soil from the Chester Test Station described above is being used in all cultures, and there is a sufficient layer of top soil on the reserved plot to make parallel cultures for a good many years. Until such time as more definite and positive information on the effect of minor variations in the soil type are deter- mined it is generally agreed that all of the comparative tests in any given series at least should be run on the same soil. Experimental work is now under way to determine the possible advantage of the addition of Krilium* to the soil in the culture bottles to maintain porosity and an even, high moisture holding capacity. After the test blocks are placed in the culture bottles and during the course of the ninety-day incubation period the screw caps are left loose. The general technique followed in making up the soil cultures, as far as moisture is concerned, parallels that used at Madison. The moisture content is close to that recommended by Flerov and Popov,^^ namely 40-50 per cent of the weight of the soil plus the feeder block, with dis- tilled water added during the test period, if necessary to maintain good ♦ An acrylonitrile product of the Monsanto Chemical Company. 138 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 21 growth of the test fungus. Breazzano thoroughly saturated the sand base in his test cultures. Leutritz and Harrow^^' ^^ working with tightly closed culture jars found a 25 per cent level in the soil to be satisfactory. Flerov and Popov state after special control tests "that replacement of (the) sand by soil had no effect on the results of the tests and only shortened their duration." Duncan'* has found from her tests that vari- ations in moisture content and soil type affect the degree of fungus attack only and that they do not change the determination of the treat- ment threshold concentration in any given set of test blocks. Even-Aged Cultures The thickness of the feeder blocks has been gradually increased to about Jf 6 inch, or between 4 and 4.5 millimeters. This provides food for the fungus to estabhsh itseff in the bottle. The inoculum pieces are roughly 1 cm square, cut from Petri dish cultures that are 15 zb 1 days old. The planting routine is carefully scheduled so that even-aged soil- block cultures — 13-15 days — are ready to receive the treated blocks when the latter are ready to be placed in test. This principle of using even-aged cultures has been stressed by the Madison investigators, and it is considered to be a factor of major importance in the proper culture technique. Standard Test Organisms There have been continuous discussions since the beginning of labora- tory tests in Europe, as well as in this country, about what test organ- isms should be used. Conforming to the experience and practice at Madison the following three numbered strains of wood-destroying fungi are recognized as the ''standard" strains for the testing of oil type pre- servatives in coniferous wood : Lentinus lepideuSy Madison 534 Lenzites trabea, Madison 617 Poria monticola, Madison 698 All three are known to be associated with the decay of treated timber. Lentinus lepideus is particularly tolerant of creosote,^^' ^^ and relatively susceptible to pentachlorophenol. It has frequently been isolated from decaying creosoted southern pine poles and other creosoted coniferous timber in contact with the ground. Lenzites trabea is generally an ''above ground" fungus. It also has been isolated from decaying creosoted tim- ber; and it is the principal cause of "shell rot" in the above ground sap- EVALUATION OF WOOD PRESERVATIVES 139 wood of western red cedar poles. It is relatively susceptible to creosote and quite tolerant of pentachlorophenol in the block tests. Porta monti- cola is relatively tolerant of pentachlorophenol and of copper compounds under laboratory test conditions, and relatively susceptible to creosote. It is of special interest also because it may be identical with some of the fungi tested in Europe under the name of Porta vapor aria, and thus its use may facilitate comparisons of a sort with results obtained by other investigators. For instance, information has reached the Division of Forest Pathology at Madison, from Findlay at the Princes Risborough laboratory in England, that Harrow's Porta vaporaria^^ is the same as Liese's,^^ and that it has been identified as a strain of Porta moniicola by Miss M. Nobles of Canada. Within the last few years another fungus, characterized by the forma- tion of conspicuous saffron 3^ellow strands, has been found associated with decayed specimens of creosoted pine poles. ^ The writer has seen the tell-tale strands in old cull dumps only. It has been identified as Porta radiculosa. Whether it is truly a primary attacker or a secondary organism is not yet clear. Soil-block tests are under way at Madison to determine its significance as a possible species to supplement Lentinus lepideus in the evaluation of creosote. In connection with the use of the three numbered ''standard" strains listed above, there may always be some reasonable doubt as to whether the cultures employed in different laboratories have the same virulence. To answer this question precisely involves a lot of careful biological check testing, and such tests are already being made in the Division of Forest Pathology at the Plant Industry Station, Beltsville, Md. It is assumed for the time being that the numbered strains are virulent and satisfactory test organisms for such preservatives as creosotes and pen- tachlorophenol-petroleum solutions. The Scope of the Soil-Block Evaluation Test For a complete understanding of the scope of the soil-block evaluation test it is necessary to consider this test as having two functions. The first function involves the use of the soil-block test per se (without weathering) to measure the reaction of the test organisms to various quantities of a given preservative, and to compare these reactions against different preservatives. In this function the test has been used in lieu of the agar Petri dish test,'' and is considered to be much more satisfactory as a screening test by workers at the Laboratories. It has been employed at Madison for testing the natural durabiUty of wood, plywood, fiber board, etc. 140 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 The second — and more important — function of the soil-block eval- uation is that incorporating a weathering or aging procedure. This puts the test in the more practical category of testing the wood preservative properties, viz., toxicity and permanence. In this respect it has something in common with the German Standard DIN DVM 2176^^ for short time mycological testing of wood preservatives by the block method and covers a broad concept from the treating through partial aging of the blocks. Separate German standards cover procedures for leaching^^ and volatihty tests,^^ (p. 264 and Fig. 42 of Reference 78). The Laboratories' concept of the scope ol the soil-block test including the weathering procedure is definite. It must be appreciated that this method which employs manipulative procedures involving both toxicity and permanency yields significant data in a period of only a few months. The data derived must be correlated subsequently, of course, with the data covering the results of tests of % inch stakes six to seven years later, with the data on test posts some ten to fifteen years later, and with data on poles in line some twenty-five years later. It is only being realistic to say that the Bell System cannot afford to wait for physical life tests of new materials under natural conditions of exposure before recommending them where techniques and extensive experience permit acceptable estimates to be made from accelerated evaluation in relatively short periods of time. Preparation of the Test Blocks — Manufacture Southern pine sapwood, free from stain or decay, is used as a base material for the test blocks. The process of manufacture begins at the saw mill, where freshly cut logs selected for the purpose are carefully sawed into one inch boards. Straight grain material is most desirable. The boards are kiln-dried immediately and shipped as soon as practicable to the Laboratories. It has been the practice to store the boards in a steam heated basement where the humidity is low enough to hold the moisture content of the boards down to about 5 to 7 per cent. The sap- wood only is used, which means that any small heart wood portions must be marked out for rejection. The blocks are accurately cut %-inch cubes. A J^-inch hole is drilled through the center of the tangential surfaces of each block. It has been found that drilling the hole through the trans- verse surface, which was the early practice with Waterman, Leutritz and Hill^^* is a difficult procedure; and sometimes it amounts to an impossibihty because the harder summerwood layers deflect and break the drills. In any event, drilling through the tangential surface opens up more paths for longitudinal absorption and penetration, as well as EVALUATION OF WOOD PRESERVATIVES 141 evaporation, of the preservative. The feeder blocks and the %-inch test blocks are usually made at the same time, from parts of the same boards. The blocks are kept clean, and reserve stocks are carefully stored in a dry room. The blocks in storage reach an approximate moisture equilib- rium of 6 to 7 per cent, on an oven-dry weight basis. Test Block Selection for Density Random samples of the blocks are weighed and segregated into groups at 0.1 -gram intervals, 4.10 to 4.19 grams and 4.20 to 4.29 grams, for example. Blocks of practically equal weight can be chosen for the com- parison within any given series of different concentrations of a preserva- tive. The weighed groups of blocks are kept in convenient lot sizes in a dry place. Since the blocks are accurately cut the segregation by weight amounts to a segregation by density. It has not been found necessary or practicable to separate the blocks into groups mth the same numbers of annual rings, although in some instances an approximation to this ideal has been attempted. Further- more, it has not been found practicable to separate the blocks on the basis of the direction in which the rings run across their transverse faces. From experience to date it does not appear that either ring direction or ring count has any material effect upon the behavior of the blocks in the culture as far as determination of preservative thresholds are con- cerned; but experiments are under way at Madison to determine the effect of density on the relative degree of decay. Inasmuch as all of the blocks are placed in culture with the transverse surface down, so that alternate spring- and summerwood layers are exposed directly to the test organism, the latter can enter either springwood or summerwood in accordance with its ability to resist the concentration of the preserva- tive present in these two parts of the annual ring. Average Block Volume The average volume of the oven-dried blocks, determined from ran- dom samples by a mercury displacement technique, was found to be 6.484 cc, with a standard deviation of 0.0831. This represents a coeffi- cient of variability of 1.28 per cent. The minimum-maximum range of volumes ran from 5.93 cc to 6.87 cc. These extreme deviations are nor- mally detected in handling the blocks and both high and low volume blocks are rejected. The variation in density and volume of the test blocks will be discussed separately in the paragraphs dealing with the treatment of the blocks. 142 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Treatment of the Test Blocks The blocks selected for any given treatment are numbered serially with India ink on the upper half of one of the radial faces. All blocks are then oven-dried for 24 hours at 105°G to an approximate constant weight. The blocks are removed from the oven, and placed in a desiccator over P2O5. Check tests of blocks held under these conditions show that they do not change weight by more than one hundredth gram within the period they are held for weighing. The cooled, oven-dried blocks are weighed to the nearest hundredth gram. The weighed blocks are placed in beakers and arranged with a tan- gential face down so that the transverse surfaces do not touch and the holes are vertical. This refinement in placing the blocks may not be nec- essary to obtain satisfactory absorption, but the procedure has worked out well, and it has been followed consistently. For any given concen- tration a sufficient amount of creosote, for example, and toluene are combined by weight to leave in the blocks, after treatment, the desired retention of preservative. Experience has indicated the concentration required, which depends to a certain extent upon the type of vacuum equipment that is available as well as upon the density of the blocks to be treated and the nature of the treating solution. Actually the process of treatment is simple. The beaker containing a given lot of weighed oven-dried blocks is placed within a bell jar and subjected to a vacuum of 3 to 4 millimeters of mercury. When this vacuum has been reached the line to the vacuum pump is shut off, and the preservative is run into the beaker from a separatory funnel^^^ fitted into a rubber stopper on the top of the vacuum chamber, the blocks being weighted down below the level of the preservative. The absorption and distribution of the oil within the blocks seems to take place very rapidly. Generally speaking, the beaker containing the blocks and the preservative are removed from the vacuum chamber as soon as practicable to permit continuing the treatment of another group of blocks. However, the blocks are usually held in the preservative solu- tion for an hour or two, which apparently is long enough to bring about essentially complete saturation. When all the treatments for a given group of concentrations have been finished (a) the treated blocks are wiped to remove the excess oil, and (b) they are weighed immediately to 0.01 gram. The retentions of creosote or pentachlorophenol, for ex- ample, are determined on a gain in weight basis by calculations from the amount of material picked up during the treatment and the concen- tration of the preservative in the treating solution. f EVALUATION OF WOOD PRESERVATIVES 143 Retention Gradients In the hope of setting at rest some of the doubts and criticisms that have arisen about the accuracy of the treatments and the retention of preservative in the blocks, some results of the treatment process just described will be presented in rather elaborate detail. The success of the treatments depends upon experience, as indicated previously, with the particular type of vacuum equipment available. However, once the level of performance to be expected from the vacuum equipment is learned, one has to take into account the variations that are introduced by the density of the blocks and by the specific gravity of the treating solution. It is the intent in all of the treatments at the Laboratories to arrive at a series of gradient retentions, on as accurate a line as possible, and as nearly as possible equal gradients, so that the fairest comparison can be made of the behavior of the different preser- vatives. Fig. 4 shows the gradient obtained by plotting the data shown in Table I for retention of creosote and retention of pentachlorophenol solution over the concentration of these preservatives in the treating solution. The analysis of the creosote — BTL 5340 — is shown in Table II. The slopes of the two gradients are considered to be about as close as the experimental procedure will permit. Fig. 5 shows the gradient 0 ? 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 PRESERVATIVE CONCENTRATION IN PER CENT IN TREATING SOLUTION Fig. 4— Gradient retentions for comparative soil-block tests of a creosote (BTL No. 5340) and a penta-petroleum solution (4.92 per cent pentachlorophenol in Standard Oil Company of New Jersey No. 2105 Process Oil). The preservatives were used in toluene solution. 144 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Table I — Full-Cell Treatment Soil block tests with creosote (No. 5340, see Table II) and with pentachloro- phenol-petroleum (4.92 per cent in Standard of New Jersey No. 2105 Process oil) in toluene; absorption and retention of preservative data for parallel com- parative tests. Average Oven-dry Average Absorption* Average retention Whole Charge No. n ct preservative Creosote Penta Weight Density Total Perec Creo- sote Penta Sol. Total' P// Total Per cc (gms) (gn tis) (percent) (Ib/cu ft) (gms) (gms) 9 3.77 .584 30 3.28 .509 8.18 2.49, — — .28 .043 — — 10 3.78 .586 30 3.40 .527 11.45 3.48 — — .39 .061 — — 6 3.80 .589 30 3.44 .533 16.11 4.96 — — .55 .085 — — 5 3.83 .594 30 3.46 .536 19.33 5.99 — — .67 .104 — 8 3.80 .589 30 3.51 .544 22.55 7.08, — — .79 .123 — — 1 3.74 .580 30 3.49 .541 25.00 7.81 — — .87 .135 — — 4 3.77 .584 30 3.51 .544 25.55 8.03 — — .90 .140 — — 3 3.80 .589 30 3.49 .541 27.50 8.58 — — .96 .149 — — 2 3.78 .580 30 3.50 .543 30.00 9.40 — 1.05 .163 — — 11 3.70 .574 30 3.33 .516 3.25 1.05 .052 .005 .0008 12 3.79 .588 30 3.33 .516 6.25 2.01 .099 — — .010,-0016 13 3.79 .588 30 3.29 .510 9.25 2.95 .145 — — .015'. 0028 14 3.75 .581 30 3.31 .513 12.25 3.96.193 — — .020 .0031 15 3.71 .575 30 3.34 .518 15.00 4.84.238 — — .025 .0039 16 3.70 .574 30 3.34 .518 18.00 5.811.286 — — .030 .0047 17 3.72 .577 30 3.38 .524 23.75 6.79|.334 — — .035 .0054 18 3.73 .578 30 3.38 .524 23.75 7.77 .382 — — .040 .0062 * Absorption is the total amount of the treating solution picked up at treat- ment, that is, the gain in weight, including both preservative and the toluene carrier. t C is the concentration of the preservative, e.g., creosote or penta petroleum, in the treating solution, in grams per 100 ml. for pentachlorophenol alone, without regard to the petroleum carrier, also plotted from data in Table I. The scale on the abscissa represents the concentration of either the creosote or the pentachlorophenol solu- tion. The ordinate represents pounds per cubic foot retained by the blocks, calculated from the pickup during treatment and from the con- centration of the creosote or pentachlorophenol in the treating solution. The Amount of Preservative in the Blocks The use of these gradient concentrations is a continuation of the procedure worked out in the earlier stages^^' *^ of the Madison tests. EVALUATION OF WOOD PRESERVATIVES 145 Table II — Analyses of Creosote BTL No. 5340, Water-Free Basis Specific gravity 38/15.5°C Distillation, per cent, cumulative to 210°C 210-235 235-270 270-300 300-315 315-355 Residue above 355°C Total Sulph. res., gm/100 ml, Tar acids, gm/100 ml . , Benzol insol., per cent. Specific gravity (38°C) 235-315°C.. . . 315-355°C 1. Fall, 1946 2. Spring, 1952* 1.088 1.102 0.00 0.00 0.80 0.00 12.87 13.59 42.12 54.30 52.03 79.10 78.05 20.90 21.64 100.00 99.69 0.51 0.59 4.10 4.44 0.07 0.59 1.053 1.118 * Average of 2 analyses. It should be noted that the retentions are calculated as averages for the respective charges. Attention is called to the small quantity of preserva- tive material involved. Even in calculated retentions of creosote, for example, 9.40 pounds per cubic foot (Table I), the retention means 1.05 grams in the whole block, or 0.163 grams in each cc of block volume. In 0.40 o u- 0.36 0.32 0.28 0.24 0.12 0.08 2 ^ 0.04 / / / ir > Y / / y X ^ :f / y / ^ / o pentachlorophenol; dry salt / ^ — ,. 0 2 4 6 8 10 12 14 16 18 20 22 24 26 PENTA PETROLEUM CONCENTRATION IN PER CENT IN TREATING SOLUTION Fig. 5— Gradient retention of pentachlorophenol, calculated for the material alone, without the oil carrier. See Fig. 4. 146 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 the highest retention employed for the penta petroleum solution the cal- culated net retention averaged 7.77 pounds of the penta solution per cubic foot, or 0.382 pounds of pentachlorophenol per cubic foot; and these figures represent, respectively, 40 miUigrams of pentachlorophenol in the average block, or 6.2 milligrams per ec of block volume. Exact data on treatment are discussed in the following paragraphs. The use of carefully calculated gradient retentions in each case makes it possible to detect any wide variation in the normal behavior of the blocks either with respect to pickup during treatment or in the reaction of the test fungus to the preservative. Data are included in Table I on average oven dry weight of the blocks. Table III — Full-Cell Treatments Soil block tests; treating solution components, per cent by weight. (See Table I). Charge No. Penta-petroleum Creosote Toluene 19 4.65 4.65 90.70 20 10.50 10.50 79.00 21 15.00 15.00 70.00 71 23.00 77.00 72 — 23.00 77.00 73 — 24.75 75.25 74 — 25.50 74.50 75 — 30.00 70.00 76 — 32.00 68.00 average density on an oven dry weight and volume basis, average pickup of creosote or penta solution in pounds per cubic foot and in grams per block, the concentration of the preservative materials in the toluene preservative solution, and the average grams of preservative per cc of block volume. All of the blocks in these two groups of charges were chosen within a narrow density range. Block Density and Preservative Absorption It will be noted that in the charges in Table I there is a general trend upward in the grams absorbed at treatment per cc of block volume, as the specific gravity of the treating solution increases. This is, of course, one of the results of increasing the concentration of creosote, for example; and furthermore, as would be expected, the higher gravity solutions represented by the creosote treatments show a higher pickup in terms of total grams as well as in grams per cc. The make-up of the EVALUATION OF WOOD PRESERVATIVES 147 Table IV — -Full-Cell Treatment with Penta-Petroleum- Creosote in Toluene Relation of variable block density to absorption of treating solution, grams per cc of block volume; density and volume on oven-dry basis. (See Tables V and VI). Charge 19 Av. retention 2.99 Ib/cu Charge 20 Av. retention 6.69 Charge 21 Av. retention 9.59 ft Ib/cu ft Ib/cu ft Density oven-dry Absorption gms/cc of block vol. Density oven-dry Absorption gms/cc of block vol. Density oven-dry Absorption gms/cc of block vol. .440 .599 .468 .589 .484 .581 .459 .583 .483 .584 .489 .576 .459 .586 .533 .549 .505 .564 .475 .575 .543 .542 .520 .550 .507 .561 .544 .527 .529 .555 .517 .545 .557 .537 .541 .541 .533 .544 .564 .529 .549 .538 .535 .525 .567 .530 .554 .532 .543 .530 .569 .569 .555 .526 .543 .537 .604 .510 .573 .538 .571 .508 .606 .511 .582 .528 .574 .524 .611 .504 .582 .530 .608 .488 .616 .499 .592 .517 .609 .495 .618 .491 .602 .512 .611 .500 .618 .497 .610 .497 . .611 .533 .624 .503 .612 .507 .623 .483 .625 .493 .614 .502 .624 .493 .627 .492 .614 .508 .637 .476 .628 .488 .615 .540 .638 .477 .644 .481 .617 .495 .641 .474 .645 .488 .617 .504 .645 .478 .646 .482 .627 .495 .646 .473 .651 .482 .633 .497 .657 .462 .662 .433 .647 .488 .660 .453 .674 .461 .650 .487 .663 .466 .676 .465 .664 .416 .663 .468 .674 .452 .672 .473 .668 .462 .690 .456 .684 .470 .691 .447 .703 .445 .687 .464 .695 .450 .738 .427 .723 .442 592 .0726 ,507 Average 614 .501 Standard deviation .0435 .0616 .0405 .598 .0591 Coefficient of variability— per cent ,512 .0370 12.26 8.58 10.03 8.08 9.88 7.23 148 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 treating solutions for Charges 19-21 and 71-76, inclusive, are shown in Table III. The relation of the density of the blocks to the pickup, i.e., the absorption at time of treatment, is illustrated in Tables IV, V and VI and in Fig. 6. The data have been split up to facilitate reference. Table IV shows the complete data for oven dry density and for ab- sorption in grams per cc of block volume for Charges 19, 20 and 21, with values for the average, for the standard deviation, and for the coefficient of variabiUty. The pickup varies inversely as the density, which is to be expected when random blocks instead of selected density blocks are employed. The coefficient of variability in the density figures is evidently greater than it is in the pickup figures; and a lower figure for the latter is related to a lower figure for the former. 0.62 0.61 0.60 til I 0.59 -I O > 0.58 :«: o 3 0.57 CD fe 0.56 0.55 0.54 0.53 0.52 0.51 0.50 049 0A& 0.47 0A6 0.45 0.44 ^ > ^ \ \ L \ ^. ^ k ^ =:^ V ^^ ^ ^ N \ ^ N ^ \ \ \ '^ X v^ % \ N ^ '^v. N N \ V \ s\ <\ ^\ >^ ^ ^ \ \ N ^ ^ \ \ ^^ \ V \ THE NUMBERS REFER TO THE CHARGES SHOWN IN TABLES Y AND 21 ^ ^ k. ^e ^^ \^ S. \ 0 ^> \ 74 7l\ 0.720 0/400 0.440 0.460 0.520 0,560 0.600 0.640 0.68 DENSITY (OVEN-DRY BASIS) Fig. 6 — Regression lines for absorption at treatment, in gms/cc of oven-dry block volume, on oven-dry density of the %-inch cube test blocks. EVALUATION OF WOOD PRESERVATIVES 149 These same values for these three charges, 19, 20 and 21, and similar values for charges 71-76, inclusive, are incorporated along with average and range of retention data in Table V, and with statistical data for regression lines in Table VI. The data serve to illustrate the degree of variability in treatment results that may occur when random blocks are used. The best indices of these variations are in the columns showing Table V — Retention Data for Full-Cell Treatment Soil block tests; average and range of preservative retention, by charges, at treatment. n Penta pe- troleum Ib/cu ft Pentachlorophenol Creosote Charge No Ib/cu ft gms Ib/cu ft gms Av. Min. Max. Av. Min. Max. Av. Min. Max. Av. Min. Max. 19 20 21 71 72 73 74 75 76 30 30 30 30 30 30 30 30 30 1.51 3.35 4.79 .093 .163 .237 .064 .136 .192 .097 .219 .269 .008 .017 .024 .007 .014 .020 .010 .023 .028 1.48 3.34 4.80 8.58 8.52 9.06 9.46 11.12 11.37 1.29 2.77 3.90 7.65 7.67 8.43 8.61 10.07 11.05 1.97 4.44 5.46 9.04 9.20 9.57 10.27 12.06 11.94 0.154 0.348 0.500 0.892 0.886 0.943 0.984 1.158 1.184 0.134 0.288 0.407 0.817 0.798 0.876 0.895 1.047 1.149 0.205 0.462 0.569 0.941 0.957 1.025 1.068 1.254 1.242 Table VI — Full-Cell Treatment Soil-block tests with (a) penta-petroleum creosote, and with (b) creosote, in toluene; relation of variable block density to absorption of treating solution. Av. retention] Ib/cu ft 4, Density oven-dry Absorption gms/cc vol. n Correl. coeff. r Charge No. Penta- absorption Y on density X Creo- sote petro Creo- sote Av. «r* c.v.t Av. (T* c.v.t 19 30 _ 2.99 .592 .0726 12.26 .507 .0435 8.58 -.3444 .628&-.2066X 20 30 . 6.69 614 .0616 10.03 ..501 .0405 8.08 -.4282 .6736-.2820X 21 30 — 9.59 .598 .0591 9.88 .512 .0370 7.23 -.4718 .6893-.2957X 71 30 8,58 ,477 ,0428 8.97 .597 .0293 4.91 -.9589 .9109-.6580X 72 30 8.52 .486 .0457 9.40 .594 .0326 5.49 -.9233 .9143-.6589X 73 30 9,06 499 0456 9.14 .586 .0275 4.69 -.9426 .8701-.5692X 74 30 9 46 489 0402 8 22 .595 .0279 4.69 -.8916 .8969-.6181X 75 30 11 12 510 0538 10.55 .596 .0302 5.07 -.9206 .8595-.5170X 76 30 11.37 — .557 .0095 1.71 .570 .0114 2.00 -.4583 .8781-.5543X * 355°C Ib/cu ft <355°C Ib/cu ft 1 1 1.065 18.5 9.8 53.1 5.2 4.6 39.4 1.81 2.79 2 7 1.077 20.5 9.0 47.8 4.3 4.7 39.3 1.85 2.85 3 2 1.081 30.6 10.2 47.1 4.8 5.4 57.8 3.12 2.28 4 6 1.093 34.2 9.0 37.8 3.4 5.6 55.0 3.08 2.52 5 3 1.108 50.4 12.2 30.3 3.7 8.5 72.3 6.15 2.35 6 8 1.115 53.2 9.4 25.5 2.4 7.0 71.4 5.00 2.00 7 9a 21.2 5.7 40.4 2.3 3.4 35.6 1.21 2.19 8 9 1.001 20.0 5.8 43.1 2.5 3.3 35.1 1.16 2.14 9 10a 14.4 6.7 50.9 3.4 3.3 29.4 0.97 2.23 10 10 1.068 15.2 6.9 52.2 3.6 3.3 31.8 1.05 2.25 11 11 1.038 18.0 6.5 47.7 3.1 3.4 34.4 1.17 2.23 12 Ml 1.107 41.9 8.0 33.8 2.7 5.3 63.3 3.35 1.95 13 M2 1.070 18.1 8.3 50.6 4.2 4.1 36.6 1.51 2.59 14 BTL 5340 1.088 20.9 7.5 46.6 3.5 4.0 39.1 1.57 2.43 * Creosotes 1, 2, 3, 6, 7, 8, 9, 10 and 11 are those in use in the Cooperative Creosote Tests (see Bibliography, References 12 and 39. Oils 9a and 10a are samples from the same lots as numbers 9 and 10. (See Bibliograph>', Reference 36.) For oils Ml and M2 see Bibliographj^, References 37 and 38. Creosote 5340 is shown in Table II. 160 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 operational losses and through the points representing weight losses. Data from a repetition of these tests is desirable in order to establish the thresholds more definitely from actual weight loss or observational data taken close to the assumed threshold points. At Bell Telephone Laboratories a check series of tests is now under way on cooperative creosotes 6, 7 and 8, domestic oils, and creosotes 9, 10 and 11, British oils; and comparison tests are also being run on creosote BTL-5340 and on 5 per cent pentachlorophenol in the 2105 process oil. The aim has been to treat the blocks to a series of retentions that vary narrowly around the thresholds set by the Madison investigators. Across the Threshold Fig. 13 is an illustration of representative blocks from the creosote series, line A (creosote BTL-5340) in Fig. 10, just at and below the threshold. Fig. 14 shows the character of the attack by Lenzites trahea on blocks treated with a 4.92 per cent solution of pentachlorophenol in Standard Oil Company of New Jersey's 2105 Process Oil in toluene. The blocks are represented at twice their original linear dimensions. The exact nature of the decay is difficult to show. The experimenter has to learn a system of diagnosis that involves both visual observation and the "feel" of the blocks for distortion and firmness that supplement weight loss data. For example, in Fig. 14, a threshold between 0.20 and 0.25 pound of penta per cubic foot (Blocks C and D) is indicated and this conforms closely to the results with the same penta-petroleum solution at Madison .^^ The Significance of the Results of Laboratory Soil-Block Tests on Oil-Type Preservatives The main conclusions from this discussion of the results of soil-block tests on weathered creosoted wood conducted at Madison are (a) that in general, under the test conditions, at least 8 and sometimes 9 pounds or more of creosote per cubic foot is a necessary treatment to prevent attack by Lentinus lepideus on %-inch cube blocks of southern pine sapwood; and (b) that a penta petroleum solution is much more effective than creosote against this same organism. As will be emphasized later, this general conclusion about Lentinus lepideus and creosote corresponds with the conclusions to be drawn from the interpretation of results of the small stake tests and from the test of pole-diameter posts in the Gulfport test plot. The creosote tested is a better preservative against Lenzites trahea than the penta-petroleum, but the creosote threshold for this organism EVALUATION OF WOOD PRESERVATIVES 101 is below what one would have to use commercially in order to insure protection against Lentinus lepideus and to preserve wood exposed to ground contact. As far as Lentinus lepideus is concerned, the best overall preservative combination from the laboratory test would appear to be a 50/50 by volume blend of the creosote and the penta-petroleum solution, since such a blend appears to contain the best in both components. However, there are certain practical considerations, principally relating to the incompatibility of some creosotes and petroleums, which make the com- bined preservative a difficult one to operate with commercially. (a) iCj (d) ng. 13-Across the threshold-creosote. Test fungus LenfmusZeptJ^eus,^^^^^^^^^^ sote concentration at treatment: (a) 11.70; (b) 6.92; (c) 5.45; and (d) 4.15 Ib/cu ft. Fig. 162 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 (a) (b) Fig. 14— Across the threshold— pentachlorophenol solution. Test fungus Len- rnH%HTfei7/^^?. p^rJI'^'J^T,^,^ treatment: (a) 0.052, (b) 0.099, (c) 0.193; and (d) 0.238 Ib/cu ft. Photo by A.M. Hearn. EVALUATION OF WOOD PRESERVATIVES 163 Relatively speaking the soil-block test procedure is much more rapid than the test plot experiments that are to be discussed next, but since the inferences with respect to retention requirements for creosote appear to be the same for the laboratory and the field tests, the former have a direct and immediate application in practical pole preservation. BIBLIOGRAPHY 1. Alleman, G., Quantity and Character of Creosote in Well -Preserved Timbers. Circ. 98, Forest Service, U. S. Dept. Agric, May 9, 1907. 2. AUiot, H., Methode d'Essais des Produits Anticryptogamiques. Inst. Nat. du Bois Bull. Techn. 1, 1945. 3. Amadon, C. H., Recent Observations on the Relation between Penetration, Infection and Decay in Creosoted Southern Pine Poles in Line. Am. Wood Preservers' Assoc, Proc, 35, pp. 187-197, 1939. 4. Baechler, R. H., Relations Between the Chemical Constitution and Toxicity of Aliphatic Compounds. Am. Wood Preservers' Assoc, Proc, 43, pp. 94r- 111, 1947. 5. Baechler, R. H., The Toxicity of Preservative Oils Before and After Artificial Aging. Am. Wood Preservers' Assoc, Proc, 45, pp. 90-95, 1949. 6. Bateman, E., Quantity and Quality of Creosote Found in Two Treated Piles After Long Service. Circ 199, Forest Service, Forest Products Laboratory Series, U. S. Dept. Agric, May 22, 1912. 7. Bateman, E., Coal -Tar and Water-Gas Tar Creosotes: Their Properties and Methods of Testing. U. S. Dept. Agric, Bui. No. 1036, pp. 57-65, Oct. 20, 1922. 8. Bateman, E., A Theory on the Mechanism of the Protection of Wood by Preservatives. Am. Wood Preservers' Assoc, Proc, 1920, p. 251; 1921, p. 506; 1922, p. 70; 1923, p. 136; 1924, p. 33; 1925, p. 22; 1927, p. 41. 9. Bavendam, W., Die pilzwidrige Wirkung der im Holzschutz benutzten Chemi- kalen. Mitteilungen d. Reichsinst. f. Forst und Holzwirtschaft. No. 7, 20 pp. Aug., 1948. 10. Baxter, D. V., Pathology in Forest Practice. 2nd Ed. XI + 601 pp. John Wiley and Sons, New York, 1952. 11. Bertleff, V., Prufung der Fungiziden Eigenschaften Ostrauer Steinkohlenteer Impragnierole und ihrer Bestandteile. Chem. Zeitung, 63, p. 438, 1939. 12. Bescher, R. H., and others. Cooperative Creosote Tests. Am. Wood Preservers' Assoc. ,Proc., 46, pp. 68-79, 1950. 13. Bienfait, J. L., and T. Hof, Buitenproeven met geconserveerde palen. I ste mededeling. Circ. 8 Ser. Ill, Conservering en Veredeling No. 3, Central Instituut voor Materiaal Onderzoek. Delft, Holland, Dec, 1948. 14. Bland, D. E., A Study of Toxicity of Australian Vertical Retort Creosote Oils to Lentinus Lepideus Fr., Polystictus Versicolor (L) Fr., and Madison 517. Australian Council Sci. and Ind. Jl. 15, pp. 135-146, 1942. 15. Blew, J. O., Comparison of Wood Preservatives in Stake Tests. Am. Wood Preservers' Assoc, Proc, 44, pp. 88-119, 1948. 16. Blew, J. O., C. A. Richards, and R. H. Baechler, Evaluating Wood Preserva- tives. For. Prod. Res. Society, Proc, 5, pp. 230-238, (Wood-block tests, pp. 235-236), 1951. t. o j 17. Blew, J. O., Comparison of Wood Preservatives in Mississippi Post Study (1952 Progress Report). Report No. R1757, Forest Products Laboratory, U. S. Forest Service, Madison, Wis. ,,«^r» t^ 18. Blew, J. O., Comparison of Wood Preservatives in Stake Tests (1952 Progress Report). Report No. D1761, Forest Products Laboratory, U. S. Forest Service, Madison, Wis. ^ 19. Boyce, J. S., Forest Pathology. 550 p. McGraw-Hill Book Company, Inc., New York, 1948. 164 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 20. Breazzano, A., Metodo Biologico di Controllo dei Sistemi di Preservazione, dei Legnami Adottato dalP Institute Sperimentale delle FF. SS. (English translation: Biological Method of Testing Systems of Wood Preservation Adopted by the State Railway Experiment Institute.) Tech. Rev. Italian Railways, 4, No. 5, pp. 3-8, Nov., 1913. Forest Products Laboratory, Madi- son, Wis. 21. Breazzano, A., Metodi Normali di Prova sulla Putrescibilitd, dei Legnani. (English translation: Standard Methods of Testing the Putrescibility of Wood.) Extract from Report of the Ninth Meeting of the Italian Associ- ation for the Study of Building Materials; Turin; April, 1922. Forest Prod- ucts Laboratory, Madison, Wis. 22. Breazzano, A., Osservazioni sul Metodo dei Blocchetti di Legno in Usa nell Analisi Tossimetrica delle Sostanze Conservatrici del Legno. Revista Technica Delle Ferrovie Italiane. 47, No. 6, June 15, 1935. (Observations on the Wood Block Method in the Toximetric Analysis of Wood Preserva- tives. Excerpt from Technical Review of Italian Railways. English trans- lation; Forest Products Laboratory.) 23. Breazzano, A., Methods for Determining the Fungicidal Power of Wood Preservatives. International Assn. for Testing Materials, Proc, London Congress, Group C — Organic Materials, Sub. group 3, pp. 484-486, April, 1937. 24. British Standard Method of Test for the Toxicity of Wood Preservatives to Fungi. British Standards Institution. British Standard No. 838, 17 pp., April, 1939. 25. Broekhuizen, S., Onderzoekingen over de Conserverende Waarde van een Aantal Houtconserveermiddelen. Rapp. Comm. Gebruikw. inh. hout. Deel II, Bijlage II, pp. 89-122, The Hague, 1937. 26. Cartwright, K. St. G., and W. P. K. Findlay, Decay of Timber and its Preven- tion. VI plus 294 p. London: His Majesty's Stationery Office, 1946. Re- printed 1948. 27. Colley, R. H., The Effect of Incipient Decay on the Mechanical Properties of Airplane Timber. (Abstract) Phytopathology, 11, p. 45, 1921. 28. Colley, R. H., T. R. C. Wilson, andR. F. Luxford, The Effect of Polyporus Schweinitzii and Trametes Pint on the Shock-resistance, Compression Paral- lel to Grain Strength, and Specific Gravity of Sitka Spruce. Forest Products Laboratory, Project L-243-J1, Typewritten Report, 29 p., 32 figs., 4 plates. July 3, 1925. 29. Colley, R. H., and C. H. Amadon, Relation between Penetration and Decay in Creosoted Southern Pine Poles. Bell Sys. Tech. JL, 15, pp. 363-379, July, 1936. 30. Colley, R. H., Some Observations on the Selection and Use of Modern Wood Preservatives. Reports, Twenty-fourth Session, Communications Sec- tion, Association of American Railroads, October, 1947, pp. 17-25. 31. Colley, R. H., Wood preservation and Timber Economy. Forest Products Institute of Canada. Papers presented at the First Annual Convention, Ottawa, Oct. 30-31, 1950. 32. Curtin, L. P., B. L. Kline, and W. Thordarson, V— Weathering Tests on Treated Wood. Ind. and Eng. Chem. 10, No. 12, pp. 1340-1343, Dec, 1927. 33. DIN (Deutsche Normen) DVM 2176, Blatt 1. Priifung von Holzschutz- mitteln. Mykologische Kurzpriifung (Klotzchen Verfahren). Berlin, Aug. 1939. (New Edition DIN 52176). 34. DIN DVM 52176, Blatt 2. Prufung von Holzschutzmitteln. Bestimmung der Auslaugbarkeit. Berlin, Mav 1941. (Reprinted Oct., 1948). 35. Duncan, C. G., and C. A. Richards, Methods of Evaluating Wood-Preserva- tives: Weathered Impregnated Wood Blocks. Am. Wood Preservers' Assoc, Proc, 44, pp. 259-264, 1948. 36. Duncan, C. G., A comparison of Two English Creosotes Produced from Coke- oven Coal Tar and Vertical -retort Coal Tar. Mss. Office Report, Division of P'orest Pathology, Bur. PI. Ind., Forest Prod. Lab., Madison, Wis., Jan. 21, 1949. 37. Duncan, C. G., and C. A. Richards, Evaluating Wood Preservatives by Soil- block Tests: 1. Effect of Carrier on Pentachlorophenol Solutions; 2. Com- EVALUATION OF WOOD PRESERVATIVES 165 parison of a Coal Tar Creosote, a Petroleum Containing Pentachlorophenol or Copper Naphthenate and Mixtures of Them. Amer. Wood Preservers Assoc, Proc, 46, pp. 131-145, 1950. 38. Duncan, C. G., and C, A. Richards, Evaluating Wood Preservatives by Soil- Block Tests: 3. The Effect of Mixing a Coal Tar Creosote and a Penta- chlorophenol Solution with a Petroleum; a Creosote with a Coke Oven Tar or Pentachlorophenol Solution. Amer. Wood Preservers' Assoc, Proc , 47, pp. 264-274, 1951. 39. Duncan, C. G., and C. A. Richards, Evaluating Wood Preservatives by Soil- Block Tests: 4. Creosotes. Amer. Wood Preservers' Assoc, Proc, 47, pp. 275-287, 1951. 40. Duncan, C. G., Evaluating Wood Preservatives by Soil-Block Tests: 5. Lignite-Tar and Oil-Tar Creosotes. Amer. Wood Preservers' Assoc, Proc, 48, 1952. 41. Duncan, C. G., Soil-Block versus Agar-Block Techniques for Evaluation of Oil-Type Preservatives: Creosote, Copper Naphthenate and Pentachloro- phenol. Mss. Division of Forest Pathology, Bur. PI. Ind., Forest Products Lab; Madison, Wis., April 29, 1952. 42. Eden, Johan, och Erik Rennerfelt, Undersokninar enligt klotsmetoden av nagra fraimpregneringsmedel. (Studies on Wood Preservatives, According to the Block Method) Meddelanden fr&n Statens Skogsforskningsinstitut, Bd. 35, No. 10, 1946. 43. Eden, Johan, och Erik Rennerfelt, Fait och rotkammarforsok avsedda att utrona skyddsverkan hos olika fraimpregneringsmedel. (Field and Decay- chamber Experiments to Ascertain the Protective Effect of Various Wood Preservatives.) Meddelanden fritn Statens Skogsforskningsinstitut. Bd. 38, No. 4, 1949. 44. Falck, R., Die wichtigsten reinen Holzschutzmittel, die Methoden ihre Wertzahlen, Eingenschaften und Anwendung. Hausschwammforschungen, 8, pp. 18-20, 1927. 45. Findlay, W. P. K., A Standard Laboratory Test for Wood Preservatives. British Wood Preserving Assoc, Jl., 5, pp. 89-93, 1935. 46. Finholt, R. W., Improved Toximetric Agar-dish Test for Evaluation of Wood Preservatives. Anal. Chem., 23, No. 7, pp. 1039-1039, July, 1951. 47. Finholt, R. W., M. Weeks, and C. Hathaway, New Theory on Wood Preserva- tion. Ind. and Eng. Chem. 44, No. 1, pp. 101-105, Jan., 1952. 48. Flerov, B. C, und C. A. Popov., Methode zur Untersuchung der Wirkung von antiseptische Mitteln auf holzzerstorende Pilze. Angew. Bot. 15, pp. 386- 406, 1933. 49. Frosch, C. J., V — The Correlation of Distillation Range with Viscosity of Creosote. Physics, 6, pp. 165-170, May, 1935. 50. Gillander, H. E., C. G. King, E. O. Rhodes, and J. N. Roche, The Weathering of Creosote. Ind. and Eng. Chem., 26, No. 2, pp. 175-183, Feb., 1934. 51. Harrow, K. M., Toxicity of Water-Soluble Wood-Preservatives to Wood- Destroying Fungi. New Zealand Jl., Sec B., 31, No. 5, pp. 14-19, Mar., 1950. 52. Harrow, K. M., Note on the Soil Moisture Content Used with the Leutritz Technique for Testing Toxicity of Wood Preservatives Against Fungi. New Zealand Jl. Sci. and Tech., 4, pp. 39-40, Jan., 1951. 53. Hatfield, I., Information on Pentachlorophenol as a Wood Preserving Chemi- cal. Am. Wood Preservers' Assoc, Proc, 40, pp. 47-65, 1944. 54. Holzschutzmittel Prtifung und Forschung III. Wissenschaftliche Abhand- lungen der Deutschen Materialpruf ungsanstalten : Berlin-Dahlem. II Folge, Heft 7, 132 p. Springer-Verlag, Berlin, Gottingen, Heidelberg, 1950 1. Schulze, B., G. Theden u K. Starfinger, Ergebnisse einer vergleichenden Priifung der Pilzwidrigen Wirksamkeit von Holzschutzmitteln. pp. 1-40. 2. Becker, G., Ergebnisse einer vergleichenden Prtifung der insektotenden Wirkung von Holzschutzmitteln. II Teil. pp. 40-62. 3. Becker, G., Priifung der "Tropeneignung" von Holzschutzmitteln gegen Termiten. pp. 62-76. , tt i u x -x* i 4. Becker, G. u B. Schulze, Laboratoriumsprufung von Holzschutzmitteln gegen Meerwasserschadlinge. pp. 76-83. 166 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 55. Hopkins, C. Y., and B. B. Coldwell, Surface Coatings for Rotproofing Wood. Canadian Chemistry and Process Industries. N. R. C. No. 1256, Dec, 1944. 56. Howe, P. J., Weathering and Field Tests on Treated Wood. Amer. Wood Preservers' Assoc, Proc, 1928, p. 192. 57. Hubert, E. E., A study of Laboratory Methods Used in Testing the Relative Resistance of Wood to Decay. Univ. of Idaho Bulletin, 34, No. 15, July, 1929. 58. Hubert, E, E., An Outline of Forest Pathology. 543 pp. John Wiley and Sons, New York, 1931. 59. Hudson, M, S., and R. H. Baechler, The Oxidation of Creosote— Its Signifi- cance in Timber Treating Operations. Amer, Wood Preservers' Assoc, Proc, 36, pp. 74-112, 1940. 60. Hudson, M. S., Poria radiculosa, a Creosote Tolerant Organism. For. Prod. Res. Soc, Jl., 2, No. 2, pp. 73-74, June, 1952. 61. Humphrey, C. J., and R. M. Flemming, The Toxicity to Fungi of Various Oils and Salts, Particularly Those Used in Wood Preservation. Bui. No. 227, U. S. Dept. Agric, Aug. 23, 1915. 62. Hunt, G. M., and G. A. Garratt, Wood Preservation. VIII, 457 p. McGraw- Hill, New York, 1938. (In course of revision.) 63. Hunt, G. M., and T. E. Snyder, An International Termite Exposure Test — Twentv-First Progress Report. Amer. Wood Preservers' Assoc, Proc, 48, 1952. 64. Jacquiot, C, Controle de I'Efficacit^ des Fongicides Utilises pour I'lmpregna- tion des Bois. Etude Critique de la Technique Standard Anglaise et de la Norme Allemande DIN DVM 2176. Principes pour I'Etablissement d'une Norme Frangaise. Extr. d'Ann. I'Ecole Eaux Forets, 8, pp. 185-206, 1942. 65. Kaufert, F. H., A Survey of Laboratory Methods Used in the Evaluation of Wood Preservatives. Report of Committee P-6, Appendix A. Amer. Wood Preservers' Assoc, Proc, 45, pp. 55-59, 1949. 66. Leutritz, J., Jr., The Toxic Action of Various Compounds on The Fungus Lentinus Lepideus Ft.). Unpublished Thesis, Columbia University, Nov., 1933. 67. Leutritz, J., Jr., Laboratory Tests of Wood Preservatives. Bell Lab. Record, 16, No. 9, pp. 324-328, May, 1938. 68. Leutritz, J., Jr., Acceleration of Toximetric Tests of Wood Preservatives bv the Use of Soil as a Medium. Phytopathology, 39, No. 10, pp. 901-903, Oct., 1939. 69. Leutritz, J., Jr., Outdoor Tests of Wood Preservatives. Bell Lab. Record, 22, No. 4, pp. 179-182, Dec, 1943. 70. Leutritz, J., Jr., A Wood-Soil Contact Culture Technique for Laboratory Study of Wood-Destroying Fungi, Wood Decay and Wood Preservation. Bell Sys. Tech. Jl., 26, No. 1, pp. 102-135, Jan., 1946. 71. Liese, Nowak, Peters, Rabanus, Krieg, and Pflug., Toximetrische Bestimmung von Holzkonservierungsmitteln. Angew. Botanik, pp. 484-488, Nov .-Dec, 1935. 72. Liese, J. et al., Toximetrische Bestimmung von Holzkonservierungsmitteln. Angew. Chemie, 48, Beihefte 11, 1935. 73. Loseby, P. J. A., and P. M. D. Krog, The Persistence and Termite Resistance of Creosote and Its Constituent Fractions. Jour. South African Forestry Assoc, Jl., No. 11, pp. 26-32, June, 1944. 74. Lumsden, G. Q., and A. H. Hearn, Greensalt Treatment of Poles. Amer. Wood Preservers' Assoc, Proc, 38, pp. 349-361, 1942. 75. Lumsden, G. Q., Proving Grounds for Telephone Poles. Bell Lab. Record, 22, p|5. 12-14, Sept., 1943. 76. Lumsden, G. Q., A Quarter Century of Evaluation of Wood Preservatives in Poles and Posts at the Gulf port Test Plot. Amer. Wood Preservers' Assoc, Proc, 48, 1952. 77. Lutz, M. L., M^thodes Permittant de Determiner la Resistivity des Bois Brute ou Immunises Soumis a PAttaque par les Champignons Lignicoles. Ann. I'Ecole Nat. Eaux For6ts 5, pp. 317-327, 1935. 78. Mahlke-Troschel-Liese, Holzkonservierung (Wood Preservation), 3rd Ed. XII + 671 p. Springerverlag, Berlin/Gottingen/Heidelberg, 1950. EVALUATION OF WOOD PRESERVATIVES 167 79. McMahon, W., C. M. Hill, and F. C. Koch, Greensalt— A New Preservative for Wood. Amer. Wood Preservers' Assoc, Proc, 38, pp. 334-348, 1942. 80. Martin, S. W., Characterization of Creosote Oils. Amer. Wood Preservers' Assoc, Proc, 45, pp. 100-130, 1949. 81. Mayfield, P. B., The Toxic Elements of High Temperature Coal Tar Creosote. Amer. Wood Preservers' Assoc, Proc, 47, pp. 62-85, 1951. 82. Narayanamurti, D., V. Ranganathan, Ragbir Singh, T. R. Chandrasekhar, and A. Banerjee, Studies on Coal Tar Creosote as a Wood Preservative, Part II. Indian Forest Bulletin, No. 144, 1948, 7 + 43 pp., Jl. of India Press, Calcutta, 1950. 83. Peters, F., W. Krieg andH. Pflug, Toximetrische Priifung von Steinkohlen- teerol. Chem. Zeit., 61, pp. 275-285, 1937. (English Edition, Pub. Int. Adv. Off. Wood Pres. The Hague. 1937.) 84. Preservative Treatment of Poles. (Condensed from report by American Telephone and Telegraph Company of Aug. 3, 1931.) Am. Wood Preservers' Assoc, Proc, p. 237, 1932. 85. Preservatives Committee; Report of Committee P-6. Am. Wood Preservers' Assoc, Proc, 48, 1952. 86. Rabanus, Ad., Die Toximetrische Priifung von Holzkonservierungsmitteln. Angew. Bot. 13, p. 352-371, 1931. (Partial translation in English. Am. Wood Preservers' Assoc, Proc, pp. 34-43, 1933.) 87. Reeve, C. S., Comment on Creosote-Permanence Toxicity Relationships. Am. Wood Preservers' Assoc, Proc, p. 78-79, 1934. 88. Rennerfelt, Erik, och Bo Starkenberg., Traskyddskomittens fait- och rot- kammarforsok. (Field and Decay-Chamber Experiments with Wood Pre- servatives.) Meddelanden fr§,n Statens Skogsforskningsinstitut, Bd., 40, No. 4, 1951. 89. Rhodes, E. O., J. N. Roche, and H. E. Gillander, Creosote Permanence- Toxicitv Relationships. Am. Wood Preservers' Assoc, Proc, pp. 65-78, 1934. 90. Rhodes, E. O., History of Changes in Chemical Composition of Creosote. Am. Wood Preservers' Assoc, Proc, 47, pp. 40-61, 1951. 91. Rhodes, F. H., and F. T. Gardner, Comparative Efficiencies of the Com- ponents of Creosote Oil as Preservatives for Timber. Ind. and Eng. Chem., 22, No. 2, pp. 167-171, Feb., 1930. 92. Rhodes, F. H., and I. Erickson, Efficiencies of Tar Oil Components as Pre- servative for Timber. Ind. and Eng. Chem. 25, pp. 989-991, Sept., 1933. 93. Richards, A. P., Cooperative Creosote Program; Preliminary Progress Report on Marine Exposure Panels. Am. Wood Preservers' Assoc, Proc, 48, 1952. 94. Richards, C. A., Methods of Testing Relative Toxicity of Wood Preserva- tives. Am. Wood Preservers' Assoc, Proc, 19, pp. 127-135, 1923. 95. Richards, C. A., and R. M. Addoms, Laboratory Methods for Evaluating Wood Preservatives: Preliminary Comparison of Agar and Soil Culture Techniques Using Impregnated Wood Blocks. Am. Wood Preservers' Assoc, Proc, 43, pp. 41-56, 1947. 96. Richards, C. A., Laboratory Decay Resistance Tests— Soil-block Method. (In Cooperative Creosote Tests by R. H. Bescher et al.) Am. Wood Pre- servers' Assoc, Proc, 46, pp. 71-76, 1950. . 97. Scheffer, T. C, Progressive Effects of Polyporus Versicolor on the Physical and Chemical Properties of Red Gum Sapwood. U. S. Dept. Agnc, Tech. BuL, No. 527, Sept., 1936. , ^, ^«. , 98. Scheffer, T. C, T. R. C. Wilson, R. F. Luxford, and Carl Hartley, The Effect of Certain Heart Rot Fungi on the Specifid Gravity and Strength of Sitka Spruce and Douglas-Fir. U. S. Dept. Agric, Tech. Bui., No. 779, 24 pp., May, 1941. ^ „, ^ „ 99. Schmitz, H., Laboratory Methods of Testing the Toxicity of Wood Preserva- servatives. Ind. and Eng. Chem., Anal. Ed., 1, No. 7, pp. 76-79, April, 1929. 100. Schmitz, H., and Others A Suggested Toximetric Method for Wood Preserva- tives. Ind. and Eng. Chem., Anal. Ed., 2, p. 361, 1930 , t j 101. Schmitz, H., and S. J. Buckman, Toxic Action of Coal-Tar Creosote. Ind. and Eng. Chem., 24, No. 7, pp. 772-777, 1932. , ^ ,. ,.,.„.,., 102. Schmitz, H., W. J. Buckman and H. von Schrenk, Studies of the Biological 168 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Environment in Treated Wood in Relation to Service Life. Changes in the Character and Amount of 60/40 Creosote-Coal Tar Solution and Coal Tar and Decay Resistance of the Wood of Red Oak Crossties after Five Years Service. Am. Wood Preservers' Assoc, Proc, 37, pp. 248-297, 1941. 103. Schmitz, H., H. yon Schrenk, and A. L. Kammerer, Studies of the Biological Environment in Treated Wood in Relation to Service Life, III. Am. Wood Preservers' Assoc, Proc, 41, pp. 153-179, 1945. 104. Schulze, B., und G. Becker, Untersuchungen liber die pilzwidrige und insek- tentotende Wirkung von Fraktionen und Einzelstoffen des Steinkohlen- teerols. Holzforschung., 2, No. 4, pp. 95-127, 1948. 105. Sedziak, H. P., The Wood-Block Soil Method of Accelerated Testing of Wood Preservatives. Report of Committee P-6, Appendix B. Am. Wood Pre- 'servers' Assoc, Proc, 46, pp. 55-58, 1950. 106. Sedziak, H. P., The Evaluation of Two Modern Wood Preservatives. For. Prod. Res. Soc, Proc, 1952. 107. Snell, W. H., The Use of Wood Discs as a Substrate in Toxicity Tests of Wood Preservatives. Am. Wood Preservers' Assoc, Proc, 26, pp. 126-129, 1929. 108. Snell, W. H., and L. B. Shipley, Creosotes — Their Toxicity, Permanence and Permanence of Toxicity. Am. Wood Preservers' Assoc, Proc, 32, pp. 32- 114, 1936. 109. Standard of N. W. M. A., Method for Testing the Preservative Property of Oil-soluble Wood-Preservatives by Using Wood Specimens Uniformly Impregnated. Nat. Wood Mfg. Assoc, M-1-51, April 27, 1951. 110. STAS 650-49, Incercarea Toxicitatii Substantelor de Impregnat Contra Ciupercilor. Comisiunea de Standardizare (Rumania) April 1, 1950. 111. Suolahti, Osmo, Uber Eine das Wachstum von Faulnispilzen Beschleunigende Chemischen Fernwirkung von Holz. Statens Tekniska Forskningsanstalt. 95 p. Helsinki, Finland, 1951. 112. Tamura, T., New Methods of Test on the Toxicity and Preservative Value of Wood Preservatives. Phytopathologische Zeitschrift, 3, No. 4, pp. 421- 437, 1931. 113. Teesdale, C. H., Volatilization of Various Fractions of Creosote After Their Injection into Wood. Circ. 188, Forest Service, Forest Products Laboratory Series, U. S. Dept. Agric, Oct. 17, 1911. 114. Tippo, O., J. M. Walter, S. J. Smucker, and W. Spackman, Jr., The Effective- ness of Certain Wood Preservatives in Preventing the Spread of Decay in Wooden Ships. Lloydia, 10, pp. 175-208, Sept., 1947. 115. Trendelenburg, R., tJber die Abkurzung der Zeitdauer von Pilzversuchen an Holz mit Hilfe der Schlagbiegepriifung. Holz als Roh- und Werkstoff. 3. No. 12, s. 397-407, Dec, 1940. 116. van den Berge, J. Beoordeeling van de Waarde van Fungicide Stofifen voor Houtconserveering. 183 p. N. V. Technische Boekhandel, J. Waltman, Delft, Holland, 1934. 117. van Groenou, H. Broese, Weatheringsproeven met Houtconserveermiddelen, (Weathering Tests with Wood Preservatives). Materiaalenkennis, 7, No. 10, pp. 63-65, Oct., 1940. 118. van Groenou, H. Broese, H. W. L. Rischen and J. van den Berge, Wood Pre- servation During the Last 50 Years. XII -f 318 p. A. W. Sijthoff, Lei- den, Holland, 1951. 119. von Pechmann, H., u. O. Schaile, Uber die Xnderung der dynamischen Festig- keit und der Chemischen Zusammensetzung des Holzes durch den Angriff Holzzerstorender Pilze. Forstwissenschaftliches Zentralblatt, 69, No. 8, 8. 441-466, 1950. 120. Verrall, A. F., Progress Report on Tests of Soak and Brush Preservative Treatments for Use on Wood off the Ground. Southern Lumberman, Aug. 1, 1946. 121. Verrall, A. F., Decay Protection for Exterior Woodwork. Southern Lumber- man, June 15, 1949. 122. Warner, R. W., and R. L. Krause, Agar-Block and Soil-Block Methods for Testing Wood Preservatives. Ind. and Eng. Chem., 43, No. 5, pp. 1102-1107, May, 1961. EVALUATION OF WOOD PRESERVATIVES 169 123. Waterman, R. E., Forecasting the Behavior of Wood Preservatives. Bell Lab. Record, 11, No. 3, pp. 67-72, Nov., 1932. 124. Waterman, R. E., F. C. Koch, and W. McMahon, Chemical Studies of Wood Preservation. III. Analysis of Preserved Timber. Ind. and Eng. Chem., Anal. Ed., 6, No. 6, pp. 409-413, Nov. 15, 1934. 125. Waterman, R. E. and R. R. Williams, Chemical Studies of Wood Preserva- tion. IV. Small Sapling Method of Evaluating Wood Preservatives. Ind. and Eng. Chem., Anal. Ed., 6, No. 6, pp. 413-418, Nov. 15, 1934. 126. Waterman, R. E., J. Leutritz, and C. M. Hill, A Laboratory Evaluation of Wood Preservatives. Bell System Tech. Jl., 16, pp. 194-211, April, 1937. 127. Waterman, R. E., J. Leutritz and C. M. Hill, Chemical Studies of Wood Preservation: The Wood-block Method of Toxicity Assay. Ind. and Eng. Chem., Anal. Ed., 10, No. 6, pp. 306-314, June, 1938. 128. Weiss, J. M., The Antiseptic Effect of Creosote Oil and Other Oils Used for Preserving Timber. Soc. Chem. Ind. JL, 30, No. 23, Dec. 15, 1911. 129. Williams, R. R., Chemical Studies of Wood Preservation. I. The Problem and Plan of Attack. Ind. and Eng. Chem., Anal. Ed., 6, No. 5, pp. 308-310, Sept. 15, 1934. Motion of Gaseous Ions in Strong Electric Fields By GREGORY H. WANNIER (Manuscript received August 20, 1952) This paper applies the Boltzmann method of gaseous kinetics to the prob- lem of charged particles moving through a gas under the influence of a static, uniform electric field. The particle density is assumed to be vanishing low, and the ion-atom collisions are assumed elastic, but the field is taken to be strong; that is the energy which it imparts to the charges is not assumed negligible in comparison to thermal energy. In Part I, the formal framework of such a theory is built up; the motion in the field is describable by the drift velocity concept, and the smoothing out of density variations as an aniso- tropic diffusion process. In Part II, the "highfield^' case is treated in detail; this is the case, for which thermal motion of the gas molecules is negligible; the equation is solved completely for the case that the mean free time between collisions may be treated as independent of speed; complete solutions are also presented for extreme mass ratios of the ions and the molecules; special attention is given to the case of equal masses, which has to be handled by numerical methods. In Part III, information about the ''intermediate fiM^^ case is collected; with the help of a convolution theorem the case of constant mean free time is solved; beyond this, only the case of small ion mass {electrons) is available. In Part IV, the diffusion process, whose existence was proved in Part I, is pushed through to numerical results. Part V discusses the scope of the results achieved and demonstrates the possi- bility of extending them semiquantitatively beyond their original range. Part I — General Theory of Strong Field Motion lA. qualitative discussion It is well known that if we consider a mixture of gases under no ex- ternal forces the steady velocity distribution which establishes itself in the mixture does not depend on the interactions between the gas molecules; we have always a Maxwellian distribution for each species 170 MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 171 with a temperature common to all. This result arises from statistical mechanics; the derivation of it is simple and requires few assumptions, yet it enjoys a wide degree of generality. As soon, however, as a non- equilibrium feature is imposed upon the system this simplicity vanishes, and the subject acquires ramifications. Results must now be derived by kinetic theory. The amount of labor required increases, while, at the same time, the result achieved becomes less general. A mixture of charged particles (ions or electrons ; in the following often simply referred to as ions) and gas molecules can in principle never be in equilibrium since the presence of the former in itself represents an instability. However, one might expect, that equilibrium exists in a restricted sense, for instance, as regards motion. Even this is rarely the case under actual conditions of observation. The non-equilibrium fea- tures of greatest importance for analyzing ion motion are a constant force (electric field) acting upon one species but not the other (mobility theory), and a concentration gradient for one particular species (diffusion theory). It is the purpose of this paper to apply kinetic theory to these problems, and to compute with its help the most important properties which such a gas of charged particles possesses. The work will be dis- tinguished from similar ones in that the electric field will not be sup- posed weak; velocity distributions which have no resemblance to the Maxwellian distribution will thus make their appearance. Furthermore, the mass of the charged particles will not be assumed small, which means the possibility of getting results for gaseous ions as well as elec- trons. Magnetic fields, plasma and A.C. phenomena will, however, be excluded. The quantities of interest under those conditions are the drift velocity of the ions, their energy, energy partition and diffusion con- stants. These quantities will be calculated by assuming plausible mechan- ical models. The work just outlined has been published in part in abbreviated form in the Physical Review;^ the exposition to follow will, however, proceed independently from these articles. Much of the work which concerns itself with transport processes in gases makes use of perturbation theory. This method permits us to predict the behavior of a gaseous assembly under an electric field or a concentration gradient in the limit when the field or the gradient are vanishingly small. The result of so perturbing a Maxwellian distribution can be expressed through certain constants, such as the mobility or the diffusion coefficient, which involve the Maxwellian distribution and the internal interactions, but not the perturbation itself. 1 Wannier, G. H., Phys. Rev., 83, p. 281, 1951 and Phys. Rev., 87, p. 795, 1952, 172 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 The limits of such a procedure can easily be estimated. In the case of an electric field, perturbation techniques apply if the kinetic energy acquired by the ion from the field is small compared to thermal energy. This means at least that the energy acquired in one mean free path be small, i.e., eE\ « kT where e is the electronic charge, E the electric field, k Boltzmann's con- stant, T the absolute temperature, and X the mean free path. Actually the situation is not even that favorable. If the mass of the ions and the molecules is very different, the energy transferred upon collision is small, and hence the ions possess the ability to store the acquired energy through many collisions; for this reason, the inequality reads more properly (M m\ \m] "^ Mj eE\ « kT, where m is the mass of the ions and M the mass of the gas molecules. After some substitutions this estimate becomes eE « p dC The equation is linear in the unknown function d{c, r, t) ; this is due to neglect of ion-ion collisions, as stated earlier. The negative term on the right hand side actually reduces to a known function of c multiplying rf(c, r, t). The positive term is a genuine integral term; it has been shown by Pidduck that the number of integrations in it can be brought down from five to three; this reduction will not be made use of in the following. If there were no terms on the left hand side of equation (10) then the solution of it would have the equilibrium form d{c, r, t) = nm{c) (11) where n is a constant. This result is a direct consequence of equation (8) which makes the curly bracket in (10) vanish identically when Maxwell ian functions are inserted. '* See Reference 4. " Pidduck, F. B., Proc. Lond. Math. Soc, 16, p. 89, 1915. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 183 The function m(c) is not the solution of our problem because of the presence of the second and third term on the left which arise from an electric field and a density variation respectively. These disturbances will be assumed of different relative importance. The density variation will be assumed sufficiently small so that the third term can be treated by perturbation theory; the field term, on the other hand will be taken so large that the equilibrium distribution (11) no longer represents a first approximation to the solution. In consequence, the equation is solved in two stages. In the first, only the second term on the left is retained, and the resultant equation is treated rigorously; in the second, the full equation (10) is used, but the new terms are taken as perturbations. The first stage describes those properties of the ion gas which it po- sesses when assumed of uniform density. Since the field is also assumed uniform and not changing in time, the dependence on r and t drops out. We may then write die r, t) = n/(c) (12) where n is a constant and /(c) is a velocity distribution function. The equation for / reads a.| = ^ // |M (C')/(c') - M(C)/(c) l7^ 30 40 ACCELERATION 80 100 150 200 X 10"* NUMBER DENSITY Fig, 3 — Drift velocity in an electric field of He+ ions in helium gas. Compari- son of observed results with an "asymptotic" straight line of slope \i. 190 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 50 40 30 20 15 10 X10~'* cf A >^ ^ Y p^ ^ y "f^ y ^ 'y L '<^ ^ f ^ 8 ^^' / ^"^ "^ A r ^ r 5 >^ /^ ^ NEON PRESSURE IN MM Hg D 7.50 A 4. 10 o 0.724 3 2 1.5 1 > / / 1.5 6 8 10 15 ACCELERATION 20 30 40 60 80 100 X 10" NUMBER DENSITY Fig. 4 — Drift velocity in an electric field of Ne+ ions in neon gas. Comparison of observed results with an "asymptotic" straight line of slope 3^^ in the parent gas, observed for He+, Ne+, A+, Kr+ and Xe+. The plot is a log-log plot of these quantities against a/'N ^ a variety of fields having been used to determine each point. The data are taken from measurements of J. A. Hornbeck^^' ^^ and R. N. Varney.^^ These data verify in the first place that the drift velocity depends on a and N only in the combination aIN . Beyond this we see that the curves consist of two straight line portions: in the lower field portion (c^) is proportional to aIN , in the higher is proportional to -s/a/N. We recognize in this latter region the high field dependence predicted in equation (26b). We learn from this that the collision cross section between noble gas atoms and their ions is approximately constant in the experimentally significant velocity range. To determine these collision cross sections the computation of only a single number, namely the one entering into »• Hornbeck, J. A. and G. H. Wannier., Phys. Rev., 82, p. 458, 1951. " Hornbeck, J. A., Phys. Rev., 84, p. 615, 1951. » Varney, R. N., Phys. Rev., 88, p. 362, 1952. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 191 10^ a/ ^ ' ^ .^ ^ fo> ,f ,' ^ > "•^ y • ^ A ^k <^- ^ ARGON PRESSURE IN MM Hg o 0.668 A 0.823 • 2.97 D 6.29 m A y °. V 6 8 10 15 ACCELERATION 30 40 60 80 100 X10"6 NUMBER DENSITY Fig. 5 — Drift velocity in an electric field of A+ ions in argon gas. Comparison of observed results with an ''asymptotic" straight line of slope H. NUMBER DENSITY Fig. 6 — Drift velocity in an electric field of Kr+ ions in krypton gas. Com- parison of observed results with an "asymptotic" straight line of slope }^. 192 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 \ \ \\ \\ \\ \\ XENON URE IN MM Hg 5 O 7.011 7 ■ 8.245 6 + 8.637 8 S 10.162 4 V 10.273 9 * 12.057 4- ▼ 13.693 0 0 15.672 4 \ \ \ \ [2 (O •^^ O' in o> ID O '^ ro V ^ < \ \ \ ' ^Oi < \ \ \ \ ^^o \ \ \ .s «. \ \ \ \ \ \ \ ^, o ^ \ V 00>fl0t^»0in Tt ro u ~ r ONODas aad Sd3i3iMiN3o NI AiiDonaA idiaa oo 3 Tt- X ^ 1 «4-l O o fl (\l o .2 in S o O o fl o o> V ^ 007 H C o V) -5 z. CO ' o UJ to _J III + m o .« 2 •- CO o o fO (VJ Q CO > 43 in ^5 1 § MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 193 (26b) is required. The linear range of this plot is not as informative as the high field one. The slope unity is common to all formulas (27), and the temperature dependence of the mobility is needed to give the correct interpretation with the methods developed here. There is a certain likelihood that the parameter a of equation (25) drifts from 0 to 1 as the speed of the ions is reduced ; this was pointed out for the special case of He"*" in He in section lA. A qualitatively similar situation appears to prevail for the other noble gases. Part II — The Motion of Uniform Ion Streams in the High Field Case II A. formulations of the boltzmann equation The dimensional analysis of the last section shows that there is an intrinsic simplicity to the high field case which is comparable to the low field case, while the intermediate case is more difficult. With one excep- tion,^ however, theoretical analysis has occupied itself with the low field case only. We shall try to remedy this in the following. To begin with, a tractable but accurate formulation of the problem has to be found. Such a formulation cannot treat the field term of equation (13) as a perturbation term, but must try instead to make use of the basically simple features of the problem, notably those exhibited by the dimen- sional analysis of Section ID. The equation governing the high field properties of the ions is ob- tained simply by substituting 5-functions for the Maxwellian velocity distributions in equation (13). This gives a • + dcz _L /i(c) = i- ff 5(C0«c0 -^ n(x) d%> dO (31) t{c) 47r J J t{c ) A reduction of the number of integrations from five to two must be possible in the integral term of (31), owing to the presence of the 6-f unc- tion. To achieve this we must transform the variables of integration so as to make three of the differentials equal to dC. We do this in the following way. First observe that r = c - c and that c is a constant vector. Hence we may replace dC by dy. The five-fold integration reads then dQy' dC = 1 dy dOy dQy (32) that is, it goes over the magnitude y which the two vectors have in 194 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 common, and their orientations, for which they are independent. It is knowTi that in integrating over the two angles defining an orientation the polar axis may be chosen freely. We shall, in the following, adopt c as our polar axis with yp, k being pole distances of y and y' to c and ^, co the corresponding azimuths. In Fig. 8 these angles are exhibited on the unit sphere. All vectors are assumed to be plotted from the center of the sphere, and show up through their piercing points. The angles between the vectors then show up as sides and the azimuths as angles. The ex- pression (32) becomes then 7 dy sin \}/ d^ dip sin k dK dw The main transformation consists now in introducing the three com- ponents of C in the place k, \f/ and p cos k cos {(p — oo)] = c sin k The curly bracket is exactly the one occurring in the Jacobian which therefore reduces to , — 12— ±L = c c sm ^ sm k L ^('c,^,^) Jc'=o {M + mY and hence 7' dy d% dQy = ^^,;^ ^^ dC' ^c' ^co ' ^ ^ Mmc Substituting finally this expression into (32) and (31) we get the Boltz- mann equation in the form dh{c) . 1 , , . dCz t{c) M^ (34) AirMmc Jc r{c) Jo The equation is in need of additional elucidation as regards the exact meaning of c' as a vector and as regards the auxiliary variable x- As to the first point we may describe the integration as occurring over a sur- face in velocity space. This surface is obtained from the relation C = 0 c' = Y (35) which substituted into (33) becomes (M + m)c - my' = My (36) 196 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Squaring this and using (9) we get (M - my'' + 2mc'-c - (M + m)c = 0 (37) This is the equation of a sphere in velocity space which passes through the point c' = c. For all other points c' is bigger than c (colHsion with a stationary object always brings energy loss). The center of the sphere lies on the line joining c to the origin; it lies on the side of the origin from c when m < if, at infinity (making the sphere a plane) when m = M and away from the origin when m > M. We make use of (37) to express the polar angle k of c' with respect to c (which does not occur as an integration variable in (34)) in terms of c'. We get (M + my -(M - m)c" ,„^^ COS K = ^r -. (,oo; 2mcc The angle of scattering in the center of mass system also results from squaring of (36) if the term my' is first taken to the right. We find °°^^ 2Mm c'2 2Mm ^ ' There is a more useful form of equation (34) which results if x is taken as one of the integration variables rather than c'. Substitution is made from the equation (39) above; it yields a ^ + -I- Ho) = 1 [' sin X d^ 5W («:)' r ,(c') d. (40) dCz t{c) Ait Jq t{c') \c / Jo The magnitude of c' and its polar angle with respect to c are now auxiliary parameters; the first is obtained from (39) c' = c ^ + ^ (41) V M2 + m2 + 2Mm cos x and the second from (38) and (41) m + M cosx .,^. cos K = (4^) VM2 4- m2 + 2Mm cos x As previously, the azimuth co of c' about c is an independent variable. The simplifications of the equation (31) exhibited in (34) and (40) still leave a double integral in the fundamental equation. The integration over do) will now be eliminated by decomposition of h{c) in spherical harmonics about the field direction. There is no loss of generality in this step. h{c) = E K(c)P, (cos t?) (43) MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 197 We have now to consider simultaneously the three vectors c, c' and a as well as the angles between them. These angles are defined in Fig 8. We study equation (34) or (40) term by term in order to see what be- comes of it upon substitution of (43). Starting with h{c') under the inte- gral sign we get from Fig. 8 and the addition theorem for spherical har- monics hie) = E hXc)lP, (cos T»P. (cos k) + 2 X; ^/ 7 ^l\ P". (cos ^)K (cos k) cos M M=l (»' + m) I For this expression, the integration over co is elementary and gives \ ' h{c) do, = 2irY. Hc)Pv (cos t?)P. (cos k) (44) Further, we get for the derivative in (34) or (40) #- (Z K ic)P. (cos t») OCz \f=0 / = ± ^ _!_ { (, + DP^i (cos ,» + yP^i (cos ,» 1 (45) Po dc Zj/ -t- 1 c 2i' H- 1 Through the equations (43), (44) and (45), all terms in equation (34) or (40) are developed in spherical harmonics with respect to the angle t? between c and the field direction. We can therefore annul separately the coefficient of each Legendre polynomial in cos t?. This gives the follow- ing set of equations (M + mf (mi'h^p, (eos .) n(x) dc' -"^ = ^^ /^^^::lW __ y- 1 h^^ic)) (46) 2v - l\ dc c I {y + l)a IdK^ , v_±2 ;l_(c)^ '^ 2v^Z\ dc c J 198 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 or hp{c) va \dhy_\{c) v — 1 •(c) 2v — \ \ dc c {v + l)a fdK+iic) ,v+2 K-iic)"! (47) where V = 0,1,2,3 •'• . The auxiliary parameters entering are given by (38) and (39) for equa- tion (46), and (41) and (42) for equation (47). The equations (46) or (47) obtained by Legendre decomposition still are, in general, mixed integral-differential equations in one independent variable. Further simplification is possible only in special cases some of which will be discussed later. An even more simple and tractable form of the Boltzmann equation can be achieved in general, however, if one gives up the idea of determining the velocity distribution function and concentrates instead on its moments. In other words, the Boltzmann equation can be looked upon as a system of relations between velocity averages, and as such it becomes a linear algebraic system. To carry out this reduction we multiply equation (47) by c'^^ and integrate from 0 to co. The second term on the left is then a simple velocity average. The same is true on the right hand side if two integra- tions by part are permissible and leave no integrated out part, s ^ — 1 is probably adequate for this. The integral over the integral term at first looks as follows I f c'^' dc r ^ (^'Y P, (cos k) n(x) sin X dx L Jo Jo t{C ) \C / In this expression we pass from c to c' as the independent variable. From (41) we see that dc _ dc 7" "7 Hence the expression becomes f c"« ^-gj dc' \ /; {^^ P. (cos .) n(x) sin . .X p MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 199 From (41) and (42) it is seen that this is actually the product of two independent integrals if the angular distribution 11 (x) is independent of the velocity of encounter c' . The first integral is then identical with the one arising from the second term in (47), and the second is a collision integral having no connection with the velocity distribution. Even if this is not the case, the second integral is still a dynamic average which can be evaluated as a function of c' previously to any knowledge of /i(c'). We express this by introducing the abbreviation Ib,v = V-\Py (cos k) Using (41) and (42) we see that 7s. ^ is the following function of x "'^^ '' + " ^' (48a) m -\- M cos X \ VM^ + m2 + 2Mm cos x / which, for the particular case of equal masses, takes the simple form Is Ax) = cos' \x p. (cos ix) (48b) With this definition the integrated equation (47) reads f I L^il^ \ kXcV^' dc = ''^''+^V^ f h,.Mc'^' dc Jo \ aTic) I 2v — \ h •(c) 2v -\- 6 Jo or in terms of averages {2v + 1) ( ^^ c'P. (cos ^)) \ aric) I (49) = K^ + s + 1)(c'"'Pk-i (cos ^)) + (^ _|_ 1)(5 _ v){(r^Fv-x (cos t?)) I believe that equation (49) contains all possible derivable relations between averages as special cases. Some of the most notable ones are 200 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 listed below s = 1, p = 1 /I — cos X o\ M + m ,^_ . \ r\ — c cos ^) = — — — (50a) s = 2, V = ^ /I - cos X 2\ {M -\r mY , . . . \ -r\ — c / = TJ \^ cos d) (50b) \ aric) I Mm s = 2, V = 2 /m sin X + 4m(l - cos x) ^.p^ (^^^ ^A ^ \ otCc) / (50c) = —j^ (c cos tf> While the averages entering into (49) are not always the desired ones, it remains true nevertheless that all solution methods evolved in the following use this equation system as a starting point rather than other forms of the Boltzmann equation. IIB. THE MEAN FREE TIME MODEL AT HIGH FIELD If the angular distribution in the center of mass system is independent of speed and the collision cross section varies inversely as the speed then the developments of the previous section permit actually a solution of the Boltzmann equation. It is a solution in the sense that all signifi- cant velocity averages can be obtained directly without the knowledge of the velocity distribution function. Before developing these facts from the equations of the last section, I should point out that the derivation to follow is in a sense artificial. It has been shown already by Maxwell^ ^ for related problems that if the mean free time between collisions is assumed constant specially simple techniques may be employed to get constants of experimental im- portance. These techniques can be employed here ; they consist essentially in multiplying (13) by a suitable multiplier, followed by integration over c. However, if we were to follow this procedure we would have to duplicate for a special model in an unsystematic way the work done systematically for all laws of interactions in the preceding section. A further advantage of using systematic procedure is that we can see at a " Maxwell, J. C, Collected Papers, Vol. II, p. 40. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 201 glance what averages can or cannot be obtained, and what the relation- ship is between the high field and the general averages. For this reason we limit ourselves at present to the high field averages obtainable from (49) . For the special case under discussion this equation system takes the form (2, + 1) zyiiiM) (c-p, (cos &)) ar I = K^ + s + l)(c*~'P^i (cos t?)) (51) + (^^ + l)(s - ^)(c'"'Ph-i (cos t?)> that is we have a system of linear relations connecting the averages (c'P, (cos ??)). The connection between these averages is made apparent in Fig. 9. Each average {cPy (cos t>)) is marked in this figure as a dot in an s-i'-plane if s is integer. The equations (51) connecting these averages are shown as lines with different equations leading to the same dot shown in different outline. These equations generally have the shape of a V; Fig. 9 — Interconnection established by the Boltzmann equation among the averages l^c*F, (cos «?)); case of constant mean free time. 202 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 there are two notable exceptions to this rule, however, which make the recurrence method possible, the equations v = 0 have no left leg and the equations s = v have no right leg. Starting out with the average s = 0, V = 0, which equals unity by definition one can thus proceed systematically as shown in Fig. 10, to get other averages. The averages reached are the ones for which s and p are non-negative integers of equal parity with the restriction s ^ v. One verifies easily that this set is equiva- lent to the set of all products of integer powers of the velocity compo- nents. The first three relations one uses in the path outlined in Fig. 10, are the simplified forms of the three equations (50). We find 2p / XV _ 4(M + my (c P, (cos t?)) - ^^ /3ji^ si^2 ^ ^ ^^(^ _ ^^^ ^^y^ _ ^^^ ^^ \ ar A ar I or, more conveniently with the help of (53) {M + mf /^ sii^' X 4- 4m(l - cos x)\ lr^\ = \ ^ /__ C54) ^'^ 3P^ /3M sin^ X + 4m(l - cos x)\/l - cos xV \ ar /\ ar / The three equations (52), (53) and (54) give the drift velocity, the total energy, and the energy partition of the travelling ion. Equation (52) gives a constant mobility and can actually be derived from a low field theory. Formula (52) thus states that for problems involving a constant mean free time the high field and low field mobilities are numeri- cally identical. One would suspect that the intermediate field value would have to fall in line too. This is indeed the case as will be shown in Section IIIA. A convenient interpretation of (53) may be had by combining (52) and (53) in the following way (,mc^) = m{c^)'' -h M^ \ r / (73) This is a Maxwellian distribution with elliptic distortion and shifted origin, that is, the type shown in Fig. 1 (b). The result (73) indicates the main features of the solution for heavy ions. Because of the neglect of derivatives higher than the second in /i(cO it is not certain that (73) is correct in all details, even in the limit of very large m/M. IID. THE CASE OF EQUAL MASSES; IONS TRAVELLING IN THE PARENT GAS The developments of the previous section show that if the ion mass is either large or small in comparison to the molecular mass, analytical methods can be applied successfully to determine the velocity distribu- tion of the ions. No such possibility was found for the mass ratio unity, which one would judge to be of particular interest because it applies to ions travelling in the parent gas. There exist isolated fragments of such an analytical theory; for instance, if we assume isotropic scattering in (46), that is n(x) = 1, then the zeroth equation (46) becomes explicitly integrable and yields Uc) = -\ar(c)''J^ (74) This is a curious reversal of the differential relationship (63) derived for electrons and implies a rather strong condition on the structure of hi{c). However, I have not been able to consolidate these fragments into something which can be used successfully in computation. The high field distribution function for mass ratio unity appears, however, suffi- ciently interesting to warrant the use of other methods. A numerical determination of the velocity distribution function was undertaken in cooperation Avith R. W. Hamming by the so-called Monte Carlo method. The Monte Carlo method is a way of gaining 212 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 statistical information about a system by following an individual member through a large number of random processes. The result of such a pro- cedure is knowledge about one member of the assembly for a long period of time. Time averages of various kinds can be obtained from such data ; these time averages are then set equal to instantaneous averages over the assembly, in accordance with ergodic theory. In our case, an ion was followed through 10,000 collisions. On an average, the collisions were isotropic in the center of mass system (11 (x) = 1) and obeyed a mean free time condition r = const. Actually, both the free time and the scat- tering angles varied from collision to collision; the angles varied in a random fashion over a unit sphere and r was random within an ex- ponential distribution. A Monte Carlo calculation of this type consists of three parts. In the first part the random numbers having the required distributions are obtained and recorded. In the present problem there were three such random numbers required for each collision, namely a time and two angles. These numbers were placed on 10,000 IBM cards, along with suitable identification. In the second part a calculating machine simulates the successive collisions and keeps a record of the initial and final veloci- ties for each one. The third part consists in analyzing statistically the numerical material accumulated in the second. For the first part of the calculation particular values must be chosen for the acceleration a and the mean free time t. These values were a = 1 r = logio e = 0.43429 However, the dimensional analysis of Section ID shows us at this point that these two constants enter into the problem only through their product ar which scales all velocities. It is therefore convenient at the statistical stage to remove these factors and to analyze the results in terms of a dimensionless variable which by (26c) we take in the form w = -^ (75) ar In view of the a priori information for mean free time problems which is gathered in Section I IB we can use the statistical data from the Monte Carlo calculation in two ways. We may (a) check the numerical computation itself or (b) gain new information not available otherwise. (a) The check of the numerical calculation by statistical analysis proceeds as follows. From deductive reasoning we have obtained the MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 213 averages (52), (54) and (57) for Ca , cl and cl . These formulas ought to be verified in the Monte Carlo calculation. This is indeed approxi- mately true. A sampling covering 9492 out of the 10,000 collisions gives by Monte Carlo by deduction (w^) 1.912 2 (wl) 0.801 ^=0.889 (wl) 5.165 ^=6.222 The agreement is essentially there but deviations are noticeable. In judging these we have to realize that fluctuations are quite large in this problem. For instance if the calculation is broken down into ten runs of approximately 1,000 events each one finds the following time averages for the partial runs: {w,} {w%) (w|) 1.821 0.837 4.453 1.975 0.748 5.531 1.915 0.785 5.132 1.954 0.829 5.441 1.868 0.731 4.794 2.003 0.807 5.581 1.766 0.793 4.811 1.962 0.827 • 5.360 2.003 0.901 5.471 1.914 0.727 5.200 predicted 2.000 0.889 6.222 Among these runs there are some having averages higher than the pre- dicted values, but the data clearly show that the Monte Carlo averages are generally lower. In the search for reasons it was first felt that perhaps the desired mean value for r is not actually reached, perhaps through systematic errors introduced by the operator when rejecting certain runs. This seems indeed to be the case. The mean free time obtained from the 9492 runs mentioned above is r = 0.4269 which is slightly low. Indeed it is observed that the runs with high r were particularly troublesome in the calculation and were preferably rejected by the operator. It seems doubtful however that this error 214 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 could account for the entire discrepancy, particularly in the mean squares, although it must be emphasized that the runs with high r make a more than proportional contribution to the total average. The angles of scattering have not been subjected to a similar analysis so that we cannot make a statement whether the aimed at isotropy in the law of scattering was realized or not. We conclude therefore by saying that while the Monte Carlo calculation gives results in general agreement with the deductive theory there are small but noticeable systematic errors in it whose origin is only partly explained. Similar errors must exist in the new results which cannot be compared with theoretical predictions. (b) In this part we will discuss the velocity distribution function which may be constructed from the Monte Carlo results. In constructing such a function we make use of the fact that, between collisions, the velocity is accelerated at a uniform rate. Thus, in each period between two collisions, the velocity vector traces out a straight line parallel to the Wz axis covering equal distances in equal times. The Monte Carlo calculation furnishes us with a number of such straight lines as shown in Fig. 11. The density of these straight line tracks in velocity space is the velocity distribution function. The actual procedure used to obtain it was to lay a grid with a mesh of 0.23 in a half-plane with coordinates Wz and Wp = ■\/wl + wl and to count the number of lines crossing each hori- zontal square edge. When the resultant count is converted to density w. o o o o p ' W^ = VWx^ + Wy2 Fig. 11 — Straight line pattern in the w, — w, half plane from which the veloc- ity distribution is constructed; the Monte Carlo calculation furnishes the initial and final velocities (dots and rings). MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 215 and normalized to 1 we get a distribution in the Wz — w^ half plane which is shown in Fig. 12. Division by 2TrWz will transform it into a conventional distribution function in velocity space; a plot of this function is shown in Fig. 13. What distinguishes this distribution function from functions previously proposed is the elongated probability contours. This feature is not unexpected in view of the unequal energy partition apparent in the equations (58) and C60). The probability contours sho^\Ti in Figs. 12 and 13 give a reliable general picture but we must not expect from them fine detail. Indeed we will now prove that the distribution function is infinite along the entire positive Cz-axis, a feature which is not obvious from inspection of Fig. 13. Fig. 12 — Motion of ions through the parent gas in a high field; distribution of velocities in the w, - w^ half plane resulting from the Monte Carlo calculation. 216 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 -3.0 -2.5 -2.0 -\.5 -1.0 -0.5 0 0.5 1.0 1.5 2.0 2.5 3.0 Wx Fig. 13 — Velocity distribution function of ions moving through the parent gas in a high field; contours constructed from the Monte Carlo results of Fig. 12. A simple physical proof of this statement goes as follows. Suppose an ion and a molecule make a collision which is almost central, but has a small impact parameter h. The collision will bring the ion almost to rest because the atom was originally at rest by hypothesis. Because the col- lision was not quite central, however, the ion will have a small residual velocity C/ at right angles to its original velocity c,- . For any reasonable law of scattering this quantity C/ will be proportional to d and to h. Cf oc cr6 The probability for a value of b between h and b + db is proportional MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 217 to h db. Thus even if all c,'s were equally probable the probability for Cf would vary as c/ dc/ . Actually very small d's may be specially probable as the theorem states and this fact may or may not increase the prob- ability for small c/. This means that P{cf)dcf probably varies as C/dc/, and may perhaps even contain a smaller power of C/ . When such a probability function is plotted in velocity space it will vary as 1/c/ . Thus we know that the distribution function (p{c) for ions immediately following a col- lision has a singularity at the origin at least as 1/c/ . The actual distri- bution function h(c) is derived from this one by spreading each point out in the forward direction as shown in Fig. 11. For the mean free time case we can write this out explicitly in the form 1 r - He) = - / 0 with suitable relationships existing between these quantities. A defect of all three approaches is that they give no information concerning the nature of the infinity for Cz > 0. One is tempted to conclude from Fig. 13 that it cannot be very strong. Something like a singularity is discernible at the origin, particularly if the contour 0.1 is drawn back to cut the Wz-Sixis at a negative value ; this is perfectly com- patible with the available information. For large positive Wz , on the other hand, the picture almost contradicts the theorem just proved. One con- cludes from this that the singularity, for large Cz , becomes a weak and narrow ridge rising more or less abruptly in an otherwise well behaved function. nE. THE CASE OF EQUAL MASSES ; A NEW COMPUTATIONAL PROCEDURE The foregoing sections have accumulated substantial evidence that there are many analytical details involved when one discusses the structure of a velocity distribution function. These details are of little interest to the experimenter who may want nothing but a formula for the drift velocity or the average energy. In view of this situation it appears very desirable to find a method whereby such quantities can be derived directly and accurately from the Boltzmann equation without a full knowledge of the entire distribution. Maxwell's original work shows us how to achieve this for molecules obeying the mean free time condition of Section IIB. In the following, a general method is described which will permit determination of such MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 219 averages for an arbitrary law of force between the ions and the gas molecules, and an arbitrary mass ratio. The application will be limited to the case of the mass ratio 1 whose study was begun in the preceding section. The basis of the method is an observation on the equation system (46) or (47), which is the form taken by the Boltzmann equation after in- serting the Legendre decomposition (43). It would appear at first sight that these recursion relations are of such a structure that an arbitrary function }U)(c) could be substituted into the "zeroth" equation and that the relations would then successively determine /ii , /i2 , /^a • • • . Upon closer inpsection this is found not to be the case. Suppose we have obtained somehow functions ho , hi , h2 - - • hn and we are trying to use the nth equation to determine hn+i . This equation is of the form — ^ + hn+i = known material - (77) dc c We solve for hn+i by multiplying with c"""*"^ and integrating. This gives c'^^^n^iic) = I (known material) dc The left-hand side is of such a structure that it must vanish both for c = 0 and c = oo . It follows that the right-hand integral when taken between the limits 0 and oo must equal zero. This condition is indeed obeyed for any ho{c) when n = 0. The integral condition reads in this case iro^.crn(x)sinx.x'^(^Y-r^c^)) etc and finally {c"). Thus we end up with the set of even moments of ho{c) which may be used in succession to determine ho(c) more and more closely. There is no guarantee that this procedure converges mathematically, since the general theorems usually require the knowledge of all integer moments.^ " Shohat, J. A., and J. D. Tamarkin, The Problem of Moments. Am. Math. Soc, 1943. The original three-dimensional formulation appears a little more favorable for a proof because, in this case, we know indeed all integer moments. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 22 1 The justification for the method rests therefore on an empirical basis at this point. Assuming isotropic scattering, as in the "Monte Carlo" calculation we express our results in terms of the dimensionless variable w defined in (75). The equation system (51) becomes then (2. - i)(i - (isAxms, v) = (78) = v{v + s-\- l)(s -l,v-l)-\-{p+ l){v - s){s - 1, 1/ + 1) where the abbreviation (s, p) has been introduced for (ly'P, (cos t>)) and the quantities (/«.f(x)) are simple numbers computable from (48b) and the assumption of isotropic scattering. The first truncated relation is s = J/ = 1. It yields (1, 1> = 2 Reducing it with the relation (78) for which s = 2, j/ = 0 we get (2, 0) = 8 (79) The next truncated relation is s = v = 2^ which yields and the reduction gives Similarly in the next stage (2,2) 16 3 <3, 1> 92 3 = 128 7 <4, 2) = 18112 133 <5, 1) = 421600 399 (6, 0> = 3372800 QQQ (80) (81) As an example of an average which cannot be had explicitly we may take the mean absolute value of the speed, that is (1, 0). We find this 222 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 value by picking a sequence of trial functions for ho(w) with the ap- propriate number of parameters and imposing successively (79), (80) and (81) upon this sequence; this leads us to a sequence of values for (w) which can then be examined. In such a procedure careful consideration of the trial functions is an important element. The following information is available. It was proved in Section IID that ho(w) is logarithmically infinite at the origin. At infinity, on the other hand, ho(w) falls as e""" times some power of w. One way to check this is to drop the terms con- taining 1/c as factor in (46) ; the solution of the recursion system becomes then K(w) ^ {2v + l)e"V where k is some unknown exponent. Armed with this fore-knowledge, we shall use the following sequence of trial function for l%{w) ho(w) = pEi{w) + qKo{w) + re~^ + swKi{w) (82) where Ei{w) = / — Jw U du and Kq(w), Ki{w) are the modified Hankel functions of order zero and 1.^^ We find in zeroth approximation from normalization only (83a) ^ 2 {wY^^ = 2.2500 = s<°> = 0 in first approximation, using (79) ^ =6 « =9 /'> = s<" = 0 (wY" = 2.3818 (83b) 2' This definition, which is in accord with the tables of Jahnke-Emde, differs from the usual one by a factor 2/7r. This change is suggested by Watson, Bessel Functions, p. 79, and proves convenient in the following. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 223 in second approximation, using (79) and (80) p''' = 7 12 e' = 32 45 ,(2) = 1 20 s''' {wf^ = 2.3858 (83c) and in third approximation, using (79), and (80) and (81) p''' 7980 ,^^> _ 202544 209475 /3) 2507 18620 s''' 3152 209475 (wY = 2.3864 (83d) Appearances indicate strongly that the sequence (83) for (w) approaches a limit which one would guess to be (w) = 2.3865 (84) More evidence that the conclusion drawn is correct can be obtained by using the set of trial functions ho{w) = pKo(w) + qe'"" + rwe~'^ We find then the following sequence of values for (w). {wf^ = 2.546 (wf^ = 2.395 {wf^ = 2.388 This descending sequence confirms (84) by approaching this same value from above. Further evidence for the correctness of the procedure can be obtanied by deriving a function hiw) from the Monte Carlo function /i(w) dis- cussed in Section IID and comparing it with our trial function. The function was constructed by covering Fig. 13 with a grid of concentric 224 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 circles and horizontal lines and replacing the integration hoiw) = 2t / h{w) d (cos t?) by a summation over grid points. The function ht(w) so obtained is compared in Table I with ho^^\w)j ho^^\w) and ho^^\w) as defined by (82) and the numbers following. We observe that the first approximation Table I Comparison of the Monte Carlo hoiw) for hoiw) with successive approximations obtained by the new method. w Kiw) h^'Hw) h^'Hw) h^'Hw) 0 00 00 00 0.5 0.74 0.8397 0.7281 0.7144 1 0.29 0.3291 0.3019 0.3002 1.5 0.15 0.1500 0.1438 0.1440 2 0.081 0.0734 0.0730 0.0733 2.5 0.0412 0.0374 0.0384 0.0387 3 0.0199 0.0196 0.0207 0.0208 3.5 0.0118 0.0105 0.0114 0.0114 4 0.0063 0.0057 0.0063 0.0063 4.5 0.0030 0.0031 0.0035 0.0036 5 0.0014 0.0017 0.0020 0.0020 6 0.0004 0.0005 0.0007 0.0006 7 0.0001 0.0002 0.0002 0.0002 is an improvement over the zeroth one, while the second one makes little difference, considering the accuracy to which h'^(w) is given. In indi- vidual cases the sequence drifts away from htiw) ; this is not surprising because the latter function is very rough ; this is to be expected from its mode of derivation. The application of this method to the hard sphere model of ion-atom collisions offers no new feature of principle. The actual working out of results is somewhat more complicated, mainly because the connection diagram for the recursion system (49) is more involved. According to equation (26b) the dimensionless variable to be used in this work is w = \/a\ (85) We denote its averages (w'P, (cos t?)) by {s, v) as previously. The equa- MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS tion system (49) then takes the form (2. + l)(l-(/.,(x)))(s +!,.> = = v(v+ s+ l)(s - 1, ^ - 1) + (^ + l)(s - v){s - 1, V + 1) 225 (86) The numbers {Ig.vix}) were already discussed in connection with the system (78). What distinguishes (86) from (78) is the way in which the variables are connected; the new connection diagram which replaces Fig. 9 is shown in Fig. 14. The truncated relations no longer dovetail into each other as they did before. Only the first stage proceeds in a similar way, yielding explicit expressions for (2, 1) and (4, 0). In the next stage we Fig. 14 — Interconnection established by the Boltzmann equation among the averages {c*P, (cos t?)); case of constant mean free path. 226 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 start out with a relation between (1,1) and (3, 2). By the use of regular recursion formulas we can successively transform this into a relation between (3, 0) and (3, 2), then (3, 0) and (5, 1) and finally between (3, 0) and (7, 0). Here we have for the first time the normal situation in which we do not get the actual value of a moment of ho(c) but only a relation between two or more of such moments; the reason for this is that the system fails to connect up with (0, 0) which equals unity a priori. A similar situation prevails for the next truncated relation; it is originally a relation between (2, 2 ) and (4, 3 ) and is finally reduced to one between (2, 0), (6, 0) and (10, 0). Similarly, the next truncated relation reduces to a relation between (5, 0), (9, 0) and (13, 0) and so forth. The first three of these reduced relations come out to be {w') = 10 (87) S{w') = 112(^') (88) fV) = 27V> + ^^(.'») (89) These formulas will now be imposed upon a sequence of trial functions for ho(w) suitably chosen. Again, we may make use of the information of Section IID, according to which hoiw) is logarithmically singular at the origin. For large w we proceed as previously from (46) leaving off the terms of 1/c. We get then K(w) ^ (2v + l)e-^ " 2 k This suggests the following trial function for }h{w) h{w) = vEiiW) + qKoihiv") + re~" "' + svl'Kiihw') (90) The best zero order approximation is actually obtained by the function K,{W)' We find r =2r^(?i) = ^-^^^^^ ^'°' = = r''' = s''' = 0 In first order we get, using (87) p(^^ = -0.46543 q^'^ = 1.45285 r''' = s''' = 0 In the second order, using (87) and (88) p^'^ = -0.80856 g^'^ = 1.88127 s''' = 0 r^'^ = -0.09804 MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 227 111 third order, using (87) and (88) and (89) p"' = -1.15071 9"> = 2.37034 s"> = 0.02062 r»' = -0.29016 These successive approximations lead to the following sequence for the drift velocity {w cos ^) (w cos r^f^ = 1.04605 (91a) {w cos i^Y'^ = 1.14256 (91b) (w cos t}f^ = 1.14616 (91c) (w cos t}f^ = 1.14661 (91d) We conclude from this sequence that {w cos ^) = 1.1467 (92) In addition to the drift velocity there is some interest in the energy and the energy partition. For the energy the following numbers are obtained {wY^ = 2.1884 (93a) {wY^ = 2.3395 (93b) {wY^ = 2.3511 (93c) (wY' = 2.3531 (93d) giving (w^) = 2.353 (94) A zero order value for {uf^ cos^ ??) cannot be said to exist because the first truncated relation is the condition that a distribution function Jhiw) exists at all. Thus, we can get only three numbers in a sequence approximating {w cos t>) {'i/cos^i^f^ = 1.8005 . (95a) {w' cos' ^f^ = 1.7696 (95b) {w' cos' ^f = 1.7685 (95c) 228 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 giving {w^ cos' t?) = 1.768 (96) We can understand the results (92), (94) and (96) by giving the frac- tion of the total energy in ordered motion and the fraction of the energy in motion along the ^-direction. We find for the first ratio <^ '""if = 0.559 (97) and for the second {^1^0^ = 0.751 (98) The ratio (97) equals 0.5000 for all mean free time models; the ratio (98) is 0.778 for the mean free time model with isotropic scattering. Thus, the deviations from the earlier results are not drastic. However, in certain derived relations the difference is more noticeable. For instance, a good measure of the anistropy of the diffusion process is furnished by the ratio of the random energy along the field to the energy at right angles.^ From (97) and (98) we find for this number {vl^ COS^ d) - {W cos df ^ . - . ,QQS KM - (w' cos2 1?)) ^ ^ For the mean free time case this number equals 2.50. Hershey^ in his work assumes this number to be 1.000. A comprehensive list of velocity averages is attached in Table II. As a comment I may add that the obvious mode of constructing such a table, namely by computing the column v = 0 from (90) and then using the recursion system (86) for the others, runs into some difficulty. First of all, a series of cancellations reduces the accuracy as v increases; finally, at the positions marked "impossible" we find the missing third members of the truncated relations. These elements cannot be com- puted by recursion at all, but would require an explicit solution of the equation system (47) for K+i(c). In the table, this more arduous path is not followed. Instead, the recursion method is used for the numbers in italic type and a few numbers are added by extrapolation. The numbers so obtained will be needed in Section IVB. The calculations on the hard sphere model are immediately applicable to the experimental data of Figs. 3 to 7, which exhibit the drift velocity of " See below, equations (147) and (165) . MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 229 the noble gas ions in the parent gas as functions of the parameter a/N. These data have a high field range in which the drift velocity varies as the square root of a/N. This is the variation for a model with constant mean free path, as seen from the dimensional formula (26b). It was indicated furthermore in the Section lA that we have good reason to think of the scattering between an ion and an atom as nearly isotropic.^' These two features characterize uniquely the hard sphere model whose treatment we have just completed. To the extent that they are verified Table II Dimensionless high-field velocity averages (s, v) for the hard sphere model and mass ratio unity. V 0 1 2 3 4 s 0 1.0000 0.7845 impossible 1 1.S923 1.1467 0.8022 impossible 2 2. 35 84 2.0000 1.4759 0.990 impossible 3 4.5868 3.9853 3.0578 2.134 4 10.0000 8.8353 6.992 5.0602 3.474 5 23.912 21.405 17.330 12.84 6 61.847 55.97 46.177 35.36 7 171.241 156.3 130.91 8 503.7 462.81 9 1563 1445 10 5090.9 4750 the model is applicable to the experimental data. The formula to apply is (92) in combination with (85) : (100) fe) = 1-147 \/ ^ In the logarithmic plot of {cz) vs a/N the intercept of the straight line of slope 1/2 which fits the high field data thus equals 1.147 The values for o- which result from this are shown in Table III. For comparison are shown the corresponding atomic cross section as deter- mined from viscosity data.^* It is interesting to observe that the ratio of 23 A quantitative discussion of this point for the polarization force will follow in Section IIIB. 2* Landolt-Bornstein, 1950 edition, Vol. I, part 1, page 325. 230 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 the two retains very nearly the constant value 3 throughout the table. The fact that the ratio is substantially larger than unity is explained by. the resonance feature of the ion-atom scattering process as discussed in Section lA. The fact that it is constant is perhaps an indication of the fact that both processes are governed by overlap conditions of essentially the same wave functions. I would like to point out in connection with the calculations of this section that the method developed is potentially of very wide applica- tion. One question that comes up, for instance, is whether a careful kinetics calculation is necessarily restricted to certain models or whether an ion-atom cross section known numerically could be used to derive therefrom kinetic properties. This is indeed possible. Suppose, for instance, that the cross section (t{c) were available as a function of c for collision of He "'"-ions and He-atoms and suppose that this cross section were to satisfy the condition of isotropy 11 (x) = 1 to a good approxi- Table III Cross sections for ion-atom and atom-atom collisions for the noble gases. Gas ion-atom cross section X lO" cm^ atom-atom cross section X W^ cm* He 54 15.0 Ne 65 21.0 A 134 42.0 Kr 157 49 Xe 192 67 mation; we may then derive for this eventuality conditions on ho(c) which are more general, respectively, than (79) or (87), (80) or (88), (81) or (89). Since we are outrunning here the experimental evidence we shall limit ourselves to the derivation of the first of these relations. The first truncated relation is exactly (50a) which, for isotropy and equal masses, reads / c.\_ \aT{c)/ = 2 (101) The reduction of this formula to a condition on /io(c) requires the rela- tionship 1/ == 0 of the set (46). This relation is always integrable to yield hi{c) in terms of /io(c), as was pointed out early in this section. For the special circumstances assumed the integrated equation equation (74) IS h{c) = 3 I ar(7) dy (102) MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 231 The elimination of hi(c) is achieved by forming the average (101) on the function (102). This leaves the required condition on ho{c); it may be given the following form (^/\VW<^.)=.2|, (103) The equations (79) and (87) are manifestly special cases of this more general relation. Adaptations of this procedure to other cases are clearly possible whenever the need arises. The calculations of this section are meant to suggest that it is possible to compute reliably average values from a Boltzmann equation without solving it completely. The method employed here for this purpose resembles a Ritz method in that it works with trial functions which must be guessed at, and like that method it is capable of indefinite improve- ment. The numerical results suggest strongly that we are converging toward a definite answer; however, a mathematical proof of this fact has not been presented. The method will be applied once more in the section on diffusion. Part III — Motion of Uniform Ion Streams in Intermediate Fields IIIA. A convolution THEOREM Wlienever we deal with the motion of a given type of charged particle in a gas of given composition, then there exists a wide range of densities n and N as discussed in Section lA in which the motion of these particles depends only on a/N and kT. For this range the motion is governed by equation (13). Since deriving that equation, all our efforts were dealing with the "high field" equation (34) or (40), in which the gas temperature is taken to be zero and the electric field often scales out, as in (26), (75) and (85). The accomplished solution of this restricted problem, together with the low field solutions available in the literature, brings us back to the more general equation (13) and the question what can be done with it. The topic of Part III so defined is definitely inferior in importance to the one in Part II. For we are studying here an intermediate range of variables which can be handled qualitatively, both in concept and practice, by some sort of interpolation between the high and low field regimes. For precise measurements, conditions can always be chosen so as to satisfy one or the other of the two extremes. For this reason the intermediate field case will only be pushed as far as it will go con- veniently, without appeal to numerical methods. In this Section IIIA we shall give a complete solution of the inter- 232 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 mediate field problem for the mean free time models discussed in Section IIB. This solution is achieved by the following theorem: Given the general equation {IS) for constant mean free time "'■ ^ + ^^""^ ^hrH ^(c')/(c')n(x) da,, dc (104) and the '^ high field'* equation derived from it by setting the gas temperature equal zero dh(c) dc. m y + h{c) ^ ^If ^(C')/i(c')n(x) dQy dC (105) and the Moxwellian equation derived from {104) ^V dropping the field term (c) = i- ^ ilf (COm(c')n(x) d^y dC (106) then the solution /(c) of {104) ^s Ihe convolution of the solution h{c) of (105) and the solution m{c) of {106) : /(c) = f h{u)m{c - u) du (107) We carry through the proof by constructing explicitly the equation satisfied by the convolution. We replace the running variables c, c', C, C' in (105) by u, u', U, U' and multiply in m(c — u). We get aT ^ m{c — u) 4- h{u)m{c — u) = dUz ^^11 ^(U')^(^')^(c - ^)n(xu) ^fl,' dJj We now define f{c) by the relation (107), and integrate the above equation over u. The second member on the left comes out to be /(c). For the first member, we carry out an integration by parts : f dh{u) f . , { r.( \ dm{c - u) , / -~-^ m{c - u) rfu = - / h{vL) — ^ du J OUz J dUz d{m{c - u)) ^^ dCz = +fh{n) = — / /i(u)m(c — u) du __ df{c) dc» MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 233 For the right hand member we observe that we have the eightfold integration that is an integration over the collision angles and all final velocity components. By a general principle of kinetic theory ^^ we can invert in this integration the final and the initial quantities and write dfi,^ dU du = d^ dU' du (108) This puts us in a position to eliminate the 5-f unction by integration. We find '''' ^ "^ ^^""^ ^hll ^(^')^^^ - ^)n(xj dn, du' (109) with the side condition that u, U, u', U' form a quadruple of vectors in the sense discussed in Section IB for which in addition U' = 0 If we substitute (107) into (104), denoting the dummy variable by u' instead of u, then the two equations (104) and (109) take on a very similar look. A proof of their identity hinges upon proving the identity of the integral terms : j hiu') dn' J m(c - u)n(xn) dQ, (110) = j h{n') du jj M(C V(c' - u )n(xc) do,' dC The form of this relation suggests the assumption that the expressions are identical before integration over u; this assumption is proved by the events below. The complicated function h(u) thus disappears from the problem. The other such function, namely 11 (x) disappears then also; for it is by assumption arbitrary, hence could be replaced by a 5-function for a fixed, but arbitrary x- The two sides of (110) must therefore be equal before we integrate over Xu or Xe , and the two x's are to be taken equal and fixed. Defining angles as shown in the spherical diagram Fig. 15 we thus get (110) in the form I m(c -u)d€^ jl MiC')m{c - u') d dC (Ilia) 2' See Reference 4, Section 3.52. 234 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 This is to be true with the side conditions c = fixed (lUb) u = fixed (111c) u' = 0 (Hid) = Xu = X = fixed (llle) Equation (111) is an identity involving only elementary functions. Thus the relation itself is in a sense elementary. Those who wish to believe it, may consider the theorem proved ; for completeness, however, the proof of (111) will now follow. Call the left side of (11 la) X, the right side Y. To determine X, we substitute from (7) and (Hid) u = m with M -\- m u' + M M -\- m 2 /2 t] = U Fig. 15 — Definition of the angles occurring in the proof of the convolution theorem. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 235 because of (9). This yields with the angles as shown on Fig. 15 X = r m{c - n)de = (^^' ' exp ["- ^mc' ,2 M^ -\- rn -{- 2Mm cos x , o/^^../ ^ + ^^ ^'os x ^_ ,1 — amu {M + m)' + 2^mcu M + m COS ^ f exp 2^ cu sin X sin ^ cos e de The integral is evaluated by a formula known from the theory of Bessel functions and yields v3/2 X = —7= (8mr" exp - /3mc - /3mi^ ,2 AT ^ + m^ + 2Mm cos x (112) (M + m)' , ^^ , m + M cos X , + 2Bmcu Tir— cos i/^ M + m ^ / 2^Mm , . .A • ^0 ( ir^—; ^^ si^ X sm ^ 1 (113) \M + m Passing now to the right hand side of (11 la) we may replace in the first place dC by dy, because of (111b). c' and C' are then replaced by the expressions c = c — C' = c M ^ M , Y + .r ■ Y ilf + m M -\- m M M Y - M + m ' M + m With the angles defined in Fig. 15, we thus get for Y Y = (^)" (^)" exp [- ^Mc^ - ^mic - uTl • ffffl^ dy sin e dB db d4> expf- jSMt' + ^McT (cos i/^ cos ^ + sin yj/ sin ^ cos 5) 2^ mM M -\- m uy (cos = 5Si.t)t)C) (116) The second integral is a thermal average, the first a high field average computable by the method of Section IIB. Thus the average (116) is a finite sum of products of computable averages and is itself computable. When formula (116) is applied to the averages (52), (53), (54), (57) and (59) very simple results are found because of the symmetry of the function m(v). For the drift velocity (cz) we get from (52) This is the same formula as (52) which is thus proved to hold inde- pendently of the gas temperature. In the energy formulas we find simple addition of the thermal and high field values because the middle term in (116) drops out by symmetry. Inserting (53), (54), (57) and (59) we find {mcl) = kT+ j-^ ^^-T— r (119) 2 /3M sin' X + 4ot(1 - cos x)\/l^^c^x\ \ ar 238 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 {rncz) — miczY = kT + (mcl) = kT + ^^ + "^^ \ ar I (120) ^ /SM sin^ X + 4m(l — cos x)\/l — cos xV \ ar /\ ar I ^ /3M sin^ X + 4m(l - cos x)Vl - cos xV \ aT l\ ar I The interpretation of these formulas is implicit in the discussion of the high field formulas given earlier. In particular the combination of the equations (55) and (116) can be given the elegant form m(c') = Miff) + m(c.)' + ilf(c.)' (122) It states that the energy of an ion is obtained by adding the energy of a gas molecule, the energy visible in the drift motion and a storage term which is M/m times the energy in the drift motion; this term becomes important for electrons in a gas. A low field approximation to this for- mula (in which the second term on the right may be neglected) has been published in the article of Kihara.^^ IIIB. RESULTS FOR THE POLARIZATION FORCE AND THE ISOTROPIC "MAXWELLIAN" MODEL The polarization force between ions and molecules which predominates over other forces at sufficiently low temperature satisfies the mean free time requirement of the preceding section. It follows that the complete theory given for those conditions applies to this force. The magnitude of force was given in (4). Its potential equals 1 eF F = i ^ (123) Classical theory is usually applicable to the scattering by the potential (123) because angular momentum quantum numbers run as high as 30 or 50 in normal situations.^^ This classical type theory, first developed by Langevin,^ follows standard elementary methods for computing the " Reference 27, formula 5.12. " Holstein, Theodore, private communication, see also Reference 11. " Langevin, Ann. de Chim. et do Phys., 6, p. 245, 1905. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 239 angle of deflection x due to a potential of the type (123). The result is du TT - 2 / /I _ e'PjM + m) 4V^' (124) Here b is the ''impact parameter", and Ui is the lower of the two positive roots of the polynomial in the denominator; if the polynomial has no real root, the integration goes from 0 to oo . The question whether the denominator has a real root or not is tied up with the nature of the orbit. If b is sufficiently large a root exists and the orbit looks like a hyperbola, Fig. 2(a); for small b no root exists and the two particles are ''sucked" towards each other in a spiralling orbit as shown in Fig. 2(b). The two regimes are separated by a limiting orbit in which the particles spiral asymptotically into a circular orbit. This limiting orbit is found by setting the discriminant of the square root in (124) equal to 0. We find From this value of bum a cross section and a mean free time r« for spiralling collisions can be derived. We find "' = 2^^ t(K+1)p} ^^^^^ This is indeed a constant mean free time as stated, the speed of encounter 7 having dropped out. 1/r is the dimensional quantity entering into the averages {{x)b db [' (k Jo Jo = irNyb'u^ f vix) d{f) Jo From (125) and (126) the factor in front of the integral just equals I/ts ; the integral on the other hand is a computable pure number independent of 7 which is obtained by inserting into it the relationship (127) between X and jS. Hence we may write /^\ = 1 f " ^(x) di^r (128) \ r / Ts Jo The three equations (126), (127) and (128) completely define the nature of the averages appearing in previous sections. The integral (128) has to be computed by numerical methods. It is seen in the course of the evaluations that it naturally decomposes into two parts. The part for which ^ varies from 0 to 1 deals with spiralling collisions and exists for any ^(x). For /? between 1 and oo we get the contribution of the hyper- bolic collisions to the average. This part is only finite if ^(x) vanishes for small angle deflections. The averages (52), (53), (54), (57) and (59), as well as (117) to (121) contain numerous averages of the form (128) all of which satisfy the predicted condition <^(0) = 0. They are obtained by linear combination of two basic types: ((1 — cos x)/t) and (sin^ x/r). The first average is given in Hass^.^ Separating the parts due to spiralling and hyperbolic collisions we find Jo cosx)^(/3') = 0.8979 J (1 - cos x) d(^^) = 0.2073 This combines to give ^l_-^^ = i. 1.1052 (129) MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 241 The analogous result for sin x was obtained by the computing group of Bell Telephone Laboratories f sin' X d(^) = 0.511 Jo / sin' X d(jf) = 0.261 which gives /sin' X^ _ 1 \~7-/ = --0.772 (130) Ta We may now rewrite the major results of Section III A for the polariza- tion force. From equation (117) we get , . _ 0.9048 /I ^ 1 E (131) This formula may be found in the literature.^^ What is new about (131) is the realization that it is exact at high as well as low electric field. The formula for the total energy needs no discussion for a special model ; it does not involve the angular distribution law when written in the form (122). Thus we would obtain, for instance, for an ion travelling in the parent gas that its total energy is obtained by doubling its ap- parent energy observable in the drift and adding to this the thermal energy %kT, For the partition of the high field component of the energy in the three coordinate directions we have two formulas, formula (58) parti- tions the entire field contribution of the kinetic energy, formula (60) only its random component. The first formula gives e,:ey:e^ = M:M:{M + 6.73m) (132) Formula (60) gives e,:ey:e* = (ikf + m): (M + m): (M + 3.72m) (133) It is convenient to apply the general formulas also to the case of con- stant mean free time, coupled with the assumption of isotropic scattering. This combination of assumptions represents, strictly speaking, an im- " The formula is equation (3), p. 39 of Reference 2, in the limit X = 0; or also the last unnumbered equation on p. 919 of Reference 6. 242 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 possibility; for we know of no mechanical force which realizes this ar- rangement. This model was already taken as the basis of the Monte Carlo calculation in Section IID. It will be seen now that it has a wider sig- nificance than one might anticipate. The necessary angular averages are (1 - cos x> = 1 (134) (sin' x) = M ■ (135) This yields for (117) , , M + m , . \Cz} = — -^ — ar (136) As usual, the formula for the energy does not involve the law of scattering if written in the form (122). If we choose the form (118) instead we get in agreement with (79) W) = 2kT + ^^^^ aV (137) The partition formula (58) becomes e^:ey:e^ = M:M:(M + 6m) (138) the partition formula (60) which counts random energy only becomes e.ieyiet = (M + w):(ilf + m):(ilf + 4m) (139) Comparison of these expressions with the ones for the polarization force shows that the difference between it and the isotropic model is remarkably small from a kinetic standpoint. We may see this by com- paring (132) and (138) or (133) and (139). For the other formulas, we may compare more specifically the polarization results with an isotropic case having its mean free time r given by T = 0.9048 Ts (140) Equation (136) becomes then identical with (131) and because of (122) the same identity persists for the energy formula (137). In the light of this we may say that it is very nearly correct to state that scattering is isotropic for the polarization force. This qualitatively correct fact was repeatedly made use of in the preceding sections of the paper. The reason for it is chiefly the predominant effect of spiralling collisions. Indeed, equation (140) shows that a modification of Ts by only 10 per cent takes into account the main influence of hyperbolic collisions. From the discussion in Section lA it may be seen that the results MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 24J} obtained for the polarization force have a potentially wide field of ap- plication when measurements of ion drift are extended to low tempera- ture. In the meantime, the results apply occasionally at room temperature, whenever we deal with a small ion and are not bothered by special scattering mechanisms having large cross section. An example of this are the molecular noble gas ions in the parent gas whose drift velocities were measured by Hornbeck^^' ^^ and Varney.^^ Table IV shows the measured mobility at standard gas density measured for these ions, in comparison with a value obtained from equation (131). The field range from which the observed mobility was obtained is intermediate. There is not only good numerical agreement, but the experiments follow the theory also in that there is little variation of the observed value Table IV Mobilities at standard density of the noble gas molecular ions. Comparison of the experiment with a formula based on the polarization force only. Gas cm2 ^*'^^- Volt sec cm2 '^^'^ Volt sec He 18 18.2 Ne 6.5 6.21 A 1.9 2.09 Kr 1.2 1.18 Xe 0.7 0.74 with the field. The discrepancy between the two columns can be used to determine a hard collision cross section which is to be superimposed on the polarization force, as is suggested in the so-called Langevin model. inc. VELOCITY DISTRIBUTION FUNCTION FOU ELECTRONS We have almost exhausted the results achieved for intermediate field conditions. For the sake of completeness I shall mention shortly the intermediate field distribution function for electrons whose derivation we owe to the ingenuity of Davydov. The derivation does not differ in principle from the one presented in Section IIC for the electrons in the high field case. The distribution function is first expanded in spherical harmonics. For group theoretical reasons the scattering term in the Boltzmann equation is diagonal in 32 Davydov, B., Phys. Zeits. Sowjetunion., 8, p. 59, 1935. See also Reference 4, pp. 349-350. 244 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 such a decomposition even in the presence of molecular agitation. Thence a generalized form of (47) may be derived containing essentially the same terms. Finally, all but the first two spherical harmonics are dropped and two equations analogous to (63) and (64) are obtained. In fact, it is found that equation (63) is maintained entirely. An extremely compli- cated reasoning is required, on the other hand, to find the generalization of (64). The result is /i(c) = SkT dfo Combining (63) and (141) we find (141) ' I , 3fcr\ dfo m /I - cos x\ + W]dE^^M'^'''^ ,\ ar(c) / / and hence /o(c) = exp — m / c dc M /I — cos x\ \ aric) I + fcT (142) This is the so-called Davydov distribution which is a generalization containing within itself the Maxwellian distribution as well as the high field distribution (65). The mean energy and the drift velocity of electrons may be calculated from (63) and (142). They are obtainable from the literature and will not be discussed here any further. Equipartition of the energy exists at all field conditions. Part IV — Diffusive Motion of Ions IVA. diffusion for mean free time models It was proved in Section IC that if there are spatial inequalities in the distribution of the charge carriers then a smoothing out process sets in which can be described as diffusion. This derivation of principle can be supplemented for "Maxwellian" molecules by an explicit computation of the two components of the tensor (24), that is an evaluation of the integral (23). We shall do this by following the method of Maxwell^* MOTION OF GASEOUS IONS IN STKONG ELECTRIC FIELDS 245 rather than by generalizing the formal procedure of the Sections IIA, IIB and IIIA. Such a generalization would no doubt be possible, but would increase unduly the bulk of this paper. We shall operate therefore directly on equation (20). To get out the integral (23) we multiply the equation vectorially with c and integrate over dc. This operation makes the first term vanish completely. This is obvious from symmetry for the components Cx and Cy of the multiplier c. For Ct we have •/ dCz An integration by parts brings this in the form (18) and thus makes it equal to zero. Temporarily, we may break the integral term of (20) into two parts, using some artificial procedure to eliminate small angle collisions. The first half of the integral term reads then simply J /.(c) c do, This is already the desired average (23). On the second half we use the identity (108) to give it the form — Ij M{C')g{c)cU{x) dQy dC' dc 47rT We now use (7) to replace c by the expression m t , M ^, , M c = c' + ^^^^- C + M + m M -\- m M + m Only Y is affected by the integration over dQy which we take up first. Using y as the axis of a polar coordinate system we may write r = Til + r-L For every value of x, Yi i has the fixed value y cos x- On the other hand the average of yj. vanishes through integration over all azimuths. Thence we may write -L f cn(x) d^y = -^ c' + -^ C + ^^ (c' - C) (cos x) 47rJ ^ M + m M -\- m M -{- m ^ m + M (cos x) / , M(l - cos x) ^/ M -{- m ^ M + m We now multiply with M(C')g{c) and integrate over dc dC' . The inte- gration of the term containing C obviously vanishes for two independent reasons. The integration of the term in c', finally, yields again the average 246 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 (23) . Combining the two pieces, we find M /I — cos x\ M -\- m\ T / /.(c) cdc In this expression, the artificial exclusion of small angle scattering is no longer necessary and can be dropped. Completing the integrating of equation (20) we see that the right hand side gives averages over the unperturbed velocity distribution /(c). Combining pieces, using (23) and indices 1, 2, 3 for the x, y and z components we get M -{- m ji(r, t) = -n(r, t) J^ k, M /I cosx\ ; / l(CiC,) — {Ci){Cy)] (143) According to (16) and (24), the square bracket in (143) is the diffusion tensor. It has two distinct components which equal respectively i), D^ = M M -\- n \~ m - cos x\ r / M /I \" - COS x\ T / (144) (145) The velocity averages entering are (120), and (121), that is the directional components of the random part of the energy. Substituting we get finally {M + m)kT Du = Mm /I - cos x\ V "/ (M + m)'/^^^H^±Ml + a' cos x)\ / (146) ^,^ /3M sin^ X + 4ot(1 - cos x)\/l - cos xV D^ = \ (M + m)kT /\ Mm /I cos x\ ; / + a= (M + m)*(?i^) (147) Mhn /3itf sin' x + 4m(l - cos x)\/l \ r /\ cos xV r / The diffusion coefficients have the simple property that they are ob- tained by adding the low field and the high field limiting expressions. MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 247 This is a consequence of the limited form of the convolution theorem proved in Section III A; it probably implies also that the theorem can be extended in some form to include the case of diffusion. It has been mentioned in the Section ID that the Nernst-Townsend relation (30) applies only to ions moving in a low field. We are now in a position to examine possible extensions of it to general fields. Equations (144), (145) and (117) suggest the form Dn _ 2 X mean random energy along n (\aq\ mobility e where n stands for one of the principal directions of the diffusion tensor. This formula contains equation (30) as a specialization to the low field case. Formula (148) is one of the formulas obtained in this study of ion motion in which model parameters do not appear. It is valid (a) for all interactions at low field and (b) for the mean free time case at all fields. It also holds dimensionally at high field for models obeying (25); this may be seen from (26a) and (28a). It appears a reasonable conjecture that (148) is approximately true for any law of interaction; the question will be taken up again in the next section. Let us, in conclusion, write down the formulas resulting from (146) and (147) for the two special mean free time models studied in detail in Section IIIB: the polarization force and the isotropic model. The necessary averages are (129), (130), (134) and (135). They yield for the polarization force Z) = ^^-i^0.905T«./cT " Mm 1 {M + m)\M + 3.72m) . (149) ^ ^ M + m 0905^^.j^y + 1 (^+7)' ^ a^(0.905r.)' (150) Mm 3 M^m{M + 1.908m) and for the case of isotropic scattering M + m , ^ , 1 (M + mf{M + 4m) ^"=nJf^^'^^ + 3 MMM + 2m) "' ^^^^^ M + m 1 {M + mY 23 ,,.0) Just as in the earlier study the results for the two models do not differ appreciably. 248 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 IVB. LONGITUDINAL DIFFUSION FOR THE HARD SPHERE MODEL Whenever the mean free time condition for collisions is not fulfilled, then the computation of diffusion coefficients requires a procedure analo- gous to that of Section HE. Since this entails some numerical work the calculation was only carried out for a case which was thought to be of experimental interest, namely for longitudinal diffusion of ions in the parent gas. In other words, we are extending the numerical computation at the end of Section HE to include longitudinal diffusion. The computa- tion to provide us with the undetermined constant of equation (28b) for the special case when m and M are equal ; it also offers, incidentally, a good test case for applying the method of Section HE outside the area for which it was designed originally. Since the equation is only to be solved in the high field case we may apply to (20) the reduction method of Section HA. If we introduce also the specialization warranted by the hard sphere model and unit mass ratio then, in analogy to equation (40), we get the following starting equation dg{w) , , . 1 f" w'^ sin x dx , , V 1 r wsmxdx f f f^ . dWz 47r Jo w^ Jo (153) = —\k{wz — {wz)}h(w) Here the dimensionless variable w defined by (85) has been employed instead of c. Equation (153) is the fundamental equation of our problem; it is an inhomogeneous version of equation (40). We solve the equation in the same way as we did previously, namely by decomposing ^(w) into spheri- cal harmonics and forming moments. In other words we follow step by step the procedure of Section HA, the only difference being the presence of an inhomogeneous term. We shall not enumerate all these steps again. We shall only note in passing the inhomogeneous form of (47) which is 74 2 Jo — g,{w')Py (cos k) sin xdx - wg^w) dgy-i 2v — 1 \ dw w g^iiw) > w J V -f 1 jdg^i ,y-{-2 . S\ MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 249 Having introduced moments in the manner described earlier we arrive at the inhomogeneous version of (49) or (86) Hs + ^ + l)ls - 1, V - 1} + (y + l)(s - y){s - 1, ^ + 1} -(2. + 1)(1 - {Is,.)){s +l,v] = -{2v + 1)(1, l)(s, v) (154) + v{s + 1, V - 1) + (v + l){s + 1, ^ + 1> Here the curly brackets {s, v} are normalized moments over ^(w) defined as follows {s,v} = ^ / qMw'P, (cos t?) dvf (155) The equations (154) show that the quantities {s, v] are numbers, the variable density gradient nk having been eliminated by the definition (155). The system does permit that arbitrary amounts of the pointed averages be added to the curly ones. This indeterminacy is removed by the supplementary condition (18) which, in the present notation reads {0, 0) =0 (156a) The connectivity of the equation system (154) is the same as that of (86). Hence it will have the same properties as that earlier system. We may, therefore, reduce it in the manner followed previously and get inhomogeneous versions of the equations (87), (88) and (89). They read {4,01 = -|<4,1) + |(1, 1X3,0). (157a) -f (2,0) -^(2,0) +10(1,1)^ 112{3, 01 - 3!7, 01 = 4(7, 1) - 4(1, 1)(6, 0) + f (5,0) + il?(5,2) -^-|?(1,1)(4,1) _1^^3,l) + ^^3,3) (^^«^) + *|?(1,1)(2,0)-^(1,1)(2,2) 250 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 54!2,0! - f {6,0! + j| !10,01 = -^ (10, 1) + ^^(1, 1X9,0) + 6{6, 1) - ^ (6, 3) - ^ (1, 1)(5, 0) + 7<1, 1X5, 2> ^^^^^^ + ^ <4, 0) + 15? (4, 2) - 5 (4, 4) -1^(1, 1X3, 1)4-^(1, 1X3, 3> The pointed averages over the distribution h{w) may be found in Table II. Substituting them we get {0, 0) = 0 (156b) {4,0) = -10.494 (157b) 112{3, 0) - 3{7, 0) = 647.8 (158b) 2Q^ 17 54{2, 0} - ^ {6, 0) + ^ {10, 0) = -566.4 (159b) The form (90) that was assumed for ho(w) will again be taken for go{w) with new undetermined coefficients p, q, r, s and a factor k\ evident from (153) or (155): Qo (w) = k\[pEiiW) + qKoiiw') + re"^"' + sw'Kidw')] (160) This is a rather poor assumption because the form (90) was adopted for hoiw) after an extensive study of the properties of the distribution function /i(w). For ^(w) we know little beyond the fact that it is some kind of distorted p-type function. The go{w) derived from this is not likely to resemble hoiw) very closely. Thus the choice (160) is mainly based on ignorance and convenience; this explains the slower convergence observed here than in (91), (93) and (95). To start with, the zero order is completely lost because (156) yields a zero coefficient. We find in first order, using (156) and (157) p''' = 4.8842 (1) (1) _ q'^' = -4.2689 r =s -0 MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 251 in second order, using (156), (157) and (158) p(2) ^ -10.542 g(2) _ +14.993 5^2) _ Q /^^ = -4.408 in third order, using (156), (157), (158) and (159) p(3) = -0.8710 q^'^ = +1.1754 r^'^ = +1.0140 s^'^ = -0.5809 The longitudinal diffusion coefficient results from these numbers by the use of (23), (24) and (160). With the notation (155) the formula becomes £>,, = -a^^V^{l, 1} (161) The formula (154) yielding {1, 1) from go(w) is s = 2, v = 0 {1,1} = m, 0) + 4(3, 1) - i(l, 1)(2, 0) (162a) or numerically from the Table II {1, 1) = i{3, 0) + 0.6433 (162b) The result is {1, 1}^'^ = -0.3695 (163a) {1, 1)^'^ = -0.2075 (163b) {1, 1)^'^ = -0.2198 (163c) The numbers do not extrapolate too reliably but one would guess that {1, 1) = -0.22 is essentially correct. Hence we have i),l = 0.22a" V (164) In order to gain an appreciation of the value obtained it is worthwhile to compare it with the value that would have been predicted from the generalized Nernst-Townsend relation (148). The mobility concept is ambiguous for all but the cases discussed then. It would seem that the appropriate concept here is the differential mobility because comparison 252 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 is made between a small density gradient and a small change in the applied field. Thus we would interpret (148) to mean Du^^-^l{c.')-{c,f] (165a) da which, with (85), (92) and (96), becomes Du ^ 0,26a"V^ (165b) The error of formula (165) is thus 18 per cent, when compared to (164). Part V — Concluding Observations The present article is supposed to contain the essentials of a kinetic theory of charged particles moving through a gas in the presence of an intermediate or high electric field. An effort was made to make the theory general, yet many irksome restrictions will become apparent to those who will try to apply it to their particular problem. Especially those who have in mind application to electrons will find the article unsatisfactory. It is true that many sections leave the masses variable; however, the assumption of elastic collisions, which is made throughout, is almost fatal to all but the most elementary applications. Thus most of the material is slanted for ions. Within this domain, numerous awkward restrictions are still found here. The most important ones are pre- sumably the restriction to D.C. conditions, the assumption of ''low" ion density, and the omission of all magnetic effects. It is my general impression, which I gained from the convolution theorem Section IIIA and which is confirmed by a recent publication^^ that much can be done to remove these three restrictions provided the mean free time assump- tion is made for collisions. To many the adoption of the mean free time condition will in itself appear an awkward restriction. In a rigorous sense this is true, and calculations are made in this article for the more ap- propriate hard sphere model when quantitative comparison with experi- ment is contemplated (equations (100) and (164)). Indications are even given for a treatment which dispenses altogether with the use of models (equation (103)). However, for rapid advance and easy handling, the mean free time assumption does appear essential. It is therefore im- portant to point out that in a wider semiquantitative sense, the use of this model is no barrier to application. In other words, there is in the mean free time formulas information which suggests a wider validity. This is particularly true for equations which do not contain model parameters, such as (55), (56), (122) and (148). Even formulas which MOTION OF GASEOUS IONS IN STRONG ELECTRIC FIELDS 253 do contain the mean free time yield to judicious treatment. For example, we have the hard sphere formula (100) for the drift velocity of an ion. This formula happens to be limited to the high field case and mass ratio unity. On the other hand we have formula (136) which holds for all fields and all mass ratios, but assumes constant mean free time. We now adopt this formula as a general guess for the hard sphere model, interpreting r as previously as the mean free time between collisions; this quantity is now no longer a constant, but should be taken as " = VWTW) ^^^^^ The denominator is the root mean square relative velocity which is familiar from other applications. The interpretation (166) yields a tenta- tive formula for the drift for all mass ratios and for all fields. Specializing to the high field case, we may neglect {C ) in (166) and then substitute for (c) from (55). This yields the high field formula This is indeed a very successful formula. For ions in the parent gas it differs from (100) by only 4 per cent. For electrons it checks the result of Dru3rvesteyn^ to within 12 per cent. Finally, for heavy ions in a light gas, we find exact agreement with equation (71). As a second specializa- tion we msiy apply (166) to the low field case. We must then set and get from (136) and (166) 1/1 IV'' eE\ ,_„, All dimensional factors in this formula are correct. Numerically (168) is somewhat inferior to (167) ; for the factor differs from the correct one by 20 per cent. Nevertheless, the combination of (136) and (166) gives results which are semiquantitatively correct in all relevant limiting cases. This makes it a reliable interpolation formula for intermediate field con- ditions; for this case (c) would have to be substituted from (122) and the resultant quadratic equation solved for (cz). From the examples given we may conclude that the mean free time formulas contain in essence information applicable to other types of elastic scattering. 3^ See Reference 2, page 40, second equation. 254 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Part VI — Acknowledgements This article is the outcome of years of fruitful cooperation with the gas discharge group of Bell Telephone Laboratories, and Dr. J. A. Horn- beck in particular. At one stage of the work I enjoyed the stimulation of Dr. R. W. Hamming who is responsible for all the details of the Monte Carlo calculation. Further acknowledgements are due to Miss C. L. Froelich who carried out the computation of the number in (130), and Miss M. Murray who carried a good share of the burden in the prepara- tion of the manuscript. Finally, I express my thanks to Dr. K. G. McKay for his critical perusal of the manuscript. Abstracts of Bell System Technical Papers* Not Published in This Journal Experimental Verification of the Theory of Laminated Conductors. H. S. Black^ C. O. Mallinckrodt^ and S. P. Morgan^ LR.E., Proc, 40, pp. 902-905, August, 1952. Clogston has discovered that if a conductor is properly laminated, there exists a particular phase velocity along the conductor for maximum penetration of the fields and minimum loss due to skin effect. An experimental coaxial Hne was con- structed whose center conductor was laminated and whose phase velocity could be varied by changing the dielectric constant of the main dielectric. As predicted by theory, the measured attenuation was critically dependent upon phase velo- city. With optimum phase velocity the attenuation, though greater than pre- dicted by theory, was less than that of a conventional coaxial cable of the same dimensions and same main dielectric. A theoretical analysis of the experimental laminated conductor is described in an append x. ASTM Standards — Their Effect on Plastics Technology. R. Burns^ A.S.T.M. Bull, No. 183, pp. 78-80, July, 1952. Typical Block Diagrams for a Transistor Digital Computer. J. H. Felker^ A.I.E.E., Trans., Commun. & Electronics Sect., No. 1, pp. 175-182, July, 1952. The first electric digital computers were built around the properties of relays. The superior speed capabilities of vacuum tubes has led in recent years to their use in new computer designs to replace relays. Because of the small size, low power consumption, and expected long life of transistors, it now appears that the transistor will replace the vacuum tube as a computer element. This paper presents a study of binary computer functions with recommended mechaniza- tions that were selected because they appeared to be readil}^ attainable with transistors now under development. Block diagrams are presented of switches, memory units, arithmetic units, and other basic components. Estimates are given for the number of parts required in the units. It is concluded that a high- performance all-semiconductor computer can be built with germanium diodes and transistors. * Certain of these papers are available as Bell System Monographs and may be obtained on request to the Publication Department, Bell Telephone Labora- tories, Inc., 463 West Street, New York 14, N. Y. For papers available in this form, the monograph number is given in parentheses following the date of pub- lication, and this number should be given in all requests. ^ Bell Telephone Laboratories. ^ Hughes Aircraft Company, Culver City, California 255 256 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 A Broad-Band Interdigital Circuit for Use in Traveling-Wave-Type Amplifiers. R. C. Fletcher^ I.R.E., Proc., 40, pp. 951-958, August, 1952. Because of its high power-handling capacity, the interdigital circuit has been considered for use in traveUng-wave-type amplifiers. An analysis is presented here which indicates that this type of circuit can be arranged to give constant phase velocity over a wide bandwidth (30 per cent), as required to give constant gain. The analysis is qualitatively checked experimentally. The impedance parameter (proportional to the cube of the gain in db) is approximately the same as the flattened helix (such as has been used for the magnetron amphfier) and about one-third that of the conventional circular heUx. Copper as an Acceptor Element in Germanium. C. S. Fuller^ and J. D. Struthersi. Phys. Rev., 87, pp. 526-527, August 1, 1952. Properties of Thermally Produced Acceptors in Germanium. C. S. Fuller^, H. C. Theuerer^ and W. Van Roosbroeck^ Phys. Rev., 85, pp: 678-679, February 15, 1952. Initial Permeability and Related Losses in Ferrites. J. K. Galt^ Ceramic Age, 60, pp. 29-33, August, 1952. Three-Phase Power From Single-Phase Source. A. L. Holcomb^, S.M.P.T.E., JL, 59, pp. 32-39, July, 1952. Described is the development of a nonrotating device for the conversion of single-phase 115-volt power to a three-phase 230-volt form for the synchronous operation of cameras, sound recorders and other film puUing mechanisms asso- ciated with production of motion pictures. Dominant Wave Transmission Characteristics of a Multimode Round Waveguide. A. P. Kinqi. I.R.E., Proc, 40, pp. 966-969, August, 1952. This paper presents some dominant wave transmission characteristics of multimode round waveguide lines in the 4-kmc range of frequencies. The use of such waveguide lines offers the advantages of lower transmsision losses than ob- tainable with single-mode rectangular waveguide, and relative ease of making good joints. Possible mode conversion effects, including dominant mode eUiptical polarization, have been examined and found to be innocuous. As a result, cross- polarized dominant waves can be used to provide two reasonably independent signahng channels at the same frequency in one pipe. The experimental results obtained with a straight line 2.812-inch inside diameter and length of 150 feet are given. Superconductivity in the Cohalt-Silicon System. B. T. Matthias^ Letter to the Editor. Phys. Rev., 87, p. 380, July 15, 1952. * Bell Telephone Laboratories • Westrex ABSTRACTS OF TECHNICAL ARTICLES 257 Simple Phase-Angle Measurement Technique. J. A. Rudisill, Jr.'. Electronics, 25, pp. 228, 232, 236, September, 1952. Solid State Physics in Electronics and in Metallurgy. W. Shockley^ Jl. Metals, 4, pp. 829-842, August, 1952. (Monograph 2011). Impurity Effects in the Thermal Conversion of Germanium. W. P. Slichteri and E. D. Kolb^ Phys. Rev., 87, pp. 527-528, August 1, 1952. Traffic Engineering Design of Dial Telephone Exchanges. J. A. Stew- art*. Midwest Engr., 4, pp. 3-5, 17-18, May, 1952. Single-Crystal Germanium. G. K. Teal^, M. Sparks^ and E. Buehler.^ I.R.E., Proc, 40, pp. 906-909, August 1952. Significant advances have been made in the development of new types of transistors, photocells, and rectifiers and in the improvement of the reproduci- bihty and rehability of the point-contact transistor. A key factor in this de- velopment has been the use of single-crystal germanium having a high degree of lattice perfection and compositional control. Of particular interest to the device- development engineer is the fact that the rectifying barriers between the p-type and w-type sections behave in a manner predictable from the measured properties of each section. The exceptionally long lifetime of injected carriers observed in the material and the high degree of control over its chemical composition make it ideally suitable for the production of p-n structures. The ranges of properties of germanium single crystals which are now reahzable are given, as well as their present degree of control. Lead-Acid Stationary Batteries. U. B. Thomas^ Electrochem. Soc. JL, 99, pp. 238C-241C, September, 1952. Polymorphism of NDJ)2P0^. E. A. WoodS W. J. Merz^ and B. T. Matthias^. Phys. Rev., 87, p. 544, August 1, 1952. Lightning Protection for Mobile Radio Fixed Stations. D. W. Bodle^ I.R.E. Trans., P.G.V.C.-l, pp. 122-133, February, 1952. Equipment in fixed stations of a mobile radio system is susceptible to damage from Hghtning strokes to either the antennas or the connecting power and land communication facilities unless special protection is provided. The problem, however, is not alone one of protecting the station equipment, but consideration must also be given to the protection of these connecting facihties to insure their continuity of service. The causes of and factors affecting Hghtning damage are discussed, including the probable incidence of strokes to the antennas. General protection principles are outlined, and the appHcation of specific protection methods is described. 1 Bell Telephone Laboratories ' Western Electric' Company * Illinois Bell Telephone Company 258 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Matching Coax Line to the Ground-Plane Antenna. R. T. DeCamp-. QST, 36, pp. 18-19, 120, 122, September, 1952. This article describes a method for predetermining antenna and matching- stub dimensions for matching an}^ selected transmission line. Although applied particularly to the ground-plane antenna, the curves are useful for half-wave dipoles if allowance is made, when necessary, for the effect of ground on the antenna characteristics. The Telephone System in National Defense. C. M. Mapes^. Military Engr., 44, pp. 375-377, September-October, 1952. Multi-Element Directional Couplers. S. E. Miller^ and W. W. Mum- FORD^ I.R.E., Prcc, 40, pp. 1071-1078, September 1952. It is shown that the backward wave in a directional coupler is related to the shape of the function describing the coupling between transmission lines by the Fourier transform. This facilitates the design of directional couplers for arbitrary directivities over any prescribed frequency band. Tightly coupled directional couplers are analyzed in simple terms, and it is shown that any desired loss ratio, including complete power transfer between lines, may be achieved. The theories are verified using waveguide models operating at 4,000, 24,000, and 48,000 mc, and it is indicated that the work is applicable to many types of electrical and acoustic transmission lines. Segregation of Two Solutes, with Particular Reference to Semiconductors. W. G. Pfann^. Jl. Metals, 4, pp. 861-865, August, 1952. (Monograph 2020). The simultaneous segregation of two solutes during the directional solidifica- tion of an ingot is treated mathematically on the basis of simplifying assump- tions. Expressions are derived for the difference in concentration of two solutes, and for the location and concentration gradient of a pn barrier formed in a semi- conductor by the segregation of a donor and an acceptor. Nonsynchronous Time Division with Holding and with Random Sam- pling. J. R. Pierce^ and A. L. Hopper^. I.R.E., Proc, 40, pp. 1079-1088, September, 1952. There is a general type of system in which an indefinitely large number of transmitters can have access to any of an indefinitely large number of receivers over a medium of limited band-width. In these systems, signal-to-noise ratio goes down as more transmitters are used simultaneously. This paper describes a particular system which sends samples by means of coded pulse groups sent at random times. The signal-to-noise ratio is good in the absence of interference and the effect of interference is minimized by holding the previous sample if a sample is lost. An experimental system worked satisfactorily and gave close to the predicted signal-to-noise ratio might be used to provide communication and automatic switching in rural telephony, or for other applications. ^ Bell Telephone Laboratories * American Telephone and Telegraph Company ABSTRACTS OF TECHNICAL ARTICLES 259 An Improved Electrolysis Switch. V. B. Pike^ Corrosion, 8, pp. 311- 313; disc. pp. 322-323, September, 1952. (Monograph 2021). An improved electrolysis switch has been placed in use in the Bell System for mitigation of electrolysis of cables by stray currents. It consists of three relays operating in sequence, namely control, intermediate and drain relays. It auto- maticalh^ closes a drainage bond between a lead-covered underground cable and a power return ground when stray current is picked up by the cable to such an extent as to make some of the sheath positive with respect to its environment. It also opens bond when the drainage current falls to zero and drainage is no longer required. Two sizes are used capable of draining 200 amperes and 400 am- peres, respectively. Its action is very fast, only 0.015 second elapsing from the time the control circuit releases until opening of the drainage bond. Separate ad- justments are available for setting the voltage at which the switch closes and opens the drainage bond. A capacitive voltage booster enables switch operation over longer battery power supply wires than is possible if the booster is not used. A power supply unit consisting of a stepdown transformer and selenium rectifier is also available to operate the switch with power drawn from an AC power source if the switch must be installed beyond reach of the battery power. A sealed steel housing enables the switch to withstand submersion and permits installation in very damp locations. Statistics of the Recombinations of Holes and Electrons. W. Shockley* and W. T. Read, Jr.^. Phijs. Rev., 87, pp. 835-842, September 1, 1952. (Monograph 2022). The statistics of the recombination of holes and electrons in semiconductors is analyzed on the basis of a model in which the recombination occurs through the mechanism of trapping. A trap is assumed to have an energy level in the energy gap so that its charge may have either of two values differing by one electronic charge. The dependence of lifetime of injected carriers upon initial conductivity and upon injected carrier densit}' is discussed. Motiort, of Gaseous Ions in a Strong Electric Field. G. H. Wannier^ Phys. Rev., 87, pp. 795-798, September 1, 1952. (Monoj;raph 2023). This paper continues an earlier one on the same subject. Its object is to eluci- date the nature of the random motion of an ion about its drift. In Section F it is shown that this motion can be described as a diffusion with a diffusion tensor axially symmetric about the field. If the mean free time between the collisions of an ion with molecules is independent of speed, then explicit expressions may be deprived for the two diffusion coefficients; these expressions are written down without proof in Section G; they are connected with the mobility by a natural extension of the Einstein relation. In Section H, the longitudinal diffusion co- efficient is computed numerically for the hard sphere model, high field, and mass ratio 1; the method of computation is the same as in Section D. Finally, it is shown in Section I how approximate formulas of wider validity can be inferred from the ones obtained. Bell Telephone Laboratories 260 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 High-Frequency Crystal Units for Primary Frequency Standards. A. W. Warneri. I.R.E., Proc, 40, pp. 1030-1033, September, 1952. A new approach to the design of crystal units for primary frequency standard use has resulted in crystal units in the 3- to 20-mc frequency range characterized by high Q and low capacitance in the series arm of the equivalent electrical circuit. By utilizing the overtone frequency of specially shaped AT-cut quartz plates, both Q and the rate of impedance change with frequency are enhanced together, and in addition the stability with time of the crystal unit is increased because of a larger frequency-determining dimension. Additional characteristics of the crystal units include small size, stability under conditions of vibration and shock, and low-temperature coefficient. Crystal-oscillator stabilities of one part in 10^ per month have been achieved without recourse to stabilized circuits. Bell Telephone Laboratories Contributors to this Issue Wallace C. Babcock, A.B., Harvard University, 1919; S.B., Harvard University, 1922. U. S. Army, 1917-19. American Telephone and Tele- graph Company, 1922-34; Bell Telephone Laboratories, 1934-. Mr. Babcock was engaged in crosstalk studies until World War II, when he studied radio count ermeasure problems for the N.D.R.C. Since then he has been concerned with antenna development for mobile radio and point-to-point radio telephone systems and has been engaged in other systems studies. Member of I.R.E. and Harvard Engineering Society. John Bardeen, B.S., in E.E., University of Wisconsin, 1928; M.S. in E.E., University of Wisconsin, 1929; Ph.D., Princeton University, 1936. Gulf Research and Development Company, 1930-33; Harvard University, 1935-38; University of Minnesota, 1938-41 ; Naval Ordnance Laboratory, 1941-45; Bell Telephone Laboratories, 1945-51; University of Illinois, 1951-. At Bell Telephone Laboratories, Dr. Bardeen, co-in- ventor with Dr. Walter Brattain of the point-contact transistor, was primarily concerned with theoretical problems in solid state physics, in- cluding the study of semi-conductors, diffusion in solids, and supercon- ductivity. Associate editor of The Physical Review, 1949-51. Stuart Bal- lantine Medal of the Frankhn Institute, 1952. Fellow, American Physical Society; member, American Association for the Advancement of Science. Walter H. Brattain, B.S., Whitman College, 1924; M.A., University of Oregon, 1926; Ph.D., University of Minnesota, 1929. Bureau of Standards, 1928-29; Bell Telephone Laboratories, 1929-. Dr. Brattain, co-inventor with Dr. John Bardeen of the point-contract transistor, has been primarily concerned with the study of semi-conductors at Bell Laboratories. During World War II he worked for the Division of War Research of Columbia University and is currently spending the fall term of the academic year 1952-53 as a visiting lecturer at Harvard Uni- versity. Stuart Ballantine Medal of the Franklin Institute, 1952. Fellow, American Physical Society and American Association for the Advance- ment of Science; member, Sigma Xi and Phi Beta Kappa. 261 262 THE BELL SYSTEM TECHNICAL JOURNAL, JANUARY 1953 Kenneth Bullington, B.S. in E.E., University of New Mexico, 1936; S.M., Massachusetts Institute of Technology, 1937; Bell Tele- phone Laboratories, 1937-. Until World War II, Mr. Bullington was oc- cupied with systems engineering work on wire transmission circuits. Since 1942, he has been concerned with transmission engineering on radio systems, especially with radio propagation studies. Member of I.R.E., Phi Kappa Phi, Sigma Tau, and Kappa Mu Epsilon. R. H. CoLLEY, A.B., Dartmouth College, 1909; A.M., Harvard Uni- versity, 1912; Ph.D., George Washington University, 1918; Austin Teaching Fellow in Botany, Harvard University, 1910-12; Instructor in Botany, Dartmouth College, 1909-10 and 1912-16; Pathologist, Division of Forest Pathology, Bureau of Plant Industry, U. S. Department of Agriculture, 1916-28. Bell Telephone Laboratories, 1928-1952. Dr. Col- ley w^as chairman of Committee 05 — Wood Poles, of the American Standards Association for nearly twenty years. He was president of the American Wood-Preservers' Association 1943-44. During his years with the Laboratories he worked particularly on development and research problems connected with material and preservative treatment specifica- tions for poles and other timber products used in outside plant. His more recent activities were directed tow^ard improvement of laboratory tech- niques for evaluating wood preservatives, and toward the development of a coordinated plan for fundamental research on oil preservatives. He was Timber Products Engineer for the Laboratories from 1940 to 1950, and Timber Products Consultant from 1950 to 1952. His article in this issue of the Journal was prepared before his retirement on May 31, 1952. Karl K. D arrow, B.S., University of Chicago, 1911. He studied at the Universities of Paris and Berlin in 1911 and 1912, specializing in physics and mathematics; Ph.D., University of Chicago, 1917. He then joined the staff of Bell Telephone Laboratories, at that time known as the Engineering Department of Western Electric Company. Here his work has included the study, correlation, and representation of scientific information for his colleagues, keeping them informed of current ad- vances made by workers in fields related to their own activities. As a corollary to his work, Dr. Darrow appears from time to time before scientific and lay audiences to lecture on current topics in physics and the related sciences. He has taken an active interest in education, teach- ing physics during summer and other sessions at Stanford, Chicago, and Columbia Universities and at Smith College. From 1944 to 1946, he served CONTRIBUTORS TO THIS ISSUE 263 as consultant to the Metallurgical Laboratory in Chicago. Dr. Darrow is the author of Introduction to Contemporary Physics (1926 and 1939), Elec- trical Phenomena in Gases (1932), Resonance of Physics (1936), and Atomic Energy (1948) and of many articles in this and other journals. He is a member of the American Physical Society, which he has served as secre- tary since 1941, the Physical Society of London, Soci^te Francaise de Physique, the American Philosophical Society, of which he was a coun- sellor for four years. From 1949 to 1951 he was vice-president of the In- ternational Union of Pure and applied Physics. In 1949 he received an honorary doctorate from the Universite de Lyon and was made Cheva- lier de la Legion d'Honneur in 1951. John Riordan, B.S., Sheffield Scientific School of Yale University, 1923. United Electric Light and Power Company, now a part of the Consolidated Edison Company, 1923-1926. Department of Development and Research of the American Telephone and Telegraph Company, 1926-1934. Bell Telephone Laboratories, 1934-. With the American Telephone and Telegraph Company, his work was largely on circuit and transmission theory, particularly in relation to inductive interference from electrified railways. This work was continued in Bell Telephone Laboratories after the Development and Research Department was consolidated with it in 1934. Since 1940 Mr. Riordan has been engaged in mathematical work: Boolean algebra in switching, number theory in cable splicing, and combinatorial and probability studies of traffic. Mr. Riordan is a member of the American Mathematical Society, the Mathe- matical Association of America, the Institute of Mathematical Statistics, and is a Fellow of the American Association for the Advancement of Science. Gregory H. Wannier, Louvain University, 1930-31; University of Cambridge, 1933-34; Ph.D., University of Basel, 1935. Assistant, Uni- versity of Geneva, 1935-36; Swiss-American Exchange Fellow, Prince- ton University, 1936-37; instructor, University of Pittsburgh, 1937-38; assistant lecturer, Bristol University, 1938-39, instructor. University of Texas, 1939-41; University of Iowa, 1941-46. Socony -Vacuum Labora- tories, 1946-49. Bell Telephone Laboratories, 1949-. Mr. Wannier, a theoretical physicist in the Physical Research Department, has worked on photoconductivity and related phenomena, and the motion of ions in gases. Also Mission to Germany, 1945. Member of American Physical Society and Schweizer Physikalishe Gesellschaft. THE BELL SYSTEIM / mcai ourna devoted to the scientific ^^^ and engineerinc Aspects of electrical communication rOLUME XXXII MARCH 1953 NUMBER: Ferrite Core Inductors h. a. stone, jr. 265 A Tlirowdown Machine for Telephone Traffic Studies G. R. FROST, WILLIAJM KEISTER AND ALISTAIR E. RITCHIE 292 J Working Curves for Delayed Exponential Calls Served in Random Order ROGER I. WILKINSON 360 Magnetic Resonance : Part II — Magnetic Resonance of Electrons KARL K. DARROW 384 A Study of Non-Blocking Switching Networks charles clos 406 The Evaluation of Wood Preservatives — Part II REGINALD H. COLLEY 425 Abstracts of Bell System Papers Not PubHshed in this Journal 506 Contributors to this Issue 519 COPYRIGHT 1953 AMERICAN TELEPHONE AND TELEGRAPH COMPANY THE BELL SYSTEM TECHNICAL JOURiNAL A D V I S O R \ IJ O A H I) S. B R A c K E N. President. H rs/crn /J/ectnc Company F. n. k A l> i» K L. lice Prcsidcif. A/neriran '/'clcpl/one and 'VeU'<>:rnf)h ('oniparn- M. J. K ]•: L L N . Prexideitt. I>
^ -" / y y //x y ^ y 0; i-'- y / / /^ ^y ^ y ''/^ ' y y ^/ y y' / /^ y / y^ i/ / y, / / / / / / / 10 20 30 40 60 80 100 200 300 400 600 1000 Qo(Q FOR DISTRIBUTED CAPACITANCE = O) Fig. 3 — Effect of distributed capacitance on Q. maintain small values of distributed capacitance. It will be necessary to increase the separation of windings from each other and from the core and this will result in a lower winding efficiency and a higher dc resistance. Since the limitations imposed by ac losses in the wire and by dis- tributed capacitance depend on the absolute magnitudes of the design parameters and the mechanical details of the inductor structure they do not lend themselves to representation in practicable generalized for- mulas as do the core losses. However, the following information on some model inductors will illustrate the magnitudes of these Hmitations. A ferrite core coil was constructed similar to that shown in Fig. 4, but having a core volume of only 0.04 cubic inches. It was a 5 mh coil for use at 100 kc and had a distributed capacitance of 10 mmf. It was wound with a single strand conductor and the winding efficiency, k^ , was 0.4. Since it was intended for use at very low power levels the hystere- sis loss was negligible. The permeability was optimized in accordance with case 1, above, for residual loss predominating. The measured Q was very close to 300. Thus, D = 0.0033. From the above data we note that C/Cg = 0.02, and from (25) we can calculate that Do = 0.0030, 276 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Fig. 4 filters. The 1509 type adjustable ferrite inductor for Type-0 carrier system using the value 0.01 for d. Thus the "geometric Q", that which would obtain were it not for the distributed capacitance, is about 330. The 1509 type inductor, shown in Fig. 4, was designed for use in the type-0 carrier system.^ It has a volume of 1.4 cubic inches, 35 times that of the small coil discussed above. From the standpoint of dc re- sistance and core loss alone, (7) indicates that the dissipation factor should be A = 3^3 (0.0030) = 0.00092, or that the Q should be greater than 1000. In order to keep the capacitance to the same order as that of the small coil, and to use stranded wire to avoid excessive ac loss, the winding efficiency, ky, , had to be reduced from 0.4 to 0.13. The effect of this on dissipation can be shown from (17), remembering that k^ varies inversely fc. Do (Sh) 1/2 1.73 Do = (1.73) (0.0092) 0.00159. Thus, the Q (still neglecting the effect of the 10 mmf distributed capaci- tance that remains after reducing the winding efficiency) should be 630. FERRITE CORE INDUCTORS 277 From (24) and A = ("00159 +^001)0-02 ^ 0.00024 D = i)o + I>c = 0.00183 or the actual measured Q of the large inductor is 550, about half that indicated by core loss considerations alone. EFFECT OF CORE PROPORTIONS ON DISSIPATION FACTOR In the foregoing discussion of coil volume it has been assumed that the core porportions remained fixed as the volume was changed. It is now of interest to know what these proportions should be to insure the lowest dissipation. The general type of structure under consideration consists of a closed cylindrical container and a center post of magnetic material, and an air gap which might be anywhere in the magnetic path. It is assumed that the thickness of the shells is such that the area of the magnetic path is uniform and equal to the area of cross section of the post. It is also as- sumed that the flux is uniformly distributed within the cross-section of the core. Fig. 5 represents such a core schematically. Given a fixed over-all coil volume it is desired to know what proportions should apply to the outside diameter, the diameter of the center post and the axial height of the structure, to provide the lowest dissipation factor. This will be examined first for the case where residual or eddy HR I I [*- PR--H I ^ R -^ Fig. 5 — Optimum proportions for post and shell core assembly. (29) 278 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 current losses are large compared to hysteresis loss, and then for the situation where hysteresis loss predominates. Referring to Fig. 5, the following relationships are noted. t (thickness of shell) is given by {R - t) = R\/r^^P'^ (26) A (magnetic cross-section) = ttP^R^ (27) I (mean magnetic path) = R{Zy/l - p2 - i j^ 2H) (28) W (available winding cross-section) = RWi^'P' - P)(2\/l~="P"' - 2 + F) X (mean length of turn) = irRi^/l^^P^ + P) (30) Vc (coil volume) = 'kR^H (31) 1. DC Resistance, and Residual or Eddy Current Losses, Predominate The dissipation factor, due to the dc resistance, is seen from (4) and (6) to be where ku = (p/2Trkkw) is a constant with respect to the core propor- tions, assuming the winding efficiency is not materially affected by changes in shape. Since, regardless of shape, we will want to adjust the air gap so that H is optimum, and since we know from the previous discussion that this will occur when the dc resistance is equal to the sum of the core losses, we have, from (10) and (12) D,c = De + Dr = (hf + h)fl (33) Eliminating n from (32) and (33) : Z>.c = A-. y/A^ (34) where kn — Vknik^f + A:?) is again a constant with respect to the core dimensions. The dissipation factor is D = 2D,e = 2kn |/|^ (35) FERRITE CORE INDUCTORS 279 Putting in the values, from (27) to (30), for X, I, W and A: D ^ ^ A/ (Vl - P' + P)(SVl - P^ - 1 + 2i/) (3g) R y P\Vl - P'- P)i2Vl - P'- 2 +H) Since the volume of the coil, given in (31), is assumed to be con- stant, we can eliminate R from the above equation: D = k^^ Y r2/3 (Vl - P' + P)(3V1 - p2 _ 1 _|, 2H)H' (37) PWi - P^ - P)(2\/l - P"" - 2 + ^) where fcis = 2ir"'k 12 Fi/3 We now have an expression for the dissipation factor in terms of the two variables, P and H, which determine the proportions of the coil structure. The effects of these proportions on the dissipation factor are shown in Fig. 6. It will be seen that best results are achieved when the radius of the post is approximately 0.45 of the outside radius and the axial height is about 1.2 times this radius. Fig. 5 is drawn to this scale, and the 1509 type coil shown in Fig. 4 approximates these proportions. 2. DC Resistance and Hysteresis Loss Predominate We have noted that for this case optimum permeability is that which results in the following relationship between the dc and hysteresis dissipations : Wh = 2Ddc (38) From (11) and (13) 7 7 3/27-1/2- D. = hB^, = ^^9^* (39) Putting the values from (32) and (39) in (38), we can eliminate fx and express Ddc in terms of the dimensional variables of the coil: >3/5^2/5 where ku = {^k^kghn^^L^ ^if ^ The total dissipation factor is ^^3/5,2/6 D = D,e-\- Dh = iD,c = iku ppTsjiTi (^1) 280 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Using (27) to (30) we can express D in terms of the coil proportions D = iku a/- (Vl - P' + P)'(3Vl - P2 - 1 4- 2Hy (42) P'PV(Vi - P2 - P)\2Vl - P^ - 2 + Hf Again, eliminating R by use of the constant volume relationship, (31) D = kn ^ (Vi - p^ + pyisVi - p' - 1 + 2Hy p'(Vi - P2)'(2Vl - P' - 2 + i/)' (43) _2/5 where ku — .26 .24 1.20 tr o 1 z o V) .08 1.06 1.04 1.02 1.00 37//« 1 1 1 1 1 \ / \ \ H: -oaJ / \ \ \ y / \ \ V \ \, y / y \ V -K \v ■ __ ^ 2^0^ y y \ \, y -y X ^ y y 0.34 0.36 0.38 0.40 QA2 0.44 0.46 0.48 0.50 0.52 0.54 0.56 P Fig. 6 — Effect of core proportions on dissipation factor when hysteresis can be neglected. FERRITE CORE INDUCTORS 281 1.30 1.28 .26 1.24 1.20 Q 1.18 5 1.12 .06 .04 1.02 .00 1 1 h 1 1 1 1 1 1 1 1 1 1 1 / ' // / // ^ / / / \ / ' // \ \ \ / ' // \ \ \ \ \ '11 \ \ \ \ o7 / // V \ \ \ 7 c 0/ , iCO—i // V \ \ / h 9/ \ 1 // 'i •^ ^ "^ — -• "A / ^^J / / / ^ yy 0.34 0.36 0.38 0.40 0.42 0.44 0.46 0.48 0.50 0.52 0.54 0.56 0.58 0.60 0.62 P Fig. 7 — Effect of core proportions on dissipation factor when hysteresis loss is predominate. This information is shown graphically in Fig. 7. It will be seen that when the hysteresis losses are important both the post diameter and the axial height should be about half of the overall diameter. These propor- tions are not far different from those derived as optimum from (37). This is fortunate since it means that cores of the same proportions will be suitable in different applications regardless of which core losses pre- dominate. 282 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 METHODS OF INDUCTANCE ADJUSTMENT The fact that ferrites can be molded in a variety of shapes, can be machined, and can be used with discrete air gaps permits considerable freedom in the mechanical methods that can be used to provide contin- uous adjustability of inductance. Whether the need is for factory ad- justment to prescribed values or for adjustment after the coil is as- sembled in its equipment, the basic methods are the same. We have noted, in (4) that the inductance of a magnetic core coil is given by L = ^ (4) in which A; is a constant depending on the units used, and the other four quantities are design variables. Any of these, or any combinations of them, can be manipulated to produce variations in the inductance. They are: length of magnetic path, l\ cross section of magnetic path, A\ nimaber of turns, iV; and effective permeability, \i. There is an additional variable hidden by the formula's assumption that perfect coupling exists among all the turns in the winding. Inductance can be adjusted by chang- ing the coupling between parts of the winding. Each of these five var- iables, which are of widely differing practical value, will be commented upon at least briefly. 1. Adjustment of Inductance by Change in Magnetic Path Length, I Adjustment by a change in I without an accompanying change in /x requires that the air gap also be changed since the effective permeability is a function of the ratio of the air gap dimensions to those of the total structure. ^ " 1 , Q (44) 1 + Mm - a where m m = permeability of the core material g — ratio of the length of air gap to total length, I a — ratio of effective cross section of gap to core cross section, A. It would be mechanically possible to design an adjustable inductor of this sort but it is unlikely that there would be any practical advantage in maintaining constant permeability over the adjusting range. On the FERRITE CORE INDUCTORS 283 contrary, it would probably be more advantageous as well as mechani- cally simpler to have the air gap length constant so that the increase in inductance due to shortening the magnetic path would be augmented by an increase in permeability. In a device such as shown schematically in Fig. 8, as the two windings approach each other the over-all magnetic path decreases and at the same time the area of the air gap increases. Actually, the preponderant effect is due to the air gap change and the effect of variation in length of path becomes of secondary importance. This is hkely to be the case generally, when both path length and air gap change simultaneously, since most of the reluctance is in the air gap. 2. Adjustment of Inductance by Change in Magnetic Cross Section, A To provide for change of inductance by constriction of a part of the magnetic cross section is apt to be undesirable since it forces a concen- tration of flux in the constricted part of the magnetic circuit and may introduce unduly high hysteresis losses. Even if the levels are low enough so that this is not a consideration the amount of inductance variation that can be achieved even by a large constriction in part of the core is relatively small. To illustrate this we will consider a structure such as shown schematically in Fig. 9, in which one sector of the core can be varied in effective cross section. The reluctance of the structure is equal to the sum of the reluctances of the fixed part of the core, the sector of length nC, whose cross section aA can be varied, and the air gap: ^ ^ (1 - n)i _^ _rU_ ^ g{ ^rnA fJ-maA A (45) = -^\l + n(-- 1) + guJ. y,mA \_ \a J Let us assume the arbitrary but reasonable values of 2000 for jijn, 0.01 for g, and 0.1 for n. Then approximately R= ' 2000A (--^)- It will be seen that as a is varied from 1.0, corresponding to the full cross section of the main core, to one-tenth of that, the change in in- ductance, which is inversely proportional to reluctance, will only amount to about 5 per cent. 284 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 3. Adjustment of Inductance by Change in Number of Turns , N An adaptation of turns adjustment for use in a shell type structure is shown in Fig. 10.^ The center post with the winding on it can revolve and turns can be removed by pulling on the outer lead, or added by rotating the knob at the end of the shaft. Continuous adjustment is possible since it is not necessary to add or remove integral numbers of turns. Although an inductor of this sort involves some mechanical com- plexity it has the advantage that the core parts do not have to be pre- cisely machined. The turns adjustment can be used to compensate for sizable variations in the dimensions of the core parts and the resulting air gaps. 4. Adjustment of Inductance by Change in Permeability , n Adjustment by change in permeability, that is, change in length or area of air gaps, can be accomplished in a variety of ways to meet dif- fering design needs and can be mechanically simple and economical. For most purposes permeability adjustment will offer more advantages than any of the other methods. Whatever the method of adjustment used its effectiveness will de- pend on proper correlation between the mechanical motion that pro- duces the change and the inductance itself. For most filter and network apphcations the following considerations will apply: (a) The slope of the line showing inductance plotted against the dis- placement that produces the adjustment should be reasonably con- stant over the adjusting range. Fig. 8 — Inductance adjustment by decrease in magnetic path length and in- crease in air gap area. FERRITE CORE INDUCTORS 285 (b) The slope should not be negative at any position in the adjusting range as this would introduce the possibility of false or double balance when the coil is being adjusted by null or peaking methods. (c) The slope should not be so small that play in the mechanical parts will cause changes in slope of the same magnitude as the slope itself. (d) Within the above limitation the smaller the slope the more pre- cisely the adjustment can be made. (e) Conversely, the greater the slope the greater the range of ad- justment for a given amount of mechanical motion. (f ) It follows from (d) and (e) that the amount of mechanical motion available determines the product of range and precision. Where the mechanical motion is rotary, such as with adjustment by turning a screw, it is possible to adjust a coil to a precision of about 1/400 of a revolution without undue difficulty. If the range is covered by N re- volutions of the adjusting screw, and the over-all range is =b i? per cent of the mean value : 2R R 400i\r 200N where P = the precision of adjustment in per cent of the nominal value. In the coil shown in Fig. 4, six turns of the adjusting screw are effec- tive in producing an over-all change of ±15 per cent in the inductance. The precision with which the adjustment can be made is, therefore, very close to ±0.01 per cent of the value desired. AIR GAP-' Fig. 9 — Inductance adjustment by constriction in magnetic cross section. 286 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 (g) It would appear from the above that by gearing or other means of increasing the mechanical motion that governs the inductance change it should be possible to improve the precision. This is true up to the point where consideration (c) is violated, to the point where play in the mechanical parts permits variations of the same order as the nominal precision. An obvious form of permeability adjustment would consist of a means for varying the distance between two plane magnetic surfaces, as shown schematically in Fig. 11 (a). One disadvantage of this, or of any other means that involves a change in air gap length, is in the nonlinearity of adjustment. The effective permeabiHty of a coil is given in (44). In most voice and carrier frequency applications of ferrite the magnitudes of g and a mil be such that, approximately M = - (46) and, if the cross section of the air gap is the same as that of the core 1 9 A typical adjustment curve resulting from this inverse relationship is shown in Fig. 11 (b). In addition to its nonlinearity the simple butted gap has a short- coming in that the mechanical motions involved are of the order of only a few hundredths of an inch, which requires that parts be very accurately fitted. This can be somewhat alleviated by using cone or wedge shaped Fig. 10 — Inductance adjustment by addition or removal of turns. FERRITE CORE INDUCTORS 287 gaps, as illustrated in Fig. 11 (c). Here the distance d, travelled by the screw is longer than the effective air gap g, by the amount d 9 sm 9 One means of overcoming the nonlinear characteristic of gaps such as these is to introduce compensation in the form of a secondary gap that opens as the main gap closes.'^ Such an arrangement is shown in Fig. 12 (a). When there are two gaps in series their reluctances are additive and their effect on permeabihty is given approximately by 9i + 92 Fig. 12 (b) shows the inductance characteristics that would result from either of the two gaps alone, and the effect of the two gaps in series. From (46) it can be seen that the effective permeability varies ap- » AIR GAP LENGTH (b) Fig. 11 — Inductance adjustment by variation of air gap length, (a) Gap formed by parallel plane surfaces, (b) Adjustment characteristic, (c) Cone or wedge shaped gaps. 288 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Lo "7GAPS (b) AIR GAP LENGTH Fig. 12 — Inductance adjustment with partially compensating air gaps. proximately directly as the ratio of air gap area to magnetic core area. Adjustment by variation of the area of the gap, therefore, is not subject to the nonlinearity that results from manipulating the air gap length. A simple method for providing adjustment by varying the effective area of the air gap is shown in Fig. 13. As one shell is rotated with respect to AIR GAP" Fig. 13 — Inductance adjustment by variation in air gap area. FERRITE CORE INDUCTORS 289 the other the area of registration between the semicircular faces of the center posts is reduced or increased. A simple arrangement such as this provides good linearity but has two limitations: (1) The total mechanical motion available for adjustment is only 1/2 revolution. (2) As the cores are rotated to reduce the air gap area and the inductance, the magnetic flux in the cores tends to concentrate more and more in the vicinity of this reduced area, and may under some conditions give rise to high hysteresis loss. The method of adjustment used in the 1509 type inductor, Fig. 4, is essentially a variable area method but it is designed in such a way as to overcome these two limitations.^ The air gap arrangement, visible in the cutaway view, consists of two gaps in parallel. The main annular gap is fixed and is large enough in area to insure that under no antici- pated conditions of operation will the flux concentration be too high. The screw adjustment moves a cylindrical ferrite part into a depression in the center post, as shown. The effective cross section of this adjustable gap is approximately determined by the amount of surface of the cyl- inder within the depression. The total useful adjusting range corresponds to about six full turns of the adjusting screw. 5. Adjustment of Inductance by Change in Coupling The overall inductance of two coils of equal inductance connected in series is L = 2(1 + k)Li (48) where L = series inductance Li = inductance of either coil k = coupling coefficient k may have any value between —1 and +1, these values correspond- ing to complete coupling and the windings connected in series opposing and series aiding, respectively. It is practicable to make inductors whose coupHng can be continuously adjusted from very high positive values through zero to equally high negative values. This type of design is especially useful where a very wide range of inductance variation is desired. It will be seen from (48) that with couplings of plus and minus 90 per cent, respectively, in the extreme positions a range of 19 to 1 in inductance variation would result. A coil of this type is shown in Fig. 14. 290 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Fig. 14 — Magnetic variometer. Inductance adjustment by variation in mutual inductance. CONCLUSION The combined magnetic and mechanical characteritics of ferrite per- mit the design of inductors superior to those using metallic cores in at least three important respects: 1. Higher values of Q are obtainable than have ever before been prac- ticable. In the range from about 50 to 200 kc, frequently used for tele- phone carrier, it is very difficult to realize a Q above about 300 in a FERRITE CORE INDUCTORS 291 metallic core coil. Ferrite coils, on the other hand, have been made having Q's as high as 1000, and inductors with Q's between 500 and 600 are available commercially. 2. Formulas have been derived which show how ferrite can be used to realize the best inductor characteristics in the smallest volume. The 1509 type inductor, for instance, is only about 1/3 as large as the nearest equivalent permalloy core coil, yet its Q is over twice as high. 3. It is physically practicable in ferrite coil designs to include in- ductance adjustment facilities to meet a wide range of requirements. It should not be concluded that the day of metallic cored inductors is over. For power applications, especially those involving direct current, present-day ferrites are inferior to silicon iron and permalloy. At voice frequencies there are many applications for which ferrite is, at best, no better than some of the older materials, although in others it has dis- tinct advantages. In higher frequency ranges, however, and especially for low power level applications, the advantages of ferrites are out- standing enough to justify the expectation that they will largely replace the older iron and nickel-iron powders. ACKNOWLEDGEMENTS The development of ferrites and their application to inductors has been carried out by several teams in Bell Telephone Laboratories, in- cluding J. H. Scaff, F. J. Schnettler and their associates in the metal- lurgical department, V. E. Legg and CD. Owens in the magnetic ap- plications group, and S. G. Hale, R. S. Duncan and others in the inductor development area. I don't know who was first to conceive of using ''dissipation factor" instead of "Q" to simplify his mathematics, but it was not the author. Equation (24) is derived by inserting l/Dr, for Q in an expression originally worked out by P. S. Darnell. REFERENCES 1. Legg, V. E., Survey of Magnetic Materials and Applications in the Telephone System. Bell Sys. Tech. Jl., 18, pp. 438-464, July, 1939. 2. Legg, V. E., and F. J. Given, Compressed Powdered Molybdenum Permalloy, Bell Sys. Tech. JL, 19, pp. 385-406, July, 1940. 3. Snoek, J. L., Non -Metallic Magnetic Materials for High Frequencies, Philips Tech. Review, 8, pp. 353-360, December, 1946. 4. Wien, M., tlber den Durchgang schneller Wechselstrome durch Drahtrollen, Ann. D. Phys. [4], 14, P. I, 1904. 5. Edwards, P. G., and L. R. Montfort, Type-0 Carrier System, Bell Sys. Tech. JL, 21, pp. 638-723, July, 1952. 6. Legg, V. E., and C. D. Owens, Patent Application Serial No. 184602, Filing Date — 9/13/50. 7. Hale, S. G., and C. W. Nuttman, Patent Application Serial No. 263564, Filing Date — 12/27/51 . 8. Duncan, R. S., and H. A. Stone, Patent Application Serial No. 262248, Filing Date — 12/18/51. A Throwdown Machine for Telephone Traffic Studies BY G. R. FROST, WILLIAM KEISTER AND ALISTAIR E. RITCHIE (Manuscript received December 3, 1952) In order to study the traffic-carrying characteristics of the No. 5 crossbar switching system, a machine has been built to simulate the operation of the system. This machine, known as a throwdown machine, is controlled by a team of Jour operators. Its input is a statistically accurate representa- tion of telephone traffic and its output is a detailed record of the course of each call through the system. This paper discusses the design principles of the throwdown machine, its operation, and the type of results obtained. INTRODUCTION Existing analytical methods are inadequate for investigating many- statistical problems in which a large number of variables and their inter- actions must be considered. The problem of evaluating the performance and traffic capacity of a large automatic telephone switching system is one example. Others involve logistics, air and highway traffic control, and certain phases of military and naval strategy. All these require the assimilation of large quantities of data, processing the data according to certain procedures which are often empirical, and producing final information from which performance of the system or the excellence of the procedures can be judged. These problems fall in the general category of "systems evaluation." The types of systems considered are those that are capable of a large number of variations depending on the nature of the input data, and must be judged on a statistical basis. One method of study might be to operate and observe an actual system. There are a number of objections to this. Operation may be so slow that the accumulation of sufficient data may require excessive time or, as in the case of a telephone switching system, so rapid that it is impractical to make the necessary observations. Operating the system under controlled conditions in these cases may be 292 THROWDOWN MACHINE FOR TRAFFIC STUDIES 293 too expensive or indeed impossible. Then, too, the system may be pro- posed only and not yet exist. One solution is to devise a method of simulating the performance of the actual or proposed system which through the use a suitable time scale will permit the necessary information to be obtained. The simulation may be done entirely on paper by recording each state of the system and modifying this state mth each bit of input information according to the system plan, as though a log were being kept of the performance of the actual system. Since the general problem involves large quantities of input data which are statistical in nature, all possible variations cannot be studied. A sufficient number of typical situations must be tried to obtain statisti- cally reliable results. These methods have been extensively used in tele- phone traffic studies and are called ^'throwdown" studies. The name stems from the use of dice in the early study of telephone traffic problems. Each die is designated to represent a particular independent event and the faces of this die are designated according to the probability of the event taking place. By repeatedly "throwing down" a number of such dice and observing the results, the probability of a particular combi- nation of events taking place can be estimated. Other similar methods based on selections from lists of random numbers have been used in telephone traffic studies for a number of years. Recently, mathematicians, using digital computers, have employed similar statistical methods in problems relating to the diffusion of gases, electron ballistics and the solution of certain types of differential equations. They have called this the "Monte Carlo" method. Various mechanical aids can be used in running a throwdown study. This paper wdll discuss the techniques of throwdown studies and will describe a semi-automatic throwdown machine which was constructed for studies of the neAV No. 5 crossbar switching system used in local telephone central offices. A general view of this machine is shown in Fig. 1. It is a system of electrical switching circuits, signal lamps and mechanical devices which simulates a large telephone switching system and its associated subscribers. The machine is controlled by a team of four operators. Artificially generated telephone traffic is processed by this machine in a manner analogous to the action of the actual system. Detailed records are made of the progress of each call and the traffic situations encountered. After a sufficient number of calls have been processed, the recorded information can be analyzed by statistical meth- ods to obtain desired information. The action of the system is simulated in sufficient detail to insure that results are representative of actual 294 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 THROWDOWX MACHINE FOR TRAFFIC STUDIES 295 system performance. The level and character of submitted traffic can be varied, and a wide range of system sizes with varying quantities of control circuits can be tested. Before the throwdown machine was built a ''paper" throwdown trial of a small No. 5 crossbar system installation had been conducted by Mr. R. I. Wilkinson. This was run by a team of girls using card files, ledgers and written records, and using dice to make certain random decisions. The machine is basically a mechanization of these early methods to make possible the testing of larger installations in a reason- able time. The methods of generating data for the machine were devel- oped by Mr. Wilkinson and many of the decisions relating to telephone traffic and statistical problems which were encountered in designing the machine were solved in consultation Avith him. THE TELEPHONE TRAFFIC PROBLEM A large automatic telephone switching system of the common control type is not a simple mechanism nor is evaluating its performance and traffic canying capacity a simple problem. The economy of these systems depends upon the efficient use of relatively small groups of circuits on a time-sharing basis to serve a large number of subscribers. Each group of circuits is specialized to perform certain of the functions necessary in establishing a connection, and circuits from several groups must co- operate to handle every call. A sequence of actions with appropriate alternatives at several stages where busy conditions may be encountered is completely prescribed for every call. However, this sequence is subject to interference due to simultaneous requests to use the same control circuits. Competition is resolved by preference arrangements which cause some rec[uests to be delayed while others are being served. Delays will increase the holding time of circuits with the possibihty of causing traffic congestion at other points in the system. With a number of subscribers originating calls at random, it becomes difficult to predict what the reactions of the system mil be at various traffic levels. Although some parts of the problem can be solved by analytical methods employing probability theory, it is doubtful that mathematical means, beyond rough approximations, are available for evaluating the performance of an entire system. Where systems have been built and placed in operation, the per- formance can be judged from observations of the working system. There are obvious weaknesses in this procedure. Only by collecting large quantities of information can the performance at various load levels 296 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 be determined. Practical systems, as a matter of economy, are not equipped with indicating mechanisms to show performance at all stages. Events take place rapidly and the causes leading up to a particular traffic situation cannot be easily observed. Traffic loads cannot be repeated under controlled conditions. Size of installation and quantities of working equipment can be varied only by small amounts in a working office. To test offices of various sizes requires that these offices be built and installed. Variations in the system operation require actual changes in a working system. This can be done only to a limited extent. In the case of newly developed systems, estimates of performance can be made on the basis of engineering judgment and experience with older systems of a similar type. This can be followed by a trial installation of a working system of a modest size which will test the system for flaws in design as well as provide information on traffic capacity which can be extrapolated to indicate approximate quantities of equipment for larger installations. At best this is a slow and expensive process and engineering data must be continually revised as experience is gained Avith larger installations. When the system is to be used extensively in installations of various sizes, methods of evaluating system performance in advance of actual construction are desired. This situation occurred in the develop- ment of the No. 5 crossbar system and was the occasion for building the present throwdown machine. THROWDOWN TECHNIQUES Since throwdown techniques have played an important part in the development of telephone traffic theory, a brief discussion of the basic principles will be given. A single throwdown test will indicate the performance of a system under a specific set of conditions. In a typical telephone traffic study, a given traffic load would first be assumed and a simulated system installa- tion to handle this load would be engineered on the basis of the best available information. The test run then will show the performance of the system under these particular conditions, and indicate both the adequacy of the initial engineering procedures and possible improve- ments. To obtain a proper balance between equipment quantities and traffic load may require several additional runs, varying the equipment quantities, traffic load or both. The procedure in a throwdown study i§ to first obtain data represent- ing the traffic to be handled by the system. The traffic data can be generated artificially by the use of random numbers. The method is THROWDOWN MACHINE FOR TRAFFIC STUDIES 297 based on a knowledge of the statistical behavior of the various factors entering into the composition of real traffic. Random numbers are drawn for each factor. These numbers are assigned values according to fre- quency distributions obtained analytically or from field observations. The regulating data are combined to produce a description of a sample of traffic which would be encountered under the assumed conditions and then processed by methods which simulate the performance of the actual system. As a simple example of the throwdown procedures, suppose that it is desired to determine how often on the average an ''all trunks busy" condition will occur in a particular group of trunks handling inter-office calls. A certain period of time is first selected and the number of calls expected within this period is determined. Two random numbers are then drawn for each call. One random number specifies the time, within the period, of origination of the call. The other random number, weighted according to an exponention distribution which will be discussed later, gives the holding time of the call. With the data of call origination times and holding times prepared, the throwdown run can be started. The calls are listed in the order of their originating times. The first call is assigned to the first idle trunk. A record that this trunk is busy is made and the time at which it will become idle determined by adding the assigned holding time to the time of origination. This is also recorded. The call which follows in time of origination is then assigned to the next idle trunk and the process continued for succeeding calls. Before each call is established the release times of all busy trunks are scanned to determine whether any busy trunk should be made idle. In setting up each call idle trunks are chosen from the group in the same order of preference that would be used in the system being simulated. Thus, the performance of an actual system is reproduced with con- siderable accuracy and detailed records of this performance can be made. From a study of these records the desired information can be determined. The probability of encountering an "all trunks busy" condition can be found, the average number of trunks busy can be determined, or a frequency distribution chart showing the percentage of the time the number of busy trunks is above any given number can be constructed. If proper records are kept such information as the average number of trunks searched over in locating an idle trunk can be determined. If the trunks were reached through a graded multiple or if they were in sub- groups with a common overflow group, simple extensions of the above procedures would be foUoAved. This particular problem can be solved by analytical methods and is 298 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 presented here only to illustrate the application of throwdown tech- niques. However, as problems become more complex it becomes in- creasingly difficult to apply analytical methods. Even in relatively simple systems the interplay of variables may be so involved that existing theo- retical methods are entirely inadequate. Various simplifying assump- tions must be made and there is often the doubt that some important factor has not been overlooked in formulating the mathematical theory. The most fruitful use of throwdown methods has been to check results obtained by theory and to obtain data upon which mathematical theories can be based. Throwdown techniques, of course, can be used to obtain direct results, but the functioning of a system will be better understood if there is at least some attempt to develop a theory which explains how various forces act together to produce observed results. Such theories may suggest modifications of the system which will improve its per- formance. It can be seen that the planning of a throwdown study requires a thorough knowledge of both the functioning of the system being simu- lated and the characteristics of the input data being processed by this system. The validity of results will depend upon the faithfulness with which the artificially prepared input data represent real data and the accuracy with which the throwdown routines represent the real system performance. The following two sections of this paper will describe the No. 5 crossbar system and the characteristics of the subscribers using this system. Later sections describe the methods used in the machine for simulating the dynamic performance of an operating system with its subscribers. THE NUMBER 5 CROSSBAR SWITCHING SYSTEM* No. 5 crossbar is a marker-controlled system designed primarily for local central office application in the residential sections of large cities and the fringe areas surrounding these cities. In regions of this type a relatively large proportion of all calls are completed to subscribers within the same office. Since the surrounding offices to which connection must be made are likely to be of widely diversified types, the system is designed to interconnect with any existing type of central office. No. 5 is also capable of serving isolated centers from about 3,000 lines up, and multioffice areas including the largest *F. A. Korn and James S.Ferguson, Trans. A.LE.E., 69, Part 1, pp. 244-254, 1960. THROWDOWN MACHINE FOR TRAFFIC STUDIES 299 metropolitan business exchanges. Although the system includes facilities for toll and tandem switching, these were not included in the throwdown studies and ^\dll not be discussed. The Switching Network The No. 5 system is built around an interconnecting network con- sisting of two types of switching frames utilizing crossbar switches and known as line link frames and trunk link frames. This is illustrated in block diagram form on Fig. 2. Each frame is double-ended and provides means for connecting any point on one side of the frame to any point on the other side. The connecting paths are known as links. All subscriber lines in the office connect to one side of the line link frames, each of which can serve, roughly, 300 to 500 lines; and all trunks connect to one side of the trunk link frames, each of which has 160 trunk appearances. The other sides of line and trunk link frames are connected together in such a manner that each line link frame has access to all trunk link frames over several paths. These interconnecting paths are known as junctors. The basic maxi- mum number of line link and trunk hnk frames is 20 and 10 respectively, and this is the size embodied in the throwdown machine. However, in practice, multipling arrangements can be employed to double the number of frames to give greater subscriber and traffic capacity. With the system described above, any subscriber line can be con- nected to any trunk over one of several paths, each consisting of two links and a junctor and known as a channel. On connections to outgoing or incoming trunks, a single channel is required; on connections through intraoffice trunks (connection between two local subscribers), two chan- nels, one from each end of the trunk, are required. The method of com- bining links and junctors to form channels will be described later. Dial Registration — Originating Registers The circuits that receive and store the dialed signals from the sub- scriber are known as originating registers. These circuits, in quantity as determined by desired quality or grade of service, are distributed over the trunk link frames as equally as possible. A connection between subscriber and register is set up through the switching network just as between subscriber and trunk. The registers call in control equipment for setting up the subscriber's connection when dialing is completed. 300 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Common Control Circuits — Markers The switching of all connections in the office is performed by a group of common control circuits known as markers, any one of which may be utilized on a particular call. The principal functions of the marker are (1) to determine or receive the specific location of a calling circuit; (2) to translate input signals into the specific location of a called circuit or group of circuits; (3) to test for availability and seize a called circuit or one of a group of circuits; (4) to locate, test and seize a s\vitching path between calling and called circuits; (5) to set up the connection; and (6) to take alternative action in case of trouble or busy conditions. A marker performs these functions in a very short period of time so that a few circuits can handle the requirements of an office. In the original design of No. 5 crossbar, a single type of marker handled all connections. This was the arrangement specifically handled by the throwdown ma- chine. Later design has introduced three types of markers; dial tone, completing, and combined. As an example of the function of the control circuits, when a sub- scriber originates a call, a connection is automatically established from the subscriber fine circuit on a line link frame via a marker connector to an available marker. The marker identifies the location of the line and establishes the fact that it is a new call requiring a register. It tests all registers and trunk link frames and chooses an idle frame with idle registers. The marker then gains access to the correct line link and trunk link frames via the frame connectors, chooses an idle register, tests all usable channels, picks a particular channel and operates the crossbar switch magnets to close the connection between line and register. After storing the line location in the register for later use, the marker discon- nects itself. When the subscriber completes diaUng, the register connects itself to an idle marker via the marker connector. It transfers to the marker the location of the originating line and the called number. If the call is local to the office, the marker determines the location of the called line from the number group circuit (a translating device) and tests and chooses an intraoffice trunk. The marker then gains access to the link frames through the frame connector, tests the called line for busy, picks a channel, and establishes the connection, thereafter removing itself and the register from the connection. During the course of the foregoing events, a call may encounter var- ious delays beyond the minimum circuit operating time in setting up the connection. Delay may be caused by meeting a temporary busy condi- THROWDOWN MACHINE FOR TRAFFIC STUDIES 301 tion of markers, registers or the particular number group link frames required. Busy condition of lines or trunks may result in rerouting the call. In general, the grade of service is measured by the delay in con- necting a subscriber to a register (dial tone delay) and the probability of not finding an idle trunk or channel. Intraoffice Trunks The intraoffice trunks, used on locally originated and completed sub- scriber connections, include the supervisory, charging and ringing func- tions. The trunk is held on a connection for the duration of the call in distinction to markers and registers which have short holding times. Outgoing Trunks and Senders Calls to other offices are completed via outgoing trunks which incor- porate the supervisory and charging functions. In order to transmit information to the distant office, an outgoing sender is connected to the trunk for a short period of time by means of an outgoing sender link. In establishing the call, the marker first connects itself to the sender via the sender connector in order to register in the sender the called number, and then sets up the trunk-sender linkage. When the sender, which may be one of several varieties depending upon the type of signaling required by the distant office, has transmitted the number to the distant office, it drops off the connection. Incoming Trunks and Registers Calls from a distant office are completed through an incoming trunk, which includes supervisory and ringing functions. In order to receive signals from the distant office, the trunk connects itself to an incoming register by means of an incoming register link. When the register, which again may be one of several varieties, has received the called number, it obtains a marker through the marker connector for setting up the connection. When its functions have been accomplished, the register disconnects. Tone Trunks This group of trunks includes those to which calls are routed when line busy or all trunk or channel busy conditions are met. Subscriber error or excessive delay in diaUng also result in routing to these trunks. The tone trunks return distinctive tone signals to the subscriber. If 302 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 the marker is unable to set up a connection to a tone trunk, the originat- ing register is able to return the tone signal. Number Groups Each number group is a translating circuit which provides terminating information for a consecutive block of 1000 directory numbers. When a marker transmits the called line number to a number group, it re- ceives back the line location on a line link frame, the type of ringing required, whether or not the line is in a PBX group, and certain minor information. The number group also includes facilities for selecting an idle line in a PBX group. As with link frames, only one marker can connect to a number group at a time, but markers can connect to dif- ferent number groups simultaneously without interference. Connectors The marker connector, providing access from line link frames and register to markers, include circuits which assign preference of all line link frames and registers for specific markers. This helps distribute marker usage. In addition, when several line link frames and registers are com- peting simultaneously for busy markers, a gating arrangement allocates the order of service to reduce excessive delays. Although shown as one block on Fig. 2, the marker connector circuits are provided one per fine link frame and one per sub-group of ten or less registers. The frame connector, which provides access from markers to link frames, includes preference and lockout features since only one of several competing markers can connect to a given frame at one time. Although shown as one circuit on Fig. 2, the frame connectors are actually dis- tributed over the frames. Outgoing senders of a particular type are divided into two groups and the sender connector permits connection from two markers, each to one sender in each group simultaneously. Preference and lockout features obtain. The number group connectors, provided one per number group, are similar to frame connectors and give access from markers to number group. Handling a Typical Call An understanding of the operational intricacies of a telephone switch- ing system cannot be gained by a discussion of components and can only THROWDOWN MACHINE FOR TRAFFIC STUDIES 303 be developed by a study of the course of events in setting up calls through the system. As has been intimated earlier, the establishment of calls is largely a matter of marker operation. In order to illustrate the marker functions, Figs. 3 and 4 show two charts indicating the order of events in establishing a connection, first, between a caUing or originating sub- scriber line and a register, and second, after dialing, between two sub- scriber lines within the same office. These two types of connection are known as a dial tone call and an intraoffice call, respectively. Figs. 3 and 4 are drawn as sequence charts with time flowing do^\^l- ward. There is no attempt to maintain an accurate time scale; the x marks on the vertical line merely represent the relative order in which important control functions take place. In actuality, of course, the time between x marks is known mth fair precision. Brief descriptions of the control functions are listed to the left of the vertical lines. The call il- lustrated is presumed to encounter no difficulties in completion. How- ever, points at which blocking might occur are marked with an asterisk to the right of the lines. If any of the difficulties noted were to develop, the marker would have to take alternative action which will be illus- trated later. Also shown to the right of the lines are potential points of delay, where a call may have to wait until a connector, a marker, or a desired frame becomes idle. It must be remembered that during mod- erate or heavy traffic, several or all of the markers are working simul- taneously and tending to interfere with each other. In a Avell-balanced and soundly engineered central office, the aggrega- tion of parts are nicely adjusted to give on the average no more than certain preassigned values of delay and blocking at some average busy hour traffic level chosen as a base. A typical example of permissible delay is no more than 1 per cent of calls having a dial tone delay greater than three seconds. When traffic is heavier than the engineering base, the percentages of delay or blocking will increase. A summation of all the possible alternative sequences which a marker may have to take when trunk busy, line busy and channel busy con- ditions are encountered becomes extremely complex. Although no at- tempt will be made to discuss this in detail, a chart shomng the opera- tional variations of a marker on an intraoffice call is presented on Fig. 5. Even this figure does not include all possible variations since, for ex- ample, the contingency of all tone trunks being busy is not shown on the diagram. This chart, similar in form to Fig. 4, will later be found useful in discussing the throwdo^\Ti machine. In order to simplify the presentation, some of the control events are combined in time with the frame seizure which precedes the event. The normal course of a call SUBSCRIBER LINES MAX -500 PER LINE LINK FRAME TOTAL MAX -10,000 LINE LINK FRAMES MAX -20 JUNCTORS TRUNK LINK FR- MAX -10 r L 1 PER 1000 NUMBERS NUMBER GROUP L J GROUP CONNEC- TOR MARKER CONNECTOR Fig. 2— Principal cor 304 ORIGINATING REGISTERS note: EQUIPMENT LIMITATIONS SHOWN APPLY TO THE THROWDOWN MACHINE AND NOT NECESSARILY TO THE NO. 5 CROSSBAR INTRA-OFFICE TRUNKS d i TONE TRUNKS OUTGOING TRUNKS TO DISTANT OFFICE INCOMING TRUNKS -^ 1 ORIGINATING REGISTERS I INTRA-OFFICE TRUNKS D TONE TRUNKS OUTGOING TRUNKS INCOMING TRUNKS ri tt-zii •4 INCOMING REGISTER LINK FROM DISTANT OFFICE TO DISTANT OFFICE FROM DISTANT OFFICE OUTGOING SENDER LINK INCOMING REGISTER n I INCOMING REGISTER OUTGOING SENDER OUTGOING SENDER OUTGOING SENDER CONNECTOR ri- .1 5 crossbar office. 305 306 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 is in a vertical direction and it is only when a busy condition or a special function such as PBX hunting is encountered that the call shifts to the right or left. On a particular call, the only control functions performed are those marked with an x. In the situation illustrated, the size of the office is assumed to be such that two groups of intraoffice trunks are provided. The marker makes a more or less random choice of the first group to be tested, and, if this group proves to be all busy, will automatically test the second group. This brief description of the Number 5 Crossbar System is not intended to be exhaustive. It discusses only the important features of the system and those which will help make the description of the throwdown ma- chine more intelligible. Certain more involved items will be discussed in greater detail in later sections concerned with the functioning of the throwdown machine. TIME I POSSIBLE DELAYS CONTROL FUNCTIONS i AND DIFFICULTIES RECEIVER OFF HOOK :^ DELAY FOR MARKER (INCLUDES LINE LINK FRAME MARKER CONNECTOR ^ qelAY TILL CONNECTOR ACHIEVES CONNECTS TO MARKER I PREFERRED POSITION IN GATEJ MARKER HUNTS FOR IDLE TRUNK LINK vv ^ ., , pp^.-Tpp- „, .^^ FRAME WITH IDLE REGISTER 1^~ * ^^^ REGISTERS BUSY MARKER SEIZES CHOSEN TRUNK LINK FRAME-X TRUNK LINK FRAME DELAY MARKER LOCATES IDLE REGISTER X MARKER SEIZES LINE LINK FRAME ^X LINE LINK FRAME DELAY MARKER IDENTIFIES CALLING LINE -L MARKER MAY IDENTIFY ANOTHER I CALLING LINE MARKER CONNECTS TO LINKS AND i^ CHANNEL MISMATCH JUNCTORS AND MATCHES CHANNEL ^ * CHANNEL MISMATCH MARKER OPERATES CROSSPOINTS X MARKER RELEASES LINE LINK FRAME X AND TRUNK LINK FRAME ^ MARKER RESTORES TO NORMAL X DIAL TONE TO SUBSCRIBER ^X DIALING 1 I END OF DIALING-- ^X * MARKER MUST TAKE ALTERNATIVE ACTION IF IT ENCOUNTERS THIS CONDITION Fig. 3 — Establishing a dial tone call. THROWDOWN MACHINE FOR TRAFFIC STUDIES 307 CHARACTERISTICS OF SUBSCRIBERS AFFECTING DATA Since the purpose of the throwdown machine is to evaluate per- formance and traffic capacity of a simulated switching system under conditions met in service, subscriber data fed into the machine must represent, as nearly as possible, the characteristic actions of real sub- scribers. Fortunately, telephone subscriber characteristics can be studied in working switching systems which are similar to the one under throw- do^\^l evaluation. Little error is introduced by such subscriber data TIME I POSSIBLE DELAYS CONTROL FUNCTIONS AND DIFFICULTIES REGISTER SEIZES MARKER CONNECTOR X DELAY FOR MARKER CONNECTOR I DELAY FOR MARKER (INCLUDES MARKER CONNECTOR CONNECTS TO MARKER X DELAY TILL CONNECTOR ACHIEVES I PREFERRED POSITION IN GATE) MARKER RECORDS TERMINATING NUMBER X AND LOCATION OF ORIGINATING LINE ^ MARKER TRANSLATES OFFICE CODE ^X MARKER HUNTS FOR IDLE TRUNK LINK i ^ All TR1IMK-LINE BUSY MIS- MATCH -X— -X — ^x-- SET REGISTER FOR TONE RELEASE TRK LK FR, LLF MARKER RELEASE ■MISMATCH I MISMATCH i MIS- MATCH X---X — ^x- — MISMATCH LINE BUSY MIS- MATCH I MIS- MATCH -R— :^-' MISMATCH — f MISMATCH i "f MIS- MATCH -X---X — ^x Fig. 5 — Possible variations in handling an intraoffice call. THROWDOW^ MACHINE FOR TRAFFIC STUDIES 309 inasmuch as subscriber behavior is very slightly influenced by the type of system ser\dng their telephones. Subscribers, although they are indi\dduals, exhibit many ''group char- acteristics" dictated not by the requirements of telephone conmiunica- tion but by their mode of life. This fact allows statistical treatment of many observed action distributions Avithout introduction of significant error. However, these group actions also present problems of congestion in telephone plant which require detailed throwdown study for solution. As an example of group characteristic, subscribers do not originate a steady barrage of calls over the twenty-four hours of the day. During mid-morning and mid-afternoon hours traffic is built to a peak value, whereas during certain of the remaining hours it is reduced to a mini- mum. In some residential areas peak traffic may also occur during the early evening. Throwdown evaluations of simulated smtching systems, however are primarily concerned with the busy hour, the hour in which the greatest number of calls are originated, regardless of its actual time of day occurrence. Useful datum obtained from busy hour field observations is the calling rate per subscriber (calls per hour) which can be used to set up traffic load conditions on the simulated smtching system. The calling rate characteristics can be measured as average calls per hour placed by subscribers in a number of group classifications. An example used in a particular throwdoA\Ti study of simulated system response to a given traffic load is given in Table I. The values given in this table represent average day to day calling rates. Weather conditions, pre-holiday peri- ods or special events have been found to raise substantially the average calling rate in affected classifications. Values adjusted for these condi- tions are useful in projecting percentage of overload that can be offered to systems engineered for average daily loads. Subscribers, however, in originating calls, act independently Av-ithin their classified group in maintaining the average calling rate. Originating times of calls, therefore, occur at random A\dthin the hour. ThrowdoA\Ti input data representing subscriber originating time behavior are produced by assigning to each call, of the total Anthin the studied hour, a six digit number from a list of random niunbers. If the hour is divided into one million parts the assigned random number determines the miUionth part of the hour in which each call aaiII originate. Observations made in the field have shoAATi that subscribers, upon receiving dial tone, do not always follow through to dial a full code. Among possible causes are failure to hang up after completion of a call, answering the wTong telephone where two or more are adjacent, diahng 310 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table I Group Classification Heavy demand individual line subscribers (such as doctor and professional services Medium demand individual line subscribers (such as small business and some residential subscribers) . . . Light demand individual line subscribers (mainly resi- dential subscribers) P.B.X. line subscribers (such as large businesses, hotels, railroads, etc.) Average'^ Calling Rate Per SubscriberjDuring Busy Hour 5.0 0.85 0.02 3.0 before dial tone, and forgetting the number. Such actions produce waste usage of equipment within the switching system, and their studj^ is pertinent to producing throwdown data. To simpUfy this study, all subscriber actions involving the alerting of central office equipment are designated "subscriber starts" and divided into four categories as given in Table II. Since "no dials" and "partial" dials are largely due to subscriber errors in properly originating calls, many of these calls wdll be originated upon discovery of the error. False starts, on the other hand, are attributed to accidental origination with no intent to place a call. A flow chart illus- trating these actions is show^n in Fig. 6. The importance of this sub- scriber behavior is indicated by the percentage of waste usage calls (FS, ND, and PD) to ultimate good calls. Pen recorder tapes taken at par- ticular central offices showed waste usage calls at 30% and good calls at 80.5% of ultimate good calls. For these specific cases false starts repre- sented 7.5%, partial dials 7.5%, and no dials 15% of ultimate good calls. Table II Category Description Good calls False starts No dials Partial dials Calls on which the subscribei; waits for dial tone and then dials the required full code Calls, on which a sender or register is seized, but which are abandoned in less than two seconds without dialing Calls lasting longer than false starts but on which no dialing occurs. These calls may exceed a certain length "time- out" period and be given a distinctive tone Calls on which less than the required full code is dialed. These calls may be held beyond a certain length "time- out" period and be given a tone THROWDOWN MACHINE FOR TRAFFIC STUDIES 311 It was also found that approximately 90% of the partial dials and the abandoned no dials were reoriginated. These percentages are quoted only to illustrate subscriber behavior under certain conditions of cen- tral office load at a particular office. Type of service, load conditions on the central office, and location can effect these percentages. For a more detailed analysis, see "Dialing Habits of Telephone Customers."* Fig. 6 illustrates only the group behavior of subscribers. Individually, the subscribers will hold equipment on abandoned no dials, abandoned partial dials, and false starts for varying amounts of time. These varying individual holding times can be quantized into several average values which are equally likely to occur, or may be averaged to one value as shown on Fig. 6 depending upon the throwdown study requirements. The holding times on calls receiving tone are usually assumed to cease a few seconds after tone is received. Subscribers, as indi\'iduals placing ultimate good calls, spend varying amounts of time, after receipt of dial tone, before start of dialing and INITIAL SUBSCRIBER STARTS GOOD CALLS FALSE STARTS ALL END WITHIN TWO SECONDS AFTER DIAL TONE IS RECEIVED NOT REORIGINATED NO DIALS PARTIAL DIALS RECEIVE TONE AFTER TIME-OUT ABANDONED BY SUBSCRIBER BEFORE TIME-OUT TONE RECEIVE TONE AFTER TIME-OUT ABANDONED BEFORE TIME-OUT TONE HELD W SECONDS HELD X* SECONDS i 1 NOT REORIGINATED HELD y SECONDS REORIGINATED HELD p SECONDS HELD a^ SECONDS I i REORIGINATED r I HELD Z SECONDS REORIGINATED *UNDER CONDITIONS OF ALL REGISTERS BUSY IN THE NO. 5 SWITCHING SYSTEM, THE TIME-OUT PERIOD IS AUTOMATICALLY DECREASED ULTIMATE GOOD CALLS Fig. 6 — Simplified characteristic action of subscribers in converting initial subscriber starts to ultimate good calls. * Charles Clos and Roger I. Wilkinson, "Dialing Habits of Telephone Cus- tomers," Bell System Tech. J., 31, pp. 32-67, Jan. 1952. 312 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 in dialing a full code. This behavior affects the holding time of registers receiving the dialed digits and must be considered in throwdown studies. Data collected on combined waiting and dialing time characteristics show a frequency and time distribution that can readily be quantized into a number of values, each equally hkely to occur. When the number of values for a particular throwdown study are determined, each quan- tized dialing time is represented by a number. Each ultimate good call is then assigned a number from a random list of the representative numbers to estabUsh the dialing time of the call. Ultimate good calls will develop one of three terminating conditions attributable to subscriber behavior: 1) DA, called subscriber does not answer, 2) busy tone because of called subscriber line busy, or 3) answer by called subscriber. It is assumed from analysis of ''don't answer" studies that, for certain throwdown evaluations of the smtching system, approximately 10% of the ultimate good calls meet the DA condition. The nimiber of busy tone terminations, of coiu*se, Avill develop during the throwdown study as a result of the average originating and termi- nating calling rate per subscriber served by the system. Most subscribers, upon encountering a line busy condition make subse- quent attempts to reach the called line. The number and frequency of attempts made depend upon the individual characteristics. A detailed analysis of this characteristic, suitable for use in throwdown studies, has appeared in a paper by Charles Clos.* When calls are answered by called subscribers, the connections will be held for varying amounts of time. It has been determined from field observations that the frequency distribution of these holding times is closely approximated by an exponential distribution. For throwdown purposes a simplifying assumption can be made that holding time is not a continuous variable but is quantized so that a particular holding time will have one of several values. To determine these values an ex- ponential distribution having the proper average is plotted as shown in Fig. 7. The area under the curve is then divided into the required number of equal parts (ten, for this example). A central value of holding time is determined to represent each subarea. Ten holding times are thus pro- duced which are weighted according to the exponential distribution and which are equally likely to occur. These holding times can be designated 0 to 9 and assigned to the calls by choosing one-digit numbers at random for each call. * Charles Clos, "An Aspect of the Dialing Behavior of Subscribers and Its Effect on the Trunk Plant," Bell System Tech. J., 27, pp. 424-445, July, 1948. THROWDOWN MACHINE FOR TRAFFIC STUDIES 313 0.6 O >- =! 0.4 CD < to a 0.2 \ \ \ \ hs \ \ \ V v^ L ij i±j L± ii i i i 1 i i 1 ' _ >^ 300 400 TIME, T, IN SECONDS 500 Fig. 7 — Distribution of holding times. GENERAL PLAN OF THE MACHINE The broad plan of the throwdown involves a division of work be- tween the team of operators and the circuits of the machine. The circuits keep track of system events, resolving complex sequences of actions concurrently taking place. Their chief purpose is to present signals to the four operators so that they mil be able to perform the right actions at the proper time and thus dovetail together the events for a large number of calls in progress. The operators keep manual records of the busy-idle states of items which occur in large quantities, such as lines, trunks and links. They also perform the searching operations necessary to locate these items when they are required to be made busy or idle. In general the circuits signal the operators to perform actions; the operators in turn signal the circuits that the action is completed or some appropriate alternative taken. The circuits then determine the next action and present corresponding signals to the operators. Thus the circuits largely control the sequencing of events. However, in some cases where the sequencing would require extensive circuitry, the operators assist in determining sequence. For example calls returning to the control circuits after a subscriber completes dialing are interleaved with new calls according to written records maintained by the operators. Releasing 314 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 connections are also placed in proper time sequence by the operators according to written records. The actions of the operators are checked electrically in many cases. An improper setting of certain switches, which is inconsistent with the state of the system at the time will in some cases give an alarm and in others, block the progress of the machine until the error is corrected. Certain actions performed by the operators involve the insertion of plugs into jacks. Where alternative actions are possible in response to a given signal, separate groups of jacks correspond to the several alternatives. Insertion of a plug into a particular group will signal the circuits as to which alternative is taken. The circuits make one of several keys effective and the operator must then depress this key (corresponding to the action she has taken) before the circuits will advance. Where several operators must cooperate to perform a given action, signals are passed between the operators by means of keys and all opera- tors must respond before the circuits can advance. Wherever possible, overlap operation is employed. Signals are presented to several operators simultaneously and each operator starts the indicated action, signaling the circuits w^hen it is completed. When the signal from the last opera- tor is received, the circuit advances. In some cases an operator is allowed to start a particular action before the stage has been reached at w^hich this action is required. For example the information necessary to choose links and determine a suitable path through the interconnecting network of the No. 5 crossbar system is available before it is necessary to establish the connection. Since this search is time consuming, the operator is allowed to start as soon as the information is available. At the proper time she is given a signal to complete the record of this connection, or if the call has been blocked before reaching this stage, she is given a ''back out" signal instructing her to restore her records to their previous con- dition. Since this is a rare condition occurring less than one time in 100 tries, little useless work is done and the overall action is speeded up. A block diagram showing the relations between the operators and the various major components of the machine is shown in Fig. 8 The CLOCK controls the flow of time in the machine and gives an indication of simulated present time in the traffic run being tested. The time de- tectors, of which three are used, provide a means for the operators to indicate a future time at which some action must be taken and be sig- naled when this time arrives. The block labeled control circuits, which present action signals to the operators, represents the main body of circuits which maintain a current record of the states of the various THROWDOWN MACHINE FOR TRAFFIC STUDIES 315 complex units of the system, such as markers. The gate provides a means for operators to feed traffic into the machine and determines the order in which each working item of traffic is taken up by the control circuits. The RANDOM CIRCUIT is an electronic ''roulette wheel" which permits the operators to make random choices in disposing of certain traffic items. It provides random selections varying from one out of three to one out of ten. The individual record of each call is made on a card of the form shown in Fig. 9 which is called a ''call slip." These are of various types identified by distinctive designations and colors for the several varieties of calls which may occur. The basic types are: dial tone, intra-office, incoming, and outgoing. As the call progresses through its various phases the slip is passed from one operator to another, each operator retaining the slip while it is in the phase mider her control and recording on it in designated spaces the nature and time of the events taking place. When the call is com- pleted, the shp carries a complete record of the call including the designa- tion number and time of seizure of the various circuit units used in estab- lishing the connection and the nature and duration of any delays encountered. GATE CONTROL CIRCUITS STOP- START TIME DETECTORS TIME PULSES CLOCK Fig. 8 — Communication between the machine components and the operators, 316 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Fig. 9 — Call slips on which records are made in the throwdown machine. DESCRIPTION OF POSITIONS The throwdown machine consists of five positions as shown in the plan view of Fig. 10. A photograph showing all but one position appears on Fig. 1. This division into positions results from the requirements of simulating the components of the No. 5 crossbar system and equalizing the work load on the attending operators. The arrangement of positions, as shown, provides a continuous clockwise flow of records and other items that must be passed from operator to operator. The five positions are known as: originating position, gate position, marker position, match position, and assignment position. Four operators attend the five posi- THROWDOWN MACHINE FOR TRAFFIC STUDIES 317 f --^ a> > a 03 ^ \^^^^ X a. ceo lUI- bC aia: < HI ^ 318 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 tions, the gate position being jointly served by the operators stationed at the originating and marker positions. Each operator is given the same name as the position at which she is stationed. The general plan of traffic flow is such that all calls are originated or restarted at the originating position. For this reason, the major equip- ment items of this position include: the subscriber line array, which represents by pegs all of the subscriber lines associated with the switch- ing system under throwdown evaluation; the trunks incoming to this system; and certain of the system components associated with incoming calls. The call slips, sorted in sequence as to originating times, are stored at the originating position for use in originating calls. Means are provided for presetting the times at which the next origi- nated or restarted calls will enter the throwdown machine. When actual time coincides with a time thus preset, the system stops and signals the originating operator to serve the waiting item. The originating position does not provide for actually entering an item into the machine, but indicates the time of entry, stops the machine, and supplies the items to be entered. For example, the operator, when signalled to originate a dial tone call in accordance with information furnished on the call slip, selects, from the line array, a peg representing the subscriber line, asso- ciates it with the call slip, and passes the two items to the gate for entry. Since the means for determining the busy or idle condition of a sub- scriber line are provided by the presence or absence of that line's peg in the array, the originating position also enters into the operation on incoming and intraoffice calls. The gate position, which simulates the marker connectors of the No. 5 crossbar system, serves as the entry point for all calls. The originating operator inserts call slips and pegs into the gate position from one side to start the call. The marker operator removes them at the other side for processing in the marker position. Relay circuits associated with the gate position control the flow of traffic through the gate in accordance with actual No. 5 crossbar operation. The marker position provides means for associating the call slips with the individual simulated markers of the switching system. Since these markers control processing of the calls, the principle records of the calls' progress are obtainable through this association. The records are kept as time entries on the call slip and marked adjacent to action lamps determining these entries. At the top of each marker unit in the marker position are cords which provide access to the switching system components under control of the marker. These components are line link frames, number groups, sender subgroups, and marker connectors. As the call progresses, the THROWDOWN MACHINE FOR TRAFFIC STUDIES 319 marker operator will connect and disconnect the marker with these components as directed by the action lamps pro\ided. The principle purpose of the assignment position is to provide equip- ment for test and choice of a trunk on each call. A trunk jack array which includes all trunks and registers of the s^\'itching system therefore ap- pears on the face of this position. Since, in the actual Xo. 5 crossbar system, testing and choosing of trunks are performed by the markers, an extension of each marker also appears in the assignment position. The assignment operator, when required, makes a visual test for idle trunks to the proper destination. She then chooses and associates one of these idle trunks with the active marker by means of a marker cord which is plugged into the trunk jack. In addition to this principal function of trunk choice, the assignment position is equipped to determine the disposition of calls when the marker has finished setting them up. Means for ascertaining conversation time, dialing time and other types of holding times is provided. The positions so far described have simulated only the two ends of a connection, the subscriber line and the trunk or register. To complete the connection a channel must be set up between these ends. In the actual switching system the marker matches a line link, a junctor, and a trunk link to form the channel. The match position is provided to simulate this action. The match operator, through visual inspection of a set of channel cards, tests and makes busy the channels for each connection. Information as to the originating subscriber line reaches her through the pass box in the form of the upper portion of the call slip. Fig. 9. Information as to the trunk or register choice is passed ver- bally by the assignment operator. Since this operation is a function of the marker, marker extensions appear at the match position. These extensions are pro\'ided Avith action lamps to guide the match operator and with, keys to inform the simulated marker circuits of the results of the match. A pass ^\dndow is cut between the marker position and the assign- ment position for passage of the call slip and originating peg at the time of marker release. Similarly, a pass box is provided for passing the upper portion of the call slip (match ticket) from the match operator to the assignment operator. The assignment operator is charged ^\ith the disposition of these items. THE TIME SYSTEM The timing system of the throwdo^\'n machine is based on a start- stop system of time pulses. Pulses generated by the clock. Fig. 11, drive indicators wluch display time and also drive circuit elements which 320 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 count time units to cause events to occur in the proper sequence. When these time counters reach a stage where some action is to be per- formed by the operators, "stop-time" signals are produced. These sig- nals lock in and cause the clock to halt. Simultaneously, action signals are displayed to the operators. When each operator completes the in- dicated action she depresses an appropriate key at her position which extinguishes the action signals and removes the stop-time signal. When the last operator has responded, all stop-time signals are removed, the time lock is opened and the clock again advances. Thus the clock, in effect, takes time-out while the operators perform the various manual searching and recording actions necessary to simulate and tabulate the performance of the crossbar system. In the throwdown machine, each clock pulse represents one millionth of an hour or 3.6 milliseconds. This quantitization of time is based on two considerations. The first is that it is convenient to represent a particular time during an hour by a six-digit decimal number. The second is that the time represented by one time unit (3.6 milliseconds) is well under the average acting time of the relays and switches used in the system being simulated. Thus the events taking place in the actual system can be reproduced in sufficient detail. Some events of longer time duration Fig. 11 — Block diagram of the time system. THROWDOWN MACHINE FOR TRAFFIC STUDIES 321 are timed in less detail. For this purpose the clock is also arranged to deliver pulses at one tenth and at one hundredth of the basic pulse rate. The clock, circuit wise, is a form of free running relay pulse genera- tor. It consists of a series of relays in which the first, in a released con- dition, causes the remainder of the series to operate in sequence. When the last relay of the series operates, it causes the first relay to operate. The remainder then release in sequence. When the last relay releases, it causes the first relay to release. This cycle, if not interfered with, is repeated continuously to produce pulses representing units of time. Time, thus, can be stopped by the simple expedient of allowing the stop time signals to hold or ''lock" the first relay in the operated state. Time is visually indicated in units and hundreds of units at the opera- tors' positions by a group of telephone type message registers termed the time counters. Certain of the registers indicate present machine time for action recording purposes. Others are set at a specified number of units ahead of present time to indicate future times at which held items will be released or re-entered into the system: To drive these counters and to safeguard their integrity, the counters are substituted for the last relay in the clock pulser relay series. The operating windings of all units counters are connected in parallel. Con- tacts, which make on each counter when the individual coimter is ad- vanced, are all connected in a series circuit to form the last relay con- tact. Failure of any units indicator to advance ^^dll, therefore, interrupt the pulse generator cycle and stop time until the trouble is cleared. The integrity of the hundred units time counters is guarded in the same manner. On each hundred pulse when these counters are advanced, their windings and contacts also form a part of the pulser circuit. The basic pulse repetition rate of the clock is approximately four pulses per second, being determined by the acting time of the counters and the various circuit elements which the clock pulses must drive. Since each pulse represents 0.0036 seconds, the ratio of basic machine time to real time is in the order of 70 to 1. The clock pulses, Fig. 11, are counted by two types of time switches. One type, the control circuit time switch, is associated directly with the component control circuits of the throwdo^vn machine. These time switches are not continuously driven but are automatically connected to the clock as required to time events in the progress of a call. The con- trol circuit time s^vitches, as discussed in more detail later, are returned to zero after each usage in preparation for timing the next event. The secoud type of time switch, designated the master time counter, 322 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 is continuously driven by the clock. This counter records simulated present time throughout the entire running of a throwdown test. The master counter does not, in itself, generate stop time or action signals. Its primary purpose is to furnish correct present time to the time comparator circuit where such signals are generated. It also con- trols a pulse divider which furnishes pulses at one tenth and at one hundredth the basic pulse rate. Since failure of the master time counter to advance on each pulse would introduce present time errors which are cumulative over the throwdown run, the integrity of this circuit is rigidly guarded. Checking circuits verify that each basic clock pulse advances the master time counter. The checking circuits, not shown on Fig. 11, upon detecting a failure to advance, produce a stop-time signal and an alarm signal which can be released only after the counter is brought to correct time. The master time counter consists of pulse driven rotary switches arranged so that each switch represents one decimal digit of time. To count one hour of simulated time, as provided in the master time counter, six switches are necessary. These switches record 000000 to 999999 units of 0.0036 seconds. Means are provided for presetting such items as subscriber originating times, incoming call originating times, and hold release times by the time detectors. When a time so set coincides with simulated time, the clock is stopped by a stop-time signal and an action signal indicates that a call is to be originated or a held item released. The time detectors consist of sets of ten-position manually controlled switches with each switch representing a decimal time digit. Since simulated time is divided into millionths of an hour, these switches are preset to the millionth interval, say 003162, in which an event is to occur. Information from each switch is transmitted to the time compara- tor, Fig. 12. Also transmitted to the time comparator, from the master time counter, is the simulated machine time, say at present, 003159. When the master time counter advances to a time 003162 which coin- cides with the detector time, the time comparator generates a stop- time signal to stop the clock. An action signal, associated with this particular time counter in coincidence is also lighted. The operator, after taking the appropriate action, resets the time counter to a future time — the time of the next waiting item in the category — and depresses the start time key. Checking circuits prevent advancement of the clock should the time detector accidentally be set to a time value which has already passed in the throwdown run, in the example, to a value less than 003162. THROWDOWN MACHINE FOR TRAFFIC STUDIES 323 Three time detectors are provided in the throwdown machine. One each is used for setting originating times of subscriber starts and of incoming calls. The third is used for releasing held items. Since the holding times of these items are measured in hundreds of time units the last two digits of the time interval are dropped, and only four switches are required. It has been mentioned that the ratio of basic machine (clock) time to real time is in the order of 70 to 1. However, in operation, the flow of time is halted frequently to permit actions by the operators. The average interval between stops in the traffic runs which have been processed is less than 10 time units. Thus the machine time is only a small fraction of the time consumed in processing a traffic sample. The ratio of total processing time to real time has turned out in practice to be betw^een 1000 and 2000 to 1 depending on the nature of the traffic sample being tested. GENERAL PLAN OF THE CONTROL CIRCUITS The major part of the throwdown machine circuitry is associated with the marker sequence and timing controls and the gate preferencing arrangement. The circuit plan followed in these two cases will be briefly described in order to illustrate how the throwdown functions were implemented. The gate circuits simulate the action of the marker connector circuits of the No. 5 crossbar system which control the access of line link frames and registers to the markers. These circuits assign traffic to idle markers according to the preference rules used in the actual system. The circuits resemble corresponding circuits of the system. They employ two relays per connector and one relay per marker and are arranged with cross- connection terminals so that the preference order and number of con- nectors can be varied as required. Each call handled by a marker consists of a series of events occurring in time sequence with time intervals between events corresponding to the "work time" consumed by the marker in performing required func- tions. The sequence of events is not fixed at the start of a particular call but may be altered from stage to stage depending on the particular busy and idle conditions encountered. A block diagram of the control circuits used for simulating this action is shown in Fig. 12. They con- sist of a number of individual circuits provided on the basis of one per marker together with common circuits whose use is shared by all markers. The fundamental plan is based on the use of two rotary stepping 324 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 switches. One of these, the sequence switch, determines the sequence in which events take place, while the other, the timer switch, measures work time between events. In the throwdown machine each possible event which may occur in setting up a call is represented by a position on the sequence switch. Circuits through the switch cause action signals for the event represented by its position to be displayed to the operators. All of the events for one type of call are arranged in order on the switch terminals so that by omitting certain positions (events) all of the vari- ations of this type of call can be represented. At the conclusion of an event the switch is directed to the position corresponding to the next appropriate event according to the setting of ''memory" relays located in each marker unit. These relays operate at various stages of the call in response to key signals from the operators indicating the conditions they have encountered in attempting to respond to action signals. Thus, significant events are recorded in order to control the future progress of the call. To provide for all types of calls it was found necessary to furnish three sequence switches for each marker. These switches have 22 positions and six arcs. Five arcs are used for displaying signals and for control while the sixth arc is used in conjunction with the timer switch to control the time at which signals are displayed. A timer switch is individual to each marker. It is a 22-position, six-arc switch arranged with auxiliary relays to count a maximum of 105 clock pulses. Terminals of the timer switch --o O TIME SIGNAL ARC ARCS SEQUENCE SWITCH Fig. 12 — Block diagram of marker progress circuits. THROWDOWN MACHINE FOR TRAFFIC STUDIES 325 are cross-connected to the time arc of the sequence switch to fix the work time preceding the event represented by each sequence switch position. In the general operating scheme, the sequence switch stands at a posi- tion representing the next event to take place. The timer switch is started from normal and counts clock pulses until it reaches the terminal cross- connected to the position on which the sequence switch stands. This initiates signals which stop the clock and cause action signals to be dis- played. When the operators respond, the sequence switch advances to the terminal for the next event in the call, the time switch returns to normal and the clock restarts. The time switch then counts time units leading to the next event. Since several markers may be in use at the same time in different stages of their calls, two markers can reach an action point during the same clock pulse. A lockout circuit insures that only one marker at a time displays its action signal. At any time that a marker stops the clock, the timer switches of all other markers halt but continue their count when the clock is restarted. Thus, relative time relations are maintained while a true count of time consumed by an operating system is obtained. The circuit action can be illustrated by a discussion of the events in a dial tone call. This call represents an attempt by a marker to establish a connection from an originating subscriber line to a dial register. The possible sequences of events are diagramed in Fig. 13. As indicated, this class of call employs eight sequence switch positions in addition to the normal position. The call starts when the gate circuit assigns an idle marker to a call which has been originated at the proper time and placed in the gate. The assigned marker is prepared to process this type of call by the operation of an associated "class" key which selects and advances the sequence switch which carries the events of a dial tone call. Advance of the switch is from normal to Position 1 to control, at the proper time, signals for the first event, namely, seizure of a trunk link frame and selection of a register. The timer switch is set at zero and in a condition to step one terminal at a time in response to clock pulses. Terminal 1 of the sequence switch time arc is cross-connected to a terminal of the timer switch representing the marker work time between the time the marker is first seized and the time it attempts to seize a trunk Hnk frame. The operator in control of the marker now operates a start key and the clock 'starts pulsing, each pulse causing the marker timer switch to advance one step. When the specified work time has elapsed the timer switch reaches the terminal connected to Position 1 of the sequence switch. This passes a signal from the timer switch through the sequence 326 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 smtch causing the clock to stop and the signals associated with Position 1 of the sequence switch to be displayed. Signals at the assignment position identify the particular marker and instruct the assignment operator to obtain a trunk Hnk frame and a register. Three conditions may occur: (1) a trunk hnk frame frame and register may be idle and available, (2) a register may be idle but the frame on which it appears is busy, be- ing held b}^ another marker, or (3) all registers may be busy. In condition (1) the assignment operator obtains the frame and register according to the operating procedures and operates an ok key at her position. This extinguishes her signals and passes a signal to the marker operator instructing her to record the present time in the space provided for trunk seizure on the call slip. The proper space is indicated by one of a row of lamps beside the call slip. The marker operator then operates the START key for this marker causing the marker timer to return to normal and the sequence switch to advance to Terminal 2. The clock starts if no other marker is waiting to display signals. In condition (2) the assignment operator observes that there will be a delay in obtaining a trunk link frame. She inserts a connector cord for this marker in a ''delay" jack and operates a delay key. This extin- guishes her signals and passes a signal to the marker operator to record SEQUENCE SWITCH POSITION 0 1 EVENT MARKER SEIZURE SEIZE TRUNK LINK FRAME TEST REGISTERS ALL REGISTERS BUSY SEIZE LINE LINK FRAME MAKE CHANNEL MATCH release line link frame and trunk link frame seize trunk link frame test registers (second trial) seize line link frame make channel match release after all registers busy release after second mismatch MISMATCH ALL REGISTERS BUSY NORMAL RELEASE )- — Fig. 13— Possible sequences of a dial tone call. THROWDOWN MACHINE FOR TRAFFIC STUDIES 327 on the call slip that a delay has been encountered. The marker operator then operates the start key for this marker. This returns the timer to normal but since a delay condition has been established in the marker circuit by the actions of the assignment operator it does not advance the sequence switch but allows it to remain in Position 1. The marker in a delay condition does not permit its timer switch to step but removes the halting signal from the clock to allow time to advance and other markers to be ser^'ed. All signals for this marker are extinguished during this time. As time advances some other marker ^vill release a trunk link frame. This passes a signal through the delay jack to the delayed marker causing it to display again the signals requesting a trunk link frame and a register associated with sequence switch Position 1. The operators and circuits then proceed exactly as when these signals were first displayed. In condition (3), the assignment operator observes that all registers are busy and operates her all busy key. This operates a memory relay in the marker circuit recording that the all busy condition has been encountered and that the future progress of the call should follow the sequence indicated in Fig. 13 by the side branch at Position 1. With all registers busy it is impossible for the marker to establish a connection. As the side branch shows, the alternative is to release and restore to normal. (Later attempts to serve this subscriber will be made until an idle register is obtained.) Thus, with the all busy condition recorded on the memory relays, the circuit will cause the sequence switch to advance, running over positions representing intermediate actions and coming to rest in Position 6 where it is prepared to display signals for the release of the marker. The operation of the all busy key also started the clock and restored the timer switch to normal so that it could measure suit- able work time before displaying release signals. The call progresses through successive events in a manner similar to that described above. After obtaining a trunk link frame and a register as in condition (1), the sequence switch stands in Position 2 while the timer switch counts work time preceding a request for a line Unk frame. During this time other markers may request service causing the clock to halt, interrupting the advance of time in all circuits until the opera- tors have completed the required actions. After this marker has counted the specified time, a signal is passed through Position 2 of the sequence switch causing a request for a line link frame to be displayed at the marker position. A delay is handled as before, the marker waiting until the busy frame is released by some other marker. When the frame is obtained, a signal at the match position requests that operator to match 1^^^^ iiiii^ii^iiiaiii^ii^ Fig. 14— Relay and switch cabinets. 328 THROWDOWN MACHINE FOR TRAFFIC STUDIES 329 a channel before allowing time to advance. In case of a mismatch the match operator depresses a mismatch key. This operates a memryo relay which will cause the marker to follow the alternate sequence in- dicated by the side branch at Position 2 in Fig. 13. As a safety pre- caution, the MISMATCH key is made effective only at the time a request for a match is made so that accidental operation at other times will not disturb the circuit action. With a mismatch recorded, the sequence switch advances to Position 3 and at the proper time displays signals to release the line link frame, trunk link frame and register. Thus the call advances with the possible alternates of all registers busy at Posi- tion 4 or a second mismatch at Position 5. The call proceeds to one of its possible conclusions where frames are released and the marker be- comes idle. The dial tone call which has been described is the least complex call handled by the machine. The intraoffice call which was diagramed in Fig. 4 is the most complex. It requires 22 sequence switch positions to represent events and may have 92 possible sequences depending on conditions encountered. To take care of variations in sequence in all calls a total of nine memory relays is provided in each marker. The circuit equipment consists, largely, of telephone type electro- magnetic relays and rotary stepping switches. Approximately 800 relays are used. The total number of switches is 57. Of these 47 are of the 22- position, 6-circuit type while the remainder are of the 44-position, 3-cir- cuit type. This equipment is mounted in the two cabinets shown in Fig. 14 and in additional units located within the operating positions. Time indications are given by four-digit message registers, approximately 40 being used. The random circuit consists of a gas tube counting ring with several control relays. Output indications are given by miniature neon lamps. Signals are given to the operators by telephone switchboard lamps, 822 being used. Manual equipment used by the operators in sig- naling to the circuits consists of keys, switches and telephone plugs and jacks. The machine contains 187 keys and switches, 60 cords equipped with plugs and 509 jacks. PREPARATION FOR A THROWDOWN RUN There are two phases in the preparation for a throwdown run. One of these is the tentative engineering of an office of the size to be tested. The other is the preparation of data to represent traffic handled by this office. The first of these follows rather closely the general procedure that would be used in planning the installation of a new switching office. 330 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 The chief difference is that prehminary decisions concerning the size and general characteristics of the office to be tested will be made. For example it may be decided to test an office in the twenty line link frame size range serving mixed business and residential subscribers with a high percentage of calls completed within the office. Based on general knowledge of subscriber behavior, the number of subscribers necessary to present a suitable traffic load to this number of link frames will be determined. Values for average holding time and calling rate associated with these subscribers must also be developed. These may, in part, be determined from estimates or specific knowledge of the traffic capacity of the line link frames, taking care that the figures are typical of such a group of subscribers as determined by field observations. Since we are usually concerned with determining the maximum safe capacity of the system, full load or overload conditions will be assumed and the usual margins for future growth considered in engineering an actual office will be omitted. Decisions will also be made as to the number of other offices to which this office has trunks and the percentages of the total traffic originated and terminated in each of these offices. Having made these preliminary assumptions concerning the nature and environment of the office to be tested, the office is then engineered according to the best available information. The numbers of registers and markers are determined and arranged in connector groups accord- ing to standard procedures. The sizes of the trunk groups to various connecting offices are determined and the placement of trunks on the trunk link frames chosen according to the usual practice. All similar factors concerning quantities and arrangement of equipment are deter- mined. The throwdown machine is then set up to simulate this office. This will involve crossconnections in the gate circuits and arrangement and designation of the facilities provided for keeping the busy-idle records of such items as lines, trunks, registers and links. The second phase in the preparation is to produce data representing calls presented to the system during the time interval to be studied. This is accomplished by choosing a random number for each call. This number must contain a sufficient number of digits to specify all the pertinent data necessary to describe the call. These digits are assigned to represent certain factors. For example, the first six digits represent the time of origination. The next two digits specify the type of call. This is done on a percentage basis. For example, if 40 per cent of the traffic is to be locally completed, the numbers 00 through 39 in these places would indicate an intraoffice call, if 25 per cent is to be outgohig to other offices the numbers 40 through 64 would indicate this type of THROWDOWN MACHINE FOR TRAFFIC STUDIES 331 call, and so on for the remaining types of calls. If divisions of less than one per cent are to be made, three or more digits could be used for this purpose. The meanings of certain of the remaining digits will depend upon the type of call. If the call is originated within the office, the next five digits will give the identity of the calling line in terms of its frame location. If the call is incoming from another office certain of these five digits will be used to specify the office of origination and the trunk num- ber. This again is on a percentage basis depending on the fraction of total incoming traffic expected from each office. Five other digits give the identity of the called line if the call is completed within the office; if outgoing they specify the terminating office. Other digits give the percentage of calls which will result in partial dialing by the calling subscriber. Since, as will be discussed later, there is a possibility that these will be re-originated later as good calls, all of the information for a good call is also determined for these calls. Additional random numbers determine various other aspects of the call such as the identity of the number group which will be used. As a suggestion to those who would undertake a throAvdown study, it is advisable to include a number of surplus "utility" digits in the original random number. It invariably happens that some factor is overlooked in the initial planning or arises during the course of a test and these digits can be used in making de- cisions in these cases. It should be noted that the random circuit is used for making certain random decisions in the course of a call at the time that a need for these decisions arises. For example, the holding time of an established con- nection and the probability of the called subscriber not answering is determined by this circuit. This circuit could have been eliminated by including digits in the original number for every possible situation of this type which might be encountered. This would cause much useless work in preparing the data since all situations are not encountered on every call. A quantity of random numbers must be drawn to provide the desired load on the office. This is not a simple process of determining the number of calls expected during a given period and drawing this number of random numbers. One factor is that in generating data by the above procedures it will be found that certain numbers will represent calls originated by lines Avhich are busy at the indicated time on a connection established previously. The number of such cases can be estimated from the expected number of busy fines in the office and a corresponding number of additional numbers drawn. When this situation is encountered 332 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 in the course of the run the call is discarded. The other important factor is a ''feedback" effect due to the calls meeting a situation which prevents successful completion and the probability that these calls are originated later. A simple example is where the called line is found busy. As previously mentioned, there is then a possibility that subsequent attempts will be made at a later time. All attempts make use of circuits and control equipment and should be considered in determining the load capacity of the office. Other situations which produce this effect are illustrated in Fig. 15. Lines entering the figure at the left represent classes of calls which enter the system. A certain number of calls will be partial dials, no dials and false starts regardless of the performance of the system at the time the call is originated. The partial dials represent cases where the subscriber makes an error in dialing or does not wait for dial tone and dials a digit before he is connected to a register. These may be aban- doned before the register obtains a marker or may "time out" in the register and be connected to an overflow tone trunk. In either case the subscriber may re-originate the call later. The probability of re-origina- tion has been estimated from field data and the dotted lines in the figure represent these reoriginated calls. In the throwdown machine the random circuit is used to determine which calls will re-originate and the elapsed time before the second at- GOOD CALL STARTS X GOOD CALLS COMPLETED REORIGINATED CALLS r,*r,-r,*, ABANDONED PARTIAL ^ ^_. DIAL^^' ~^ ME OUT {. NO DIAL ABANDONED ►- { TIME OUT DON'T ANSWER K BUSY i ^-K PARTIAL DIAL STARTS X ABANDONED j ^ — I I "\TIM E OUT ^-- NO DIAL STARTS X FALSE STARTS X ABANDONED ► { TIME OUT ► Fig. 15 — Composition of the load on a crossbar office. THROWDOWN MACHINE FOR TRAFFIC STUDIES 333 tempt. No dial starts represent such situations as a change of mind by the subscriber after Hfting the receiver or accidentally removing the receiver and are not considered to be correlated with later trials. The same is true of false starts which are momentary start signals often hard to explain. A certain number of partial dial and no dial calls are the result of dial tone delays which may occur during heavy load con- ditions. These branch from the ''good call" line in the figure and are due to the subscriber not waiting for dial tone and dialing part of the digits or even the entire number before being connected to a register. The probability of this occurrence will depend on the extent of dial tone delay. After dial tone delay on each call is known, successive uses of the random circuit determine whether the call is partial or no dial type and if the call is to be re-orginated at a later time. Rough estimates of the quantity of this type of traffic could have been made and included in the original traffic data. However, since they depend on the performance of the system there is a tendency toward a ''snowballing" effect and it was thought best to handle it as described in order to detect this effect. It can be seen that the exact load on a system is difficult to estimate on the basis of initial starts. The procedure then is to make the best possible estimate of initial starts necessary to produce a given load, taking into consideration all important known factors and, at the con- clusion of a run, make a count to determine the exact number of calls of various types handled. When the data for the various calls have been determined from the random numbers, the pertinent information must be transcribed on the call slips. For most calls originated in the office this will requre two call slips. One of these is for the dial tone stage of the call and is used in establishing a connection to a register. The other represents the later stage of the call where the register connects to a marker after dialing is completed and an attempt is made to establish an intraoffice or outgoing connection. The initial time of origination and the calling line number will be recorded on the dial tone slip. The slip for the second stage of the call will carry the calling line number and the called line or outgoing trunk number, but will not carry a time of origination. This is a function of dialing time and is determined at the conclusion of the dial tone stage by the random circuit. It is recorded on the second call slip at that time. The two slips are associated by recording the serial number of the associated slip on the dial tone slip. Incoming call slips carry the origination time, the calling trunk, and the called line numbers. Call slips for partial dial and no dial initial starts carry information similar to that on the dial tone slips. 334 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 In preparation for a run all slips bearing initial time entries are stacked in order of their times of origination. Slips for the second stage of a call are kept in a separate stack. At the conclusion of a dial tone action when the time of origination of the second stage is determined, the associated call slip is located, the time is entered on this slip and it is inserted in proper order in the stack of originating calls. In the case of re-originated calls, the information from the old call slips is recopied on a new slip mth the new time of origination and these slips are placed with waiting calls in the proper time order. Detailed Description of Equipment and Functions of THE Operating Positions THE originating POSITION The originating position, in relationship to the rest of the throwdown machine, appears to the extreme right of Figs. 1 and 10. Detailed views of the two wings of the position are shown on Figs. 16 and 17. The prin- cipal features of this position are arrays of jacks and wooden pegs rep- resenting subscriber lines and incoming trunks, together with associated time counters and detectors. The chief function of the originating opera- tor is to enter all calls into the system at appropriate times. This re- quires that the stacks of call slips, visible in Figs. 16 and 17, be held at this position. These call slips are arranged in order of their originating times, with latest time on top, and carry the originating line identifica- tion number so that line peg and call slip can be associated when the call is to start. The line array, split into two wings, contains jacks and pegs for all subscriber lines in the office. The jacks which are simply holes with no electrical function, are arranged in a coordinate grid to assist in quick location. Twenty frames are provided, each holding 500 lines for a total of 10,000 lines. The array is divided into 20 horizontal sections, running across the two wings, each representing a line link frame. Each frame is divided vertically into 10 subgroups of 50 lines, each representing a horizontal group. Since the directory number assigned to a subscriber line in a crossbar office is purely arbitrary and has no physical significance, it is not used in the present case for line identification. Rather, an equipment number, which represents the location of the line on a line link frame, is used. This is a five digit number, stamped on the peg, and made up THROWDOWN MACHINE FOR TRAFFIC STUDIES 335 NCOMING REGISTER ARRAY NCOMING TRUNK ARRAY as follo\Vs: Fig. 16 — Originating position — left side. HG— LLF— L where HG — ^horizontal group No. 0-9 LLF— hne link frame No. 00-19 L— Hne No. 00-49 The right upper five lines in each horizontal group of a frame (lines 45-49) are reserved for PBX use. A PBX can be assigned to hne jacks occupying the same relative location in a vertical row (20 line link frames- same horizontal group). The pegs in a PBX group are marked with a distinctive color to facilitate hunting. 336 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 RANDOM PREFERENCE CIRCUIT MONITOR Fig. 17 — Originating position — right side. Line pegs are removed from the array while they are occupied on a call. The jack thus remains empty during the busy interval. The origi- nating operator utilizes a hne peg and a call slip in originating each dial tone call. The incoming trunk and register arrays are located on the left column at this position as shown on Fig. 16. The incoming trunk array consists of pegs and jacks arranged in accordance with the trunk groups. The primary horizontal grouping is assigned a route number which is used only for identification purposes. The horizontal rows within a route rep- resent the number of the trunk link frame switch on which the trunk is located. Vertical rows represent trunk link frames. Provision is made for a maximum of 400 incoming trunks. Trunk identification is a four- THROWDOWN MACHINE FOR TRAFFIC STUDIES 337 digit number made up as follows: R— TLF— SW where R— route No. 10, 11, 12, 13 TLF— trunk link frame No. 0-9 SW— smtch No. 0-9 Jacks representing these same incoming trunks also appear at the assignment position. The incoming registers consist of short sleeves which can fit over the trunk peg. When a call is originated by a particular trunk as indicated by the time on a waiting incoming call shp, the trunk is associated with the correct type of incoming register by slipping the sleeve correspond- ing to the latter over the trunk peg. The peg and sleeve then accom- pany the call sUp into the system. Appropriate times for each action are entered on the call slip. The incoming register sleeves are held on arrays adjacent to the trunks. They are identified by a three-digit number: CONN— REG where CONN — marker connector No. 0 or 1 REG— register No. 00-19 Waiting holes are provided at the register array for holding trunks if all registers are busy. MARKER CONNECTOR OR GATE This is a bridging position between the originating position and marker position! Through it must pass all call slips and pegs requiring marker service. The originating operator places call sUps and pegs in the gate from one side and the marker operator removes then at the other. The gate, shoAvn on Fig. 18, provides jacks for pegs and slots for call sUps arranged in blocks corresponding to the individual marker connectors for line link frames, originating registers and incoming re- gisters. It is divided into two sections, a storage section in which calls wait for an idle connector, and an active section in which calls wait for an idle marker and removal by the marker operator. An associated relay circuit controls the flow of traffic through the gate, simulating actual No. 5 operation, and indicates by lamp the appropriate action. When a call shp is ready to enter the gate, the originating operator obtains the corresponding line or register peg and places the peg and slip in a jack and slot, respectively, in the correct connector block of the 338 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 storage section of the gate. The connector block number is ascertained from the peg identification or from the call slip. If the connector is free a CONN READY (r) lamp lights to indicate that the peg and slip should be advanced to the connector block in the active section of the gate. If the connector is not ready, the lamp will light at some subsequent time and the connector delay time will be noted on the call slip. If several calls pile up in a particular storage connector block, means are provided for maintaining correct preference for entry into the active section in the case of registers. With line originations, a random line preference circuit under key control at the right side of the originating position key shelf picks, on a random basis, the next entry into the active section of the gate (except in the case of preferred lines whose pegs will have a distinctive marking) and which have precedence over ordinary lines. Keys for a choice of one out of two, three, four or five are provided. With INCOMING TRUNK AND REGISTER WAITING FOR MARKER ORIGINATING REGISTER WAITING FOR MARKER LINE LINK FRAME MARKER CONNECTOI ORIGINATING REGISTER MARKER CONNECTORS- INCOMING REGISTER v-^^-R CONNECTORS—, MARKER PREFERENCE KEY ACTIVE SECTION INCOMING TRUNKS AND REGISTERS WAITING FOR CONNECTOR STORAGE SECTION SUBSCRIBER WAITING FOR CO Fig. 18 — The gate for control of marker connector preference. THROWDOWN MACHINE FOR TRAFFIC STUDIES 339 register originations, the operator controls the preference with the assis- tance of a lamp signal per connector and a written register preference record. Once in the active section of the gate, a call awaits an idle marker. When correct preference conditions are met, a conn closed (c) lamp lights, together with an attention lamp, to indicate to the marker opera- tor which is the next call to be served. The marker operator depresses the MKR PREF key in the connector block to lock in a signal at the marker position indicating the marker to be used on this call. The operator then removes the call sUp and peg from the gate and inserts them in the correct unit of the marker position. The match ticket portion of the call slip is torn off and passed to the match operator. On originating register controlled calls, the register match ticket will have been at- tached to the call slip and this also is passed to the match operator. MARKER CONNECTOR ARRAY SENDER GROUP CONNECTOR SENDER HOLD JACKS SENDER PEGS « o & o « I II li 11 H II H ll » n t MARKER UNIT Fig. 19 — ^Marker position. m^ \,> O Gi O • ©» O O • I) 15 Fig. 20 — Call slip ill marker unit. 340 THROWDOWN MACHINE FOR TRAFFIC STUDIES 341 The circuit associated with the gate assigns all calls and markers ac- cording to No. 5 preference arrangements and provides the correct gating action. This preference is set up on cross-connection field located within the position and may be changed to correspond to any desired system arrangement. MARKER POSITION A photograph of the marker position is shown on Fig. 19. The indi\dd- ual marker units, of which ten are provided, are disposed on the sloping work panel. On the array panel are jack arrays representing the Une hnk frames, the number groups, outgoing sender subgroups and senders, and the marker connectors. Connections to these jacks are made by means of plugs and cords located at the top of each marker unit. The call slip for dial tone class of call is shoT\^i inserted in a marker unit on Fig. 20. At the top of the slip, in the type and grig line boxes and opposite t grig and t cgnn, are typical entries as made by the originating operator. The marker operator has also made three time entries. The heaw dot at the top edge of the call slip indicates which of the six class keys to turn. The class keys condition the marker circuit to handle each type of call correctly. The hole at the top of the sUp lines up with a jack and is used to store temporarily the line peg (register or incoming trunk peg in the case of other types of call) as received from the gate; it also serves to locate and hold the call slip in the correct position. On the lower portion of the shp are spaces corresponding to all marker actions requiring association with other frames or circuits. These spaces line up with lamps on the marker unit which signal when and what action should be taken and indicate whether or not the time should be entered on the call slip. If a delay is encountered at an> point, a check is put in the associated delay block to assist in subsequent computations invohdng the call slip. The left hand row of lamps hghts to indicate when a direct action should be taken and a time written down. For example, when the third lamp from the left hand top lights (opposite LLF cord) the operator places the line link frame cord at the top of the unit in the correct jack in the llf connector array, and writes down current time in the space opposite the lamp. If the llf jack is occupied by another marker, the operator plugs the llf cord in a preference delay jack associated with the line link frame jack and places a check mark in the delay block. 342 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 THROAVDOWN MACHINE FOR TRAFFIC STUDIES 343 As discussed previously, certain marker functions are located at the assignment, originating and match positions. The choice of trunk link frames and trunks, for example, is made at the assignment position. Therefore certain lamps at the marker position light only to signal the writing doAvn of a time. This is true of the second lamp from top left which indicates register seizure. The lamps in the right-hand row are also used in conjunction with actions at other positions. They Hght only to indicate that a delay should be checked in the corresponding box. At the time of marker release, both left and right lamps Hght opposite one of the release categories. The specific type of release determines partially the further disposition of the call. The START key at the bottom of each marker unit is pressed after completing each action called for by the sequence lamps. Operation of this key puts out the lamp and permits the circuit to advance. ASSIGNMENT POSITION The assignment position, shown on Fig. 21, includes the trunk array, the register false start holding array, the traffic assignment equipment, the holding time book and counters, and indi\dduals units representing extensions of the markers. The chief function of the assignment operator is to test and choose trunks on each call, to determine the disposition of certain calls which encounter excessive dial tone or busy conditions, and to 'ascertain holding times of originating registers and trunks. The trunk array consists of jacks (holes) representing all the trunks, and jacks and pegs representing the registers. Line pegs are held in the jacks during conversation time to mark the trunks busy. The array is divided into vertical sections representing trunk link frames. The trunk groups are disposed in horizontal row^s so that trunk jacks of each group or route occupy the same relative position in each frame section. Thus a trunk-hunting action consists of picking a horizontal level in the array and searching along the level to the first idle trunk as indicated by an empty jack. The array provides more jacks than the total number of trunks so that most small trunk groups can be set up on individual horizontal levels. The trunk identification is a three or four digit number composed as follows : R— TLF— SW where R is route number 0-99 TLF is trunk link frame 0-9 SW is switch nimiber 0-9 344 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 The route number in some cases is required only for location purposes since many trunk groups are identified by name. It is assigned to a common group of trunks, maximum 100, or 10 per frame. However, in the case of outgoing trunks, the route number is required to identify the specific trunk group desired. The route number for originating reg- isters is the same as the marker connector number for each register and is unusual in that any route number from 0-9 represents the same reg- ister group. The register jacks occupy the lower trunk level on the array and are supphed with pegs which are used to originate the second half of out- going and intraoffice calls. During dialing time, the associated line peg occupies the register jack to mark it busy. The next higher level on the array is the group of incoming trunks. These are furnished on a two- jack per trunk basis and are, in effect, multiples of the incoming trunk array at the originating position. The two jacks are required to hold the terminating line peg and the incoming trunk peg. The route number for these trunks is significant only for the function of location since it is not necessary for throwdown purposes to segregate incoming trunks into groups as it is with outgoing trunks. The intraoffice trunks are divided into an A- and a B-group since more than 20 trunks per frame are required. The No. 5 marker can only test up to 20 trunks per frame at one time. The machine automatically allots calls between the two groups. Two jacks per trunk are required for originating and terminating line pegs. Unlike the incoming trunk jacks each jack of an intraoffice pair corresponds to a different switch location, although only one switch number is used for identification purposes. The outgoing trunk groups are disposed above the intraoffice trunks. For the most part, these trunks will be in small groups, each with its own route number. When a call is set up to an outgoing trunk, it is necessary for the marker operator to inform the assignment operator of the correct route number. Several of these trunk groups can occupy the same horizontal level to conserve space. Only one jack per trunk is required to hold the originating fine peg. Tone trunks (busy, overflow, partial dial, no dial trunks) are at the top of the array. Only one jack per trunk is required. Within the trunk link frame, the preference order in which a No. 5 marker tests trunks and the manner in which the order shifts from call to call is rather complex. In order to reduce the load on the assignment operator in trunk hunting, the actual trunk preference is approximated by reversing the direction of hunting within a frame group from call THROWDOWN MACHINE FOR TRAFFIC STUDIES 345 to call on a substantially random basis. Thus, when the operator deter- mines the trunk frame ^vithin which she ^vill hunt, she observes the LEFT and RIGHT lamps below the array for the indication as to whether to hunt from left to right or \'ice versa. TRAFFIC ASSIGNMENT One of the important functions of the assignment operator is to deter- mine the assignment of all calls at the time of marker release. For this purpose she is furnished with traffic assignment keys and indicators which appear to the right of her position. An adequate understanding of this feature requires a somwhat detailed explanation. With the aid of the traffic assignment controls and indicators, shown to the right of Fig. 21, the assignment operator performs the following specific functions : On Dial Tone Calls: The operator determines whether a call reaching a register should be classified as a good call (successful subscriber dial- ing), a PD call (partial dial — incomplete subscriber diahng) or an nd call (no dial — ^no subscriber dialing while connected to a register). A proportion of call sUps are originally marked as pd or nd (and also fs or false start) and on these this determination need not be made. If the call is of the pd or nd type, the operator determines whether it should be subtypes PDl, PD2, PD3 or NDl, ND2, ND3. The subtype affects the assumed time until the subscriber abandons the call. If the call is classified as good type, the operator determines which of several dialing times should be used. If the call is classified as pd, nd or fs type, the operator determines when the call is abandoned and whether or not it is routed to a tone trunk. On Calls Completed to a Subscriber: On calls completed via intraoffice, outgoing or incoming trunks, the operator determines whether the call is answered and which of ten holding times should be assigned for subscriber line and trunk. On Calls Routed to a Tone Trunk: On calls which are routed to tone trunks or given a tone signal from the register, the operator determines the trunk or register holding time and whether and when the call is re-originated. In making these determinations, the operator presses keys which cause a lamp to Hght either beneath a time counter or beside a designa- tion strip. The time counter, set ahead of present time, indicates trunk release time, register return time, etc., while the designation strip class- ifies the calls. The determining factors include the magnitude of dial 346 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 tone delay, probability, and whether or not all registers are busy. Where dial tone delay is concerned, it is determined by comparing the origina- ting time of a dial tone call slip with three dial tone delay time counters. These counters give present time minus 1, 2 and 9 seconds respectively so that matching the time of origination against them indicates whether the delay was < 1 sec, 1-2 sec, 2-9 sec. or > 9 sec The probability factor is obtained from a circuit w^hich is capable of lighting one lamp out of ten, one out of three, etc on a random basis when a key is depressed. By means of the circuit, calls can be assigned to various categories in correct proportion in accordance with the best available traffic information. Whether or not all registers are busy is indicated by the all register busy circuit controlled by the assignment operator. If the assignment for a trunk or a register is that it be held for a period of time and then released, the assignment operator makes use of the holding time book as described later. If the assignment is for a register to return with a bid for a marker, or a new call to return into the system (after encountering busy, for example), the time is noted on the call slip and the latter is passed to the originating operator for subsequent action. For new call return time, a letter designation associated with the signal lamp is also entered on the call slip. The letter is carried for- ward on the new call slip. If subsequent attempts of this same line meet busy or overflow, the letter designation is used to identify the same category of return time instead of using the random circuit. When a trunk call is set up, the trunk and one or two lines must be kept out of ser\'ice for one of several fixed holding times. There are ten different assigned holding times with an equal likelihood of an established call falling into any one of them as determined by the traffic assign- ment circuit. False start, tone trunk and don't answer connections provide four additional holding times. The holding time counters provided at the assignment position in- dicate present time plus a fixed holding time. Thus each counter gives the time at which a connection, set up at present time and assigned to that particular holding time, will release its elements back into ser\dce. Holding time starts for a given call at the time the release lamp lights at the assignment position marker unit associated with that call. At such a time, the assignment operator obtains the one or two line pegs of the call and plugs them in the trunk jack or jacks (identified by the frame connector cord), noting at the same time the trunk number. The operator then depresses a key which causes the traffic assignment circuit to light a lamp under one of the holding time counters, thereby assign- THROWDOWN MACHINE FOR TRAFFIC STUDIES 347 ing a release time. The trunk number and release time are recorded. At the end of the holding time, the operator removes the line peg or pegs from the trunk jacks and the pegs are returned to their home jacks. Presence of the Une pegs in the trunk jacks during holding time marks the trunk busy. The busy trunk numbers are recorded in a holding time book according to release times. Time units in the system represent millionths of an hour (0.0036 second). Each page of the holding time book covers 10,000 units of time and is divided into one hundred 100-unit blocks. An illus- tration of one page of the book is sho^^Tl on Fig. 22. Since the time unit is one millionth of an hour, each time unit in a one houi' series can be represented by a six-digit number. Of these six digits, as far as the long holding times are concerned, the last two are unimportant since 100 units is only 0.36 second. If they are dropped, the first two digits of a four digit holding time release figure give the 10,000 time unit group or page number of the holding time book and the second two digits give the 100-block on the page. Trunk numbers are entered in the book on this basis with the page and block number ob- tained from the assigned holding time counter. By this means the release times of all items appear in consecutive order in the holding time book. Holding time entries are always made several 100-blocks beyond the next release time, which eliminates con- Fig. 22— Holding time book. 348 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 fusion. Following the simple numerical order of release times, the opera- tor sets up the next release time on the holding time detector, similar to the time detectors at the originating position, which stops the system and signals when that time arrives. The operator returns the held items hsted in that 100-block to normal, crosses out the block and sets up the next following time on the detector. False start and register overflow release time entries are made in the same holding time book. A jack array is provided to hold the register peg until release time under these conditions. Since the longest holding time is six minutes (100,000 units), the opera- tor is never concerned with more than 11 pages at one times. With each page clearly labeled by a numbered margin tab, search time to make an entry is reduced to a minimum. MATCH POSITION The match operator performs the function of testing and making busy the switching channels through which each connection is set up. Her position includes ten marker units, each consisting of a group of lamps, and a set of files which bold the channel cards by means of Avhich the channel records are kept. The position is shown on Fig. 23. The No. 5 marker picks out a channel between a subscriber on a line link frame and a trunk on a trunk link frame by matching a Kne link, a junctor and a trunk link which are capable of being switched together to connect the two end points. A schematic of the system is shown on Fig. 24. Each horizontal group of subscribers has direct access to a set of ten Une links which connect to the ten junctor switches of the frame. Each Unk of the set can be given a number from 0 to 9 corresponding to the junctor switch on which it terminates. Verticals on each junctor switch connect to junctors which are dis- tributed over all the trunk link frames in the office. In a ten trunk link frame office, the ten verticals of each line Unk frame switch are distributed to all ten trunk link frames. This provides a set of ten junctors from each line link frame to each trunk link frame. If there are less than ten trunk link frames in the office, additional sets of junctors, perhaps comprising less than ten junctors per set, are provided between frames. The junctors connect between like-numbered switches and within a set bear the same number (0 to 9) as the switch. The trunk Hnks are similar to the line links except that twenty Hnks connect from each trunk switch to the ten junctor switches. The twenty THROWDOWN MACHINE FOR TRAFFIC STUDIES 349 links are subdivided into left and right sets of ten which connect to the left and right halves of the junctor switches respectively. Within each set, the links are numbered in accordance with the junctor switches on which they terminate. The junctor switches are split horizontally as shown on Fig. 24 to provide for twenty junctor connections per switch (one from each of twenty line link frames). Thus a set of junctors ter- minating on the left halves of a junctor switch must be matched with a left set of links. It can be seen on Fig. 24 that when the three sets of links and junctors capable of connecting a trunk and a subscriber are matched, the two No. 0 Hnks and the No. 0 junctor go together, the two No. 1 links and the No. 1 junctor go together, and so forth. The marker performs the LINE LINK FILE TRUNK LINK FILE JUNCTOR GROUPING SWITCHES Fig. 23— Match position. 350 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 11^ Fig. 25 — Channel cards. 351 352 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 matching function by gaining access to a set of line links, a set of trunk links and a set of junctors on the basis of its knowledge of line and trunk location. It tests for three idle elements of like number within these sets and picks the lowest numbered idle channel available for making the connection. If it is impossible to match three idle elements and there is more than one set of junctors between frames, the marker will change the junctors and make a second attempt at matching. This marker function is performed by a six-stage allotting circuit which advances one step for every match operation. Depending upon the number of junctor subgroups, the allotter is arranged to rotate the first choice subgroup on each subsequent match operation and provide an alternate sub- group if the first match attempt fails. This system tends to equalize the use of junctors. For use by the match operator, a channel card for each set of ten links or junctors is provided. Each card, as shown on Fig. 25, has ten pockets, one per link or junctor element, in which can be inserted a busy tab. When the three cards required for a particular connection are identified, they can be stacked as shown on Fig. 26 (note the difference in card size) and the idle channels are immediately obvious. In this case, channel 5 is the lowest available one and would be assigned to the call. The identity of each card corresponding to a set of links or junctors Fig. 26 — Channel matching procedure. THROWDOWN MACHINE FOR TRAFFIC STUDIES 353 is determined by switch and frame numbers. There are ten sets of Une links per Une Hnk frame and a maximum of twenty frames. The identi- fication of each set is a three digit number made up as follows: SW-LLF where SW — switch number (0-9) LLF— frame number (00-19) This provides for a total of 200 line Hnk sets or cards. The number of the line link set is incorporated in the line identification number so that from the latter can be determined immediately the particular line links available to the line. Note on Fig. 24 that line 803-24 must use the LL803 set of Hnks (card shown on Figs. 25 and 26) for any connection. There are twenty sets of trunk links per trunk hnk frame (ten left and ten right sets) and a maximum of ten frames. The identification is a two digit number plus a left or right indication. The number is com- posed as follows : TLF— SW where TLF — frame number (0-9) SW — switch number (0-9) and the left or right indication is, for convenience, one of two colors. This provides for a total of 200 trunk link sets or cards, 100 of each color. The link identification is included in each trunk number. Thus, on Fig. 24, trunks 21-49 must use trunk link set TL 49, either left or right. The number and disposition of junctors are determined by the layout of line link and trunk link frames. In a 20 line hnk-10 trunk link frame office, there is one set of junctors between each pair of frames. For fewer frames there are more sets between frames to a maximum of five for a four-line link, two-trunk hnk frame office. The junctor sets or cards are identified by the two frame numbers involved, as LLF— TLF where LLF— Line Link Frame No. (00-19) TLF— Trunk Link Frame No. (0-9) If more than one set of junctors interconnect two frames, letters A to E are added to the base number. For a particular line and trunk, the junctor number is derived from a combination of the line and trunk numbers. For example, the essential parts of line number 803-24 and 354 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 trunk number 21-49 combine to give a number 80349 where the underUned digits represent the set of junctors (see Fig. 24). In the general case, the junctors of a set may terminate on either the left or right half of a trunk link frame junctor switch. When the channel cards are made out preparatory to a throwdown study, the junctor numbers are assigned to cards of one of two colors (same as trunk link cards) depending upon whether it is left or right connection. Thus, in picking the three channel cards for a match, the choice of trunk link card depends upon, and must be the same as, the color of the junctor card. Normally the channel cards are kept in filing sections at the match position as shown on Fig. 23. Before the matching operation is physically performed, the operator has available on a match ticket the line-trunk composite number. In the example shown on Fig. 26 this number is 80349 The digits 803 identify the line link card; the digits 034, the junctor card; and the digits 49, together with the junctor card color, identify the trunk link card. The operator removes the cards from the file, stacks them, and places them in a slot associated with the particular marker until the match signal is received. At that time, the operator picks the lowest numbered free channel in the stack, marks it busy and enters the channel number on the match ticket for record. Each channel must be released at the same time that the line and trunk with which it is associated are removed from holding. Since release times for lines and trunks are entered on the assignment position holding time detector, this latter detector is used to signal release times to the match operator. The match operator maintains all her established match tickets in release time sequence with the earliest time on top of the pile. The lighting of a signal lamp indicates that the channel identified by the top ticket should be restored to normal. A channel release condition which the match operator must recognize without a special signal occurs at the marker release time on intraoffice, outgoing, partial dial and no dial calls. At this time the register channel associated with the call must be dismissed. The operator will have received from the marker operator the register channel ticket involved and must restore the channel to normal before answering any new signal. THROWDOWN MACHINE FOR TRAFFIC STUDIES 355 MASTER CONTROL PANEL A master control panel for the throwdown machine is supplied on the relay cabinet shown on Fig. 14. This panel provides: means for turning the system on and off at the beginning and end of each day's operations; a centralized alarm indicating system; a bank of marker action lamps which indicate marker status during operation; and a present time counter with a units-to-seconds conversion scale. The equipment operates on two battery supplies, one known as per- manent battery and the other as day battery. Permanent battery is on continuously during a complete throwdown run in order to hold operated certain record relays. The day battery, however, is turned off diu-ing idle periods. The equipment is arranged so that at the end of a working period all operator functions requested by signal lamps during the last working unit of time can be completed before the machine automatically stops. This is controlled by the day-night switch which is turned to the night position when it is desired to cease operation. The night lamp Ughts at this time. When the operators have extinguished the last signal lamp, the day power switch is turned to off. Certain critical portions of the machine are provided with alarm circuits which automatically stop the machine and light lamps at the control panel. In most cases, the lighting of one or more of these lamps will require troubleshooting. When the trouble has been found, operation of the key associated with the lamp extinguishes the lamp and permits the machine to start again. results of throwdown studies The throwdown machine has now been in operation for slightly less than four years. During this time 1383 seconds of equivalent central office time, divided among eleven runs, have been accumulated. The machine, of course, has not been in continuous operation, since the time required for preparation of a run and eventual analysis of output data exceeds the actual operating time for the run. In general, the same team of girls has handled both preparation and analysis of data and operation of the machine. Beyond this, there have been periods when no studies were in progress. A detailed presentation of throwdo^vn results is not properly within the scope of this article. However, a brief resume of the several runs with some mention o^ their primary objectives and typical results is necessary to conclude this picture of the throwdown machine. 356 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 It should be emphasized again that the principal function of a throw- down machine is to provide data, under controlled conditions, which can be used to develop and check a comprehensive theory of operation of a complex system. A secondary function is to study the reaction of the system to specific equipment or circuit arrangements. The No. 5 throwdown machine has been used for both purposes. All runs have furnished masses of statistical data which have been very useful in formulating traffic theories applicable to No. 5 crossbar. Some data, such as those on link matching, have a more general field of application. All throwdown runs have emploj^ed a basic size office of twenty line Unk frames and ten trunk link frames, loaded by 9000 subscribers. On the first two runs, a traffic level designed to load these frames to their normal capacity was applied to the machine. This level can be called Table III Dura- tion Quantity Sys- tem Load- Run Mar- kers Orig. Primary Purpose of Run Reg- ing isters Sec- per ends cent I 256 5 67 100 To load up machine and give normal load picture II 216 5 59 100 To study effect of higher marker occupancy due to reduced number of registers III 90 5 59 125 To study effect of 25 per cent overload IVa 86 5 59 110 To establish response of system with original gate preference for comparison with run IVB IVb 94 5 59 110 To provide data on response of system with reversed gate preference V 65 6 65 This was an intermediate run in which traffic at high level was introduced in order to build up to system equilibrium at the 120 per cent load level VI 108 6 65 120 To study system at high load level below saturation VII 288 7 60 120 To study effect of situation where registers are more severe bottleneck than markers Vlllabc 180 4 35 140 These three runs tested two proposed changes in the gate control circuit against the standard reversed preference arrange- ment. Identical traffic was used for each run THROWDOWN MACHINE FOR TRAFFIC STUDIES 357 0.4 0.3 0.1 0.08 UJ f^ 0.04 0.03 ^ 0.02 0.01 0.008 0.006 0.004 0.003 0.002 - { - I \\ \\ \\ RUNI (NORMAL EQUIPMENT) RUN n (REDUCED EQUIPMENT) V\ \ \ \ \ - V\ - \\ V \ ^ \ \ > ^X \ - \ \ \ - ^ ^ \ V \ \ Fig. 27 — Dial tone service w by the throwdown machine. 1.6 2.0 2.4 Iq 2.8 3.2 3.6^^ 4.0 TIME,T, IN SECONDS ith different equipment quantities as determined 100 per cent load. In succeeding runs, an overload of varying amounts was utilized. In the several runs, the quantities of registers and markers were varied to obtain basic engineering data. As the No. 5 system was originally engineered, the preference order in which markers were assigned by the gate control circuit to marker connectors during heavy loads was as follows: (1) line link frame marker connectors; (2) originating register marker connectors; (3) incoming register marker connectors. At a later date, this order was changed to put the line link frame marker connectors last in preference. In the dis- cussion which follows, this arrangement will be known as "reversed preference." 358 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 The essential features of the throwdown runs to date are given in Table III. The curves of Figs. 27 and 28 are representative of the type of data made available by the throwdown machine. Fig. 27 shows the overall dial tone service obtained during the first two runs. The shght degrada- tion of service in Run II is caused by the reduction in number of registers. In both cases, however, service is very good. Fig. 28 shows the spread of delays met by line fink frames in obtaining a marker on dial tone calls during the same two runs. These curves are representative of the distributions of delays encountered at individual stages of handling a call. Other examples might be line link and trunk 1.0 0.8 0.6 0.4 0.3 0.2' O.I 0.08 m ^ 0.04 0.03 0.02 0.01 0.008 0.006 0.004 0.003 0.002 0.001 - \ \\ RUN I (NORMAL EQUIPMENT) RUNH (REDUCED EQUIPMENT) - ' t \s^ \ \ \ \ \ \ \ \ \ \ - \\ - \\ \ A \ \ \ \ V^^ k \ \ 1 r 0.8 1.6 2.0 2.4 2.8x '0 TIME, T, IN SECONDS Fig. 28 — Marker delays on dial tone calls with different equipment quantities determined by the throwdown machine. THROWDOWN MACHINE FOR TRAFFIC STUDIES 359 link frame delays. Taken together, these various delays determine over- all grade of ser\^ce. An interesting example of the practical utility of the throwdown machine is furnished by the series of events which lead to the introduc- tion of reversed gate preference in the Xo. 5 crossbar system. The chang- ing quantities of line and register pegs in the gate position provide a graphic visual indication of the dynamic status of the system. After watching the gate for some time in the early overload runs, it was no- ticed that the register marker connectors were relatively less successful in gaining access to markers than the line link frame marker connectors. During all register busy periods, this reduced the call handling capacity of the system since registers were delayed in becoming available to waiting lines. The effect was compounded by wasting marker time in at- tempting to set up dial tone calls when no registers were idle. It was felt that a change in gate preference, placing register marker connectors before line link frame marker connectors, would improve this. The new arrangement was tested and confirmed in throwdown run IV and is now a system standard. Working Curves for Delayed Exponential Calls Served in Random Order By ROGER I. WILKINSON (Manuscript received December 19, 1952) Working curves of delays for waiting calls served at random are given for a considerable range of loads and group sizes. Exponential holding time calls are assumed originating at random, and served by a simple group of paths. Results of a number of throwdown tests are given to illustrate the effect on call delays of several modes of service, and particularly of service on a random basis. For random service, these results verify the theory recently developed by J. Riordan; perhaps more interestingly they show the effects on delays of certain blends of queued and random service which approximate methods of handling delayed calls in practical use {such as gating and limited storage circuits). The use of random and queued delay theory is illustrated by a number of examples. To remind the reader that these results are not limited to telephony, department store and vehicular traffic problems are included. A theory for predicting the delays which telephone calls (or other corresponding types of traffic such as vehicular, aircraft, people waiting in line, etc.) having exponentially distributed holding times would en- counter when the delayed calls are served in a random order was pub- fished in a recent issue of this Journal* by John Riordan. Mr Riordan's mathematical analysis involved a determination of the first several moments of the delay distributions. He then devised a method of com- bining elementary exponential curves in such a way as to satisfy the moments previously calculated. Since a limited number of moments were used in the above determina- tions the curves derived are approximate only, but at the same time they are believed to be good approximations. The critical cases are those of paths carrying very heavy loads, in the occupancy ranges of a = 0.80 or higher. * Bell System Technical Journal, January, 1953, pages 100-119. 360 DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 361 10" V V ^. ^ \:^ I 1 NSN \N ;. RANDOM THROWDOWN .,^- — or = 0.9 > .^ \ ^S^x ^ V N, X 5^^ v ->- U RANDOM THEORY \ V^ RANDOM THEORY .^0C= 0.9 ^^K. \\ \\ \ QUEUED THEORY /a =0.9 ^ ^ \ r I QUEUED THEORY \ V ^-S ■-^ 10 15 20 25 30 35 40 45 50 55 60 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME 65 Fig. 1 — Distribution of delays. Theory versus throwdown, delayed calls han- dled at random, c — 2 paths, a = 0.90, 3000 throwdown calls. THROWDOWN CHECKS Before calculating a field of curves for working purposes it was thought desirable to make at least a modest throwdown test, or traffic simulation, at these high occupancies to observe the agreement of theoretical delays with those determined by a trial in which the theoretical assumptions would be closely followed. This has now been performed at two trunk group sizes, c = 2 paths, loaded by approximately a = 1.8 erlangs or an occupancy of a = 0.90, and c = 10 paths at an occupancy of ap- proximately a = 0.80. For these throwdo^^^ls, random origination times were obtained through use of Tippett's Random Numbers. An hour was visualized as being composed of 100,000 (or, as in one case, 1 million) consecutive dis- crete intervals, numbered serially. Choosing 5 (or 6) digit random num- bers then provided the start times of the subscribers' bids for service. Likewise holding times were chosen by random numbers from an exponential universe by dividing it into 100 equal probability segments and assigning each a number from 00 to 99. A central value of holding time was chosen to represent the range of cases within each segment. The last segment, number 99, on the long tail was further subdivided into 100 parts in order to give more definition in the long call lengths which are believed to be critical. I 362 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 A comparison of the proportion of traffic expected to suffer delays beyond various multiples of the average holding time as given by Rior- dan's theory for delayed calls served in random order, and by the throw- down results, is given in Figs. 1 and 2. As discussed below, the cases studied are considered to give satisfactory assurance as to the adequacy of the approximations involved in the theory. The two trunk case based on 3000 calls submitted shows fairly good agreement with the theoretical distribution out to delays as large as 50 multiples of an average holding time which includes more than 99.5 4 h- 3 V\ V \\^ - \v\ - \v \ 1 Ox V' \\ \ \ queued\ THEORY \ ,\ , RANDOM ^ THROWDOWN \ -2 \ \ \ ^ - \ V \ - ^ N .\ RANDOM VC THEORY ^. ^ \ "^c. 3 \ \ k 2 ) y \ -3 \ \ \ \ \ 1 1 \ Oi 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING 0 5.5 TIME 6.0 Fig. 2 — Distribution of delays. Theory versus throwdown, delayed calls han- dled at random, c = 10 paths, a = 0.80, 1500 throwdown calls. DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 363 q/T. 1 1 1 1 1 1 ! 1 1 ! •> % 1 1 1 '1 III 1 1 1 1 1 1 1 1 1 1 I 1 !lT>~.-U" 'iiiiiTTriTr mTt ~rr- r-f-r+- u^ U — 16 CALLS CALLS BEING WAITING SERVED 20 24 28 32 36 40 44 48 52 NUMBER OF CALLS BEING SERVED OR WAITING Fig. 5 — Distribution f{x) of simultaneous calls. Theory versus throwdown, c = 2 paths, « = 0.90, 3000 throwdown calls. 366 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 line spikes correspond to observations when all calls in the system were being served, that is x ^ c. The dotted spikes show those proportions of observations when one or more calls were waiting, that is x > c. The theoretical values of f{x) are indicated by the smooth curves where they pass over discrete values of x. The theory and observations are seen to be in quite good agreement. Referring again to the theoretical delays (and the throwdown checks) on Figs. 1 and 2, very much larger delays can obviously be obtained when delayed calls are handled at random than when they are handled in a strict first-come-first-served, or queued, order, the latter distri- CALLS BEING SERVED CALLS WAITING 0.16 0.14 0.12 IZ 0.10 i-iu ><" to: 0.08 0.04 c \ k ^ N i / i V i ^ jS V / 1 ^ I Hi! iTtH^^ fcSri 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 X= NUMBER OF CALLS BEING SERVED OR WAITING Fig. 6 — Distribution (/(x) of simultaneous calls. Theory versus throwdown, c = 10 paths, OL = 0.80, 1500 throwdown calls, butions being shown by the straight lines which start at nearly the same ordinates at delay 0 as the random handling curves, and cut down across the lower part of the charts.* Although fewer very short delays occur * Delay curves for exponentially distrubuted holding time calls in systems where delayed calls are handled in order of arrival, are given by E. C. Molina in "Application of the Theory of Probability to Telephone Trunking Problems," Bell System Technical Journal, Vol. 6, p. 461, July, 1927. They are calculated from the Erlang equation P(>0 = P(>0)e-(^)' a'=e-^ c! c — a T=i x\ c\ c — a g-(c-o)< (1) where the delay t is expressed in multiples of the average holding time. Values of P(>0) = C(c,a) can be read approximately from Figure 21. DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 367 with this method of handhng than when a random selection of the wait- ing calls is followed, the very long delays are markedly reduced, and on this account the queueing procedure is generally preferred. These effects are particularly evident at the higher occupancies. As illustrated in Fig. 1, the ''queued" and "random" delay curves at an occupancy of a = 0.4 show Httle difference down to the P = 0.001 delay level. IMPERFECT QUEUEING Interest has often centered in questions as to what form the delay curves might take in a system in which queueing of the calls is main- tained to a limited extent, and beyond which the record of order of arrival would be lost. Such an instance might occur with a team of toll recording operators who were able to keep Avell in mind the order of arrival of signals up to a certain number waiting, whereupon they would lose track and not regain this ability until the number of waiting calls had again dropped below some small number. Other situations with actual or equivalent limited delay storage arrangements can readily be imagined. To study a case of limited queueing, a short subsidiary throwdown was next run on the c = 2 case, using the 1000 calls of Runs 1 and 2 of Figs. 3 and 4 (which comprised the 1000-call sequence most closely approach- ing the theoretical distribution). Three rules for delayed call handling were tested: (1) Delayed calls are served in random order. (2) Delayed calls are queued (served in order of arrival). (3) Delayed calls are queued until more than w are waiting at which time their arrival order is lost and they are served at random. When the number waiting again drops below w, newly arriving calls are queued behind those randomized calls still waiting. Note that case 1 corresponds to ly = 0, and case 2 to w = qo . The comparative results are sho^\'n on Fig. 7, with w given successively values of 0, 8, 20, 25, 30 and <» . The w = ^ curve, of course, is taken directly from Fig. 4 for Runs 1 and 2 combined. Although this curve does not agree particularly well with theory (Curve A), its movement with changes in w is nevertheless instructive. As seen, queueing as far as If; = 8 waiting calls produced practically no improvement in the delay distributions. (Perhaps with the occurrence of such large numbers of waiting calls, reaching a maximum of 35, one could not expect queueing of so few as 8 to have much effect.) The next selection oi w = 20, how- ever, still showed only a relatively sUght improvement, particularly in \ 368 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 / (VJ / / O >- o UJ X. z < 1 "^1 / (0 / o II in / %-' If) m . / / 0 o / J / .!>'' «n 1(0 /; rj .>> 5" / .^ /■ 00 ,^ ,x^ II rvi / Q__ ^^fl^ (VJ ) o / 7/ X- /^ ^ _, . (\i in < / / ^'^ ^ y^ ii-- '-" 8 "11 5 01 A ^ 9^ 4 fy / A ^ p'y^ _i c3 o a % II UJ bO o :3 CO § Q. ^ CI. ? a < ^ O <0 ) -Ho oV V ^ ^ \ ^ \, -^ ^ ^ V^l \ \ s s. ^ 1 i ' 1 \ \ \ *^ 1\ \ \ \ \ \ L 20 t/h: 30 40 50 60 70 80 DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Delayed traffic served in random order, exponential holding times, circuit system. The resultant delay distribution is shown as Curve B on Fig. 7. (It is appreciated that this hardly represents a tolerable normal operating situation, but rather illustrates what the performance might be under extremely heavy traffic conditions.) The results are very close to those obtained with perfect queueing (Curve C) and show in striking fashion the gains in service to be made in certain delay situations by providing a limited storage apparatus with a memory not subject to confusion during moments of heavy overload. When the 1000 calls of Runs 1 and 2 are submitted to the 2 paths through a simple gate in order to produce approximate queueing, the resultant delays are shown by Curve D on Fig. 7. Large improvements again occur in reducing the very long delays found with random handling. In fact by use of this simple (and usually relatively inexpensive) gating DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 371 i > A =! 0.01 <^ ft C ^ a 0- 4 II 0.001 m c = = 2 Su ^^ M \\ ^. m\ m;\ ^■-^ -^3 111 \ ^ \ SS«o ilU \ y , V \ ^*v^ ^ Hff^^ W ^ h^ Mi k= ^^"^ -^^ TH \^; ^ Vo \ \ \\ J I \ \ % \, 1] ^ V\-X- A ^ \. "TU m^ ^ ^ ^^ ' lUe fr^ \— ^S \ \ \ \ \ \, ^V, \ \ \ \ \ \, \ \ \ ^ \ \ \ N V 10 t/h 15 20 25 30 35 40 DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Fig. c = 2. Delayed traffic served in random order, exponential holding time^ scheme, delay results are obtained nearly as good as those realized by the proi'ision of 20 storage circuits (Curve B). WORKING CURVES The adequacy of the Riordan theoiy when delayed exponential calls are served at random is believed to have been established and that it may be used with confidence to solve those practical problems where the underlying assumptions are well satisfied. For working purposes, curves showing distributions of delays expected for occupancies up to a = 0.90 and for group sizes of c = 1, 2, 3, 4, 5, 6, 8, 10, 20, 50 and 100, are shown in Figs. 8 to 18. These are plotted in the customary fashion with delay in multiples of average holding time 372 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 0.1 8 5 00' < s O 6 a. 0.001 8 0.0001 C = -3 \\ — l\\\ ' s 1^ iT\M — y ■^s, :^^^ ,___.^^ \W \ s \ \\ 1 > \ \ ^■^ ^hr \ "C \ N^o s: =A-=^ rH "•s^ 1 oTl \ \ ' r"^ — ^ \o ^^-^ \ \ \ ^ ^ \o \ ** — e--o \ \ V= ^ - " Ji°\v 1 ^^ ST \m t^-N ^ \ ^^ 1 \ \ \, \ V ' \ ^ \ \ \, 111 0 5 \ \ 10 1 5 2 L 0 2 5 3 0 3 s. 5 4 0 4 5 50 t/h= DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Fig. 10 — Delayed traffic served in random order, exponential holding times c = 3. ' as abscissa, and P{>t/h), the probability of a random call meeting a delay greater than t/h, as ordinate. Estimates of average delays, t (which are the same for queued and random service), are also commonly desired, and these are shown in Fig. 19. They are calculated from the equation t/h = P(>0)/(c - a) (2) If one wishes instead the average delay, t, on calls delayed, it may be obtained from 't/h = t/h 1 P(>0) c (3) DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 373 ■0.001 m H — — — — c - = 4 % ^v 1 w \ Wf m ^-^ ^ \\\ \ \ •^ \ \\\ \ \ \ ^ ^0 ffiV .;M ^K= 1 ^^^ Mi t^ l-^ XT ^-^ ^^-^ 11 \ ^ \ \ ■^ I \ \ \ \ ^ s \. W- MX' — \'So/w^ton. The load to be carried is a = (225)(100)/3600 = 6.25 erlangs. The average delay, i, is not to exceed 30/100 = 0.3 holding time. Read- ing on Fig. 19, opposite an ordinate of 0.3 we select several trial values of trunks (operators) c, versus occupancy a, and form Table I, cal- culating the last column from the first two: DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 375 '^ p. o 6 '^ 4 ■ 0.001 if - 4- 1 c - = 6 L\ \ ^, f¥= L:== = fr- ffira tn s V J V \ ^ \, hL\- ^ \ V In > Rtvt ^-^^ \ :so: — i^^- m^ ^-^ ^° 114 \ \ \ \ 1\ \ ^ \ \^ \ \ \ s^ -rtV^- ^-^ \ s= ^ ^^ H 1 \ ^^ =^ r* f \ \, \ \ k \ \ \ \ \, 1 \ \ \ 5.0 7.5 10.0 12.5 15.0 17.5 20.0 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME 22.5 Fig. c = 6. 13 — Delayed traffic served in random order, exponential holding times, To carry the 6.25 erlangs of traffic and meet the average delay require- ment we see that 8 operators will be needed. Will 8 operators also fulfill the no more than 20 per cent delay over 1 minute requirement? Enter Fig. 14 (the c = 8 chart) with an occupancy of a = 6.25/8 = 0.78. The per cent of calls exceeding a delay of 60/100 = 0.6 holding time is about 12 per cent. A provision of 8 operators satisfies both requirements. Table I c a a = ca 7 8 9 0.78 0.81 0.82 5.46 6.48 7.38 376 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 1.0 8| Q. 0.001 0.0001 — C = 8 ,v m -^ ._S^ ' W > s |\ \ \ VI \\ ^ ^ X- -H-V-: \ ^ 1 ■ \ ^ 1 oil ^ \^c \ - \-^ \, V \ \ tVn-Tc ■M:^ \ ]m #H N ri k \ V \ ^^ ^ \ \ \ 1, \ \^ \ \ \ 2.5 5.0 7.5 tO.O 12.5 15.0 17.5 20.0 22.5 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME 25.0 Fig. 14 — Delayed traffic served in random order, exponential holding times, Example No. 3 Suppose in Example 2, the second requirement had been that no more than one of 1000 customers should be required to wait over 3 minutes. Would 8 operators then suffice? Solution. Reading on Fig. 14, with a = 0.78 and t/h = 180/100 = 1.8, P{>t/h) = 0.027. Thus 27 in 1000 calls would be expected to experience delays over 3 minutes, and therefore more than 8 operators will be required. Consulting the c = 10 curves of Fig. 15, we find that with a = 0.625, and t/h = 1.8, P(>3 minutes delay) = 0.0012 which closely meets the one in a thousand requirement. Ten operators would then be needed; and this would, of course, (from Fig. 19) reduce the average DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 377 0.1 8 6 5 ^ Q O 2 =! 0.01 < 8 O ^ 0.001 8 6 4 K _._.„ — c = 10 V \ s \ ^ ^v- 1 — ^'S K ^ 1 1 \ V \ ^ \ \ *v \ ^ s s. r^og^ ■f \ ^ <-> 1 ^ ^ 1 ! r\- p^ r s. 1 ^ ^-==^4— 1 \ — N V \ i ""^->^ \ \ \ '^ 0 ! \ \ > s. 1 N \ §£ P^ '^\r. o ^ S^ — N ^ —^ 1 — ^ — w\ \ \ N. '< \ \ 1^ ^^ i \ ^ \ s. 1 >v.! \ \ \ N S ! '^^ 10 0123456739 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Fig. 15 — Delayed traffic served in random order, exponential holding times, delay on all calls to 0.035 (100) = 3.5 seconds, an improvement in this characteristic of 7 to 1 over the 8 operator service.* Example No. 4 How much improvement in the delay service would be obtained in Examples 2 and 3 by purchasing storing or gating equipment which would substantially insure calls being handled in order of arrival? Solution. With 8 operators working at an occupancy of 0.78, the pro- * Had some number of operators been required other than those for which working charts, Figs. 8 to 18, are supplied, intermediate values could be obtained by graphical interpolation, or better still by employing the basic Riordan chart. Fig. 20, combined with P(>0) found on Fig. 21, to obtain delay versus load for any desired number of paths or facilities. This latter process is described in the Appendix. 378 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 1.0 8 0.01 8 a. 0 001 8 6 0.0001 ::it — c = = 20 U— > y..- fc# i \- \ I \ s h¥^- \ ^ V— mw ^ ^ ko p \ \ N, t V v^ \ S w=^=^ k-- S ::^ ^ 1 o \o r* ^ \ — ^ "^ -l^t \ s. 1 \ \ \ ^ _I_t_\ \ 01 23456789 10 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Fig. 16 — Delayed traffic served in random order, exponential holding times, c = 20. portion of calls delayed is found to be P( > 0) = 0.41 (Fig. 14) . The proba- bilities of exceeding delays of t/h = 0.6 and 1.8 holding times are cal- culated for calls served in order of arrival by equation (1), in the following table: / (Min.) t/h Queued P(>0) = i>(>0)e-('=-«) ^1^ Random Handling 1 3 0.6 1.8 0.143 0.019 0.12 0.027 Comparing the queued and random handling of delayed calls one finds the perhaps unexpected result that with random handling some 2 per cent /either calls are delayed longer than 1 minute than if perfect queueing had been present. This is due to the characteristic shapes of the two types DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 379 0.1 8 ^ 8 ^ 6 4 O.OOOI Fig. 17 c = 50. \ 1 \ C= 50 \ =- -\-- _.. \ L > \, > s s 1 \ \ Q_ [- =^;^= P^ ^e \ \ N, \ v >s \ \ \ N -X --A \R \. ^"^ — ^ ?" ^ =^ "^ 1 w \ ^V "^, 1 \ \ \, \ \ 1.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 2.50 2.75 3.00 3.25 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME Delayed traffic served in random order, exponential holding times, of delay distributions, random handling producing more quite short and very long delays than does queueing. When a criterion of service is set at a relatively short delay, one may often expect it to be met more easily by not providing storing or gating circuits. On the other hand a criterion of service based on relatively long delays can nearly always be more readily met by the use of devices insuring partial or total queueing. In the example above the per cent of calls delayed longer than 3 minutes would be cut by a third through the use of queueing devices. Example No. 5 Automobiles are parked in a large area adjacent to a State Fair grounds. There is one main exit through which two cars can pass at the same time. Upon leaving, drivers pay according to their parking time; and it requires, on the average, 20 seconds to complete the payment. If cars wish to leave during the afternoon busy period at a rate of 5.4 per 380 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 0.01 8 0.001 0.0001 h-- — C = 100 1 \- Mllllll 1 \-X^- s s lih ■ -\-- ' \9^ \' \ 1^ \\\ k^ ^S— - ^ V N \^ i\ S \ 2.50 Fig. = 100 0 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00 2.25 t/h = DELAY IN MULTIPLES OF AVERAGE HOLDING TIME 18 — Delayed traffic served in random order, exponential holding times, minute, what per cent of the cars will be delayed more than 5 minutes? What will be the average delay for all cars? Solution. Assume there is no traffic supervision and cars converge on the gate from many directions. Service in random order (or worse) among those delayed might then be approximated. Also the distribution of times for calculating and collecting the charge might be roughly exponential. We have then, c = 2 paths a = (5.4)(20)/(60)(2) = 0.90 t/h = 5(60) 20 = 15 Enter Fig. 9 at t/h = 15, read to the a = 0.90 curve, opposite which find P = 0.069. Hence 7 per cent of the cars would be expected to have to wait 5 minutes or more. To obtain the average delay for all cars, enter DELAYED EXPONENTIAL CALLS SERVED IN RANDOM ORDER 381 6 5 \, \, \ \ 10 \ \ Sr ' Vt \ 8 6 ^ ^^= 5 VN \ \ \\ V ^^ \ kV k^ \ ^ V9- 1.0 8 6 5 rX^ =^^= w^ A^ ¥fm v\ \ \ \s, ^ . \— - W^ « X -^-t 10-' 8 \\\J \\ \ 6 5 V \k \ m mm . \ \ -ZS \-\ \\\ \c \K \ \\ ^ 10-2 '\y w 8 6 \r\\ \r^^ =l^=='> 5 \\ \ Yt^ t u i"" \ \ \ \ \ \ v _\.. \ \ \\\ UA t 10-^ w \\\ W \ 3 4 5 6 8 10 15 20 30 40 60 C = NUMBER OF PATHS 100 Fig. 19 — Average delay on all calls, exponential holding times. 382 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Fig. 19 with the abscissa of 2 paths, read to the a = 0.90 curve and find the average delay = 4.25 average holding times = 85 seconds. Or, one may obtain the same answer by substituting in equation (2), t = P{>0)h/{c - a) = (0.85)(20)/(2 - 1.80) = 85 seconds. Example No. 6 Suppose in Example 5, an efficient corps of police had been directing traffic toward the exit so that good queueing was maintained. What per cent of the cars would then be delayed more than 5 minutes? Solution. We may now refer to other published delay curves for queued operation*, or, more generally, calculate the well known equation (1). In the present case we can read the answer from the ''queued" curve of Fig. 1 as 4.2 per cent. Thus serving customers in the order of arrival nearly halves the occurrence of very long delays. (Note that the average delay for all cars remains unchanged at 85 seconds.) If a partial queueing were maintained the improvement would be intermediate, perhaps com- parable with one of the "limited queueing" distributions shown on Fig. 7. The author is indebted to Miss C. A. Lennon for constructing the working delay curves, and to Misses C. J. Durnan and J. C. McNulta for performing the throwdown checks. Appendix calculation of delay values not found on the working curves of figs. 8-18, for delayed exponential calls served in random ORDER A master chart. Fig. 20, reproduced from Riordanf, gives in condensed form the proportion F{u) of delayed calls delayed longer than u, where the delay is now expressed in multiples of the h/c (u = ct/h) , and c = number of paths (trunks, operators, etc.) provided h = average holding time t = delay time To obtain the probability P{>t/h) of any call being delayed longer than *E. C. Molina, Ibid, t Loc. cit. .01 p 3 4 5 6 7 8 9°-' 2 3 4 5 6 7 II / 7 . / / / / / 1 1 / — --L-,' - .^ / 7 / 1 4 z ;.-;;;— z'- -1 - z t_ - > z 1 r z jr / Z f 1 / A / / / / f / / / —^=4 / — 1 =====^^y^ Hm -— -/ ' — / — -/- ^ -j- 7 > 1 / / 1 / / V [ / _J~ t _^ 1 j_ /' / 1 1 ] 1. r .01 5 6 7 8 9 0.1 2 3 4 5 6 7 a = AVERA Fig. 21 — Probability' of dela}- for exponen 2 3 4 5 6 7 8 9 1.0 - - ^ ' « ^ « ^0 GE LOAD SUBMITTED tial holding time calls handled in a "delayed" basis. DELAYED EXPONENTIAL CALLS SERVED IN RANDON ORDER 383 t/h, we have P{>t/h) = P(>0) F{u) = (7(c, a) F{u) (4) Values of P(>0) = C{c, a) are given for a wide range of a and c in Fig. 21. The appUcation of equation (4) is quite simple. Illustration 1 . Suppose it is desired to obtain the probabihty of a call being delayed more than 3 holding times on a 10 trunk group without storage or gating circuits, and which carries a = 9 erlangs. Here t/h = 3.0, c = 10,a = 0.9. Then u = ct/h = 30, and reading on Fig. 20 with this value of u, and a = 0.9, we find F{u) = 0.080. Fig. 21 provides C{c, a) = 0.67 for a = 9 and c = 10. Substituting in equation (4), P(>3 hold times) = 0.67 (0.080) = 0.053, which checks the value read directly from the c = 10 curves of Fig. 15. Illustration 2. With an occupancy of a = 0.65 on 15 paths what is the probability of meeting a delay greater than one holding time when delayed calls are served in random order? Calculate u = ct/h = 15. Enter with this abscissa on Fig. 20, and interpolating between the a = 0.6 and 0.7 curves, read F{u) = 0.022. Fig. 21 shows for a = 0.65(15) = 9.75 and c = 15, C(c, a) = 0.085. Hence P(>1 hold time) = 0.085(0.022) = 0.0019. Magnetic Resonance PART II— MAGNETIC RESONANCE OF ELECTRONS By KARL K. DARROW (Manuscript received December 24, 1952) Magnetic resonance of electrons is the analogue of magnetic resonance of nuclei, treated in the first part of this article. Though the analogy is close and the fundamental laws are identical, the two topics are remarkably dif- ferent in detail. Though electrons are the commonest of particles, they display magnetic resonance only in somewhat exceptional cases. In many free atoms and most solid and liquid substances, magnetic resonance is suppressed by what is known as the ^^anti-parallel coupling^^ of electrons two by two. The exceptional cases are those of certain free atoms, ferromagnetic substances, and a restricted class of strongly paramagnetic substances; the resonance has also been observed very lately for the conduction electrons in metals. In the cases in which it does occur, resonance is likely to occur at a frequency or frequencies very different from that which the elementary theory predicts. This is sometimes because of the orbital motions of the electrons, oftener mainly because of the electric and magnetic fields existing in solids, and the deviations of the observed cases from the ideal case shed light upon these fields. The subject of these pages is the magnetic resonance of electrons — "electron resonance" for short. Electrons being everywhere, one might expect it to be found in every substance; but for a fundamental reason it is a rare phenomenon, and this magnifies its interest. Those who search the literature for it under this its proper name will seldom find it, for it is frequently called ''paramagnetic resonance" or, in appropriate cases, "ferromagnetic resonance," These are lengthy names which tend to veil the similarities between electron resonance and nuclear resonance, which latter was the theme of Part I of this article (in the January issue of this Journal) . I will introduce electron resonance by making use of all these similarities. Magnetic resonance in general is due directly to the magnetism of 384 MAGNETIC RESONANCE. II 385 subatomic particles: nuclei and electrons. These, apart from the nuclei that are non-magnetic, may be visualized as minuscule barmagnets. The laws of resonance are determined by the fact that in a steady mag- netic field, the magnetic moments of these particles may not point in any and every direction: instead, they are constrained to a finite and small number of what are called "permitted orientations." To each of these corresponds a special value of the energy of the little magnet in the field: thus the energy also is constrained to a finite and small num- ber of ''permitted" values. These are often called ''Zeeman levels" or just "levels"; and the word "level" should be well known to those who are going to delve into the literature. H,-^ i t I i t t I t t t I I ▼ ▼ T SAMPLE ^X Fig. 1. — Scheme of the apparatus for observing magnetic resonance. The high- frequency circuits are omitted. The arrows within the sample maj^ be taken as portraying the magnetic moments of either protons or electrons: their orientations are as given by the old quantum-theory. Consider two orientations or levels of different energy-values. It will take work to turn the tiny magnet from the one of lesser energy to the one of greater energy. Magnetic resonance — • and now I ought perhaps to speak specifically of magnetic resonance absorption — is such a turning. The agent of the turning and the source of the work is an alternating or oscillating magnetic field. The simplest cases are those in which the particle in question has only two permitted orientations. Many nuclei, among them the proton, belong to this class, and the electron belongs to it also. It is the analogy between proton and electron which I mil develop. Fig. 1 of this part is also Fig. 1 of Part I. The central rectangle depicts the sample, which for the study of proton resonance must be hydrogen or a compound thereof. The big arrow on the left represents a big mag- netic field, of the order of several thousand gauss, which pervades the 386 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 sample; it is vertical and its strength is denoted by H. This is the field wdth respect to which the protons are oriented. These magnetic particles are represented by small arrows within the rectangle, the point of each arrow corresponding to the north pole of the corresponding proton. Slightly more than half of them are pointed in what I call the "up" orientation, which is that of lesser energy. The rest are pointed in the "down" orientation, that of greater energy. Magnetic resonance absorp- tion of protons is the turning of ''up" protons into the ''down" direction. The field which does the turning is an oscillating magnetic field with frequency (denoted by v) in the radio-frequency range. It is horizontal, thus at right angles to the big field. It is produced either in a solenoid (the usual method for nuclear resonance) or in a resonant cavity (the usual scheme for electronic resonance) which encloses the sample but in Fig. 1 is left to the imagination of the reader. Magnetic resonance occurs when the quantum-energy hv of the oscil- lating field is equal to the work required to turn the proton from the up orientation to the down one: hv = work of turning (1) h standing for Planck's constant. In Part I it was shown that the "work of turning" or energy-difference between the two orientations is equal to 2upH: here /Xp stands for the magnetic moment of the proton, soon to be more carefully defined. Thus : hv = 2n^H (2) When V and H are related by this equation one finds proton resonance absorption, which manifests itself by a splendid peak in the curve of absorption versus H for constant v or the curve of absorption versus v for constant H. For the frequency 42.6 megacycles the peak is found at H = 10,000 gauss. To arrive at the basic formula for electron resonance we simply take (2) and substitute into it /x« , the magnetic moment of the electron, for hv = 2m^ (3) The magnetic moment of the electron is about 660 times that of the proton. Therefore if one works with such a field strength as brings the proton resonance into the radio frequency range, the electron resonance is to be sought in the microwave range. One might think that now I have said all that there is to be said about electron resonance ; but this is only the beginning. MAGNETIC RESONANCE. II 387 Much was said in Part I about the magnetic resonance of nuclei having more than two permitted orientations. We may seem to be wan- dering off the course if we revert to these, but this case is very pertinent. There are nuclei with three, four, ... up to ten or maybe more allowed orientations. One would expect them to display a multitude of peaks; but there is never more than one. This is for two reasons, which I give after introducing the symbol (27 + 1) for the number of orientations. First, it is impossible to turn a nucleus from any orientation to any other except the nearest to the original one. This reduces the number of possible peaks to one fewer than the number of orientations. But second, all of these 27 possible peaks are of the same frequency for given H, or at the same field strength for given v, so that they all coalesce into a single peak. The formula for this apparent single peak which is strictly 27 coinci- dent peaks has been derived in Part I, and this is it: hv = {n/I)H (4) Now it is necessary to interpret 7 and /x; and the interpretation is dif- ferent according as one uses the old quantum theory or the new quantum mechanics. The old quantum theory deals more simply with these prob- lems, and would be preferable if this field could be isolated from all the rest of physics; but the new quantum mechanics is worth the extra trouble that it causes. In the old quantum theory, there are two definitions of 7 that reduce to the same thing. First, 7 is the angular momentum of the nucleus in terms of the unit /i/2x; that is to say, the angular momentum of the nucleus is 7/i/2x. Second, Ih/2Tr is the maximum possible projection, upon the field-direction, of the angular momentum of the nucleus. This is because, among all of the allowed orientations of the nucleus, the one which is most nearly parallel to the field-direction is exactly parallel to the field-direction. So it was shown in Fig. 1. In the new quantum mechanics, the second of these definitions re- mains valid and the first does not. This is because the orientation which is most nearly parallel to the field-direction is not exactly parallel thereto. It is inchned, in fact, to the field-direction by the angle arc cos I/\^I{I + 7), and the angular momentum of the nucleus is V7(7+ l)(V27r). Thus there is one definition of 7 which is valid under both theories, and that is, that 7 is the maximum possible projection upon the field- direction, of the angular momentum of the nucleus in terms of the unit h/2Tr. Similarly it is always correct to say that m is the maximimi pos- 388 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 sible projection upon the field-direction, of the magnetic moment of the nucleus. But since these phrases are intolerably long, one avoids them by saying that / is the spin and ju the magnetic moment of the nucleus. In this sense, which to the users of quantum mechanics is a distorted one, the words "magnetic moment" shall be used hereafter. The spin of the proton is 1/2, and so is the spin of the electron. Equa- tion (4) degenerates into (3) for the electron and into (2) for the proton. These we had already; what was then the point of introducing here the general case? Well, the point is that two, or three, or several electrons may col- laborate in what is known as "parallel coupling," though in the new quantum theory it is not quite parallel. They behave as though they formed a rigid unit, of which the spin is the sum of their spins and the magnetic moment is the sum of their magnetic moments. Thus if there are N of these electrons welded together (metaphorically speaking) it comes to the same as though there were a single particle of spin A^ times 1/2 and magnetic moment N times Me • On putting these values of / and /u into equation (4) we find ourselves right back at equation (3), which is that for the individual electron. There is a single peak of magnetic resonance composed of N coinciding peaks, and it is just where the peak for a single electron would be. Thus in the ideal case, N electrons coupled parallel behave just like one electron by itself. Such a conclusion may seem hardly worth the trouble of arriving at it; but note the stipulation "in the ideal case." This refers to what has been tacitly but obviously assumed till now, to wit, that no force acts upon the electronic magnet except the big field H. But there are also what I will call "local forces," forces due to fields within the sample arising from other particles in the sample. These forces may, and they often do, separate the N peaks which in the ideal case coincide. Often one finds a flock of resonance lines where, or near where, there should be only one; and if this is the explanation (which is not always the case, for there are other causes of "splitting") then the number of lines in the flock is the number of electrons coupled parallel. This illustrates one of the great contrasts between the electronic resonance and the nuclear. Nuclear resonance is a "textbook phenom- enon." The ideal case and the actual case are close together; the devia- tions due to the local fields are neither trivial nor useless, but they are not large enough to distort the simple laws, and it is quite permissible to leave them out of a first presentation. But the phenomenon of elec- tronic resonance is liable to be distorted almost beyond recognition; and if one were to present only the cases in which the local fields are ncgli- MAGNETIC RESONANCE. II 389 gible in effect, one's story would be relatively short and it would be grossly inadequate. But here the physicist, true to the tradition of his science, turns hindrance into help, and analyzes the distortions for the knowledge they are capable of giving about the fields prevailing in the sample. Thus whereas nuclear resonance is largely used for getting light on nuclei, the electronic resonance is largely studied for the information that it yields about the solid state. Another of the great contrasts is due to what are called the "anti- parallel couplings" between electrons. Generally speaking (and this means: conceding an occasional exception) any type of nucleus of non- zero magnetic moment will display a detectable resonance if there are enough of them in the sample. Were this so A\dth the electron, every substance whatsoever would display electron resonance. Experience shows that electron resonance is rare, usually conspicuous by its absence. This is because electrons may, and not only may but usually do, pair off with one another in such a manner that the spin of such an "anti-parallel" pair is zero and so is the magnetic moment. There is no resonance for such a pair; and the customary absence of electron reso- nance signifies that in most solids, all the electrons are joined two by two into antiparallel pairs (this was known before magnetic resonance was first produced). 1 will call such electrons "compensated"; in this language, the substances in which magnetic resonance is to be sought for are those with uncompensated electrons. Mostly these belong to one or the other of two classes: the ferromagnetic bodies including the anti-ferromagnetic, and the "strongly paramagnetic salts." But there are a few other cases, and among these are those which are closest to the (unattainable) ideal of the perfectly free electron subjected. THE NEARLY IDEAL CASES Nearest of all to the ideal case are presimiably the atoms which con- tain uncompensated electrons and are available for study by the molecu- lar-beam method. Outstanding among these is the hydrogen atom, whose single electron must remain uncompensated because there is no other in the atom. About or quite as good are the atoms of sodium, potassium, and the other alkali metals, each of which contains a single uncompen- sated electron not to speak of several which are compensated. Moreover, these atoms are normally in a "ground state" in which the uncompen- sated electron has no orbital angular momentum. This hints at a complexity which is not always without influence on electron resonance, and must be mentioned here at the price of a detour. 390 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Going back to ancient theory, let us imagine an electron revolving with frequency/ in a circular orbit of radius r. It is equivalent to a cur- rent ef running continuously in the circular loop. According to the old theorem of Ampere, its magnetic moment is equal to the area of the circle multiplied by the current-strength; but the current-strength is to be expressed in electromagnetic units, so that the magnetic moment M equals {e/c)fTrr^. The angular momentum p is mr times the speed of the electron, and therefore equals 27rmr^/. For the ratio of the two we find: ijl/p = e/2mc (5) This is what has lately been miscalled the ''gyromagnetic ratio," a name which was originally applied and ought still to be applied to its reciprocal. It would be good to follow Gorter's suggestion of calling it the "magneto-gyric ratio." I now state equation (5) in another fashion so as to introduce a symbol which is really a word, and is the technical word of this field of physics: it ought to be a word all spelled out, but it is just the letter g. (Morb/Pcrb) = g{e/2mc), 9 = ^ (6) Thus g is the ratio of magnetic moment to angular momentum given in terms of e/2mc as unit, and its value for the orbital motion of an electron is one. Note also that though we have arrived at (6) in a very old-fashioned way, it is one of the results that have stood firm through all the mutations of quantum theory. The study of what are known as "multiple ts" in optical spectra led some thirty years ago to the conclusion that for the spin of the electron the magneto-gyric ratio is such that g = 2: (M8pin/Pspin) = g{e/2mc), g = 2 (7) This belief was substantiated by the "Dirac theory," and was not upset until measurements were made of the magnetic resonance of electrons in atoms by the molecular-beam method. The first such measurements were made upon atoms containing uncompensated electrons which had orbital motion as well as spin. I pass them over, and come direct to the most recent experiments on hydrogen atoms in their ground state, where there is no orbital motion of the electron to complicate matters. These are so recent that they came into print as these words were being written. The hydrogen atom is a good example to take, not only for the reasons that I have given already, but also because it may be compared with the hydrogen molecule H2. The two electrons of the hydrogen molecule compensate one another, and there is no electron resonance. The two MAGNETIC RESONANCE. II 391 nuclei — ■ protons — ■ of the molecule compensate one another in some of the molecules, enter into the parallel coupling in others. There are always some of these last in a beam of hydrogen molecules, and they produce the proton resonance of which so much was said in Part I. The atoms produce the electron resonance. Look now again at equation (4), and remember that p is Ih/2Tr — and remember that p is to be interpreted as the maximum permitted com- ponent, along the field-direction, of the angular momentum. Consider now the experimenter with molecular beams of hydrogen molecules and hydrogen atoms at his disposal. In a magnetic field of field strength H he finds the proton resonance of the former at frequency Vp , and ascertains (m/Z) of the proton by putting his data into equation (4): (m//)p = hvp/H (8) In the same field he finds the electron resonance of the latter at fre- quency Ve , and ascertains (fji/l) of the electron similarly : (M//)e = hVe/H (9) Now he has both values; but the accuracy of both is contingent on the accuracy of the measurement of H, and this is not so good as he desires. However he can dispense with the measurement of i7 at a price — the price of getting his value of the magnetic moment of the electron in terms of units other than c.g.s. units. This is not a great sacrifice; Nature does not share our affection for c.g.s. units; there are others which are more suitable to the enterprises of the theorist. If we divide (8) into (9) we get rid of both H and h. This means that if the experimenter measures vp and Ve in one and the same applied field, he can evaluate {/jl/I) for the electron in terms of (fx/I) for the proton without bothering about the values of H and h. Since I is the same for both particles, he obtains the ratio of the magnetic moments of electron and proton. The value of this ratio would be precious in itself, even if one had not the faintest idea of the value of either moment in c.g.s. units. It is 658.2288 d= 0.0006. It is also feasible to get the value of (m/Z) for the electron in terms of the ''unit" eh/^irmc. This entity is so important that it has a name of its own: it is called "the Bohr magneton." There is also a combination of experiments by which (/x//)e may be evaluated in terms of the unit (eh/4:Trmc) . This unit is so important that it has a name of its own: it is called "the Bohr magneton." The reader can easily show for himself that (n/I) in terms of this unit is none other than the quantity g, of which this is a second definition (not identical with that of g in Part I). 392 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 The frequency of the proton-resonance, Vp , is compared in a special experiment mth what is known as the "cyclotron frequency," VcfOi the electron. A free electron, projected at right angles to a magnetic field H, describes a circle in the plane perpendicular to the field. The frequency with which it makes the tour of this circular orbit is given by the equa- tion:* vc = 2iHe/4Tmc) (10) If this frequency is determined in the same field as has served or is to serve for the location of the proton-resonance, we have : (m//)p = 2(eh/A7rmcXpp/vc) (11) and consequently: {n/I)e = 2(ve/pp){eh/^Tmc)Mvc) (12) So here is the value of (ti/I) for the electron expressed in terms of the Bohr magneton, determinable by measurements on ratios of frequencies only! At this point the reader may well wonder why I did not eliminate Vp from (12) by simply dividing it out. The reason is that one group of experimenters has determined (ve/j^p) at one fieldstrength and another group of experimenters has determined (vp/vc) at another fieldstrength, so that Vp does not have the same value in the two brackets: this is trivial. The old belief, as I remarked above, was that (fi/I) for the electron amounted to exactly two Bohr magnetons. But the combination of two experiments which I have just so sketchily described has led to the following result for the electron in the hydrogen atom : (/i//). = gieh/^rmc), g = 2.002292 ± 0.000024 (13a) But is this truly the ideal case? Defining the "ideal case" as that of the free electron, remembering that the electron in the hydrogen atom is bound even though lightly bound, and making what is deemed the appropriate correction, one elevates the foregoing value of g by 35 parts in a million, and obtains: Ideal (ji/I)e = g(eh/4irmc)y g = 2.002327 ± 0.000024 (13b) Thus the old belief was wrong by about one part in a thousand. Be it mentioned in passing that the Dirac theory which led to ^ = 2 has been modified in the meantime by what is known as "quantum electrody- namics", which gives a good account of this result. * To be derived by equating the force Hv{e/c) exerted bv the field upon the electron to the "centrifugal force" mv^/r; here v stands for the speed of the elec- tron and r for the radius of the circle. MAGNETIC RESONANCE. II 393 Since / is J/^ for the electron (as it is for the proton) the magnetic moment of the free electron is: M. = (3^)^(6/47rmc) = 1.001146 ± 0.000012 Bohr magnetons (14) This is the value which is 658.2288 times the moment of the proton. Another case very near to the ideal is afforded by the electrons of such atoms as manganese widely dispersed in a phosphorescent solid. Thus, there exists a measurement of g made upon "zinc sulphide phos- phor" containing manganese atoms in a concentration of 0.001 per cent. The value is 2.0024 =b 0.0004. It must be said that the resonance in question is complicated both by fine structure and by hyperfine struc- ture, terms to be explained in following sections. It is therefore neces- sary to use theory to locate, among the complex of peaks, the frequency which corresponds to the appropriate value of g. Still another case which is close to the ideal is provided by the "F- centres" in colored crystals, mention of which was made in Part I. An F-centre is a cavity in a crystal lattice occupied by a free electron batting around, as I said in Part I, like a wild animal in a cage. Several physicists have found their resonance, present when the crystal is colored and absent when the crystal is bleached. One, who produced the colora- tion by neutron-bombardment, located the peak at gr = 2.00. Others report 1.995 =b 0.001. Still another case which is close to the ideal is afforded by the con- duction electrons in a metal. These are so numerous that one might expect that the electron resonance that they produce must be extremely prominent. Yet the first such peak to be observed has been reported only as these lines are being written! The reasons for its inconspicuous character are two: most of the conduction-electrons are coupled anti- parallel, and the skin-effect confines the oscillating field in a conductor to a very narrow region close up against the surface. The second of these hindrances is overcome by using a colloidal dispersion of the metal, of which the spherules are less than 10~^ cm in diameter. Data are avail- able (though not yet all in print) for hthium, sodium and potassium. The values of g are within a few promille of 2.000; the differences between these and the ''ideal" value are small but not trivial, and in the case of lithium have been explained. ELECTRON RESONANCE IN PARAMAGNETIC SOLIDS There are paramagnetic soUds that display the electron resonance. A magnificent illustration is sho^vn in Fig. 2, belonging to an organic 394 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 substance of which the name and the structural diagram are included in the figure. This is one of the strongest and sharpest electron-resonance peaks on record. The gr- value is 2.0064 d= 0.0002; it is therefore almost an ideal case, but the difference from the ideal value is sure and signifi- cant. It would however be misleading to suggest that such a case is t3^ical. What are called the ''strongly paramagnetic salts" form a group with several features in common. They tend to have long names, and they have complex chemical formulae; crystal lattices or at any rate unit-cells which are non-cubic; and atoms some of which belong either to the rare-earth elements or to the elements of the "first transition group," iron or cobalt for instance. These atoms are likely to have two or more uncompensated electrons in parallel coupling. I now recall what was said about such coupled electrons in the introductory passage. 8520 8540 8560 8580 8600 8620 8640 8660 8680 STATIC MAGNETIC FIELD IN OERSTEDS Fig. 2. — Electron resonance of porphyrexide. This is one of the strongest and sharpest peaks of electron -resonance yet observed. The g-value is 2,0064 ± 0.0002, which makes it slightly but significantly different from the ideal case. In the structural diagram, the asterisk signifies a three-electron bond. (A. N. Holden, W. A. Yaeger and F. R. Merritt). MAGNETIC RESONANCE. II 395 Two or more electrons — ■ N electrons, let me say — may form, in effect, a rigid unit having a total spin S = N/2 and a total magnetic moment Nue . Such a unit will have (iV + 1) allowed orientations in the big magnetic field. These will engender N resonance-peaks. In the ideal case, all of these would have the same frequency 2neH/h, and would therefore coalesce into a single peak at the position appropriate to g = 2.0023. But in these crystals Ave are likely to find cases far from ideal, because of the conjoined influence of two factors. These are the presence of orbital motions of the electrons, and the presence of a big electric field within the crystal. Were the atoms in question free, we could allow for the orbital mo- tions. There would be a single resonance-peak, corresponding to a value of g which could be computed by a formula well known and much used in optical spectroscopy. Incidentally, this formula was used in interpreting the earliest molecular-beam experiments (not here described) that were the first to show that g in the ideal case is not exactly equal to 2. Now, however, we are dealing with resonating electrons that are in a strong electric field, and moreover, an electric field which is usually unsymmetrical. If the asymmetry is sufficiently great, the orbital mo- tions suffer a singular effect. This effect is known as "quenching." It is impossible to explain and difficult even to describe without invoking quantum mechanics. One may say that the orbital angular momentum is no longer constant in time, and the associated magnetic moment abnost but not quite disappears. The spin survives the quenching: but it would not be right to say that the quenching restores the ideal case. The resonance is affected by what have been called the ''remains" of the orbital magnetic moment. These have the following consequences : (a) The N resonance-peaks, which coincide in the ideal case, may be drawn apart. They then form a group of N separate peaks, which is known as a ''fine-structure pattern." The number N tells us the number of electrons coupled parallel in the atom, for these two numbers are the same. Often the number of electrons coupled parallel is known from independent evidence, and in such cases it is confirmed by the number of lines in the fine-structure pattern. Sometimes it is not otherwise known, and in such cases it is identified with the number N. (b) The value of g corresponding to the centre of the fine-structure pattern may be altered considerably from 2.0023, falling as low as 1.35 or rising as high as 6.5. This is as though a part of the orbital magnetic moment were added to or subtracted from the magnetic moment of the spin. 396 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 (c) The value of g may depend upon the orientation of the applied magnetic field with respect to the crystal. (d) The frequency of the resonance-peak or peaks may not be pro- portional to H. In fact, it may deviate so far from being proportional to H that extrapolation to ^ = 0 will indicate that even in the absence of ar applied magnetic field there would be a separation of the levels. Thus the asymmetric electric field within a strongly paramagnetic crystal may by itself produce the effect, which hitherto we have been ascribing en- tirely to the applied magnetic field. This is called "zero-field splitting." One sees only too well that the interior of a strongly paramagnetic salt is no place to look for the ideal case, and that resonance in such a salt is a theme for deep study and not for facile interpretation. As a matter of fact, electron resonance in paramagnetic salts is valued for its contribution to our knowledge of the electric fields in these crystals ; which is to say, that it is a part of solid-state physics, the details of which lie beyond the scope of this article. HYPERFINE STRUCTURE OF ELECTRON RESOXAXri: One of the most beautiful phenomena in this province of physics — and, I venture to say, not in this province only but in the whole of physics — is the "hyperfine structure" or "hyperfine splitting" of the electronic resonance. Here we see the spin and the magnetic moment of the nucleus collaborating with those of the electron to produce an ex- quisite and lucid joint effect. It is still the electronic resonance, and must never be confused with the nuclear resonance; but the single resonance- peak of the ideal case is split into a group of peaks, the number of which is determined by the spin of the nucleus. Fig. 3 relates to neodymium — not however to the metal, but to neodymium atoms in a salt of neodymium, diluted with a salt of another metal so that the neodymium atoms may not influence one another through undue proximity. Neodymium is an element with two "odd" isotopes — that is to say, isotopes of odd mass-number — and several "even" isotopes. The even isotopes have non-magnetic nuclei, and so do not perturb the electron resonance. Each of the two odd isotopes has a nucleus of spin 7/2 and non-zero magnetic moment. Such a nucleus will have eight permitted orientations in the big magnetic field. It will produce a local magnetic field in the region of the resonating electrons, and the strength of this field will depend on the orientation. The reso- nance-frequency depends on the big field compounded with the local field (we met with instances of this rule in the study of nuclear reso- MAGNETIC RESONANCE. II 397 nance). Therefore there are eight resonance-peaks for the electrons in the atoms of the isotope 143, and eight more for the electrons in the atoms of the isotope 145. This is the key to the remarkable pattern shown in the curve at the bottom of Fig. 3. In the middle of the pattern is the stump of a tall peak. This is the unperturbed peak due to the electrons in the atoms of even isotopes, those of which the nuclei have no magnetic moment. Whether it is at the position corresponding to g = 2.00 will depend on whether the dis- placement due to electric fields in the crystalline salt of neodymium, with which these data were obtained, is negligible or is not. Then, there are eight much shorter peaks. These are due to the electrons in the atoms Nd EVEN Nd 143 I I \ I \ [ Nd 145 I \ \ I \ \ I I Fig. 3. — Hyperfine-structure pattern of the electron resonance of neodymium in a salt of the metal, showing that the nuclei of each of the odd isotopes of neo- dymium have eight orientations and therefore a spin of 7/2, and that the even isotopes do not affect the resonance. (Courtesy of B. Bleaney). of the more abundant of the two odd isotopes. Then, there are eight still shorter peaks (provided we count one which is merged with one of the other group of eight) . These are due to the electrons in the atoms of the less abundant of the two odd isotopes. This is beautifully confirmed by the fact that the statures of the two groups of peaks stand to one another in the ratio of the abundances of the two isotopes! Further, the spacings within the two groups stand to one another in the ratio of the magnetic moments of the nuclei of the two isotopes. As for the two combs that stand above the curves, they are markers to identify for the onlooker the members of the two groups of peaks. Observations on the similar pattern of a (rare) isotope of vanadium — vanadium 50 — have led to the inference that this nucleus possesses a non-zero magnetic moment and a spin equal to 6 (the highest value so far known). This may seem surprising, since I have implied that nuclei of even mass-number have neither spin nor magnetic moment. Vanadium 398 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 50 is however a nucleus with an odd number of protons and an odd number of neutrons. Such nuclei, of which there are only a few stable examples, (in Part I we met with two, the deuteron and N^^), are not bound by the usual rule. FERROMAGNETIC RESONANCE Ferromagnetic bodies owe their distinctive feature to uncompensated electrons. This suggests that the magnetic resonance of electrons will be discernible in such bodies, and so indeed it is. In this case it is com- monly known as "ferromagnetic resonance." However, unless the sample is in the shape of a sphere, the resonance-peak will be found in what appears to be very much the wrong place. This is due to the magnetiza- tion of the substance, which produces a remarkable effect upon the location of the resonance. The field strength in the region occupied by the sample, which would be H if the sample were not there, is changed to a very different value ; and yet in general it would not be right to take the value of Hi the ''internal field strength" and put it in place of H in equation (3). We must understand this effect and make the proper allowance for it before we do anything else with the data (unless, I repeat, Ave confine ourselves to data obtained with spheres). The effect appears to be beyond the power of ''intuition" to conceive, and we must have recourse to the fundamental equations, which describe the pre- cession of the electronic magnets. It will be recalled that in Part I, we^ looked at nuclear magnetic resonance sometimes as the turning-over of nuclear magnets and sometimes as an outcome of precession. Now we are going to treat the electronic resonance as an outcome of precession. The fundamental vector equation, which was given in a sort of diluted form as equation (6) of Part I, reads as follows: dp/dt = fieX Hi (15) Here p and fie stand for the angular momentum and the magnetic mo- ment of the electron, and Hi for the field which operates on the electron. We have seen that m«/p is written as ge/2mc ; we denote this quantity by y; and we give it the minus sign because, for the electron, angular mo- mentum and magnetic moment are antiparallel to one another. Now we have: dfie/dt = -TMc X Hi (16) This we proceed to write as three scalar equations; but first we replace fXe by M. This will help to do away with the implication that the mag- netic moment varies in magnitude (it is the direction that changes with MAGNETIC RESONANCE. II 399 time) and will also convey the plausible suggestion that all of the resonat- ing electrons in the substance are coupled parallel, so that M can signify the magnetization of the substance. We have: dM,/dt = -yiMyHi, - M.Hiy) dMy/dt = -y{M^Hi, - MMiz) (17) dM./dt = -y{MMiy - MyHi,) Now we are to make the following important substitutions, some of which are approximations. (1) Presuming that M the magnetization of the substance will not deviate far from the z-direction, we are to write M for Mz . (2) For Hiz , the 2;-component of the field actually operating upon the electrons, we are to write {H — NzM) . Here H stands as heretofore for the applied field and Nz for the "demagnetizing factor" in the 2-direc- tion, which latter is a measure of the strength of the free poles on those surfaces of the sample which face the pole-pieces of the magnet (Fig. 1). Thus —NzM is the value of the field produced in the substance by these free poles. (3) YovHix and Hiy we are to write —NxMx and —NyMy . This means that whatever applied fields there may be in the x and the ^/-directions are negligible, and yet the components of magnetization in these direc- tions are not neghgible, so that the free poles on the surfaces perpendicu- lar to X and to y respectively are producing the internal fields of which —NxMx and —NyMy are the strengths. (4) We are to ignore terms in which the product MxMy appears, these being small. The fourth of these conditions makes dMz/dt vanish: we are left with only two of the three equations (17), a convenience. Making the substitutions allowed by the first three conditions, we find that the other two assume the forais: dMx/dt = -yMy[H - {Nz-Ny)M] (18) dMy/dt = -yMx[-H + {Nz-Nx)M] Now suppose that Mx and My are periodic functions of time, of fre- quency j/. We write them as Ml exp {2Trivt) andilfy exp (2Trivt). Substituting into (18), we find: 2irivMl + y[H + {Ny -N.)M]Ml = 0 -y[H - {N, - Nx)M]Ml + 2irivMi = 0 ^^ 400 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 These two simultaneous equations will be compatible with one an- other — • one might say that they make sense — only if they are ulti- mately the same equation. The ratio of the coefficients of Ml and Ml in the one must be the same as the ratio of the corresponding coeffi- cients in the other. In words more natural to algebraists, the determinant of the coefficients must vanish. It turns out that this condition deter- mines a specific value of v, and this value is the resonance-frequency: p = {ge/Airmc) V[H + (N, - N^)M][H -f (A^, - N^)M] (20) For reasons deriving from the history of celestial mechanics, this pro- cedure is known as ''solving the secular equation." In the most common experimental set-up, the sample is a thin layer parallel to the 2;-direction — so thin that by comparison with its breadth, the free poles at the surfaces opposite the pole-pieces of the magnet may be regarded as infinitely far away. Under these conditions Nz vanishes, and so does Nx if we lay the a:-axis parallel oo the surface of the thin layer; but Ny does not vanish, it is in fact equal to 4^. Under the radical, the first factor becomes equal to H and the second to H -\- 4x3/, which latter is by definition the induction B. We have: V = {ge/4wmc) VHB (21) Note here that since B depends upon both H and M, one cannot use the formula unless one knows the value of M, which is the magnetization of the substance at saturation. This usually requires knowledge obtained from other experiments; but we shall meet with a case in which, at least "in principle," the value of B may be found from the resonance-experi- ment itself. Equation (21) is the commonest formula for the ferromagnetic reso- nance, for it fits the "geometry" of the original and of most of the subsequent experiments. Yet there are other formulae corresponding to other geometries, and two of these are particularly important. It is feasible to orient the layer at right angles to the big appHed field. For this case we shall do well to turn the axis of z so that it remains parallel to the big field. Now A^^ and Ny vanish and Ns becomes Air, and the formula is this: V = {ge/4Tmc)(H - 47rM) (22) The quantity {H — iwM) is the internal field Hi , the field strength within the magnetized body. This is the special case in which the right result is obtained by going back to equation (4) and putting for H the actual field strength at the scene of the resonating electrons. In other MAGNETIC RESONANCE. II 401 words, this is the special case in which the naive approach does not lead the student astray. A more singular special case is that of the sphere. In this case A^x and Ny and Nz are all three of them equal — ■ equal to one another but not to zero. Nevertheless the formula is just our old formula (3), the same as though there were no magnetization at all: V = {fi/I){H/h) = {ge/4Trmc)H (23) One wonders how long it would have been before anyone set up equations (18) and derived equation (21), if all experiments had been perfomde with spheres. In the foregoing pages we have derived the resonance frequency by making certain listed approximations in the basic equations (19). Among these approximations was the neglect of the oscillating field, parallel to the axis of x. We arrive at some interesting results by introducing this field into the equations and giving it an arbitrary frequency, while con- tinuing to make all of the other approximations. It shall be denoted by Hi exp {2'wivt) ; Hi , it may be recalled, was the symbol used in Part I for the amplitude of this field. In this passage v shall signify any fre- quency that the experimenter may choose to apply, while the resonance - frequency heretofore called v shall change its symbol and become vq . On the right-hand side of the second of the equations (19) will now appear, as the reader can show for himself, —yMHi instead of zero. The two simultaneous equations now^ make sense for any value of v, instead of just the value vq . On solving them for Ml , one finds: ^°/^- - H + il^- N.)M T^hM^ (^^) The quantity on the left, and hence also the quantity to which it is equated, is the "susceptibiUty" of the substance ^\^th respect to this oscillating field which, be it remembered, is imposed at right angles to the big applied field. The quantity on the right has the well-known form of an optical dis- persion-curve. Suppose the frequency to be increased from zero. The susceptibility rises from a finite and non-zero value at j/ = 0 to positive infinity at the resonance-frequency vo ; here it jumps suddenly to nega- tive infinity, from which value it rises asjTiiptotically to zero as the frequency is increased toward infinity. In magnetics there are methods of measuring directly, not the sus- ceptibility X itself but the sum (1 + 47rx), which is called the "per- meability" and is denoted by m- It is evident that Avhile the susceptibility 402 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 is rising, with increase of frequency, from its negative-infinite value at VQ to its asymptotic value of zero at infinite frequency, the permeability is rishig from negative infinity to an asymptotic value which is equal to + 1. Somewhere along this range of frequencies it must pass through zero, at a frequency to be denoted by vi . For a stratum parallel to both the big applied field and the oscillating field, it is easily shown that vi is equal to {ge/^Trmc)B. This offers a way of determining B and conse- quently M. 80 60 - 1 1 1 - Umax 584o oe 40 20 10 8 6 1 / - \ - \ \ 4 1 \° / \ 1-1 ^ 2 1.0 0.8 0.6 / -N,. >-^ *^^ J / J f^ ■O-O- ^rPc D — J k ( fj V / 1 0.4 I n /; \ H U 0.2 0.1 > \ jf-HMlN=2O40 OE \ / - [ - 1 < 0 1 2 3 4 5 6 7 8 9 10 11 12 1 3 14 15 16 17 INTERNAL STATIC MAGNETIC FIELD, Hz, 'N KILO-OERSTEDS Fig. 4. — Ferromagnetic resonance in Heusler alloy (Cu-Mn-Al). "Apparent" permeability is plotted against H at constant frequency: the resonance-maximum (at the position corresponding to g = 2.02) is vividly shown, as is the minimum mentioned in the text. The solid curve is a theoretical curve based on a specific assumption about damping. (W. A. Yager and F. R. Merritt). MAGNETIC RESONANCE. II 403 Next suppos( that what is plotted against v is not /jl but |ju|, the abso- lute valu( of th( permeability. The portions of the n-vs-v curve which were below the horizontal axis now appear inverted and above the horizontal axis. The curve has an upward-pointing peak reaching to infinity at vo , anc' a downward-pointing peak touching th( axis with its tip at Vi . Such is the general aspect of the curve of Fig. 4, pertaining to a Heusler alloy. There are superficial differences: the curve of Fig. 4 is plotted against H for constant frequency, and the scale along the axis of ordi- nates is logarithmic. The reader can easily make allowance for these. There is also a fundamental difference : the curve reveals the presence of damping or relaxation, which broadens the peaks and prevents \^l\ from rising to infinity or dropping quite to zero. The continuous curve is derived from a theory which involves a specific assumption about the damping; one sees that it agrees well with the data excepting in a re- gion around the minimum. Curves such as these are likely to be in- fluenced by anisotropy in the ferromagnetic substance, which reversely can be evaluated from the curves. How about the values of g for ferromagnetic substances? The Heusler alloy to which Fig. 4 pertains has a value of g which, so far as the accuracy of the experiment permits us to judge, may be identical with the ideal value (the most probable value is however 2.01). This is an exception and not the rule. The range of values is rather wide, though apparently not so wide as in the strongly paramagnetic salts. Most of them lie between 2.22 (for cobalt) and 2.01 (for the Heusler alloy aforesaid); but there are instances of values still higher, including one of 3.75 for manganese arsenide. There is also at least one value lower than 2.00; it is presented by gadolinium, a very interesting element. Below its Curie point at 16° absolute, gadolinium shows a resonance-peak of which the breadth interferes with a precise location of its top; the value of g is given as 1.95 to 1.96. Above the Curie point, gadolinium is para- magnetic, but the peak persists and is sharper; the value of g is 1.95 d= 0.03. I remind the reader that when the experiment is such that formula (21) must be used, a g'-value implies an assumption about the value of M the magnetization of the substance at saturation. I must not close this topic without alluding to something which there is not space to expound. Experiments on the ''gyromagnetic effect" — something which has a much longer history than ferromagnetic reso- nance — lead to values of a quantity which has also been denoted by g. Until a few years ago it was supposed that this quantity must be the same as the g of these pages; but experiment has ruled otherwise, and theory has been successful in at least suggesting a reason. The g of these 404 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 pages is now called the spectroscopic splitting factor; the other has been set apart as the ''gyromagnetic" g, and some people have even taken to writing it as g' , which seems rather unfair to the senior g. It seems to be a general rule that when one of the two is greater than 2.00 the other is smaller than 2.00; and in the case of the Heusler alloy, they may well coincide. ACKNOWLEDGEMENT Among the many who have helped me with this article I wish to extend especial thanks to Charles Kittel, A. F. Kip, W. D. Knight, M. E. Packard, E. M. Purcell, J. H. Van Vleck and W. A. Yager; and to Messrs. Packard, Purcell and B. Bleaney for prints of the illustrations captioned with their names. REFERENCES This makes no pretense of being a bibHography of magnetic resonance : such an enterprise would cover more pages of this Journal than these articles themselves. It is somewhat, but not much, more than a listing of the sources of the data quoted in the articles, with the names which have been omitted in the belief that names tend to slow down exposition. Many relevant papers, with an abundance of foot- note references to anterior work, are to be found in what I abbreviate by P.I. C.S.R. ("Proceedings of the International Conference on Spectroscopy at Radiofrequen- cies, Amsterdam, 1950" separately published and also printed as part of Volume 17 of Physica). The locus classicus for the nuclear resonance absorption is the paper of Bloem- bergen, Purcell and Pound {Phys. Rev. 73, p. 670, 1948); the first publication of this school is by Purcell, Torrey and Pound in Phys. Rev. 69, p. 37, 1946. The locus classicus for the precession-theory and the nuclear-induction method is the paper of Bloch (Phys. Rev. 70, p. 460, 1946), followed by the first lengthy descrip- tion of nuclear -induction measurements by Bloch, Hansen and Packard (ibid. p. 464); the first publication of this school is in Phys. Rev. 69, p. 127, 1946. The discovery of nuclear magnetic resonance by the molecular-beam technique was disclosed by Rabi, Zacharias, Millman and Kusch in Phys. Rev. 63, p. 318, 1938, and a more detailed account is given ibid. 55, p. 526, 1939; consult also the paper of Kellogg, Ramsey, Rabi and Zacharias z6 id. 57, p. 677 (1940) for the resonances of protons and deuterons in molecular beams of H2 , D2 and HD. The reference to the article of G. E. Pake {Am. Jour. Phys. 18, pp. 438-452 and pp. 473-486, 1950) is here repeated to draw attention to this excellent survey of nuclear magnetic resonance and relaxation. Another survey article is that of Rollin, Reports on Recent Progress in Physics, 1948-49. The reference for the chemical shift in ethyl alcohol (Fig. 7 of Part I) is Arnold, Dharmatti and Packard, J. Chem. Phys. 19, p. 507, 1951 . For the influence of F-centres on nuclear relaxation- time see Hatton and Rollin, Proc^ Roy. Soc. 199, p. 231, 1949. For the final experiments on determination of g for electrons in hydrogen atoms see Koenig, Prodell and Kusch {Phys. Rev. 88, p. 191, 1952); references to earlier MAGNETIC RESONANCE. II 405 work on other atoms are listed there. For the deterniination of the proton moment see Gardner, Phys. Rev. 83, p. 996, 1951, and Sommers, Hippie and Thomas, ibid. 80, p. 487, 1950. For the measurementof g with F-centres see Hutchinson and Noble, ibid. 87, p. 1152, 1952, and Tinkham and Kip. ibid. 83, p. 657, 1951, and Schneider and England in P.I.C.S.R.; this last is also the source for the work ion zinc sulfide phosphor. For resonance of conduction-electrons see Griswold, Kip and Kittel, Phys. Rev. 88, p. 951, 1952, and papers yet to be published. The resonance-peak of Fig. 2 of Part II of this article, for porphyrexide, comes from Holden, Yager and Merritt, /. Chem. Phys. 19, p. 1319, 1951. The literature of resonance in strongly paramagnetic salts is extensive and tough; I refer to the papers in P.I.C.S.R. and the references they give. The basic theory is to be found in the book of J. H. Van Vleck, Electric and Magnetic Susceptibilities (Oxford, 1932). Hyperfine structure of electron resonance was discovered by the late R. P. Penrose (see Nature, 163, pp. 988 and 992, 1949). This field is almost a monopoly of Britain and in particular of Oxford; many of the papers bear the name of Bleaney with or without collaboratores. Fig. 3 of Part II of this article comes from Proc. Phys. Soc. 63, p. 1369, 1950; the statements about vanadium 50 from ibid. 66, p. 952, 1952. The discoverer of ferromagnetic resonance was J. H. E. Griffiths (see Nature, 158, p. 670, 1946). The precession-theory for ferromagnetic substances is due to Kittel; equation (21) of this article is derived in Phys. Rev. 71, p. 270, 1947; a fuller treatment appears ibid. 73, p. 155, 1948. Survey articles are those of Van Vleck in P.I.C.S.R. and Kittel in Jour, de Phys., 12, p. 291, 1951. Fig. 4 comes from the paper of Yager and Merritt, in Phys. Rev. 73, p. 318, 1949; a similar curve for supermalloy appears in the preceding paper; references to the ^-values of other ferromagnetic substances are given by Yager. The question of ''g versus g'" is dis- cussed by Kittel in Phys. Rev. 76, p. 743 (1949). A Study of Non-Blocking Switching Networks By CHARLES CLOS (Manuscript received October 30, 1952) This paper describes a method of designing arrays of crosspoints for use in telephone switching systems in which it will always he possible to establish a connection from an idle inlet to an idle outlet regardless of the number of calls served by the system. INTRODUCTION The impact of recent discoveries and developments in the electronic art is being felt in the telephone switching field. This is evidenced by the fact that many laboratories here and abroad have research and development programs for arriving at economic electronic switching systems. In some of these systems, such as the ECASS System,* the role of the switching crossnet array becomes much more important than in present day commercial telephone systems. In that system the com- mon control equipment is less expensive, whereas the crosspoints which assume some of the control functions are more expensive. The require- ments for such a system are that the crosspoints be kept at a minimum and yet be able to permit the establishment of as many simultaneous connections through the system as possible. These are opposing require ments and an economical system must of necessity accept a compromise. In the search for this compromise, a convenient starting point is to study the design of crossnet arrays where it is always possible to establish a connection from an idle inlet to an idle outlet regardless of the amount of traffic on the system. Because a simple square array with N inputs, N outputs and N^ crosspoints meets this requirement, it can be taken as an upper design limit. Hence, this paper considers non-blocking arrays where less than N^ crosspoints are required. Specifically, this paper describes for an implicit set of conditions, crossnet arrays of three, five, * Malthaner, W, A.,andH. Karle Vaughan,An Experimental Electronically Controlled Switching System. Bell Sys. Tech. J., 31, pp. 443-468, May, 1952. 406 NON-BLOCKING SWITCHING SYSTEMS 407 etc., switching stages where less than N^ crosspoints are required. It then deals with conditions for obtaining a minimum number of cross- points, cases where the N inputs and N outputs can not be uniformly assigned to the switches, switching arrays where the inputs do not equal the outputs, and arrays where some or all of the inputs are also outputs. SQUARE ARRAY A simple square array having iV inputs and N outputs is shown in Fig. 1. The number of crosspoints equals N"^ and any combination of N or less simultaneous connections can exist without blocking between the inputs and the outputs. The number of switching stages, s, is equal to 1. The number of crosspoints, C(s), is: C(l) = N' (1) c r 5 C ^ ( c rpL JTS ) C ? ■z " z ° NUMBER OF CROSSPOINTS =N^ Fig. 1 — Square Array. THREE-STAGE SWITCHING ARRAY An array where less than N^ crosspoints are required is shown in Fig. 2. This array has N = 3Q inputs and iV = 36 outputs. There are three switching stages, namely, an input stage (a), an intermediary stage (b), and an output stage (c). In stage (a) there are six 6 x 11 switches; in stage (b) there are eleven 6x6 switches; and in stage (c) there are six 6 X 11 switches. In total, there are 1188 crosspoints which are less than the 1296 crosspoints required by equation (1). Of interest are the derivations of the various quantities and sizes of switches. In stage (a) the number, n, of inputs per switch was assumed to be equal to N^^^, thus giving six switches and six inputs per switch. In a similar manner stage (c) was assigned six switches and six outputs per switch. The number of switches required in stage (b) must be suf- ficient to avoid blocking under the worst set of conditions. The worst case occurs when between a given switch in stage (a) and a given switch in stage (c) : (1) five links from the switch in stage (a) to five correspond- 408 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 ing switches in stage (b) are busy; (2) five links from the switch in stage (c) are busy to five additional switches in stage (b); and (3) a connection is desired between the given switches. Thus eleven switches are required in stage (b). The remaining requirements, namely, eleven verticals per switch in stages (a) and (c) and six by six switches in stage (b) are then easily derived. The number of crosspoints required for three stages, where n = A^^^^, is summarized by the following formula : 1/2 (7(3) = {2N"' - 1) (3N) (2) (2a) In Table I it may be noted that the number of crosspoints is less than iV2 for all cases of iV > 36. PRINCIPLE INVOLVED The principle involved for determining the number of switches re- quired in the intermediary stage is illustrated in Fig. 3. The figure is for a specific case from which one can generalize for n inputs on a given input switch and m outputs on a given output switch. In the figure it is desired to establish a connection from input B to output H. A suf- ficient number of intermediary switches are required to permit the (n — 1) inputs other than B on the particular input switch and the (m — 1) outputs other than H on the particular output switch to have connections to separate intermediary switches plus one more switch for the desired connection between B and H. Thus n + m — 1 inter- mediary switches are required. Table I — Crosspoints for Several Values of N N Square Array N* Three-Stage Array 6^"^ - 3N 4 16 36 9 81 135 16 256 336 25 625 675 36 1,296 1,188 49 2,401 1,911 64 4,096 2,880 81 6,561- 4,131 100 10,000 5,700 1,000 1,000^000 186^737 10,000 100,000,000 6,970,000 NON-BLOCKING SWITCHING SYSTEMS 409 0OQQO9OOOOC> N=36 4 OTHER INPUT SWITCHES n = 6 V P 0 0 9 OTHER INTERMEDIARY SWITCHES QOppOOpOOOO A I 4 OTHER OUTPUT SWITCHES N=36 k— 6~>H STAGE (a) STAGE (b) STAGE (c) NUMBER OF CROSSPOINTS = 6N^/^-3N (1188 CROSSPOINTS WHEN N =36) Fig. 2 — Three-stage switching array. nVE-STAGE SWITCHING ARRAY A five-stage switching array is illustrated in Fig. 4. The analysis of this array can be made in the following manner. Each input and output switch is assumed to have n = N^^^ inputs or outputs, respectively. Connection between a given input switch and a given output switch SWITCHES n INPUTS QN A PARTICULAR INPUT SWITCH A — m OUTPUTS ON A PARTICULAR OUTPUT SWITCH -D •-E -F G — H SWITCHES REQUIRED -i) + (nn-i)+i=n + m-i WHEN n = m, AB0VE=2n-l Fig. 3 — Principle involved. 410 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 is made via levels, a level consisting of three intermediary switching stages. The number of levels required is {2N^'^ — 1). Each level has N^'^ inputs and the same number of outputs. The number of crosspoints for a three-stage non-blocking array for A^^^^ inputs and A^^'^ outputs can be obtained from equation (2) by substituting A^^^^ for N in that equation. The total number of crosspoints required for the five-stage array is: C(5) = (2N'^' - if SN''' + C2N'" - 1) 2N (3) = 16A^'^' - UN -f SN'^' (3a) The number of crosspoints required for several sizes of the five-stage array is given in Table II. The results are compared to the square and three-stage arrays. SEVEN-STAGE SWITCHING ARRAY A seven-stage switching array can be analyzed by considering paths requiring five intermediary switching stages as paths via switching aggregates. The number of such aggregates is {2N^''^ — 1). Each ag- gregate has N^'* inputs and a like number of outputs. From equation (3) the crosspoints for each aggregate can be obtained by substituting ^^ 1 x^ . SV^IJCHES^^ """"'^ \v SWITCHES 1 r 2 .^ ^ V 1 : Js, A A (EA \\ ° )WITCHING LEVEL CH LEVEL CONSI F THREE STAGES s STS ) ^ 2 n 4 // ^ ' A 5 \^ 2n-2 ^/ 5 \ / ^ 2n-i ^ Fig. 4 — Five-stage switching array. NON-BLOCKING SWITCHING SYSTEMS 411 Table II — Crosspoints for Several Values of N N Square Array Three-Stage Array Five-Stage Array 64 729 1,000 10,000 4,096 531,441 1,000,000 100,000,000 2,880 115,911 186,737 5,970,000 3,248 95,013 146,300 3,434,488 N^^'^ for N in that equation. The total number of crosspoints required for the seven-stage array is: C(7) = {2N''' -If 3N''' + (2N'" - if 2N'" -f {2N'" - 1)2N (4) = 36Ar'^' - 46V + 20V'^' - 3V'^' (4a) general multi-stage switching array Equations (1), (2a), (3a) and (4a) are herewith tabulated as a series of polynomials together with the next polynomial: C(l) = N' (1) C(3) = 6N'" - 3V (2a) C{d) = im'" - UN + 3V'^' (3a) C(7) = 36V'^* - 46V + 20V'^' - 3V^'^' (4a) C(9) = reV"^' - 130V + 86V'^' - 26V'^' + 3V'^' (5) These polynomials can be determined for any number of switching stages from the following formula where s is an odd integer: C(s) = 2E; s-fl 2k / ,2F 2 8 + 1 >)^ + N' .+1 (2^.+! -■)- (6) An alternative expression equivalent to equation (6) has been sug- gested by S. O. Rice and J. Riordan. The recurrence relation used in individually deriving the foregoing polynomials can be used to directly derive the following formula : C{2t + 1) = n (2n 1) [(on - 3)(2n - 1)'"' - 2n*] (6a) n - 1 where s = 2^ -f 1 N = n'+^ Table III gives comparative numbers of crosspoints for various num- 412 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table III — Crosspoints for Various Numbers of Switching Stages, s, and Values of A^ N s = 1 s = 3 s = 5 5 = 7 5=9 100 10,000 5,700 6,092 7,386 9,121 200 40,000 16,370 16,017 18,898 23,219 500 250,000 65,582 56,685 64,165 78,058 1,000 1,000,000 186,737 146,300 159,904 192,571 2,000 4,000,000 530,656 375,651 395,340 470,292 5,000 25,000,000 2,106,320 1,298,858 1,295,294 1,511,331 10,000 100,000,000 5,970,000 3,308,487 3,159,700 3,625,165 lU- / / / y / in7 / / / A / / o S-\/ / /. :^ i,oe / X^ > >^ / ^ >V 2 ° in A / ^> ^ o . / x^ >> / / ^'' 5 105 / / S = 9 .^ ^ • S=5 ^ r^ \ Z ° <1^ y^' ?" S = 7 / IJ"^ ^' / y' >^ ^' in4 /<^ y '^yy y 103 1 1 1 1 100 200 300 400 600 1000 2000 4000 NUMBER OF INPUTS AND NUMBER OF OUTPUTS 6000 10,000 Fig. 5 — Crosspoints versus switching stages. NON-BLOCKING SWITCHING SYSTEMS 413 bers of switching stages and sizes of A^. The data of Table III are plotted on Figure 5. The series of curves appear to be bounded by an envelope, representing a minumum of crosspoints. The next section dealing with minima indicates that points exist below this envelope. MOST FAVORABLE SIZE OF INPUT AND OUTPUT SWITCHES IN THE THREE- STAGE ARRAY The foregoing derivations were for implicit relationships between n and iV, namely, n being the ( — - — \th root of N. To obtain minimum number of crosspoints a more general relationship is required. For the three stage switching array this is : C(3) = (2n - 1) (2N -f J) (7) When n = N^^^ equation (7) reduces to equation (2). For a given value of N, the minimum number of crosspoints occurs when dC/dn = 0 which gives: 2n' - nN + N = 0 (8) This equation has the following two pairs of integral values: n = 2, iNT = 16 and n = 3, N = 27 As N approaches large values equation (8) can be approximated by: N = 2n2 (9) Graphs of equations (8) and (9) are shown in Fig. 6. In Table IV the numbers of crosspoints are based on the nearest integral values of n for given values of N. Where comparisons can be made, Table IV indicates fewer crosspoints than does Table I. This fact can be realized in another manner. By eliminating n in equations (7) and (9), the result for large values of N is: C(3) = 4 (2f"N'^' - 4N (10) Equation (10) indicates fewer crosspoints than does equation (2). MOST FAVORABLE SWITCH SIZES IN THE FIVE-STAGE ARRAY If n be the number of inputs per input switch and outputs per output smtch, and m be the number of inputs per smtch in the second stage 414 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 2 3 4 5 6 8 10 20 n = INPUTS OR OUTPUTS PER SWITCH Fig. 6 — Relationship between N and n for minima in crosspoints. Three-stage array. Table IV — Crosspoints FOR Several Values of N j\r Nearest Integral Number of Crosspoints ' N' Equation' (7) 16 2 256 288 27 3 729 675 40 4 1,600 1,260 44 4 1,936 1,463 65 6 3,025 2,079 60 5 3,600 2,376 65 5 4,225 2,691 78 6 6,084 3,575 84 6 7,056 4,004 98 7 9,604 5,096 105 7 11,025 5,655 NON-BLOCKING SWITCHING SYSTEMS 415 and outputs per swdtch in the fourth stage, then the following equation gives the total number of crosspoints: 0(5) = (2n - 1) [2N + i2m - 1) (f + £,)] (11) The partial derivative of this equation with respect to m when set equal to zero yields: n = m^ (12) The partial derivative of this equation with respect to n when set equal to zero yields the following equation: ^j nrn(2n + 2m — 1) , ^^ ^ = (2m - l)(n - 1) ^^^^ Equations (12) and (13) can be solved for n and m in terms of given values of N. For example for N = 240, we obtain n = 6.81 and m = 3.56. SEARCH FOR THE SMALLEST N FOR A GIVEN U FOR THE THREE- STAGE ARRAY For a given value of n, equation (7) furnishes a means for locating that size of three-stage switching array which has N^ or fewer cross- points. This can be done by setting equation (7) equal to N^: N' = {2n - 1) (2N + J) (14) and solving for N in terms of n. The solution is: ^ s ?!^(H!^_i) (15) {n - 1)2 Minimum values of N for given values of n are hsted in Table V. This table also hsts the next highest N exactly divisible by n. From this table it appears that when iV = 24, we have the smallest switching array for which it may be possible to have less than N^ crosspoints. However for N = 25, as sho\vn in Table I, equation (2) gives more than N2 crosspoints. The problem is one of finding an array for N = 25 with fewer than N" crosspoints. For this and all cases beyond, the next sec- tion indicates that it is profitable to consider situations where N is not exactly divisible by n. 416 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 CASES IN THE THREE-STAGE SWITCHING ARRAY WHERE N = r(MOD n) Table I indicated that for AT = 25 and n = 5 a total of 675 cross- points were required. A square array requires only 625. Fig. 7 shows a layout of switches where N = 25 and n = 3. In this case one input is left over when 25 inputs are divided into threes. The lone input requires three paths to the intermediary switches. This is in accordance with Fig. 3. The lone output also requires three paths to the intermediary switches. Also from Fig. 3, the lone input to the lone output requires only one path. Hence there must be one switch capable of connecting the lone input to the lone output. The number of crosspoints required is 615 which is less than the 625 required by the square array. This scheme can be extended to any case where N = kn -\- r, where the re- mainder, r, is an integer greater than zero but less than n. The formula for the number of crosspoints where k input and k output switches of size n and one input and output switch of size r are used is: C = 2(2n - 1)(N - r) -I- 2(n + r - l)r + (n - r) + {n + r- 1) i^-^ + lY - n + i (16) I. G. Wilson has pointed out that for a lone input the crosspoints in the intermediary switches can be used to isolate its possible connections hence no crosspoints are required in the input stage. This likewise ap- plies for a lone output. With this modification the array in Fig. 7 requires six fewer crosspoints. For this case, when r = 1, the number of cross- points is: C = 2(2n - \){N - 1) + (n - 1) (^^^^^Y Table V — Minimum Values of A^ for Given Values of n n N per Equation 15 iV = 0 (mod n) 2 24 24 3 22.5 24 4 24.9 28 5 28.1 30 6 31.7 36 NON-BLOCKING SWITCHING SYSTEMS 417 J. Riordan has found a more efficient arrangement for cases where N = kn -{- r. In place of using k switches of size n and one switch of size r, he proposes that {k -\- 1 - n + r) switches of size n and {n - r) switches of size n — 1 be used. For this case the number of crosspoints is: C = 2(2n - l){k + 1 - n + r)n + 2{2n - 2) (n - r){n - 1) + (2/z - 3)(/c + 1)^ + 2(k + l){k + 1 - n + r) (17) INPUT SWITCHES o o o o o c c c p p p ° ° p f p < p p < p ^ c p p c p c p p ° ° c p ? p c p c p < p ■ p c p c p < p p ° ° INTERMEDIARY SWITCHES QpQOQQOQO oooooooon OppOOQOOQ oooooono OOOOQOpp OUTPUT SWITCHES Q O O O O 0 0 0 0 0 p < p c p < p < p " p c p c [' p " " I Fig. 7 — Three-stage array. 25 ^ 1 (mod 3). An equivalent arrangement is to provide two 8x8 and three 9x9 intermediary switches. Two of the 9x9 switches need only 80 crosspoints. 418 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table VI — Crosspoints for Various Values of N and n Three-Stage Array N Square Array w = 2 n = 3 « =4 n = 5 23 529* 540 530 556 24 576 576 560* 588 — 25 625 625 609* 633 — 26 — 663 643* 667 — 27 — 716 675* 701 — 28 — 756 730* 735 ■ — 29 — 813 766* 788 — 30 855 800* 824 864 31 — — 861 860* 911 32 — — 899 896* 951 33 — — 935* 957 991 34 — — 1002 995* 1031 35 — — 1042 1033* 1071 36 — — 1080 1071* 1128 37 — — 1153 1140* 1170 38 — 1195 1180* 1212 39 — 1235 1220* 1254 40 — — 1314 1260* 1296 » = 4 « = 5 n = 6 n = 7 n = 8 50 1819 1800* 1879 — — 60 2415 2376* 2420 — — 70 3164 3024* 3056 — — 80 3920 3744* 3764 — — 90 — 4536 4455* 4499 — 100 — 5400 5291* 5315 — 110 — ^— 6199 6100 6156 120 — 7040 7044 6975* 130 — — 8076 7923* 7947 140 — — — 8840* 8860 150 — — — 9968 9811* 160 — — — 10979 10800* * Minimum values. Equation (17) is identical to equation (16) when r = n — 1. There are two cases, namely, when n = 2 and n = 3 where equation (16a) gives fewer crosspoints than does equation (17). SEARCH FOR THE MINIMUM NUMBER OF CROSSPOINTS BETWEEN N = 23 AND A^ = 160 The equations of the preceding sections furnish a means for search- ing for minimum crossnet arrays. Table VI shows the results of such a search up to A^ = 160. Results are indicated in unit steps from N — 23 to iV = 40 and for every tenth interval thereafter. At iV = 161, a five-stage array requires the fewest crosspoints. Table VI was computed by the use of finite differences. The equations NON BLOCKING SWITCHING SYSTEMS 419 were: C[{k + l)r?] C{kn) = (2n - l)(2n + 2/c + 1) C{kn + r + 1) - C(/cn + r) = 2(fc + 3n - 1) C(kn +1) - C(/crz) = 2/cn + 1 (18) (19) (19a) Equation (18) was derived from equation (7) with N being replaced by {k + l)n and by kn as required. Equation (19) was derived from equa- tion (17) with r being replaced by r + 1 as required. This equation applies for all values of n greater than 3 and for the particular case of n = 3 and r = 2. Equation (19a) was derived from Equations (16a) and (7) and is for the particular case of r = 1, when n = 2 and n = 3. SEARCH FOR THE MINIMUM NUMBER OF CROSSPOINTS FOR N = 240 For a case where A^' is large enough to require five-switching stages, the search for the minimum number of crosspoints should be based on equations (12) and (13) and on the use of Table VI. The method is sug- gested by means of Table VII. The data in a previous section indicate Table VII — Crosspoints for N = 240 and Various Values of n Input and Output Stages Intermediary Stages ~ Total No. of Size of Cross- No. of Inputs and Outputs Cross- Crosspoints Switches Switches points Levels points 2 120 2x 3 1,440 3 120 X 120* 8 20,925 22,365 3 80 3x5 2,400 5 80 X 80* 5 18,720 21,120 4 60 4x 7 3,360 7 60 X 60* 5 16,632 19,992 5 48 5x9 4,320 9 48 X 48 4 15,120 19,440 6 40 6x11 5,280 11 40 X 40* 4 13,860 19,140 7 /30 15 7x13 5,460 2 30 X 35 3 1,826\ 11,363/ 19,369 6x12 720 11 35 X 35* 4 8 30 8x15 7,200 15 30 X 30* 3 12,000 19,200 9 11 9 X 17 7,344 2 24 X 27 3 1,230\ 10,125/ 19,467 8x16 768 15 27 X 27* 3 10 24 10 X 19 9,120 19 24 X 24* 3 10,640 19,760 11 /20 I 2 11 x21 9,240 2 20 X 22 — 880\ 9,196/ 20,116 10x20 800 19 22 X 22 — 12 20 12x23 11,040 23 20 X 20 — 9,200 20,240 Cross Cross Cross points I )er equation (3) fi\ )er equation (2) th )er equation (^ ) so '■e-stag< 3 array . ... 20,596 21,624 points I ree-sta uare a ee array . . . points J rrav 57,600 * See Table VI for minimum number of crosspoints. 420 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 that a minimum should occur for N = 240, when n = 6.81 and m = 3.56. In Table VII the minimum occurs when n = 6 and m = 4. It fails to occur at n = 7 because 240 is not exactly divisible by 7. Except for this situation, the minimum would have occurred as predicted. RECTANGULAR ARRAY Referring to Fig. 1, if there were A^ inputs and M outputs, a simple rectangular array would result which would be capable of sustaining up to A^ or Mj whichever is the lesser, simultaneous connections without blocking. The number of crosspoints is: C(l) = NM (20) N INPUTS AND M OUTPUTS IN A THREE-STAGE ARRAY For the case of a three-stage switching array with N inputs and M outputs, let there be n inputs per input switch and m outputs per out- put switch. A particular input to be able to connect without blocking under the worst set of conditions to a particular output will require (n — l) + (m— 1) + 1 available paths. Thus by providing for that many intermediary switches, a non-blocking switching array is obtained. The number of crosspoints is : C(3) = {n + m- 1)\n + M + —1 (21) Differentiating this equation first with respect to n and then to m yields two partial differential equations w^hose solution indicates that a minimum is reached when n = m. Replacing m by n in equation (21), the equation for the number of crosspoints becomes: C(3) = (2n - 1) [at + ilf + ^] (22) Solving for the minimum number of crosspoints gives the following expression : , ATM „ , NM „ ..-„, When N — M this equation reduces to equation (8). The three-way relationships of n, N and M are shown in Fig. 8. NON-BLOCKING SWITCHING SYSTEMS 421 5000 4000 3000 2000 1000 800 600 500 400 200 100 80 60 50 40 30 20 10 11 - \\\ - 1 1 \\\ V 1 \ \ \ \ \ \ 1 \ \ \ \ ^ \n = ,2 1 ^ 1 \ \ \ \ V V. 1 \ \ \V \^ \ \ ^ ^ \ \ V \ \ \ x'^ \ - - ^ — - \ \ \ y ^ ^ — -^ . ^ - \ > V \ s? \^ '^^^ — . ..^ \ \, X V \ V \ "<> ^ """~" ' — ' y V \. \ ■ — \ M V — n=2\ -^ v^ 1 1 1 —1- 1000 10 20 30 40 50 60 80 100 200 300 400 600 M= TOTAL OUTPUTS Fig. 8 — Relationship of n to iV^ inputs and M outputs for a minimum in cross- points in a three stage array. TRIANGULAR ARRAY If a case exists where all inputs are also the outputs, then an ar- rangement such as is shown in Fig. 9 can be used. The crosspoints in the intermediary switches permit connections between all switches on the left hand side. For connections between two trunks on the same switch it is assumed that one of the links to an intermediary switch can 422 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 be used to establish the connection but without affecting any of the crosspoints on the intermediary switch. The number of crosspoints for this case is: (24) ^ = (2--i)(^+£-£) where T = number of two-w^ay trunks. By differentiation, conditions for obtaining minimum numbers of crosspoints can be determined. The arrangement can also be extended to cases where extra switching stages are required, ONE-WAY INCOMING, ONE-WAY OUTGOING AND TWO-WAY TRUNKS A combination of the triangular array of Fig. 9 and of unequal in- puts and outputs is shown in Fig. 10. In this figure, one-way incoming, INPUT AND OUTPUT SWITCHES T = 24 TWO-WAY TRUNKS < ■) 5 C 5 C c ■> ^ p < p c p <: p c p ° ° ° c p I p c p p c p " <: [] p c ' p p ° ° ° c p c p c [ p c -o- n=3 o- INTERMEDIARY SWITCHES Fig. 9 — Triangular array. NON-BLOCKIXG SWITCHING SYSTEMS 423 one-way outgoing and two-way trunks can be freely interconnected without blocking. The number of crosspoints for this case is: -1 (25) The comments concerning the triangular array also apply for this case. COMPARISON WITH EXISTING NETAVORKS Few existing crossnet arrays are non-blocking. An example is the four- wire intertol] trunk concentrating system. In one of its standard sizes 4,000 crosspoints are required for 100 incoming trunks and 40 outgoing intertoll trunks. From Fig. 8, for A^ = 100 and M = 40 it may be noted that the nearest integral value for n is 5. By substituting this value in equation (22), a non-blocking three-stage switching array of 2,700 cross- INTERMEDIARY SWITCHES ONE-WAY OUTGOING TRUNKS ~" INCOMING TRUNKS T??^ ^ 1 1 '~^ 1 0 i J ' ° 1 1 1 ' ' ? Q r H ?^ 1 °~ 99999 — o *-" o o < ? ? 1 1 1 1 1 9 990 0 1 J o o c _j — o 1 o Y ^' s n * TWO-WAY TRUNKS ^ o o c o ^9?! ) A ^ — o ' — o ^ 1 ' ' 1 I O o o n n o o o o j c>4- A T-9 o- 1 ° ^ ^ 1 1 n 1 '^ o 1 1 n-3 o- 1 ^-^ y n J o- ^ M 1 1 T^^ Fig. 10 — One-way incoming, one-way outgoing and two-way trunks. 424 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 points is found which could be used for the concentrating switch. In this case the new approach to the switching network problem may prove to be of value. Comparisons with existing arrays having blocking are likely to be unfavorable because the grades of service are not the same. For in- stance, a No. 1 crossbar district-to-office layout of 1,000 district junctors and 1,000 trunks requires 80,000 crosspoints. This layout can handle 708 erlangs with a blocking loss of 0.0030. The minimum number of crosspoints with a non-blocking array is slightly less than 138,000. This, however, can handle 1,000 erlangs without blocking. By introducing blocking into the design methods described in this paper, a more favor- able comparison with existing arrays having blocking can be made. This can be done by omitting certain of the paths. If done to an array requiring 1,000 inputs and 1,000 outputs a layout can be obtained re- quiring 79,900 crosspoints with a blocking loss of 0.0022 for a load of 708 erlangs. For this example, at least, it appears that the new design methods may prove to be valuable especially for use in the development of electronic switching systems where the control mechanism may not be dependent upon the particular switching array used. CONCLUSION In present day commercial telephone systems the use of non-blocking switching networks is rare. This may be due to the large number of crosspoints required. With the design methods described herein, a wider use of non-blocking networks may occur in future developments. For the usual case of networks with blocking, new systems have generally been designed by an indirect process. Several types and sizes of switch- ing arrays are studied until the most economical one for a given level of blocking is found. With the new design methods, a straightforward approach is possible. Fig. 5 indicates that a region of minimum values exists. By first designing a non-blocking system with a reasonable number of switching stages and then omitting certain of the paths, the designer can arrive at a network with a given level of blocking and be very close to a minimum in crosspoints. The possibility of the adoption of this direct design method is important. ACKNOWLEDGEMENT In addition to those specifically mentioned in this paper, the author is also indebted to E. B. Ferrell, B. D. Holbrook, C. A. Lovell and E. F. Moore for suggestions and encouragement in the preparation of this paper. The Evaluation of Wood Preservatives Part II By REGINALD H. COLLEY (Manuscript received September 22, 1952) This paper offers a review and interpretation of laboratory and field experi- ments aimed at determining the necessary protective threshold quantities of wood preservatives. It details the procedure followed in the soil-block tests at Bell Telephone Laboratories, Incorporated. Discussion of specific criticisms of the techniques involved and replies to these criticisms are included. The paper also presents for the first time a correlation of the results obtained from soil-block culture tests, outdoor exposure tests on stakes and on pole-diameter posts as well as pole line experience. TABLE OF CONTENTS — Part II Evaluation by Treated ^ Inch Southern Pine Sapwood Stakes in Test Plots 427 Rating the Condition of the Stakes 427 Depreciation Curves for % Inch Stakes 431 Estimating Threshold Retentions and Average Life 434 Evaluation by Treated Pole-Diameter Posts in Test Plots 443 Evaluation by Pole Test Lines and by Line Experience; Service Tests. .... 449 Discussion 451 Density and Growth Rate 452 Size and Shape of the Test Blocks 452 Toluene as a Diluent for Creosote Treating Solutions 455 The Distribution of the Preservative in the Block 456 Heat Sterihzation of the Treated Blocks 456 The Weathering of Creosote and Creosoted Wood 457 General Considerations; Creosote Fractions 458 Creosote Losses 461 Creosote Losses from Treated Blocks 469 Creosote Losses from Impregnated Filter Paper 473 An Interpretation of Creosote Losses 474 The Gross Characteristics of the Residual Creosotes in Soil-Block Tests of Weathered Blocks 479 The Evaluation of Greensalt 487 The Evaluation of Pentachlorophenol 488 Swedish Creosote Evaluation Tests 489 425 426 THE BELL SYSTEM TECHNICAL JOUKNAL, MARCH 1953 >.i J H ^ P^ 41 IHI m^imm^^m^m EVALUATION OF WOOD PRESERVATIVES 427 Shortening the Bioassay Test 490 Toughness or Impact Tests for Determining Preservative Effectiveness. 491 Other Accelerated Bioassay Tests 493 Other Observations 494 Conclusions 497 Acknowledgments ■ 498 Bibliography 499 EVALUATION BY TREATED %-INCH SOUTHERN PINE SAPWOOD STAKES IN TEST PLOTS Rating the Condition of the Stakes One of the general and unavoidable difficulties in experiments involv- ing exposure of small specimens in test plots is arriving at a measure of the inspector's judgment of the condition of the individual specimens at each inspection period. Bell Laboratories' investigators have used a sys- tem of numbers beginning with 10 as the highest, and running down in single steps to 0, to define the various gradations of destruction shown in the specimens as they pass from perfectly sound to the state of "failed" units. This system has been considered by some as slightly cumbersome; but it is a truly effective method of depreciation rating in a continuous series of inspections. In such a series any minor errors of judgment in one season can be corrected in the next. Slow depreciation can be recorded in the upper ratings until progressive destruction becomes clearly evi- dent. Most observers of test plot experiments on treated wood specimens use a series of five numbers for five condition categories, about as follows : 10.0 Sound — no decay. 7.5 Surface soft — suspicious of decay. 5.0 Shght — positive decay. 2.5 Severe — deep decay. 0.0 Failed — almost complete loss of strength. Some, like Rennerfelt^^ for example, use this system upside down, with 0 for no decay, and 10 for failure. The writer has proposed^^ the use of a new 5-number depreciation system, with the same definitions, based on the logarithms of the above 5 figures, and rounded off to 10, 9, 7, 4 and 0 respectively. This simplifies the Bell Telephone Laboratories' 11 divi- sion, 10-0, system while retaining the advantage of slow depreciation at first; and at the same time it avoids the sudden, and in the writer's opinion, unjustified drop from 10 to 7.5 in the arithmetic series for suspicious-of -decay specimens. In the following presentation and inter- 428 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 pretation of the behavior of some of the ^^-inch stakes in the Gulfport test plot the recorded per cent condition of the stakes at any one inspec- tion period has been translated into terms of the proposed 5-number log base system. The stakes were carefully sawed and planed units, %-inch square in cross-section/^' ^^ Before treatment the stakes were selected so that they would represent the normal distribution of density in the material avail- able. Table IX shows the analyses of the four different creosotes used to treat the stakes in respective 4- and 8-pound groups. Both empty-cell and full-cell treatments were employed. The full-cell treatments were made with toluene as a diluent in order to provide more uniform and lower controlled retentions in the treated specimens. The empty cell specimens were sorted after treatment to retain the middle group of retentions, "svith a view to eliminating as far as practicable some of the factors in the empty-cell treatment variation. All of the stakes were treated between March 11 and March 26, 1941, and they were all placed in the plot in the approximate period from April 8 to April 22, 1941. The distribution of retentions in the 8-pound stakes set out in the Gulfport plot are shown in Table X, along with data on average retention, stand- ard deviation, and coefficient of variability. Table TX — Analyses, Water-Free Basis, of Four Creosotes Used in Treating %-inch Southern Pine Sapwood Stakes 1941 series; Gulfport test plot Creosote BTL No. 5283 S286B 5286A 5285A Snecific eravitv 38/15.5*0 . 1.055 1.053 1.068 1.111 Distillation, per cent, cumulative To 210''C 2.4 13.2 38.1 51.1 58.4 81.1 18.8 99.9 3.4 7.8 4.6 22.4 49.3 60.5 65.6 80.7 18.6 99.3 1.6 5.7 5.1 20.3 41.1 50.0 53.5 67.3 32.5 99.8 0.7 4.0 0.5 210-235 4.2 235-270 270-300 18.8 28.7 300-315 315-355 Residue Total 33.1 53.2 46.7 99.9 Suloh res. em/100 ml 0.7 Tar acids, gm/100 ml 4.0 Table X- — Distribution of Retentions,* lb/cu ft at Treatment, of Four Creosotes, 8 lb Empty-Cell (EC) AND Full-Cell (FC) Groups ^^-inch southern pine sap wood stakes; 1941 series; Gulfport test plot. Creosote, BTL No. 5283 5286B 5286A 5285A lb/cu ft EC FC EC FC EC FC EC FC 6.4 2 2 2 _ 6.5 — — — — — 2 — 6.6 6.7 6.8 — — 4 — 6 2 2 2 — — — — 2 — 4 6.9 — — — — — 2 — — 7.0 4 2 2 — — 2 — 7.1 — 1 — — — 2 2 7.2 6 3 8 — 4 4 — 2 7.3 — 2 — 2 — 2 — 6 7.4 6 4 4 — 2 4 6 2 7.5 — — — 4 — 2 2 7.6 4 2 4 — 8 6 2 3 7.7 — 2 — 2 — — — 3 7.8 — 4 -2 4 4 — 4 4 7.9 — 6 6 — — 7 8.0 2 3 2 — 4 — 6 2 8.1 — 1 — 4 — — 2 8.2 — 2 — 4 — 2 — 1 8.3 2 2 4 2 — 2 — 5 8.4 — — — 2 — — — 2 8.5 4 — — — 4 4 6 — 8.6 — 1 — — — — — 1 8.7 8.8 8.9 — 1 4 2 2 4 2 — — 2 2 2 2 2 6 9.1 2 — — 2 — — 2 9.3 9.4 9.5 9.6 9.7 9.8 12.5 8 1 4 2 — 2 — 1 1 1 2 — 2 — — — 2 — — — — n 40 44 42 40 42 40 40 50 Average Retention lb/cu ft 8.12 8.02 7.97 8.07 7.76 7.59 7.79 7.91 Standard Deviation 0.913 0.723 1.288 0.514 0.839 0.694 0.758 0.523 Coefficient of Variation 11.24 9.01 16.16 6.37 10.81 9.14 9.73 6.61 * All retentions were calculated from weights before and after treatment. The full cell (FC) treatments were made with toluene-creosote solutions. 429 430 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table XI — Inspection Ratings %-inch southern pine sap wood stakes in test 6 and 7 years; 1941 series; Gulfport test plot. Creosote BTL No. 5283 5286B 5286A 5285A 5285A Treat- ment n EC 40 FC 44 EC 42 FC 40 EC 42 FC 40 EC 40 FC 60 Average retention at treatment Ib/cu ft 8.12 8.02 7.97 8.07 7.76 7.59 7.79 7.91 Test period years Niunber and per cent of specimens rated 29 72.5 17 42.5 16 36.4 2 4.6 25 59.5 21.4 10 25.0 1 2.5 25 59.5 8 19.0 10 25.0 8 20.0 36 90.0 29 72.5 35 70.0 17 34.0 20.0 17 42.5 22 50.0 19 43.2 10 23.8 14 33.3 27 67.5 19 47.5 15 35.7 32 76.2 27 67.5 20 50.0 4 10.0 9 22.5 12 24.0 23 46.0 1 2.5 2 5.0 1 2.3 14 31.8 4 9.5 10 23.8 1 2.5 18 45.0 1 2.4 1 2.5 8 20.0 4 8.0 1 2.5 3 6.8 3 6.8 1 2.4 4 9.5 1 2.5 1 2.4 1 2.5 1 2.5 2 5.0 2 4.0 2 4.0 2 5.0 3 7.5 2 4.6 6 13.6 2 4.8 5 11.9 1 2.5 2 5.0 2 4.8 1 2.5 3 7.5 1 2.0 4 8.0 EVALUATION OF WOOD PRESERVATIVES 431 Depreciation Curves for Y^ Inch Stakes Under the conditions at Gulfport the depreciation curves for the cre- osotes — particularly the low residue oils — show an increased down- ward pitch at the 6th to 7th year of exposure. The relative proportion of the stakes rated respectively at 10, 9, 7, 4 and 0 at the 6- and 7-year inspections, are shown by number and per cent in Table XI. The change within the one year interval is particularly striking in the 10 and 9 columns. The rating and the distribution of retentions of creosote 5283, empty-cell treatment, at 6 and 7 years, respectively, are shown in Figs. 100 80 O 70 cc HI Q. 60 50 40 30 20 :!:!.-: tU :! illl I •-•-• — • — • — • EACH DOT REPRESENTS ONE SPECIMEN 3456 789 10 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 15 — Distribution of ratings in relation to fetention by weight at treatment; creosote No. 5283, ^-inch southern pine sapwood stakes; 1941 series, empty-cell treatment; six years exposure; Gulfport test plot. See text and companion Figs. 16, 17 and 18. 432 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 15 and 16; and similar data for the full cell toluene dilute treatments are shown in Figs. 17 and 18, respectively. In spite of the attempted care in selection of the specimens and in treatment of the empty eel] 8-pound group, the lot is heavily weighted (see Table X) toward the high reten- tion end, e.g., by the 12 stakes in the 8.9, 9.1 and 9.5 groups. The presence of these relatively heavily treated stakes may explain in part the position of the depreciation curves for the 8-pound treatments shown in Fig. 19. On the other hand the empty cell 8-pound (nominal) stakes treated with 100 I 1 . t 1 f. . .\ h » •-X s 80 z • , . 1 tr z - 60 l~ tu 0 <40 UJ 30 20 10 0 EACH DOT REPRESENTS ONE SPECIMEN 1 M-i- 1 4 5 6 7 8 9 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT 10 Fig. 16 — Distribution of ratings in relation to retention by weight at treatment; creosote No. 5283, empty-cell treatment, seven years exposure. EVALUATION OF WOOD PRESERVATIVES 433 creosote 5286B at 9.3, 9.7 and 12.5 pounds (see Table X) apparently have not operated to increase the average life of the group treated with this oil as much as appears to be the case in the group treated with creosote 5283. The difference in behavior at the 6-7 year interval of the stakes that were treated by a full cell process with treating solutions made by dissolving the creosote in toluene is even more marked than it is in the empty cell groups. The average per cent condition of the stakes over the 9-year test period up to 1950 is shown in Table XII. Data are included on the number of stakes in each lot, and on the average treatment retention in !• St *••< : — • — 90 80 z liJ O 70 cc z o §50 8 < 40 OJ 30 20 10 0 M 1. .if • ml J .1 . EACH DOT REPRESENTS ONE SPECIMEN .1 -:• : 34567 89 10 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 17 — Distribution of ratings in relation to retention by weight at treat- ment; creosote No. 5283, full-cell treatment (toluene dilution), six years exposure. 434 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 pounds per cubic foot for the respective groups. Two groups of stakes treated with greensalt K^^-^^ at 0.57 and 1.17 pounds per cubic foot, respectively, are included in the table for comparison. Depreciation curves for the 4- and 8-pound groups for the four creo- sotes are shown in Figs. 19, 20, 21 and 22. A depreciation curve for greensalt K specimens treated with 1.17 pound per cubic foot is included in Fig. 19. Estimating Threshold Retentions and Average Life In presenting the following discussion of a theoretical approach to the estimation of threshold retentions and average life no attempt has been 100 90 80 »- z tu •J 70 a 111 a. z - 60 Z o o Z 50 O O UJ O < 40 a: UJ 30 20 U »S**— ••! :.:-.» EACH DOT REPRESENTS ONE SPECIMEN 3456789 10 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 18 — Distribution of ratings in relation to retention by weight at treatment; creosote No. 6283, full-cell treatment (toluene dilution), seven years exposure. EVALUATION OF WOOD PRESERVATIVES 435 made to separate the possible effects of the attack by different fungi or combinations of such fungi. The basic data are the figures reported by the inspectors of the stakes; and data on the actual organisms involved are very difficult — if not impossible — to obtain at the time of inspec- tion. For the present purpose then, any differences in rate of decay by different organisms or in different parts of the test plot are all blanketed under the per cent condition averages. Table XII — Average per cent Condition of J^-inch Southern Pine Sap wood Stakes Treated with Four Creosotes, AND WITH GrEENSALT K 9 years in test; 1951 series; Gulfport test plot. Years in test Treatment n Average retention at treatment Ib/cu ft Creosote BTL No. 1 3 6 7 8 9 Aven ige per cent condition 5283 EC EC 40 40 4.38 8.12 100 100 97 99 74 92 50 86 32 74 13 55 FC FC 49 44 3.93 8.02 99 99 86 96 54 86 30 68 16 54 8 31 5286B EC EC 42 42 4.23 7.97 100 100 93 100 67 89 40 72 21 54 12 40 FC FC 40 40 4.03 8.07 100 100 86 97 56 89 29 77 10 47 1 32 5286A EC EC 42 42 3.93 7.76 100 100 96 100 76 97 57 88 42 82 28 72 FC FC 40 40 4.08 7.59 100 100 93 96 84 89 64 80 44 72 16 53 5285A EC EC 40 40 4.15 7.79 100 100 99 100 91 99 75 95 62 92 53 90 I FC FC 51 50 3.92 7.91 100 100 94 97 78 93 61 83 46 76 24 62 Greensalt K FC FC 99 100 I 0.57* 1.17 100 98 84 80 72 100 99 96 88 86 67 79 * Ib/cu ft of dry salt. too 90 80 70 60 §50 o z o O 40 UJ < m 30 20 10 «^ ■^^. ^ >-. "^ *s^ < :x ^;i . N M N, \ \ > t^ N ^. GREENSALT K \ AT 1.17 N \ \K \ \ \ \ \ \ . > ^ \ \ \ \ \ \ \ \ \ \eC at 8.12 \ \ \ \ > \ \ \ t 9 \ \ \ \ \ \ \FC AT 8.02 \ \ V \ K^ WEC AT 4.38 \ 1 FC AT 3.93 1 5 6 7 YEARS 8 9 Fig. 19 — ^Depreciation curves for ^-inch southern pine sapwood stakes treated with creosote, BTL No. 5283, empty-cell and full-cell (toluene dilution) processes, and with greensalt K; Gulf port test plot. See text, Tables XII-XIII, and com- panion Figs. 20, 21 and 22. 100 90 80 2 70 O a. UJ OL 60 50 O z o «-> 40 30 20 to "'"'*■•«. ^ ) ""^^ ^^^^ X 'x s^ > \ ( s\ \ N \ V" \ N \ \ V \ > \ ^\ \ \ \\ ^ y \\ \ ^ \\ N \\ \ , ...^ . ^ \ \ \ \ V \ \ \ \ \ \ 1 < K \ \ \ s \ \ ^ 1 \eC at 7.97 \ \ \ X \ \ y ^ \ \ A \ \ \ \ N \ FC AT 8.07 \ \ \ X \ X , y. ^ \ ^EC AT 4.23 1 1 X FC AT 4.03 V) 1 1 0 1 23456 78 910 11 12 YEARS Fig. 20 — Depreciation curves for ^-inch southern pine sapwood stakes treated with creosote. BTL No. 5286B. 436 70 60 ~^"* -^ ^ ^^^ -^ \1 ■~1 bl k \ V ^. \eC at 7.76 1 \ 1 ^ \ \ N \ \ > \ i 1 \fC at 7.59 \ \ \ \ \ \E:C at 3.93 > > \ \fC at 4.08 0 1 23456789 10 11 12 YEARS Fig. 21 — Depreciation curves for %-inch southern pine sapwood stakes treated with creosote. BTL 5286A. 90 80 p50 Q Z 8 40 ££ 30 1 r F^ i L "-. ^N, N \ ' \ ^EC Al ■ 7.79 \j \ \ M ) \ \ \ \ k ^ \ \ \ \ N AT 7.91 \ \ \ S \ ) \ \ C AT ^ .15 \ \ \ \ \ \ \ \ AT 3.< 32 20 0 1 2 3 4 5 6 7 8 9 10 It 12 13 YEARS Fig. 22 — Depreciation curves for ^-inch southern pine sapwood stakes treated with creosote, BTL-5285A. 437 438 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 The straight lines in Figs. 23, 24, 25 and 26 are drawn through the average points for per cent condition and retention of the respective 4- and 8-pound groups, and projected to intercept the 100 per cent con- dition hne. The empty-cell data are represented by the solid line and the full-cell data by dashes. The threshold concentrations necessary to pre- vent all decay are estimated from the intersection of the gradient lines and the 100 per cent condition line. This method assumes that the rela- tion of condition to treatment retention is linear at or near the threshold, providing the average condition points through which the lines are drawn are established by the logarithmic based rating system described. The method also resembles in a way the procedure of the Madison investi- gators who have used the intersection of straight lines drawn through operational losses and decay weight losses in estimating the threshold retention in creosoted blocks.^^ As in the case of the latter the method would probably be more precise if one had more points for average condition at average retention nearer the thresholds. At any event the system represented by Figs. 23-26 seems to be about the only one that indicates probable thresholds for these particular creosotes and these particular sets of data. Table XIII — Estimated Threshold Retention and Average Life ^-inch southern pine sap wood stakes; (see Tables X-XII); 1941 series; Gulfport test plot. Treatment Years in test Average life-years* Creosote No. 3 6 7 "4 lb" "8 lb' Estimate d thresholds Ib/cu ft* 5283 EC FC 9.7 9.6 9.7 9.8 9.6 11.4 7.0 6.1 9.2 8.2 5286B EC FC 8.0 9.2 9.8 9.4 11.3 10.0 6.7 6.2 8.3 7.9 5286A EC FC 7.7 12.2 8.3 Indet. 9.3 11.9 7.5 7.8 10.4 9.2 5286A EC FC 7.8 11.8 8.3 9.7 8.7 11.1 9.0 7.8 12.8 9.9 Greensalt K FC (0.67) FC (1.17) — 1.4 2.1 10.5 11.4 * See text for additional data on method used in estimating the threshold re- tentions at treatment and the average life figures. EVALUATION OF WOOD PRESERVATIVES 439 100 fiO 70 60 50 Q z o O 40 UJ o < S 30 i // f 3 —A — ^- ^^^^ ^ / /y / l^"^ -"'' ^-'' ^ / • / / GREENSALT fi^^ ^ * '-'9 y / r" ^^ / / ,'^CREOSOTES • / • / / • / r YEARS IN TEST 23456789 10 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT 12 Fig. 23— Theoretical lines for estimating threshold retention for creosote No. 5283, in ^-inch southern pine sapwood stakes, empty-cell and full-cell (toluene dilution) treatments, and for greensalt K; Gulf port test plot. See text, Tables XI and XIII, and companion Figs. 24, 25 and 26. The tendency for the 7-year Unes to fall off to the right shows the effect of increasing decay in the 8-pound group. In the higher residue creosotes the earlier decay of the toluene dilute 8-pound treated specimens — that is, specimens that were treated below the threshold retention — tends to pull the lines so far down as to spoil their usefulness as tools for estimating thresholds. Obviously the slopes of the Unes will be influenced by the depreciation rating of the 4-pound as well as the 8-pound groups. Furthermore it would appear that the utility of the specimens treated with toluene-creosote solutions for estimating thresholds does not extend much beyond the 6th year of exposure under conditions such as those prevaihng at the Gulf port plot. The estunated thresholds at the 3-, 6- and 7-year inspection periods, and the estimated average life values for the different groups are sum- marized in Table XIII. The average hfe is estimated from the intersec- tion of the depreciation curves and the 50 per cent condition Hues. There 440 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 90 80 5 70 U 50 Q Z o U 40 UJ O < ff, 30 YE IN EARS TEST 3 . .^ ^^^- ^^^ ' ^^ / / ^-'' p'"" --""* / ^^v ^/> ^ y ^ ^ y • / ;^' ^y / y /^ ' e/ / >^ / / V^ / -1/ y 01 23456789 10 n 12 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 24— Theoretical lines for estimating threshold retention for creosote No. 5286B; ^-inch southern pine sapwood stakes. S 90 u a m Q. 80 70 H 5 z o O 60 50 YEARS — ^ IN TEST D ^ -^ -—/' / P^ 6 IX^ :^-' ' — "V ,,-' ^-"' «^ X -^ /^^^ ,^ V-" '/^ ^ v^ /^ 40 0 t 23456789 10 11 12 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 25 — Theoretical lines for estimating threshold retention for creosote No. 5286 A; %-inch southern pine sapwood stakes. EVALUATION OF WOOD PRESERVATIVES 441 §90 a. UJ Q. 70 Q §60 UJ O a. 50 40 3 "-^ 2^'"" ^^ >-■ 6 ^^ '"1^ >^ 7V iy jC r-,. - YEA IN T EST 12 0 1 23456 789 10 1 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 26 — Theoretical lines for estimating threshold retention for creosote No. 5285A; %-inch southern pine sapwood stakes. will inevitably be some difference of opinion as to which level to use. In the case of %-inch treated stakes it is quite evident that the preservative is no longer functioning effectively if the stakes have reached a decay rating of 7 or less. In such small specimens it is questionable whether any purpose is served by allowing them to stay in the ground under the conditions at the Gulfport test plot until they practically fall over by being completely destroyed at the ground line. Anyone who has worked with small test plot specimens will appreciate the many difficulties in the way of establishing standard procedure for determining the "failed" point or the end point of specimen life. On somewhat larger stakes Rennerfelt^^ has used a strength testing appara- tus. To test the fitness of small poles in line some Associated Operating Companies have used a spring scale dynamometer which is slipped onto the base of a pike pole. In actual utility plant experience it is obvious that it is impossible to wait for the complete decay of the wood unit. Elaborate tables have been worked out as guides for pole line inspectors to let them know how far decay can go under given load conditions before a pole has to be removed from line. Generally speaking, such removal must occur at some period well in advance of the time that the specimen would have rotted clear through at the ground line. In the writer's opinion it might be preferable to estimate the average life for ^-inch stakes from the point where the depreciation curve passes downward through the 60 per cent condition line, leaving all the units in any given series in test until that time, except of course the stakes that may have actually rotted off earlier. Study of Table XIII indicates clearly that both threshold and average 442 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 life estimates for any given set of small specimens will vary, depending on the time at which the estimates are made. Taking the 6-year inspection data as perhaps representing the best figures from which to make thresh- old and average life estimates of this kind, it appears that in the case of all of the four creosotes something more than 8 pounds of creosote per cubic foot was necessary to protect the %-inch stake specimens against decay. There seems to be no material difference in the performance of creosotes 5283 and 5286B as far as the estimated thresholds are con- cerned. For both the empty-cell and full-cell treatments it appears to be somewhere in the neighborhood of 9.5 pounds or above. In the case of the higher residue oils 5286A and 5285A there appears to be a significant difference between ratings obtained from the empty-cell specimens and from the toluene dilute full-cell specimens. Estimates of average life from the 8-pound empty cell specimens appear to be significantly higher than estimates from the 8-pound full cell toluene-creosote specimens, except in the case of oil No. 5286B (Table XIII). The estimated thresholds for the full-cell toluene-creosote specimens lie within the same general mag- nitude as the retention at treatment thresholds found in the soil-block tests. (Cf. Tables XIII and XXXV). How far one is justified in comparing straight 8-pound empty-cell treatments and 8-pound full-cell toluene-creosote treatments in ^-inch stakes is still not clear. It certainly cannot be done without taking into account the much greater variability in the retentions in empty cell stakes and the different but difficult to describe variations in distribution of the creosote from outside to inside, particularly in the case of empty cell treatments with high residue oil. Unless the empty-cell stakes are carefully selected within limited variations from the average retention, it has been found that the empty-cell coefficient of variability for reten- tion in an 8-pound treatment, for example, may run as high as 35-40 per cent, against a coefficient of 8-10 per cent only for companion full-cell treatments with toluene-creosote solutions. The comparisons among the latter treatments appear to be more rational and fairer; and they may give a truer picture of effective threshold requirements. The analysis of the stake test data here presented is intended merely to illustrate one set of procedures that may be used in the interpretation of small stake tests. The four sets of data are part of a much larger group of results that are being worked up for publication. Among the latter there are numerous lots indicating that truly protective thresholds of creosote for %-inch stakes lie somewhere between 10 and 12 pounds per cubic foot, which is not far out of line with the estimates given above. Final analysis and publication will either confirm or modify such esti- mates. EVALUATION OF WOOD PRESERVATIVES 443 EVALUATION BY TREATED POLE-DIAMETER POSTS IN TEST PLOTS Reference was made in the introduction to the fact that Bell Telephone Laboratories' experience over the last quarter of a century in the evalua- tion of preservatives by the use of pole-diameter posts in the Gulfport test plot was reviewed by Lumsden' in April, 1952, before the American Wood-Preservers' Association. A supplementary analysis and interpreta- tion of some of the same evidence is attempted in the following para- graphs. Data for 185 of the posts, the time of placement and the number of years in test, and the general condition of the posts as of the 1950 inspec- tion, are shown in Table XIV. The data for the individual posts of these lid H Z ^ 11 1— < OJ cc in o o o o o SOUND » SLIGHT DECAY o o o • MEDIUM DECAY • FAILED 1- ^" O ^ 4 o o o Oo o« o o o o o o o CD ^ CC 7 OoO a o 0 8 oC^ 9 o o o Q. o o -l::M:i SLIGHT DECAY ^^^ MEDIUM DECAY POSTS GROUPED IN 2 POUND RETENTION INCREMENTS 12345 6789 10 II RETENTION BY EXTRACTION IN POUNDS PER CUBIC FOOT AT TREATMENT Fig. 28 — The relation of creosote retention in Ib/cu ft at treatment by toluene extraction, and the rated condition of pole -diameter test posts, by retention groups; 1950 inspection; Gulf port test plot. See text and Table XVI. eleven groups are shown graphically in Fig. 27. The relation of retention by extraction at the time of treatment to the sound, decaying and failed specimens is clearly evident. All of the failures and the very great ma- jority of the remaining decaying poles are below the 7-pound retention hne. Only two cases of medium decay occur above the 7.5 pound line. It should be borne in mind particularly at this point that these retention levels represent the retentions found by extraction in the whole cross sections of the posts as soon as practicable after the posts were treated. The significance of these over-all retentions and of the calculated reten- tions in the outer one-inch layer of the posts will be discussed later. Since the data were calculated in terms of oven-dry weight and volume of the extracted wood the values may be a little high. For Groups 5 to 10 inclusive data are available on the retention in zones, that is, in the outer quarter inch, the next quarter inch, the next half inch, the next one inch, and the remainder of the treated sapwood. These data were obtained by appropriate cutting of the samples into zones and pooling for extraction the parts that came from the same zones, 446 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 either in sectors cut from discs or in boring samples. A summary of the distribution of the retentions will be found in Table XV. If the overall retention data for post groups 1-11 are distributed in 2-pound retention lots the evidence falls into the categories represented by Table XVI and Fig. 28. The inference is clear. One would expect a satisfactory service life if the treatment retention by extraction, based on the whole cross section, was at or above 8 pounds per cubic foot. This is perhaps over-simplification. The variations shown in Tables XIV and XV in the behavior of the posts in Groups 7 and 9, and possibly to a certain extent in Group 8, disturb what looks like reasonably straight reasoning. In order to extend the reasoning and provide another method of interpretation, let it be assumed that the posts average 8 inches in di- ameter at the ground line, which is close to their actual size. Calculation of the pounds per cubic foot in the outer inch of such average posts is shown in Table XVII. The influence of posts in groups 7 and 9 on the averages is still evident. If there are differences in the efficiency of the Table XVI — Creosoted Southern Pine Test Posts Condition at 1950 inspection by retention groups; 14-19 years in test; Gulf- port test plot. n Condition of ground section, niunber and per cent Retention* Ib/cu ft at treatment Sound Decaying Failed Slight Medium 1. 10.0-11.9 10 100.0 10 100.0 0 0 0 2. 8.0-9.9 42 100.0 38 90.4 3 7.2 1 2.4 0 3. 6.0- 7.9 44 100.0 31 70.4 6 13.6 2 4.6 5 11.4 4. 4.0- 5.9 71 100.0 26 36.6 11 15.5 14 19.7 20 28.2 5. 2.0- 3.9 14 100.0 2 14.3 4 28.5 4 28.6 4 28.6 6. 0.0- 1.9 4 100.0 0 0 1 25.0 3 75; 0 Totlas 185 100.0 107 57.8 24 13.0 22 11.9 32 17.3 * The distribution into retention lots was made on the basis of both weight and extraction data, depending on information applicable to the eleven groups in Table XIV. EVALUATION OF WOOD PRESERVATIVES 447 creosotes in these groups, the differences are not at present real and tangible because they are masked by other factors, among which distri- bution of the preservative, and retention and penetration variables seem most important. The over-all indications are that in general one should insist on something more than 8, and probably more than 9, pounds of creosote per cubic foot, at the time of treatment, in the outer inch of the ground section of a southern pine pole. This, in simple terms, is in line with the conclusions reached about threshold retention requirements from the laboratory tests of creosoted J^-inch cubes and from test plot results on J^-inch stakes. It is much harder to rate treated test posts in terms of per cent condi- tion than it is to rate small specimens. Fig. 28 must be considered there- fore as a generalization from the plot inspection data. It must be inter- preted with the help of Fig. 27 and the retention by zones data in Tables XV and XVII. The reader who is at all familiar with pole line service records will recognize how much more detailed information there is for these test posts than there generally is for the ordinary pole hne. Yet it is practically impossible to draw up precise, unequivocal statements about the results in certain of the post series. The conclusions must be broad. The structure of the wood in a pole, the distribution of moisture con- tent, the variation in density of the wood from the outside to the inside annual rings, and the distribution of creosote from the outside toward the inside all go to make up a resultant that is obviously complicated. Some of these factors have been reduced to a schematic figure (Fig. 29) Table XVII — Creosoted Southern Pine Test Posts Relation between average treatment retention in outer inch of sapwood and condition of posts at the 1950 inspection; Gulf port test plot. Average retention at treatment in outer 1 inch of sapwood Group Sound Decaying Failed n Ib/cu ft n Ib/cu ft n Ib/cu ft 5-10 5 6 7 8 9 10 Totals 64 64 10.14* 10.14 2 3 4 5 3 7 24 7.13 6.09 12.11 8.71 9.38 5.95 8.03 1 2 9 9 21 11.61 7.90 9.15 5.95 7.59 * All calculations are made on the basis of an 8-inch diameter at the point of sampling. 448 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 1 , HEARTWOOD ■<- SAPWOOD >j 14 8,2 u. u to o UJ Q. to o § 8 1 1 1 \ "*" SOUND POSTS \ >' DECAYING POSTS / -FAILED POSTS .o-y / 0. z §6 1— z HI [- UJ cr 4 1- z UJ h- 0 ^ ___J5eN sivr ^ >»c >=< f^" X t a. \ / \ 1.0 0.4 2 3 4 3 2 DISTANCE FROM SURFACE IN INCHES 100 80 60 oc 40 3 h- (0 o 5 ^ Fig. 29 — Schematic diagram of a longitudinal section of a creosoted southern pine test post, showing average distribution of creosote in Ib/cu ft; average density (oven-dry weight and volume as tested) ; average moisture content, oven- dry wood basis; and the retention levels in the outer inch of sound, decaying and failed specimens. See text. representing a longitudinal section of an 8 inch diameter southern pine test post. The sound, decaying and failed retention levels calculated for the outer one inch (Table XVII) of such a diagrammatic post are also shown. When one considers all the variations in the wood itself and in the treatment, the site and environmental conditions where the pole is used, and the probability of the incidence of decay, the above general conclusions with respect to necessary retention seem reasonable and practicable. To refine and narrow the conclusions by repeating the post tests with the same creosotes — if one could get them — would certainly require the use of at least some posts with higher retentions and an obligatory assay of all of the treated specimens to be sure that the re- quired retentions were actually in the wood in the right places. Appar- ently practical answers to questions about such required retentions can be answered very much more quickly by repeated series of soil-block tests in which many of the variables can be controlled. EVALUATION OF WOOD PRESERVATIVES 449 EVALUATION BY POLE TEST LINES AND BY LINE EXPERIENCE; SERVICE TESTS In 1932 a condensed reporf^ of American Telephone and Telegraph Company experience with creosoted pole test lines, including brush, open-tank and pressure treatments, and covering northern (eastern) cedar, southern white cedar, chestnut and southern pine poles, ends with the following conclusions, among other generalizations: "Because of the fact that many of the test specimens are still in service, conclusions can be reached only in the case of some of the less important problems whose solution has been sought. The possibility of extending the life of poles through preservative treatment is abundantly demonstrated, but the capabilities of the more effective processes of treatment (studies) can as yet only be estimated. "... the indications are that in the cases not already affected, the beginning of decay attack will mainly be dependent upon changes in the quantity and the composition of the preservative retained in the individ- ual poles." This was only twenty years ago. Even in 1932 this service test report was more valuable as history than as a technical base for treatment specifications. The report reconfirmed the common knowledge that dur- able timbers like northern cedar and chestnut were better line units if properly butt-treated with creosote, and that adequate penetration and absorption of creosote were essential factors in the economy of pressure treatment of non-durable southern pine poles. However, by 1932 (a) The type of creosote used for treating the relatively large heart and small sapwood southern pine poles for the famous Washington-Nor- folk and Montgomery-New Orleans lines was no longer available com- mercially; (b) The type of virgin pine timber used was becoming scarcer and scarcer; (c) The supply of commercial chestnut pole timber was just about completely exhausted; (d) New and improved methods of butt treatment involving the use of machines for incising the ground section were being applied to northern cedar and to western cedar poles, and besides, the use of northern cedar was gradually shrinking to limited areas in telephone plant in the North- eastern and Lake States; and (e) Because of vastly increased competition in the pole treating in- dustry, and because the chestnut supply had failed, and because the excessive bleeding of creosote from the old style "12-pound" full-cell 450 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 southern pine poles — like those in the cited lines — made them unac- ceptable either as replacements for chestnut or for new construction in large sections of the northern and central telephone plant areas, and for basic reasons of economy, new specifications for creosoted southern pine poles were being issued. In preparing these specifications the Lab- oratories reduced the retention requirements to 8 pounds of creosote per cubic foot and provided for treatment by an empty cell process with a view to providing a clean pole, and set definite quality limits on penetra- tion to insure an economic service life. These moves were made after careful experiment, and after analysis of 8-pound treatment results. The experience in the pole test lines had an indirect effect rather than a determining effect on the proposed new penetration requirements. The Laboratories by 1932 had been operating the Gulf port test plot for 7 years and the Chester test plot for 3 years. Research and develop- ment programs on laboratory and test plot evaluation procedures ' ^^^ were well under way. The aim of this broad program was to determine the necessary requirements for preservatives, for retention, for treatment and for penetration before the poles were placed in line; which is quite different from the philosophy of depending on service tests or records to reveal at some later date what was done wrong in the first place. The reader should not get the idea that service tests can be dispensed with entirely. Material in service in the telephone plant is always under observation, casual or intensive, and the development of any obvious faults is corrected by the application of results of more and perhaps better laboratory experiments. In very many cases the faults are dis- covered in the laboratory or test plot before they are found in the field. However, one can never brush aside the insistent and determining effect of long and satisfactory field experience with treated wood. For example, when the behavior of creosoted poles in line is good the service tests take on what seems to be the outstanding characteristic of their present function, namely that of a comforting confirmation of previous conclusions. The results of some of the Laboratories' analyses of the relation between penetration and decay in creosoted poles were pub- lished in July, 1936, in this Journal^^ and again in 1939.^ These results confirmed the previous actions involved in specifying an empty cell treat- ment and a retention of 8 pounds of creosote per cubic foot. The poles were clean, they were being accepted generally for line construction, and the incidence of infection and probable early failure seemed to be about in line with anticipation. Emphasis in the last two papers cited was on penetration, which is easily determined, relatively speaking, and not on retention. The relation EVALUATION OF WOOD PRESERVATIVES 451 of low retention to decay showed up in the Gulf port plot/^ and it was explained by Laboratories' extraction and analysis of the creosote in the decaying or failed test stakes and test posts. One may say that such results confirmed suspicions; and they did, because creosoted poles in line — which were "related" to the decaying posts at Gulf port — were found on inspection to be behaving badly. It is believed that corrective measures in the way of supplemental ground line treatment were taken in time to give the poorly treated poles a reasonably satisfactory Hfe. In another striking set of circumstances, however, a number of cases of unprecedented premature failure of pine poles in line treated with a mixture of creosote and copper naphthenate petroleum revealed that the preservative solution had gone out of control and that in consequence the poles had not received the specified protective amounts of copper. Bell Telephone Laboratories' analyses of parts of the decaying poles showed that the decaying areas contained less than, and the sound areas more than, the Madison soil-block threshold of approximately 0.08 Ib/cu ft of copper as metal. These poles were immediately traced back to the supplier by their brand label. Present day pole treatment specifications in general require branding of each pole unit with symbols or code letters for the supplier's plant, the species of timber, the year of treatment, and the class and length of the pole. Each such pole becomes automatically a unit in a sort of universal service test in Bell System pole lines. Service tests and service records are a significant part of the over-all process of evaluating wood preservatives, but insistence on service rec- ords as the most important criterion simply perpetuates the reputation of established preservatives and forever blocks or seriously impedes the development of new and promising materials. In the very nature of the case the results of service tests can be stated in the form of broad generali- zations only. The truly technical approach must be made through better methods of measuring the effectiveness of preservative materials by accelerated field tests on a sufficiently large number of small stakes or by controlled laboratory experiments such as the soil-block tests. DISCUSSION There are a number of things in connection with the soil-block test procedure, its interpretation and the correlation of its results with the results of other evaluation methods that require further discussion. For example, questions have been raised vigorously on such matters as: (a) The effect of variation in growth rate and density of the wood; 452 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 (b) The size and shape of the test block ; (c) The use of toluene or any other diluent with creosote to get low retention; (d) The distribution of the preservative in the block; (e) The practice of heat sterilization of the creosoted blocks; (f) The whole philosophy of the weathering procedure; and (g) The methods of control assay, such as weight, creosote extraction, analysis of extracted creosote, and lime fusion chloride determinations as applied to pentachlorophenol. None of these questions can be simply brushed aside, in spite of the fact that some of the points raised have a little of the nature of quasi- technical road blocks that may temporarily slow the approach to any standard laboratory test for creosote. Experimental work now under way at Madison^^ and our own laboratories at Murray Hill will help answer some of the questions. The following discussion will at least explain the nature of the problems involved. Density and Growth Rate Data on the relation of block density and absorption at treatment have already been presented (Tables I, IV, V and VI). It is not yet evident that either density or growth rate has any material effect on the treatment retention thresholds for creosote when the tests are run on outdoor (or equally depleted) weathered blocks, and when the steps in the gradient retentions are properly spaced. When some present experi- ments are finished the matter may be resolved more conclusively. Size and Shape of the Test Blocks Some critics of the soil-block test have objected rather strenuously to the ^^-inch cube because the two transverse faces are so close to each other. The inference is that either (a) the preservative, and in this case creosote is usually meant, is lost too rapidly through the end grain of the wood or (b) that in the evaporation process, or in the course of the weathering procedures, the preservative may be concentrated on the transverse faces. The validity of these contentions is under investigation. However, the importance of block size and shape can be greatly exag- gerated. One need not assume, in order to plan for comparative tests of oil preservatives, that a block must or can be cut so that creosote will be lost from it at exactly the same rate or manner as creosote may be lost radially from the concentric annual rings of a post or pole. Block shape EVALUATION OF WOOD PRESERVATIVES 453 has varied almost as much as the test procedure employed. In a labora- tory working plan for a study of the "Efficiency of Various Wood Pre- servatives," with the subtitle ''The Efficiency of Various Wood Preserva- tives in Resisting the Attack of Fungi in Pure Cultures," signed by C. J. Humphrey, dated July 10, 1913, and approved by Howard F. Weiss, at that time director of the young U. S. Forest Products Laboratory, Humphrey proposed the use of eastern hemlock heartwood, cut into pieces measuring 1}^ x 1}^ x 2 inches, which were to be treated and quartered longitudinally "to reduce the size sufficiently to allow their intro- duction into the (Erlenmeyer) flask.'' (Writer's italics.) Untreated wood blocks were to be used as culture media, and thirty days after inoculation the test pieces were to be put into the flasks and shaken up with the inoculated blocks. The test fungus was to be Lentinus lepideus. The culture period was to be 9 months, with culture blocks and mycelium "kept in a condition moist and warm enough for active growth" throughout the test period. Coincidently, Humphrey^^ and his colleagues were developing the broad foundation for the Petri dish tox- icity test. The proposed wood block tests never reached a really satis- factory experimental level for treated wood, although they were em- ployed for comparative natural durability tests." Breazzano^"' ^^ in the same year experimented with blocks of beech wood cut from treated ties and measuring about 9 x 2 x 1 cm. As has been pointed out earlier in this paper he recommended blocks measuring 4 X 2 x 1 cm as a tentative standard, and later, in 1922, blocks 4x4x2 cm. were accepted as standard in Italy. ^ In the case of the latter the larger faces were to be transverse faces, cut across the grain. Howe^^ reports tests of small sticks (blocks) of southern pine, measur- ing % X % X 6 inches, that were treated with salt preservatives and later inserted in "8-inch sterilized test tubes containing about 10 cc of standard malt agar." He also used sets of four small sticks in Petri dishes, placing them on 10 cc of nutrient agar medium that covered the bottom of the dishes. The fungus used was called Fames annosus. To supplement these tests he mixed ground-up treated wood in different concentrations with agar media, and tested his mixes against this same Fames annosus and a number of other wood-destroying fungi. Howe and Curtin^^ were working on a broad plan aimed at correlating laboratory tests with test plot tests made, e.g., at Matawan, N. J,, and with experience in line. Snell^°^ argued that one might obtain good growth by placing thin blocks of wood in tubes with agar, but that the growth of the test fungus on the agar and on the wood might be influenced by diffusion of the 454 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 preservative into the agar medium. He proposed the use of thin plaques of wood, measuring 3^ x 3 x 3 inches. These could be soaked to any re- quired concentration of preservative and then tested by placing them over wet filter paper in Petri dishes. The test specimen would be sup- ported just above the wet filter paper on sterilized wood strips. Inocula- tion was to be by the simple process of placing a small square of agar plus the growing test fungus directly on the upper surface of the test ''block." Similarity to natural conditions and the control of moisture content of the wood were considered to be decided advantages for the method. Cislak (see discussion of Snell and Shipley^^^) used similar plaques of wood measuring 4x4 inches and about J^g-i^^ch thick for experiments on evaporation and permanency of creosote. Rhodes, Roche and Gillander^" used blocks measuring 3^ x J.^ x 3 inches. The European standard^^' ^^ block measures 5 x 2.5 x 1.5 cm, with the long axis in the direction of the grain. Schulze, Theden and Starfinger, in addition to the standard block, used ''half" blocks, i.e., blocks measuring 5 x 2.5 x 0.75 cm. All factors considered, they do not regard the thinner block as an advantage, and they have held to the standard size.^ ^^^ Lutz^^ in 1935 suggested the use of 2 x 2 x 5 cm. blocks, with the long sides dressed parallel to the fibers of the wood. He also used blocks measuring 1 x 1 x 5 cm. Alliot^ favored blocks measuring 5.0 x 1.0 x 0.5 cm. in the longitudinal, tangential and radial directions, respectively, for a French standard test. In his recent tests^^' ^^ Harrow has used IJ^ x ^^q x % inch blocks. Sedziak^"^ uses %-inch cubes cut from %-inch stakes after treatment. The National Wood Manufacturers' Association^"^ standard block size is 1.25 inches on the radial surface, 1.75 inches on the tangential surface and 0.25 inch thick, i.e., in the longitudinal direction of the grain. Various size blocks have been used at Madison in soil-block tests of natural durability, but two sizes only have been employed commonly since 1944 in the above mentioned soil-block tests and agar-block tests. The soil-block is the ^-inch cube, generally drilled with a 3^-inch hole in the center of a tangential face. The agar-block was cut with two broad transverse surfaces measuring ^xl}^ inches, and with a distance along the grain of only % inch; and it was not drilled.'*^ The Madison agar- block is basically the same sort of a block as the one described in the previous paragraph, and it resembles the Breazzano blocks as far as maximum transverse surface exposure is concerned. EVALUATION OF WOOD PRESERVATIVES 455 The list is incomplete, but it serves to illustrate one important point about laboratory evaluation tests, and that is the inevitable variation that creeps naturally into explorative research. In the writer's experience and opinion much of the variation in block size has been the result of a sort of forced adaptation, on the part of the investigator, to the size and shape of his laboratory glassware, coupled with certain practical prob- lems of block procurement and manufacture. Obviously it is easy to use thin sticks in test tubes, thin sticks or plaques in Petri dishes, and relatively flat blocks in Kolle flasks. The ^-inch cubes, which in essence, as has been stated before, are simply sections of the ^-inch stake, handle easily in the soil-block cultures. Criticism of the shape of the %-inch cube did not become pointed until after the publication of the first papers^^' ^^ on the Laboratories' cooperative work with the Madison laboratory. It is argued — as men- tioned before — that the evaporation of creosote from such blocks is unfairly rapid. However, the losses reported by Rhodes et al,^^ which will be discussed later, indicate that separation of the transverse faces will not prevent creosote evaporation. Assuming properly calculated gradient retentions, and the use of weathered blocks, it does not appear likely that the shape of the blocks — if kept constant within any given comparative test series — will affect the location of the treatment threshold retention. Toluene as a Diluent for Creosote Treating Solutions This subject is most controversial in this country. The use of acetone or chloroform has been widely accepted in Europe, but Schulze and Becker^'^'^ warn that the use of any diluent may change the rate of evaporation of given creosote fractions, and may affect the rate of evapo- ration of whole creosote. However, there are points in the debate which can be stressed, namely: (a) The treatment of test blocks to low and uniform retentions without a diluent is extremely difficult, if not practically impossible; (b) The rate of loss of creosote by evaporation, and any change in the character of that loss that may result from toluene dilution might be evident in freshly treated blocks but not in weathered blocks; and (c) The volatile fractions of undiluted creosote are lost fairly rapidly from small saplings^^^ and from test blocks; and it is assumed that the use of toluene does not cause loss in the fraction above 355°C. Discussion of Item (c) will be resumed in the section on weathering. In view of the perplexing character of this toluene dilution question steps are being taken to find out what happens. In the meantime the toluene diluent which has not been found to exert any measurable toxic effect in 456 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 the soil-block tests at Madison^^ must be used to secure gradient reten- tions down to and below the threshold level. The Distribution of the Preservative in the Block The opinion has been expressed by a number of investigators^^' ^^* ^^' ^°° that it is difficult to secure uniform distribution of the preservative in the test block, and that evaporation would cause concentration of certain preservatives on the block surface, where the toxic material might influ- ence the behavior of the blocks in the test cultures. If such a concentra- tion occurs the result would be to fix the threshold at lower over-all treatment retention than it would be if the preservative were not con- centrated, for example, at the transverse faces. One is likely to agree that uniform distribution of some creosotes might be difficult without the use of a diluent, particularly in the low retention groups. Rhodes et af ^ were apparently satisfied that they got a fairly good distribution with a 16.7- pound retention. Experience at Madison and at Murray Hill indicates that the blocks are saturated with the toluene-creosote solution or with the toluene-pen ta-petroleum solution; and this confirms the ideas of Schulze, Theden and Starfinger^ ^^^ on the way oven-dried test blocks take up the treating solutions. Data on the distribution of residual creosote in weathered blocks is presented in Table XXX, and discussed in the section on weathering. Preliminary assay of blocks treated with pen ta-petroleum, (a) just after treatment, (b) after weathering and (c) after testing show that the pentachlorophenol concentration is slightly lower at the transverse faces than it is in the middle section of the blocks. The same seems to hold true for the copper metal in blocks treated with copper naphthenate in toluene-petroleum. It should be pointed out here that preservatives are rather fortunately distributed in treated wood in such a way that the outer fibers or annual rings normally contain much higher concentrations of the toxic material than is found in the inner fibers. (See Tables XV and XVII and Fig. 29). Heat Sterilization of the Treated Blocks In pure culture test experiments some form of sterilization must be used to avoid contamination by other organisms than the test fungi. Most frequently such sterilization, for the minimum necessary time, is accomplished by flashing the treated block through a hot flame or by steaming at 100°C and atmospheric pressure. Either of these procedures EVALUATION OF WOOD PRESERVATIVES 457 might cause a measurable loss of volatile creosote fractions from freshly treated blocks; but such losses appear to be negligible in the case of weathered blocks. At Bell Telephone Laboratories control data on these possible losses are being determined by extraction of control blocks after the steriUzation phase. Such blocks are run through all the steps in the bioassay procedure up to planting in the soil-cultures. It mil be recog- nized by anyone familiar with pressure treating methods that the 100°C temperature, usually held for 15 minutes only, represents a much gentler set of conditions than the after-treatment steaming for several hours at 240-259 °F which is permitted in many specifications for creosoted poles. The Weathering of Creosote and Creosoted Wood Creosote is a remarkably good wood preservative, and nothing in the following paragraphs is intended to detract from its reputation in that respect. Sometimes the creosote oozes or bleeds from the surfaces of treated units such as poles and crossarms, especially on hot, sunny days; and when such bleeding occurs the treated material is unsatisfactory for use in mam^ parts of the telephone plant. In order to prevent bleeding difficulties and the consequent unhappy employee and public relations that result, the retention requirements have been held down to the com- mercial standard level of 8 pounds per cubic foot for southern pine poles, and the residue above 355°C limitations on the creosote have been kept in actual practice at 25 per cent or below except when war or post-war emergencies have interfered. The creosotes Ymy in the proportion of readily volatile materials they contain, and these materials are lost from creosoted wood — largely by evaporation — under many different use and exposure conditions. The general facts about such losses have been reported over and over again since the turn of the century. The significance of such losses is still not broadly understood or appreciated. Their possible bearing on the weath- ering procedure to be used in the soil-block tests is of fundamental im- portance. Paraphrasing the quotation from Schmitz^^ cited in an earlier paragraph: It is really necessary to know how much creosote to inject into wood to allow for loss by volatility and to insure a residual of the preservative, remaining in sufficient amount, to protect the wood for an economical service period. Bell Laboratories has been deeply concerned Avith the question whether it is practicable under commercial conditions to specify enough retention at treatment to provide the necessary pro- tective residual and at the same time require clean, satisfactory, treated poles and crossarms for Operating Company use. 458 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 General Considerations; Creosote Fractions Before presenting definite evidence of creosote losses this seems as good a place as any to refer briefly to investigations that have been aimed at discovering what components, volatile or relatively stable, give creo- sotes their properties of toxicity and permanence. Three articles by Martin,^^ Rhodes^^ and Mayfield^^ are the latest American papers cover- ing the general subject, the two former dealing with the technology of hydrocarbons and creosotes and creosote production and the latter re- viewing the results of tests of whole creosotes and creosote fractions. Teesdale's short report^^^ in 1911 is one of the earlier records in this country of experiments aimed at determining the loss of creosote frac- tions from treated wood. He used a creosote with 49 per cent distilling below 250°C and a residue above 320°C of 28 per cent, which would have been about 20 per cent at 355°C. This oil was fractionated into 5 parts, I, to 205°C; II, 205-250°C; III, 250-295°C; IV, 295-320°C; and the residue above 320°C. These five fractions and a sample of creosote with a similar distillation range were used to treat air-seasoned pieces of Pinus taeda, mostly sapwood, cut 2 feet long from 5 to 6-inch diameter peeled posts. The retentions were about 18 pounds for the numbered fractions, 15 pounds for the residue, and 21 pounds for the whole creosote. The treated pieces were open-piled in the laboratory for tw^o months, with temperatures running from 60 to 80°F. At the end of that period the per cent losses of the five fractions and the creosote were respectively 34.7, 21.3, 15.9, 6.2, 4.0 and 5.4. The results were in line with expecta- tions. The test period was short, and there were no outdoor weathering factors. Teesdale notes that the loss of the whole creosote was about the same as the losses in the two higher fractions. III and IV; and that a proportionately composited sample of the five fractions lost at the rate of fraction III, the total at the end of the two-month period being 15.8 per cent or about three times as great as for the whole creosote. Loseby and Krogh^^ reported in 1944 on outdoor weathering tests of creosoted wood blocks ; and they compared the weight losses in the blocks with evaporation losses from open Petri dishes. The creosote used was a relatively low residue oil produced at Pretoria. The residue above 355°C was 19.95 per cent, the specific gravity at 38/20°C was 1.088, and the amount distilling to 235°C was only 4 per cent. The test blocks were planed pieces of light weight Pinus insignis measuring 6 x 13^^ x IJ^ inches. The per cent moisture at the time of treatment was 10.3 per cent. The creosote was fractionated into four parts; Fraction I, up to 270°C; Fraction II, 270-315°C; Fraction III, 315-355°C; and. Fraction IV, the EVALUATION OF WOOD PRESERVATIVES 459 residue above 3o5°C. The blocks were very heavily treated by a hot and cold soaking process with the straight creosote and each of the four fractions to the following average retentions, respectively: 52.2, 45.6, 50.3, 50.2 and 23.7 pounds per cubic foot. These high retentions place the experiments out of line with most of the others cited in this paper; but the South African tests are unique in that they supply evidence on the rate of creosote losses from such high retentions. The relative order of losses of the materials, with the one having the highest losses first, was the same for the blocks that were hung on wire outdoors and for the open Petri dish samples, namely; Fraction I, II, whole creosote. Fraction III, and the residue. The latter actually showed no significant loss in either block or dish tests, and in the blocks there was a slight increase, possibly referable to oxidation, of a maximum of 2.2 per cent at the end of the three-year period. This gain was gone by the end of the 53^-year test. Fraction III was lost more rapidly from the blocks than from the Petri dishes; the reverse was markedly evident for Fractions I and II ; whereas the pattern for the whole creosote was very similar after the first year. The rounded figures for losses from blocks treated with the whole creosote, at the end of 1, 2, 3 and 53^ years out- door exposure, were 36, 42, 44 and 47 per cent ; and the losses from the open dishes for the same periods were 34, 40, 43 and 48 per cent respec- tively. One may conclude from these tests and conditions that a loss level of about 50 per cent would have been reached in about six years in blocks treated to a reported 50-pound per cubic foot retention with an undiluted creosote. ^lost of the laboratory toxicity tests on fractions have been run by the agar or the agar-block method. ^^ ^^^' ^^^' "' ^^^' ^^ The results obtained by Schulze and Becker^^^ are cited by Mayfield, who includes as his Fig. 1 a copy of the summary curves prepared by the Berlin investigators. The interested reader can profitably use the rather full excerpts of tabulated results of other investigators given in Mayfield 's paper as an introduction to the difficulties of testing creosotes and of interpreting the results of such tests. Some of the main controversial points are brought out by Peters, Krieg and Pflug in 1937^^ who challenge the results of Petri dish agar toxicity tests with results of the German agar-block tests — one of the first such broad comparisons to be made. Broekhuizen '' pubHshed his findings the same year in a comprehensive paper covering agar-block tests on creosotes and creosote relatives and creosote fractions. He dis- cusses his results with, different preservatives in relation to their toxic properties, their protective and preservative qualities, and the perma- nence of such quahties, and the bearing of these qualities on practical 460 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 wood preserving procedures. His discussion of the importance of ''weath- ering" tests and of his own results with such tests make up one of the best, if not the best consideration of this important phase of evaluation tests that had appeared up to that time. Numerous references to weathering techniques will be found in van Groenou, Rischen and van den Berge;"^ and van Groenou's own paper"^ in 1940 is an excellent short review of previous work, giving his views of the pros and cons of different procedures with emphasis on the essential nature of some test to determine what changes are likely to take place in a preservative as a result of leaching or evaporation. The f oUoAving examples will be interesting as illustrations of the differ- ent techniques that have been employed in testing the toxicity, or the potential preservative value of creosotes. F. H. Rhodes and Gardner^^ in 1930 determined evaporation losses of creosote fractions from thin pads of dried ground Sitka spruce pulp. The whole creosote and the fractions were introduced in an ether solution. The impregnated pads were placed on top of agar cultures in Petri dishes. The test fungus was the usual Fomes annosus, more recently called simply Madison 517. They state that: "It was found that under these conditions (the Petri dish covers loosely fitting — not water sealed) the more volatile preservatives vapor- ized from the test specimen, so that at the end of the month only a relatively small portion of the fungicide remained in the pulp." They were using a domestic creosote with a specific gravity of 1.065 at 38/15. 5°C and a residue above 355°C of 21 per cent. They tested fractions of the dead oil from which the tar acids and tar bases were removed, and fractions of the tar bases and the tar acids themselves. They determined percentage losses by evaporation for all lots by letting the treated pulp disks remain in covered Petri dishes for one month at 25°C. All of these reported losses for the dead oil occurred in fractions boiling below 316°C, for the tar acids in fractions with the same upper limit, and for the tar bases in the fractions boiling below 308°C. The losses varied inversely as the boiling range, the greater being in the low-boiling fractions, as was to be expected. The toxic limits for Fomes annosus that they determined showed a gradual increase as the boiling point increased, i.e., in agreement with other workers with agar tests they found the lower boiling fractions the most toxic. Their bibliog- raphy, with one or two exceptions, covers American articles only. Rhodes and Erickson,^^ continuing the same general technique, but substituting mechanical pine pulp for spruce pulp, showed that much EVALUATION OF WOOD PRESERVATIVES 461 higher quantities of respective creosote fractions were required to kill Fomes annosus in the pulp cultures than in Petri dish agar cultures. They concluded from their experiments that "no one compound in coal- tar creosote is primarily responsible for its preservative power. . . . The fractions from water-gas tar oil are much less effective as preservatives than are those from coal-tar creosote oil. The chlorine derivatives of phenols and creosote and of naphthalene are more toxic to fungi than are the compounds from which they are obtained." In 1933 Flerov and Popov^^ tested fractions of two creosotes by their soil-block method and compared the results Avith tests of the fractions emulsified in agar, following American Petri dish procedure in general. They found (English translation by Hildegard Kipp, Forest Products Laboratory) : "On wood the toxicity of the heavy fractions is considerably higher (compared to that of the lower fractions) and the most toxic fractions are those from 315 to 375°C. ... In tests on wood, in addition to the preservative effect, the effect of the evaporation factor, which is of great importance with oily preservatives, is determined T\dth (more or less) accuracy." In 1951 Finholt^^ revives the use of emulsion of the creosote fractions in the old Petri dish or flask method, like history repeating itself. The reader can be excused if he senses a degree of confusion. Creosote Losses Baechler's 1949 paper^ on the toxicity of oils before and after aging would be a fitting introduction to this section. The first task is to con- dense into simple statements or tables some of the available data on creosote losses from treated wood. Curtin^^ cites Bond on the latter's experimental determination in 1910-11 of creosote losses from thoroughly air-seasoned red oak and maple railroad ties with approximately the same moisture content. Bond reports that : 1. Full cell treated red oak in 200 days between November 1910, and June 1911, lost 19.0 per cent; and 2. Similar empty cell treated ties lost 52.7 per cent; and 3. Full cell treated maple ties in 105 days from March to June 1911, lost 13.4 P^f cen^; and 4. Similar empty cell treated ties lost 23.0 per cent of the creosote absorbed at treatment. The losses were determined by weight. He shows that the losses were 462 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 greater for the lighter treatments, and that in effect the losses were relatively greater for the empty cell than for the full cell treatments. His conclusions appear to hold good through all the subsequent cited data that admit of such comparisons. Bateman's early, 1912, work^ on oils extracted from two old piles that had been in service about thirty years indicated some considerable loss — in one case more than 35 per cent in the above water section — and relatively lower losses for water line and below water line sections. This confirmed a generally accepted common opinion. Losses below water line were apparently confined to the fraction distilling below 225°C in the case of the pile which he considers to have been treated with a pure coal-tar creosote. He calls light oils those distilling below 205°C. With the exception of the above water section of one pile all the samples of treated wood still contained about 17 pounds per cubic foot. He states: ^'The creosote in the pile which was perfectly preserved contained originally at least 40 per cent of naphthalene fractions, a large portion of which remained in the wood. The creosote in the pile which was less perfectly preserved contained little or no naphthalene." Service records on such oils resemble records from the Washington- Norfolk line and from the specimens examined by Alleman in that the data have little if any bearing — other than historical — on the creosote use problems of today. Schmitz et al^°^' ^°^ report a loss of 25.9 per cent after five years service in track in red oak ties treated with a 60/40 creosote-coal tar solution, compared to a loss of 17.3 per cent after three years service. The losses, determined by extraction, varied inversely as the boiling range, as was expected. There was very little loss in the 315-355°C fraction, and no loss is indicated for the fraction boiling above 355°. Bateman in 1922,^ and again in 1936 (see Discussion and ^°^) in con- nection with his explanation of the relation between the loss of the creosote fraction below 270°C and a formula for estimating the perma- nence or preservative life of creosote in treated wood, cites the earlier work of von Schrenk, Fulks and Kammerer, and Rhodes and Hosford on creosote losses from southern pine poles in the Washington-Norfolk and Montgomery-New Orelans lines of the American Telephone and Telegraph Company. Certain of the poles in question were installed in 1897 and removed in 1906 after about nine years service in line. The creosote used to treat these poles is reported to have had a specific grav- ity of 1.022-1.030 at 3°C above the melting point of the oil. The residue above 315°C was about 16 per cent; and the per cent naphthalene was "not less than 40 per cent." Using the pitch residue — the per cent EVALUATION OF WOOD PRESERVATIVES 463 boiling above 315°C — as a base for calculation, along with the change in residue determined from extracted creosote, von Schrenk, Fulks and Kammerer estimated the creosote losses, above and below ground line, for five poles from the Washington-Norfolk line, that are shown in Table XVIII . The poles were old growth longleaf pine, with a high heartwood volume. They were heavily treated — to about 16 pounds — by a full cell process. There is a wide variation in the results shown in the table Table XVIII — Creosote Losses, Based on Residue Increase, from Southern Pine Poles 9 years exposure; Washington-Norfolk line, American Telephone and Tele- graph Company — Data of von Schrenk et al. Pole No. Average loss, per cent Top Butt 1425 29 10749 2931 9700 Overall average 43.3 42.4 50.9 58.0 70.8 53.1 2.7 16.4 20.8 32.8 60.3 26.6 Table XIX — ^ Creosote Losses, by Extraction Southern pine test posts, aerial sections; 1926 and 1927 series; Gulf port test plot — BTL (Waterman) data. Exposure period, months 17 22 31 32 46 Average loss, per cent 1926 Series 8 lb empty cell 12 1b full cell "Light" oil "Mixed" oil 1927 Series "Light" oil 8 lb empty cell 12 lb full cell "Heavy" oil 8 lb emptv cell 12 1bfuircell 16 31 41 9 21 25 17 32 41 10 22 28 54 34 55 38 27 44 29 36 16 36 18 30 464 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table XX — -Creosote Losses from Southern Pine Posts Analyses of original creosote and creosote extracted from outer 1 inch below ground line of posts in test 3, 6 and 7 years; Gulfport test plot; 1925 series; 12 lb. full-cell treatments — BTL (Waterman) data. Specific gravity Distillation, water free basis, per cent, cumulative To210°C 210-235 235-270 270-300 300-315 31&-355 Residue Total Sulph. residue, gm/100 ml Tar acids, gm/100 ml Estimated losses, per cent, based on residue increase Original After 3 After 6 creosote years t years 1.037 1.056 1.057 (38/15.5) (60/60) (60/60) 4.1 0.5 0.3 32.0* 4.0 3.0 44.8 22.6 22.1 57.2 38.3 39.0 73.8 45.9 47.2 88.9* 72.2 73.9 10. 7t 26.4 25.2 99.7 98.6 99.1 3.2 4.2 3.7 8.3 3.3 2.8 59.5 57.5 After 7 years 1.059 (60/60) 0.1 3.4 26.3 43.0 50.7 75.8 24.1 99.9 2.4 6.5 55.6 * To 360°C in original oil analysis, t Above 360°C in original oil analysis. X The values for 3, 6 and 7 years are averages of data for 2, 3 and 3 posts, re- spectively. Table XXI — Creosote Losses from Southern Pine Posts Analyses of original creosote and of creosote extracted from outer 1 inch zone below ground line of posts in test 4, 5 and 6 years; Gulfport test plot; 1926 series; 8 lb. empty-cell treatments — BTL (Waterman) data. Specific gravity Distillation, water free basis, per cent, cumulative to210°C 210-235 235-270 270-300 300-315 315-355 Residue Total Sulph. res., gm/100 ml Tar acids, gm/100 ml Estimated losses, per cent, based on residue increase Original creosote After 4 yrs. After 5 yrs. 1.044 1.080 1.084 (38/15.5) (60/60) r60/60) 1.3 0.3 0.9 34.2 1.7 2.9 60.1 15.7 21.1 74.1 38.4 39.7 79.9 49.2 50.0 96.3* 79.7 77.6 3.4t 19.5 21.6 99.7 99.4 99.2 1.7 0.5 1.1 8.2 2.4 2.7 69.2 72.2 1.131 (60/60) 6.6 15.2 54.7 44.1 98.8 3.9 J6.4 * To 360*0 in original oil analysis, t Above 360 *C in original oil analysis. EVALUATION OF WOOD PRESERVATIVES 465 for the butt sections, but the top and butt figures, respectively, seem to bear some relation to each other. R. E. Waterman in a Bell Telephone Laboratories' memorandum dated January 23, 1928, reported losses of creosote from poles removed from the ]\Iontgomery-New Orleans line and in a memorandum dated March 7, 1931, reported creosote losses, determined by periodic extrac- tions, from the aerial sections of southern pine posts treated in 1926 and 1927. Companion posts are among the earliest lots reported on by Lums- den.^^ Part of Waterman's data are condensed in Table XIX. His figures confirm in general the conclusions reached by Bond,^^ namely, that the losses were greater for the light than for the hea\'y treatments, for the empty cell than for the full cell treatments, and in addition, for the lighter oil than for the heavier oil. Such conclusions are in line with what might be expected from the physical characteristics of the creosotes and general knowledge of the distribution and dispersion of the creosotes in the various treatments. The losses shown in Table XIX are rounded figures that apply to the whole cross section of the pole-diameter posts. Of more significance are data on creosote losses from the outer 1 inch of the helow ground section of companion posts in the Gulf port plot. Tables XX and XXI show distillation figures for the original creosotes and for the extracted oils, from full cell and from empty cell posts, after varying exposure periods up to about seven years. The oils were both low residue creosotes. The indicated percent losses are based on the increase in the residue above 35o°C — of which more later. The losses are greater for the empty cell treatments than for the full cell treatments. The fact that so much of the loss occurred within the first four years is extremely important in evalu- ation philosophy. Tables XXII and XXIII present data for whole cross sections of two posts that had begun to decay and that were removed for assay four years after installation at Gulf port. Table XXII shows the original analysis of the creosote and the average analysis of the extracted oil from the two posts. The indicated loss in the ground line decay ar6a, figured from the residue increase, was 61.1 per cent. Table XXIII shows the distribution of creosote at treatment by zones — from the outside toward the heartwood line — from extracted borings, and the distribu- tion of creosote after removal from test, based on extraction of sectors cut from whole cross section disks. The indicated losses, figured from average over-aU retention at treatment and after removal were 65.1 and 55.5 per cent, or an average of 60.3 per cent. This figure can be considered to be in agreement with the 61.1 per cent figure cited above. 466 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table XXII — Creosote Losses from Southern Pine Posts Analyses of original creosote and of creosote extracted from posts after 4 years in test; 1936 series; Gulfport test plot — BTL (Waterman) data. Specific gravity Distillation, water free, per cent, cumulative to210°C 210-235 235-270. 270-300. 3QO-315. 315-355. Residue . Total... Sulphonation, residue, gm/100 ml Tar acids, gm/100 ml Original creosote Avarage, exiraciea creosote* 1.055 1.134 (38/15. 5°C) (60/60) 3.4 0.2 17.0 0.5 40.4 2.1 53.3 7.5 59.6 15.0 81.2 51.5 18.3 47.0 99.5 98.5 4.7 3.7 7.9 12.5 * Average analyses of toluene extracted oils from disks cut adjacent to decay line; posts 273 and 280. The estimated average loss, based on residue increase, is 61.1 per cent. Table XXIII — Creosote Losses from Southern Pine Test Posts By zones, by toluene extraction; 1936 series; Gulfport test plot — BTL (Waterman) data. Post No. Years in Test Retention, by extraction Ib/cu ft Whole cross section Zones, inches Outer J^ Next H.Next yi Next Remainder of treated sapwood Original retentions, at treatment 273 280 — 3.8* 4.7* 6.9 11.2 3.5 4.1 3.8 3.3 3.9 1.5 3.0 3.5 Retention after removal 273 280 4 4 1.69t 1.64t§ 1.51 2.78 1.20 1.32 1.34 1.90 1.76 1.63 2.50 1.10 * Average analyses of boring samples. t Average analyses of sectors cut from a disk taken 3 inches above maximum decay line. X Average analyses of sectors cut from two disks taken 3 inches above and 3 inches below the maximum decay line. (See Table XXII for analyses of original and extracted creosote.) § Average loss, estimated from retentions at treatment and after removal: For whole cross section 60.3 per cent For outer 1 inch Post 273 70.3 per cent Post 280 64.0 per cent Average 67.2 per cent EVALUATION OF WOOD PRESERVATIVES 467 Calculated losses in the outer 1 inch of these two posts — • assuming 8-inch diameter — are 64.0 and 70.3 per cent, respectively, or an average of 67.2 per cent; and this figure corresponds very closely with the four- year loss figure of 69.2 per cent calculated for the empty cell posts in Table XXI. The posts represented in Tables XXII and XXIII were obviously treated to retentions that were too low to be effective; but they illustrate what is likely to happen when too low retentions of highly volatile light creosotes are used in wood in contact with the ground. Bateman discusses a laboratory experiment to determine creosote losses, over a 70-day period from pieces of round post sections, 5 inches in diameter and 2 feet long; and he extends his comparison to data derived from experiments conducted for the San Francisco Marine Piling Commission. He reports the following treatment data for the three creo- sotes involved: 1. 18 Ib/cu ft of a creosote with 92 per cent distilling below 275°C; 2. 10 Ib/cu ft of a creosote with 40 per cent distilling below 275°C ; and 3. 27.5 Ib/cu ft of a creosote with 42 per cent distilling below 275°C. His comparative loss data are condensed in Table XXIV. Waterman and Williams report creosote losses based on periodic extractions of comparable lots of specimens from treated round southern pine saplings exposed in the Gulf port test plot. Their data are condensed Table XXIV — Creosote Losses By weight, from round southern pine post sections*, and from pile sections- Madison (Bateman) data. Groupt Exposure period, days 18 Ib/cu ft 10 Ib/cu ft 27.5 Ib/cu ft Average loss, per cent 10 16.0 7.0 1.3 20 22.5 9.5 — 30 28.0 12.0 3.7 40 32.0 13.5 — 50 36.0 15.5 — 60 39.0 17.0 — 70 42.0 18.5 — 90 — — 7.3 222 — — 16.6 475 — — 24.5 510 — — . 25.3 785 — — 38.2 ■^ Five inch diameter posts, 2 feet long t See text for group and creosote data. i; 468 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Table XXV — Creosote Losses from Round Southern Pine Saplings B}' extraction; Gulf port test plot — BTL (Waterman and Williams) data. Years in test 1 2 3 n Av. loss, per cent n Av. loss, per cent n Av. loss, per cent Empty-cell treatments; 3.5-15.1 Ib/cu ft Above ground line Below ground line Full-cell treatments; 14.0-38 Ib/cu ft Above ground line Below ground line 14 14 9 9 35.5 40.4 22.7 13.9 10 10 7 7 55.8 58.8 44.2 39.2 3 3 4 4 55.9 61.6 59.8 42.4 Table XXVI — Analyses of Creosotes used in Weathering Wheel and Outdoor Exposure Tests Southern pine sapwood blocks — Koppers (Rhodes et al.) data. Creosote I Creosote II Specific gravity at 38/15. 5°C 1.064 3.8 21.2 41.4 59.8 80.9 18.7 0.6 1.039 1.106 1.081 Distillation, per cent, cumulative 0-210°C 1.8 210-235 13.7 235-270 30.2 270-315 .... 46.3 315-355 64.9 Residue above 355°C ... 34.9 Water 0.6 Specific gravity of fractions 235-315° 1.037 315-355° 1.104 Table XXVII — Creosote Losses from Southern Pine Sapwood Blocks Ether extraction; weathering wheel tests — Koppers (Rhodes et al.) data. Average loss,* per cent Exposure period, weeks Creosote I Creosote II 0 0.0 0.0 1 33.7 28.4 3 60.1 40.3 6 57.2 45.3 9 64.7 52.3 * Treatment retention 16.7 Ib/cu ft. EVALUATION OF WOOD PRESERVATIVES 469 in Table XXV. The loss figures are averages for all the creosotes used. It will be noted that there is a large variation in the treatment retention groups and that the losses were more rapid and definitely higher for the lower retention empty-cell specimens. Preliminary estimates from data on creosoted J^-inch square stakes at Gulf port indicate losses of about the same order of magnitude, with the trend in the direction of rela- tively higher figures than those for the round saplings. This is in line with expectation because of the use of the toluene diluent and the practice of controlling the treatments to secure lower than threshold retentions in both full-cell and empty-cell treatments. Creosote Losses from Treated Blocks E. 0. Rhodes and his colleagues have published two excellent pa- pers^°' ^^ that are most significant in a discussion of creosote losses from treated wood blocks. They used two creosotes, I and II, the analyses of which are shown in Table XXVI. Southern pine sap wood blocks meas- uring 0.5 X 0.5 X 3.0 inches, with the long axis in the direction of the grain, were treated and exposed on a weathering wheel in the laboratory, and also out of doors. The laboratory test specimens were treated to a retention of 16.7 pounds per cubic foot by soaking the blocks, heated to 105°C, in creosote at 100°C, in a sealed container. The creosote cooled during a five-hour soaking period to 40-50°C. The authors felt that the retention of about 16.5 pounds would facilitate extraction recovery of enough oil for analysis, and that ''the treated portion of a tie or pole probably contains about this amount of oil." Bell Telephone Labora- tories' experience has shown (Fig. 29) that the retention in the com- mercially treated 8-pound post averages only about 12 pounds per cubic foot in the outer J^ inch of wood and that the retention drops off rapidly in the wood farther beneath the surface. The Rhodes blocks, therefore, must be regarded as heavily treated. Losses of creosote were determined by ether extraction, and the change in character of the preservative was determined by distillation of the extracted oil. Corrections were made for resin extracted with the creo- sote. The losses of creosote from the test blocks on the weathering wheel are shown in condensed form in Table XXVII. The authors report creosote loss from blocks treated to a 15.0-pound retention and exposed outdoors during the winter as 44.4 per cent; and similar blocks treated to a 16.5-pound retention lost 47.1 per cent in a nine-week exposure period during the summer. The results of the experi- ments were taken to mean that the losses were of about the same charac- ter in the outdoor winter and summer exposure tests, and that the blocks k 470 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 weathered on the wheel in the same way they did outdoors. Rhodes reaffirms his conclusions about the weathering wheel tests in 1936 in a discussion of the Snell and Shipley paper^^^ in these words: "Our consideration of this problem convinced us, and Snell and Ship- ley agree mth the opinion, that natural weathering produced by heat and cold, rain and wind, involves not only evaporation but water leach- ing and mechanical losses of whole oil by water or by bleeding. ... To simulate these conditions, we exposed blocks of wood treated with the test creosotes to variations in temperature, to moving water and to moving air. ... In fact, we believe that our wood-block exposure tests include most, if not all, of the factors of natural weathering. ..." Now to go back a bit to Curtin's 1926^^ experiments, this time refer- ring to his own weathering tests on creosoted wood. Table XXVIII is an interpretation of the results he obtained by exposing small blocks, cut from pressure treated 2x4 inch southern pine sap wood stakes, to natural out-of-door weathering. His procedure was extremely severe, but he was aiming at an extreme accelerated test for permanency. After treatment his 2 X 4 inch sample pieces were held in storage under cover for 2 months; and then they were cut up so that the test blocks were about 2x}/^x% inches. These small pieces were exposed 15 feet above the ground on wooden trays for a four-month period from September, 1926, to January, 1927, and for a ten-month period from September, 1926, to July, 1927. Losses from the 2x4 pieces that must have occurred during their two- month storage period would increase the loss figures shown in the table. The losses are obviously greater for the lighter treatments, and for the lower residue oil. Table XXVIII — Creosote Losses prom Southern Pine Blocks* Outdoor weathering tests — Based on data by Curtin.32 Trt. No. Average retention Ib/cu ft Average loss, per cent Creosotet At treat- ment After 4 mos. After 10 mos. After 4 mos. After 10 mos. 1 1 1 2 1 2 3 1 23.18 17.13 19.38 26.19 19.69 11.69 6.50 18.75 17.13 10.88 5,44 15.00 15.1 31.8 37.4 28.4 26.1 36.5 47.6 42.7 * See text for description of blocks t Creosote number 1 was a domestic oil, specific gravity 1.056 and residue above 355°C of about 26 per cent; number 2 was a British oil, specific gravity, 1.068 and residue above 355°C of about 19 per cent. EVALUATION OF WOOD PRESERVATIVES 471 In their roof exposure, outdoor weathering tests Duncan and Rich- ards^^' ^^ have regularly found losses of creosote, by weight, in %-inch blocks treated with creosotes having residues above 355°C of 20-30 per cent, to run in the neighborhood of 45-50 per cent. All treatments were made by a full-cell vacuum process with toluene creosote solutions. The over-all exposure period consisted of three stages,^^ a three-week con- ditioning period in a constant humidity and temperature room at 30 per cent relative humidity and 80°F, a sixty-day outdoor exposure on a rack on the roof, and a three-week reconditioning to approximately constant weight in the same humidity room. The losses were figured from the original conditioned weight of the blocks, the creosote retention at treat- ment, and the calculated amount of creosote remaining as indicated by the weight of the weathered and reconditioned blocks. A condensed and simplified summary for two creosotes and two block shapes^^' ^^ is shown in Table XXIX. As might be expected under identical weathering conditions, the losses were higher in the case of the % x % x 13^ inch blocks — with twice as much end grain exposed as the %" drilled cubes — which presumably resulted in an accelerated longitudinal evaporation. In other words, a given percentage loss of creosote is arrived at sooner in the case of the block with greater transverse surface area. The loss increase amounts to about 7 per cent at the 8 to 10-pound retention level. This loss is about twice as great as it would be if calculated on the basis of the increased Table XXIX — • Creosote Losses from Longleaf Southern Pine Sapwood Blocks Outdoor weathering tests — Madison (Duncan and Richards) data. Creosote number Treatment retention Ib/cu ft Coop. No. 2 5340 5340 Creo sote loss, per cent, by weight 18* 44. Of -t -t 16 45.0 — 14 45.9 — — 12 47.0 — — 10 48.0 42.5 50.0 8 49.5 44.3 51.2 6 51.3 46.5 54.6 4 54.3 49.8 60.0 2 60.0 54.5 75.0 * Retention by full-cell treatment under vacuum with toluene-creosote solu- tions. t %-inch cubes. XH x^xl-}i inches. 1, 472 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 surface area only of the flat blocks. The losses are interpreted in general to mean that the exposure conditions were not as severe as those in the Rhodes^^ weathering wheel tests, and in line with his outdoor tests. The results of recent determination at Bell Telephone Laboratories of creosote losses from %-inch cube blocks, by toluene extraction, are shown in Table XXX. The blocks were divided into thirds before extraction (Fig. 30) and the respective parts were further divided and then pooled in the extractor. The blocks in each group of twenty were selected to represent the whole gradient of treatments with an average treatment retention for each group of 6.05 pounds per cubic foot. The average density of the blocks, oven-dry weight and volume basis, was the same for each group, namely 0.56. The oil used was cooperative creosote No. 11, a 50/50 blend of British vertical retort tar creosote and British coke oven tar creo- sote," diluted as usual with toluene. The exposure period outdoors was sixty days at the Chester Field Station between January 4 and March 3, 1952, plus several days exposure on a bench in a steam heated laboratory both before and after the outdoor period. The average loss of 40.3 per cent is considered to be in line with losses for the same creosote in the Madison tests, considering the factor of the winter chmate at Chester. The distribution of the residual creosote in the test blocks as sho^vn by the averages reported may be considered to be remarkably uniform. The differences in these averages are not regarded as statistically signifi- cant. Further discussion of losses from weathered blocks will be resumed in later paragraphs. Table XXX — Creosote Losses from Loblolly-Shortleaf Southern Pine Sapwood Blocks, and lb/cu ft Remaining By toluene extraction; outdoor winter weathering tests — BTL (Snoke and MacAllister) data. Lot No. n Average reten- tion at treat- ment, lb/cu ft lb/cu ft remaining Outer* third Middle third Outer third Whole block 1 1 20 20 6.05 6.05 3.62 3.49 3.43 3.49 3.93 3.68 3.66 3.55 Average lb/cu ft of creosote remaining; in outer thirds in middle thirds in whole blocks Average loss of creosote; 2.44 lb/cu ft, or 40.3 per cent. 3.68 3.56 3.61 ♦ ^-inch cubes. See cutting diagram in Fig. 30. EVALUATION OF WOOD PRESERVATIVES Creosote Losses from Impregnated Filter Paper 473 There have been numerous criticisms of the use of creosote loss figures obtained by evaporation from open dishes in any consideration of creo- 108 sote permanence. The use of the losses reported in this section may be criticized in a similar manner, but they represent extreme acceleration and they seem to have a bearing on the interpretation of any weathering Fig. 30 — Diagram of cutting plan for dividing a weathered creosoted block into three approximately equal parts for determination of residual creosote by toluene extraction. tests in which evaporation plays the major part. Some years ago the late Heinrich T. Boving of Bell Telephone Laboratories, ran an extensive series of evaporation experiments on eight so-called Fulweiler creosotes. He used impregnated strips of filter paper, hung on quartz springs in protecting glass apparatus and exposed to a constant flow of air with no turbulence, and under constant temperature and humidity. His ex- periments were performed with great care. Repetitions gave excellent agreement. A condensation of his loss figures, rounded off to whole num- bers, for seven- and fourteen-day exposure periods, are shown in Table XXXI. Let it be stated unequivocally at this point that it is recognized that there are important physical differences in the wood fiber combina- tions represented by filter paper, wood blocks, small round saplings, posts and poles ; but the quantitative loss data seem to fall sooner or later into a similar pattern for all these test media, under the various condi- 474 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 tions, and with the assumptions made. About the difference in the quali- tative changes that take place during creosote losses there is much, but not enough, information. An Interpretation of Creosote Losses Frosch,^^ in describing certain physical characteristics of the Fulvveiler oils, states that they may be considered as truly viscous solutions in Table XXXI — Creosote Losses, by Weight, from Impregnated Filter Paper, and Calculated Increase in Residue BTL (Boving) data Fulweiler Original residue above 335° C. per cent Per cent loss Calculated residue* creosote No. in 7 days in 14 days in 7 days in 14 days 1 43.84 31 63.5 2 40.25 32 36 59.2 62.9 3 31.11 36 — 48.6 — 4 28.52 38 — 46.0 — 5 20.85 44 — 37.2 — ■ 6 15.49 48 — 29.8 — 7 12.21 54 — 26.5 — 8 8.49 61 68 21.8 26.5 * Assuming that all loss occurs in fraction below 355°C. Table XXXII — Theoretical Changes in Creosote Loss of volatile fractions by evaporation; amount remaining of total fraction below 355°C. I 2 rime perioc 1* Time period 2 3 4 5 6 7 8 9 10 11 12 Calculated Calculated Ful- Assumed per cent More Less per cent More Less weiler treat- Per Resid- residue than than Per Resid- residue than than creo- ment re- cent ual oil above 355° 355° cent ual oil above 355° 355° sote tention loss Ib/cu ft 355°C in loss Ib/cu ft 355°C in No. Ib/cu ft residual residual oil Ib/cu ft oil Ib/cu ft 2 8.00 32 5.44 59.2 3.22 2.22 36 5.12 62.9 3.22 1.90 8 8.00 61 3.12 21.8 0.68 2.44 68 2.56 26.5 0.68 1.88 2 10.00 32 6.80 59.2 4.03 2.77 36 6.40 62.9 4.03 2.37 8 10.00 61 3.90 21.8 0.85 3.05 68 3.20 26.5 0.85 2.35 2 12.00 32 8.16 59.2 4.83 3.33 36 7.68 62.9 4.83 2.85 8 12.00 61 4.68 21.8 1.02 3.66 68 3.84 26.5 1.02 2.82 * Time periods 1 and 2 represent exposures that would result in the per cent losses determined from Boving's 7 and 14 day tests, respectively. See text. EVALUATION OF WOOD PRESERVATIVES 475 100 30 40 50 TOTAL CREOSOTE LOSS 60 70 N PER CENT Fig. 31 — Theoretical relation of the per cent increases in the residue above 355°C and the per cent losses of total creosote. which the fraction below 355°C is the solvent and the fraction above 355°C is the solute; and that this condition does not hold for any other temperature point. Let it be assumed that one may dodge the inferences of the results obtained by Hudson and Baechler^^ about the increase that may occur in the residue above 35o°C as a result of oxidation. Here one would be in agreement with Schmitz et al.^^^ ^'^^^ Let it be assumed further that all loss takes place in the fraction below 35o°C, and that the residue above 35o°C is inert, (cf. Loseby and Krogh^^). The residue in the eight creosotes used by Bo\dng would increase in their appropriate ratios to the figures sho^\Ti for seven and fourteen days in Table XXXI. Losses for creosotes of different initial residues and consequent residue increases can then be represented by a family of curves for residue increase with L 476 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 quantitative loss such as those shown in Fig. 31. Now let it be assumed that these hypotheses can be applied to treated wood, with full realiza- tion that the application may err in the direction of oversimplification. Bateman and Cislak accept the general principles of the 355 °C division point between volatile and nonvolatile creosote constituents in debating the theoretical aspects of creosote losses. ^^^ Rhodes^^ indicates some actual loss in the fraction above 355°C in his test blocks, possibly in part the result of oil displacement in the water phase of his tests, rather than any increase that might occur as a result of oxidation. ^^ For the purpose at hand in this paper it is convenient to use the relations shown in the percent loss — residue increase curves in Fig. 31. Table XXXII represents a theoretical approach to what would happen to the gross characteristics of the oils if, in some time period X, the losses from creosoted wood treated to 8-, 10- and 12-pound retentions became Table XXXIII — Creosote Losses* from Southern Pine Sap wood Blocks t Calculations of residual fractions below 355°C; weathering wheel tests. | 1 2 3 4 5 6 7 8 Creosote loss, per cent Creosote remaining By extraction By calculation! Exposure period, weeks per cent Ib/cu ft Residue above 355^C, per cent Residual fraction be- low 355°C Ib/cu ft Residue above 355°C, per cent Residual fraction be- low 355°C Ib/cu ft Creosote I 0 0.0 100.0 16.7011 19.0 13.53 19.0 13.53 1 33.7 66.3 11.07 25.2 8.28 28.7 7.89 3 50.1 49.9 8.33 34.6 5.45 38.1 5.16 5 57.2 42.8 7.15 40.2 4.28 44.4 3.97 9 64.7 35.3 5.90 48.7 3.03 53.8 2.73 1^1 66.5 33.5 5.59 56.3 2.44 56.7 2.42 Creosote II 0 0.0 100.0 16.7011 33.6 11.09 33.6 11.09 1 28.4 71.6 11.96 43.7 6.73 46.9 6.34 3 40.3 59.7 9.97 51.4 4.85 56.3 4.36 5 45.3 54.7 9.14 57.3 3.90 61.4 3.53 9 52.3 47.7 7.97 60.8 3.12 70.4 2.36 m 55.0 45.0 7.60 65.0 2.53 74.7 1.89 * Losses based on ether extraction. t Blocks 3^^ X J'^ X 3 inches, i See Bibliography, References 50 and 89. § Assuming that all loss occurs in the fraction below 355°C. Ij Retention by soaking in undiluted creosote. The 12-week figures were calculated from extrapolations of the loss curves. EVALUATION OF WOOD PRESERVATIVES 477 k (a) CREOSOTE I SFF TARI F XXVI I — 1^ — BY EXTRACTION — O— CALCULATED \\ \ ^ V ^ ^^^- =^ -A- -.V- 1^ i 10 8 6 4 i r' (b) CREOSOTE n SEE TABLE XXVI \ '-Cr— BY EXTRACTION ^C^ CALCULATED \ \ <;-. ^^^ 355°C Ib/cu ft <3S5°C Ib/cu ft 1 1 1.065 18.5 9.8 53.1 5.2 4.6 39.4 1.81 2.79 2 7 1.077 20.5 9.0 47.8 4.3 4.7 39.3 1.85 2.85 3 2 1.081 30.6 10.2 47.1 4.8 5.4 57.8 3.12 2.28 4 6 1.093 34.2 9.0 37.8 3.4 5.6 55.0 3.08 2.52 5 3 1.108 50.4 12.2 30.3 3.7 8.5 72.3 6.15 2.35 6 8 1.115 53.2 9.4 25.5 2.4 7.0 71.4 5.00 2.00 7 9a 21.2 5.7 40.4 2.3 3.4 35.6 1.21 2.19 8 9 1.001 20.0 5.8 43.1 2.5 3.3 35.1 1.16 2.14 9 10a 14.4 6.7 50.9 3.4 3.3 29.4 0.97 2.23 10 10 1.068 15.2 6.9 52.2 3.6 3.3 31.8 1.05 2.25 11 11 1.038 18.0 6.5 47.7 3.1 3.4 34.4 1.17 2.23 12 Ml 1.107 41.9 8.0 33.8 2.7 5.3 63.3 3.35 1.95 13 M2 1.070 18.1 8.3 50.6 4.2 4.1 36.6 1.51 2.59 14 BTL5340 1.088 20.9 7.6 46.6 3.5 4.0 39.1 1.57 2.43 * Creosotes 1, 2, 3, 6, 7, 8, 9, 10 and 11 are those in use in the Cooperative Creo- sote Tests (see Bibliography, References 12 and 39). Oils 9a and 10a are samples from the same lots as numbers 9 and 10. (See Bibliography, Reference 36.) For oils Ml and M2 see Bibliography, References 37 and 38. Creosote 5340 is shown in Table II. EVALUATION OF WOOD PRESERVATIVES 481 SEE TABLE XXXV , COLUMN 2 FOR NUMBER REFERENCES 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 TREATMENT RETENTION IN POUNDS PER CUBIC FOOT Fig. 33— Creosote losses in weathered, ^^-inch cube, southern pine sapwood test blocks; relation of per cent loss of total creosote to original residue above 355°C and retention at treatment, Ib/cu ft. All treatments were made with toluene creosote solutions. The total elapsed time from treatment to final reconditioned weight was about 105 days, including 60 days outdoor exposure on the Laboratory roof at Madison, Wis. See Table XXXV for number references. Table XXXV is a condensed set of data on fourteen lots of weathered, creosoted %-inch cube blocks. All of the tests were run at the Forest Products Laboratory at Madison, Wisconsin, in the Division of Forest Pathology, under the direct supervision of the same investigator. Dr. Catherine G. Duncan 36, 37, 39, 41 The test fungus was Lentinus lepideus, Mad. 534. Ten of the tests have been run in cooperation with Bell Tele- phone Laboratories, and four have been run more or less concurrently with other cooperators. The technique for handling the weathered blocks has been essentially the same, and it has been rigidly controlled, except for the vagaries of the weather itself, at all essential points. The data for the creosotes (Cols. 3 and 4), for the thresholds (Col. 5), and for the amount of residual oil in the blocks at the time they were placed in test (Col. 8), and the per cent and amount lost (Cols. 6 and 7) are all taken from the published reports or from manuscripts either ready^^ or in preparation for publication.'*^ The writer has calculated the residues in the residual oils, and the respective amounts remaining above and below 35o°C (Cols. 9, 10 and 11) in pounds per cubic foot, on the assumption that all the loss occurred in the fractions boiling below 355°C. Particular attention is directed to the figures in Col. 11 — the calculated amounts remaining of the fractions boiling below 355°C. In terms of 482 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 • -^ 1 o t» 0'-< O Sh oj O CO . ^ O 03 ^ cj O &0 ^' (D ^ 3^ o CO 2 Z 1- ^C3C3 O bC 02 r- o ILI m .- 0) -tJ u aga eath frac (£) reosotes during w ining of if) een c osses rema OO>00r^«>in^n(\J lOOd OianD b3d SQNnOd NI NOIlN313d lN3lMV3ai •- a, ook^ « a2r*S EVALUATION OF WOOD PRESERVATIVES 483 pounds per cubic foot the over-all picture of the relative threshold amounts of creosote at the time of treatment, the oil lost, and the calcu- lated proportional parts of the residual oil — after weathermg — above and below 3o5°C, are sho^vn graphically in Fig. 34. All of the data are in terms of pounds per cubic foot. If one bears in mind that the thresh- olds (Col. 5, Table XXXV) as given by the Madison investigators were located by a combination of visual observation of the blocks and extra- polation of straight lines through the weight loss data one may conclude that in all of these tests the results were essentially the same for all the 3/4 INCH BLOCKS 14 SOIL- BLOCK TESTS LENTINUS LEPIDEUS (MADISON) 13 • 12* • FULL CELL ▲ EMPTY CELL SEE TABLE Xyvv ^ COLUMN 1 FOR NUMBER REFERENCES 5« 2* 3/4 INCH STAKES 1 5283 A 1 • 1 5286 B • ▲ STAKE GULF (B- TESTS PORT 5286A ▲ ( 1 5285A ▲ • • OUTER INCH OF POSTS POST TEST GULFPORT (BTL) 1 ▲ 0 2 4 6 8 10 12 14 16 THRESHOLDS IN POUNDS PER CUBIC FOOT Fig. 35 — Relative values of creosote thresholds by soil-block tests, ^-inch stake tests, and post tests. I 484 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 creosotes with respect to the indicated threshold amounts distilling helow 35 5° C. Statistically, the figures in Table XXXV Col. 11, are not signi- ficantly different. Slight changes in estimating the thresholds would con- ceivably bring them all to approximately the same level. In any given set of experiments the level would also vary with the duration (and types) of a different weathering cycle. It will be recognized that no attempt has been made to separate or define the gross components of the fraction remaining below 355°C. This calls for more study, and for the development of refined methods of extraction and assay by weight and by distillation. Also, no attempt has been made to interpret the value or significance of the residue above 355°C, either because of its possible retardation of the evaporation of lower fractions, or because of some potential mechanical blocking effect. The whole interpretation is based on the simple division of the creosotes into two parts, the part distilling below 355°C and the part distilling above 355 °C. Refinement will depend upon better future experimental evidence. As an illustration of the application of the hypotheses discussed in the preceding paragraphs, one may reexamine the data from the weath- ering wheel experiments.^^ Rhodes, Roche and Gillander used one creo- sote retention only in their weathering wheel experiments, namely 16.7 pounds per cubic foot. In commenting on their work C. S. Reeve^' ^' ^^~^^ noted Rhodes' emphasis on "the fact that toxicity without permanence is just as worthless as permanence without toxicity". Reeve and his colleagues carried out somewhat similar weathering experiments using slabs of wood about J^ inch thick, exposing the treated pieces to some- what lower temperatures than those in the Rhodes' experiments and ''following a procedure with circulated air, and heat, and water". The plan of the experiments called for the conduct of "weathering cycles with reduced increments of various oils in order to get down finally to a percent of impregnation right at the end of a weathering cycle which would actually yield a rotting specimen of Lentinus lepideus^'. The work had not progressed far enough to accomplish this end, but Reeve says "The results . . . are in very close corroboration of what Mr. Rhodes has found. In other words, our loss curves with different oils running from relatively low residues to relatively high residues, have been almost parallel, I believe, with the loss curves which he has shown ..." Rhodes et al used essentially an agar-block method for testing their weathered blocks against Lentinus lepideus, Mad. 534, the same strain as that used in the Madison tests. The residual creosotes in the blocks treated with Oil I and Oil II (Table XXVI) are calculated to have been EVALUATION OF WOOD PRESERVATIVES 485 5.90 and 7.97 pounds per cubic foot at the end of the nine-week weath- ering cycle (Table XXXIII); and by extraction these residuals con- tained 3.03 and 3.12 pounds per cubic foot of the fractions distilling below 355°C. On the basis of residue change alone these amounts are calculated at 2.73 and 2.36 pounds respectively. Rhodes states^^ ^' ^^ "In no case was a weathered specimen attacked by the fungus". Para- phrasing his next sentence, this proved that both Creosote I and Creo- sote II at a treatment retention of 16.7 pounds per cubic foot "were affording adequate protection at the end of nine weeks, equivalent to many years of actual service". Was there any reason to expect such specimens to decay? It is easier to attempt an answer to that question now than it was in 1934. Duncan^^ shows a treatment threshold retention for "conditioned", i.e., unweathered, blocks of 1.6 pounds per cubic foot by agar-block tests, which is in close agreement with an average of 1.56 calculated from recent European tests reported by Schulze, Theden and Star- finger. ^^^^ Of more importance for the question at hand is Duncan's weathered agar-block creosote threshold, given as 5.0 pounds per cubic foot. The loss in creosote at this 5 pound level was 57 per cent, which left 2.2 pounds per cubic foot of residual creosote in the blocks. The residue is calculated to have risen, as a result of weathering losses of the lower fractions, from an original 20.9 per cent (Table II) to 48.5 per cent; and on the basis of this figure the 2.2 pounds of residual oil consisted of 1 .07 pounds per cubic foot of the fraction above 355°C and 1.13 pounds of the fraction below 355°C. The Rhodes' nine-week weathered blocks still contained roughly two to three times this amount below 355°C. One may conclude that the nine weeks weathered blocks should not have shown decay under the culture conditions; and certainly none of the more briefly weathered blocks should have shown evidence of attack. If original retentions of 10 and 8 pounds of creosote had been used in the experiments the test fungus might have attacked nine and twelve weeks weathered 8-pound blocks treated with either Creosote I or Creosote II (Table XXXIV). Using the results of the Madison tests shown in Table XXXV and Fig. 34 as indices of what might have happened if Rhodes, Roche and Gillander had used 16.7, 10.0 and 8.0 pounds at treatment and if — instead of employing an approximate agar-block technique — they had run their evaluation tests by the soil-block method, one would have expected decay to show up as indicated by the horizontal lines in the data columns in Table XXXIV. In other words, in the 16.7 pound treat- ments Lentinus lepideus would probably have attacked the blocks if they 486 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 had been weathered twelve weeks; and the 10.0 blocks would have been attacked at the end of five weeks w^eathering; while the 8-pound blocks would have been attacked after three weeks and about two weeks in the case of Creosote I and II, respectively. Under all these assumptions the treatment threshold for creosote in these block tests would probably have been set at above 8 pounds, and possibly around 9 pounds or more per cubic foot, in 1934; which would agree very well with the evidence obtained from soil-block tests, ?^-inch stake tests, and test posts that has been presented in this paper. Older records show very substantial quantities of the fraction below 355°C remaining in well treated wood after long service. Alleman's 1907 paper^ is a classic. Writing of the increasing use of creosoted wood he states: "Recent reports . . . have clearly shown that, while proper treat- ment gives remarkably good results, much of this timber was not prop- erly treated and has not lasted as it should". All of which in his opinion *^ . . makes it imperative that we should know, as completely as pos- sible, just what constitutes efficient creosote treatment. The different sorts of oils are believed to have different preservative values w^hen mjected into timber, but there is, unfortunately, a lack of uniformity in opinion". AUeman chose, as the best method for finding some of the answers, an extraction of oils from treated timber that had given good service. As solvents he used absolute alcohol and subsequently anhydrous ben- zene. He fractionated the extracted creosotes with a view to determining the character of the oils, deciding to make his cuts so as to collect the distillate as follows: I to 170°C; II 170-205°C; III 205-245°C, which he regarded as the naphthalene fraction; IV 245-270^C; V 270-320°C; VI 320-420°C; and VII the residue above 420°C. The wood from which he extracted the creosotes consisted of ties, mostly British, piles from England and the United States, paving blocks, and a section of creosoted wood duct removed in perfect condition after fourteen years service in Bell Telephone plant in Philadelphia. The EngUsh piles had been in service forty-three years, the other old samples all averaged a little over twenty years. Alleman's extractions showed — • after all these years in use — that there were on the average over 9 pounds per cubic foot of oil remaining in the ties and English piles; nearly 9 pounds remaining in the conduit; and about 16 pounds remain- ing in the American piles and the paving blocks. The writer has calculated the residue above 355°C in these extracted oils to have varied from about 23 to about 42 per cent ; and the pounds per cubic foot of oil distilling below 355°C remaining in the treated wood EVALUATION OF WOOD PRESERVATIVES 487 ran from a low of about 5.5 pounds to about 11.0 pounds. The wood had apparently remained sound. Alleman cites the difficulties of arriving at precise judgments but concludes ''that 10 pounds of creosote per cubic foot is ample for railroad ties, and that piles require from 10 to 20 pounds" according to location. Alleman 's discussion of the relative amounts of light and heavy oil that might be desirable are not applicable to present day oils and commercial conditions. His perplexities remain — in almost identical form or in mod- ernized version — and his extraction results are a long way from those reported by Lumsden/^ and those cited elsewhere in this paper. Bre- azzano was beginning the application of biological tests in Italy^^ with a view to correlating chemical and fungicidal characteristics of preserva- tives. This type of endeavor was later to be pressed vigorously by Bate- man,^ whose work has already been mentioned. If the interpretation offered is supported by additional experiments already under way the Madison data in Table XXXV and Figs. 34 and 35 mil be recognized as representing one of the most consistent series of laboratory tests for the evaluation of creosotes that has ever been run. The Evaluation of Greensalt The satisfactory performance of posts and poles treated Avith green - salt' ' has been reported in Lumsden's paper.' So far the very satis- factory results at the Gulfport test plot are accurate indices of what has been found by examination of poles in line. The only data on greensalt treated J-^-inch stake tests reported in this paper are shown in Table XII and Figs. 19 and 23. Summaries of additional stake data are now in preparation for publication. The indications are that the threshold for greensalt under Gulfport conditions is 1.42-2.1 pounds per cubic foot for J^-inch stakes. The average life estimates for the two treatment groups — 0.57 and 1.17 pounds of dry salt per cubic foot — ■ (Table XIII) compare most favor- ably with the estimates for the four creosotes in the same table. The number of greensalt specimens is large enough to warrant the conclusion that the lines used for estimating the threshold in Fig. 23, in their trend to fall off to the right soon after the sixth year of exposure, indicate that particular period as a critical .one for comparisons and interpretations. Commercially treated southern pine poles meeting the standard specifi- cation requirements for retention — 1 pound of dry salt per cubic foot — have about 2.0 pounds of dry salt in the outer inch. The agreement be- tween the stake and post tests seems good. 488 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 In soil-block tests^°* ^^ Porta incrassata, Porta monticola and Lenzites trabea have been resistant to greensalt K, whereas Lentinus lepideus is very susceptible. The writer interprets the results of these three types of evaluation procedure, by soil-block, by %-inch stakes, and by posts or poles to mean that: 1. The conditions for decay, as far as Lenzites trabea are concerned, are much more favorable in the soil-block culture bottle than they are in the above ground part of a pole under normal outdoor exposure con- ditions; 2. The incidence of attack or infection at the ground line by Lenzites trabea, Poria incrassata and Poria monticola at Gulfport is relatively rare; and that 3. The success of greensalt K in southern pine poles may be attributed in large part to the susceptibility of the ubiquitous Lentinus lepideus to the combination of salts in the greensalt preservative. Incidentally one may cite the following example of confirming results in tests of another salt preservative. Harrow's ^^ experiments with soil- block tests resulted in locating a threshold for zinc chloride on unweath- ered blocks at 0.28 pounds per cubic foot (at treatment) for Lenzites trabea, Madison 617 and 0.53 for Poria vaporaria. Richards and Addoms^^ found approximately 0.25 for Madison 617, and approximately 0.50 for Poria monticola, Madison 698. These two Porias are possibly the same species. The similarity in the thresholds appears to be the definite result of following the same technique, rather than a haphazard coincidence. The Evaluation of Pentachlorophenol The highly toxic properties of pentachlorophenol have been estab- Ushed by exhaustive Petri dish agar toximetric tests. ^^ A 5 per cent solution of penta in a light petroleum solvent is the preservative of reference in the recommended standard test for evaluating oil-soluble wood preservatives of the National Wood Manufacturers Association. ^^^ This test, as pointed out previously, is an agar-block test. Duncan re- ports*^ that threshold determinations based on soil-block tests of a 5 per cent solution of penta in petroleum (cf . Tables V and VI) against Len- zites trabea, a critical fungus for this preservative, have not varied more than zhO.2 pounds from 4.8 pounds per cubic foot in 7 series of weath- ered block experiments over a five-year period. Recent Bell Telephone Laboratories' soil-block tests confirm this result, with the same fungus and the same petroleum carrier. These results are confirmed at the Laboratories' Gulfport test plot EVALUATION OF WOOD PRESERVATIVES 489 (unpublished data). In two separate series — 1937 and 1938 — %-inch stakes were treated at retentions slightly below the threshold cited by Duncan^^ with o per cent solutions of pentachlorophenol in light petro- leum (gas oil) and with two coal tar creosotes. The performance after six and seven years was approximately the same for both the penta solutions and the creosotes. However, completely favorable results on test posts have been reported for Gulfport^^ and Saucier, Miss., test plots. Early tests on 2 x 4 inch stakes are now being critically examined, and more tests are in progress. ^^ Penta treated posts are installed in the Saucier plot, where they are under periodic observation and comparison along with posts treated with the cooperative creosotes. ^^ All of these experiments will greatly facilitate correlation of the results of different test methods. As far as pole line tests are concerned one can only echo the report^^ that up to this time not one of the tens of thousands of poles in Une that were treated with either straight penta-petroleum or with mixtures of penta-petroleum and creosote have been reported as failing because of decay. Swedish Creosote Evaluation Tests Rennerf elt and Starkenberg^^ report that of fourteen stakes measuring 1.5 X 1.5 X 100 cm, that were cut from the middle and inner sapwood of creosoted Scotch pine poles, none are sound after ten years in the test plot (May, 1950). The stakes were rated as 3 with sHght decay, 8 with medium decay, and 3 with severe decay. Apparently these results cannot be correlated with definite treatment retentions. On the other hand, the same authors state that stakes measuring 2 X 5 X 50 cm, treated to an average retention of 5.55 pounds per cubic foot of creosote (undiluted) have all decayed in a 4.5-year exposure period in greenhouse decay chamber tests. Additional experiments have been started, presumably with stakes at higher retentions, ''in order to determine whether it is possible to correlate results from such decay chamber experiments with the results obtained in field and service tests." In another series of experiments Rennerf elt and Starkenberg find after seven years (as of May, 1950) that creosoted stakes measuring 2x5 x 50 cm are showing different degrees of resistance to wood-destroying fungi in their four different test plots. The difference in behavior in differ- ent test plots — w^hich is more or less to be expected — holds true for salt as well as creosote treatments. Creosote and Bolidens (zinc-chro- mium-arsenic) are the better performing preservatives, with creosote in 490 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 the lead. However, there have been a total of three failures in the creo- soted stakes treated with average retentions of 3.6 (two stakes) and 5.6 (one stake) pounds of creosote. Stakes treated to an 8.6 pound retention are showing slight to medium decay, and in two plots slight decay has been found on a total of three stakes treated to a 12.1 -pound retention. The stakes are all treated without the addition of any diluent, such as toluene, to the creosote. The reader, bearing in mind the differences in the site conditions, can make interesting comparisons between the small stake test results obtained in Sweden and in the Gulf port test plot. Of further interest, however, is the fact that in the Swedish tests, round posts treated with average retentions of 5.37 and 5.80 pounds per cubic foot are all rated as sound after seven years exposure. One can assume that at treatment the outer annual rings of such posts con- tained 8 pounds or more of creosote and this amount has been sufficient to protect the posts in the Swedish climate. Rennerfelt has stated per- sonally to the writer that one would have to proceed with caution in Sweden in the direction of increasing the creosote retention for poles, on account of public reaction against bleeding. His test results — the only ones of their kind available from Europe to the writer's knowledge — seem to be in line with Bell Telephone Laboratories findings. They would be more interesting if he had used soil-block tests for correlation. Shortening the Bioassay Test Besides speeding up the weathering period by the use of a weathering wheel, or by the method of alternating water and controlled heat cycles now being developed at Bell Telephone Laboratories, there are two other avenues of approach to shortening the bioassay test. One is the use of thin wood veneer test units in place of wood blocks, in the methods of impregnation and exposure to fungus action proposed by Breazzano ' and Hopkins and Coldwell.^^ Breazzano claims a maximum of accuracy because of uniform distribution in thin pieces of wood, 0.6-0.7 mm, of the preservative to be tested and because the fungus attack and passage through the thin strip gives a quick visual indication of the necessary protective threshold. He also claims advantages for his Italian method because it is not necessary to use any culture medium at all — he ex- posed his wood strips over water only (cf . Waterman et al^^^) — and because no tedious weighing techniques and record making are required. His arguments are intriguing, but his method seems to be quite out of question for testing toxicity-permanence relations of volatile preserva- tives like creosote. Evaporation losses would be very rapid, close to those EVALUATION OF WOOD PRESERVATIVES 491 reported by Cislak in discussion following the presentation of the Snell and Shipley paper,^°^ and would approach those obtained by Boving in his impregnated filter paper experiments (see Table XXXI). Further- more, the results of tests such as his, which involve the principle of inoculation by placing "fungus on wood," instead of ''wood on fungus" as in the soil-block test, require a lot of translation to interpret their significance in practical wood preservation. Rabanus^^ brought this mat- ter out into the debate very clearly 20 years ago. Liese et af^ answer Breazzano's objections to the block test. There have been no further reports on the procedures followed by Hopkins and Coldwell. Their methods are subject to some of the same criticisms that have been mentioned in the preceding paragraphs, partic- ularly if one were to consider such techniques in a search for a way to speed up the culture tests. However, whether one agrees with them or not, their introduction of the idea of applying strength tests leads di- rectly to a discussion of strength losses, as against weight losses, as criteria for establishing preservative thresholds. Toughness or Impact Tests for Determining Preservative Effectiveness Trendelenburg^^ ^ published in 1940 his scheme for testing the strength of treated blocks that had been exposed to fungus attack. He was aiming at a technique that would shorten the time period of fungus tests on wood, but he was also looking for some other criterion than weight loss as a measure of fungus attack. He used the relative impact strength values of matched sound and decaying test specimens as indices of the degree of decay. Boards were carefully quarter-sawed first into pieces of double specimen width plus saw kerf tangentially, and of double speci- men length. From these blanks four test specimens were cut that meas- ured 8.5 X 8.5 X 120.0 mm. The pairs were considered to be matched laterally and vertically since every effort was made to cut them from the same annual rings. One piece from each pair was exposed to fungus attack and the other served as a control. The fungus cultures were made in Kolle flasks on malt agar. Specimens were placed radial side down directly on the growing fungus surface. In the pendulum testing machine the impact load was always appUed to the upper radial face, so that the lower more or less infected radial face represented the tension side of the specimen in the breaking test. Trendelenburg showed that the per cent change in strength caused by the test fungus in the early stages of decay was much more pro- nounced than the concomitant changes in weight or density. He called 492 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 attention to the fact that the German Standard^^ for testing wood pre- servatives contains a stipulation that a weight loss of less than 5 per cent shall not be considered significant unless there is visual evidence of actual wood destruction by the test fungus. He presents data to show that a loss of 50 per cent of the relative original impact strength in spruce and fir occurred after about fifteen days in test, and that the weight loss for the test pieces was about 3 per cent only. Impact strength seems to be affected much more than the bending or compression strength. He was confident that his method would not only shorten the time of the bioassay test but also give more reliable and more significant results than those based on weight loss alone. After Trendelenburg's death his ideas have been further tested and developed by von Pechmann and Schaile.^^^ In addition to trying out the suitability of the strength test procedure, they have explored the changes in the wood structure with the microscope, and, as decay progressed, they have determined the gross relation between weight loss and solu- bility in sodium hydroxide. They present as an example comparable data for the German Standard agar-block test and a test run by Trendelenburg's method, using pine wood (presumably Pinus sylvestris) and the test fungus Coniophora cere- hella, against a proprietary preservative. The absorption of the preserva- tive was essentially the same in each test; but not enough preservative was used to permit determination of the threshold. The main results were that in the impact strength test procedure in fifty days the strength reduction was 66.4 per cent and the weight loss 12.7 per cent; whereas in the standard agar-block procedure in four months the weight loss was only 2.0 per cent. Von Pechmann and Schaile feel that it is possible to save 23^ to 3 months time by using the strength test technique, and that with proper attention to detail the results will be more definite and just as reliable as those obtained in the longer period required for the standard agar-block tests. The precise Trendelenburg technique has not been tried out in this country in any comparative tests on wood preservatives but toughness test data on small specimens of wood and veneer, sound, fungus stained and decayed, treated and untreated, have been accumulating at the Forest Products Laboratory at Madison, Wis. The fact that strength loss begins earlier and may increase more rapidly than weight loss or than change in specific gravity in natural infections in the heartwood was shown by the writer^^ and his colleagues shortly after the end of World War I. Confirming data were secured in later tests.^^' ** Scheffer^^ showed the same results, on a more definite basis, by growing Polystictus versicolor EVALUATION OF WOOD PRESERVATIVES 493 on red gum sapwood in large test tubes and testing matched specimens for various strength properties as decay progressed. The Trendelenburg technique has possibilities, but it will be some time before one can say whether it is practicable to take full advantage of it in developing supplemental bioassay tests. The one outstanding difficulty in the way of extensive use of strength tests on small specimens lies in the procurement, in the very variable southern pine, for example, of a requisite quantity of straight grained quarter saAved wood for the manu- facture of the matched blocks. Small scale check tests are practicable, according to the writer's experience. The cost of personal supervision and manufacture of any large number of specimens would appear in advance to be exorbitant. Still, strength loss as a result of attack in treated wood is important, and Trendelenburg's ideas may win more proponents, if only as a supplemental procedure, after the soil-block technique has become more firmly established and appreciated. OTHER ACCELERATED BIOASSAY TESTS There are a number of items that must be mentioned before bringing this long paper toward its conclusion. There are, for example, other types of outdoor exposure tests on wood and of laboratory block tests than those cited. Two types only will be used as illustrations of the efforts that are being made to evaluate preservatives by other than the tradi- tional service test, namely: Verrall's^^^' ^^^ above ground outdoor testing procedure, and the experiments of Tippo et al with large block tests devised to determine effective concentrations of preservatives for preven- tion or control of decay in wooden ships. Since the soil-block test is essentially a laboratory simulation of con- trolled ground line conditions, there is a need for some other type of test that will approximate the above ground conditions to which treated wood may be exposed. Verrall treats pieces of dressed nominal 2x4 inch southern pine sapwood and exposes them to the varying wet and dry, hot and cold weather conditions at Saucier, Miss. One of two pieces has a 45° end cut. This end cut is toe-nailed to the side of the other piece, which is then nailed upright on a supporting treated or untreated rail support, with the A' up. This permits maximum hazard as far as catching water is concerned. His results are furnishing valuable information about pentachlorophenol, copper naphthenate and organic mercury com- pounds, for example. His techniques are applicable to other preservative problems, and other investigators are using his scheme in Canada for general studies,^^"' ^^^ and at Ann Arbor, Mich., for testing the amount 494 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 of preservative needed in the upper, or above ground section, of thin sap wood poles. The work of Tippo and his associates represents one phase of an ex- tensive set of experiments in which large (6 x 5 x J^ inch) and small (3 X 5 X J^ inch) specimens are made up to simulate a butt-block as- sembly, and exposed to the attack of certain critical fungi by adding to the block assembly another inoculum block (3 x 5 x % inch) that has been thoroughly infected. The assembled units are kept in a warm and practically saturated atmosphere until the reaction of the fungus to the different preservatives can be determined. This work is being expanded in view of the importance of minimizing decay in wooden ships. The point to be made here is that Verrall's tests and Tippo's tests should be evaluated carefully before the service test program is broad- ened extensively. OTHER OBSERVATIONS Some of the results of Suolahti's interesting studies"^ on the influence of Avood at a distance on the intensity and direction of growth of fungus filaments (mycelium) have been confirmed by preliminary experiments at Bell Laboratories. Small sterilized southern pine sapwood blocks en- closed in either test tube or Petri dish cultures exert a positive pull on the filaments that is effective over a distance of several centimeters. The growth of the mycelium is more luxuriant, and the filaments are definitely drawn in the direction of the wood. Without attempting any interpreta- tion of the significance of this phenomenon one may be permitted to point out that such studies strongly support the very great desirabiHty — if not the necessity — of using wood in any studies directed toward evaluation of wood preservatives. In view of the nearly forty years of prior work both here and in Europe, in which it was definitely established that certain higher fungi were the principal causes of decay in wood, it is hard to see why Weiss^^ spent so much time and such careful work on trying to test wood pre- servatives by using bacteria as his bioassay agents. Following the pre- sentation of his paper before the Society of Chemical Industries in 1911 some of Weiss's critics pointed out that his methods were unrealistic as far as oil preservatives were concerned, one of his commentators suggest- ing that the proper approach to the problem of preservative evaluation was to test treated and untreated wood, (unsterilized and sterilized) under conditions favorable for fungus growth. Tamura in 193l"^ used an assembly of two pieces of treated wood EVALUATION OF WOOD PRESERVATIVES 495 molding between which he inserted a properly sized piece of untreated wood, the whole being exposed over the surface of an agar culture of the test fungus. He did not attempt to add a block of infected wood to his setup as Trippo did; but his procedure illustrates an attempt of some twenty years ago to test the protective action of preservatives in the laboratory. His statement that sterilization might drive off a significant amount of volatile preservative from freshly treated blocks, but that the sterilizing process would probably have very little effect on the preservative residual in weathered blocks anticipated similar views ex- pressed in the present paper. For testing initial toxicities one can still use the Petri dish or agai flask method; but it is about as unreahstic as Weiss's procedure as far as tests of toxicity and permanence of wood preservatives are concerned. The results can be presented for their academic interest, and the investi- gator can keep safely aloof from the perilous practical problems of wood preservation unless he attempts to translate his data into terms of permanence and preservative value. Then his practical colleagues as well as his technical friends point out to him, truly with a vengeance, the error and unrealistic character of his efforts. It may be, as Rabanus^^ has suggested, that closely similar results can be obtained by the agar and by the agar-block method in the case of certain definite toxic chemicals, particularly water soluble ones. If so, the Petri dish or agar-flask method could be used with such preservatives, and the results of the tests could be applied in practice, after such agree- ment between agar and block tests was firmly proven and established. It is completely unrealistic to attempt to arrive at significant values for the volatile fractions of creosote by confining them within a tightly closed culture dish. If such materials are really transient their toxic function can operate only during the early life of the treated wood. In a whole creosote, for example, the relatively higher toxicity of the volatile low boiling fractions supplies an important initial power to the preserva- tive, which power is evidently diminished as the volatiles leave the treated wood. The degree of change in toxicity, as measured by the agar method, in new and aged or weathered creosotes, as shown by the works of Snell and Shipley,'^' Schmitz et al,'^'' '^' Baechler' and others is distinctly reaHstic as an index of how an oil may be altered by time and weather — at least with respect to a measure of its toxic properties. Such changes have great practical significance when minimum quantities of a preservative are employed, either for reasons of economy or in order to insure cleanhness in the treated product. The trouble is that the results of the Petri dish or agar-flask method do not indicate directly how much 496 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 preservative to use with a view to providing the necessary residual and effective preservative. The agar-block tests are somewhat better in this respect. While addi- tional proof is necessary the results of such tests may be good indices of retention requirements for certain water-borne preservatives. The com- parison of agar-block and soil-block tests made by Warner and Krause^^^ is incomplete, and therefore somewhat unsatisfactory, particularly in view of the title of their article. They do not follow through, and they repeat some of the inferential objections raised by others as to the effect of soil differences and methods of interpretation. The comparison by Finholt et al^^' *^ is also inconclusive chiefly because of the nature and design of their experiments. So also, but to a lesser degree, are the com- parisons that one may draw from the first two papers on the soil-block and agar-block tests from Madison, ' Avhich after all, were in the nature of preliminary or reconnaisance studies, introducing the first trials of an outdoor weathering technique, and developing the necessary steps in the broader and more comprehensive plans followed later. Duncan has now brought out a full scale comprehensive study of the agar-block and soil-block techniques, which, however, deals with oil type preservatives only. She shows definitely — as was indicated in the earlier Madison work — that the test fungi are more aggressive under the more natural and more realistic environment of the soil-block cul- tures. Definite evidence of the very important better control of moisture content in the soil-block tests is presented. The soil-block thresholds for the different preservatives are generally higher in the soil-block tests, although the order of effectiveness is essentially the same. Sedziak's^^^ recent paper comparing results of his tests on buried soil-blocks and results of tests by a soil-block technique approximate to that used at Madison and at Bell Laboratories is not convincing with respect to the implied superiority of the buried block method. His paper covers work begun after the early soil-block was started at Bell Telephone Laboratories, but before the extensive experiments at Madison were initiated. Satisfactory comparison of the work of the Madison and Ot- tawa laboratories is difficult because Sedziak has used a different set of test organisms, including the European Coniophora cerebella for example, and Lenzites sepiaria. which is apparently not a satisfactory discriminat- ing organism for creosote and pentachlorophenol. He has omitted the very critical Lentinus lepideus that has been employed at Madison since 1944. While he interprets threshold retentions for penta and for copper naphthenate that are close to those obtained at Madison, the steps in his gradient retentions leave one wishing that the Madison and Ottawa EVALUATION OF WOOD PRESERVATIVES 497 plans could have been more closely harmonized before the most recent Ottawa work was started, at least to the point of some tests with the same procedures and same test fungi. However, nothing in Sedziak's results negates the general conclusions reached at Madison and at Bell Laboratories about the value of the present soil-block technique for the testing of creosote and other oil type preservatives. CONCLUSIONS 1 . In the course of this paper evidence and interpretations have been presented to show that the soil-block technique incorporating a weather- ing procedure is a practical, rapid method of bioassay and that the results obtained from this method are in general agreement with acceler- ated stake and long time pole-diameter post tests on the same or similar preservatives. For example, it is shown that a creosote retention at treatment of 9-10 pounds per cubic foot is necessary to insure a satis- factory degree of preservative permanence in test blocks, in J^-inch stakes and in the outer 1 inch of test posts. There is no reason to believe that this minimum limit does not apply to the outer 1 inch of poles in Une. 2. The good reputation of well creosoted material is reaffirmed by these findings. Moreover, thej^ show why failures have occurred and indicate what should be done to forestall such failures. 3. Since the results of the block tests are essentially the same as the results of the much longer stake and post tests, the block test data can be used at once as a basis for the establishment of the necessary amounts of the respective preservatives distribution in the wood where they will do the most good. The possibilities of bleeding increase as the retention is increased, so the bioassay technique becomes an essential tool for closer appraisal of effective wood preserving power. 4. It is important now to recognize that the soil-block results with creosote, for example, reveal the fact that the results of the European agar-block tests are — in all cases — too low to represent indices of actual requirements in treated wood . Therefore, the results for creosote tabulated by Schulze, Theden and Starfinger^^ ^^^ have to be corrected upward by some multiplying factor; perhaps of the order of three or four, before they can be correlated with the results on blocks, stakes and posts presented in this paper. For the true scientific solution of the problems of these different techniques, perhaps an international task force may be required. 5. The interpretations presented in this paper indicate that the use of 498 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 the Laboratories' controlled weathering procedure will provide a means for determining truly effective threshold retentions for oil-type and salt- type preservatives for comparable service requirements. 6. These threshold determinations are supplying data that would have been most valuable in planning retention gradients in small stake and pole diameter post tests in test plots; and, in general, the bioassay tests explain and confirm the unequivocal results of experience. 7. Through the soil-block test a ready method is available for use in the quality control of a wood treater's current product at plants where large supplies of preservatives are received from one or more constant sources and are stored in bulk. No present method of bioassay control is sufficiently rapid to be effective or practicable on mixed samples taken at treating plants receiving preservatives at frequent intervals in small lots from varied sources. 8. The soil-block development may soon make it possible to reach approximately the ideal in which the long-time service tests of treated material in fine will confirm the Laboratories' rapid test results with respect to preservative requirements. Coupled with results- type require- ments (wherein the end product — not the steps of manufacture and treatment — are defined) viz., (a) retention in the wood, (b) penetration, and (c) cleanliness, there will then be even better assurance than at the present of the quality performance always expected under Bell System specifications. ACKNOWLEDGEMENTS It is hoped that the preceding pages will be accepted by the reader for what they are — condensed results of teamwork over the years — into which have been woven some individual ideas, opinions and interpreta- tions. The writer is responsible for the literature review and for the selection of the items discussed. In one way or another various members of the Timber Products Group of Bell Telephone Laboratories have contributed to the collection and analysis of the supporting data. Thanks are due especially for help on the soil-block section to J. Leutritz, Jr., L. R. Snoke and Ruth Ann MacAUister; on the %-inch stake section to J. Leutritz, Jr. ; on the text post and pole line sections to G. Q. Lumsden and A. H. Hearn; on the review and editing of the manuscript to R. J. Nossaman, F. F. Famsworth and G. Q. Lumsden; on the zealous check- ing and assembly of test, tables and figures to Jean E. Perry; and on the correlation of the Madison cooperative test data to Catherine G. Duncan. Dorothy Storm's untiring efforts as a secretarial task force are deeply EVALUATION OF WOOD PRESERVATIVES 499 appreciated. Without such help from these people the paper could not have been prepared in its present substantial form and substance. BIBLIOGRAPHY 1. Alleman, G., Quantity and Character of Creosote in Well -Preserved Timbers. Circ. 98, Forest Service, U. S. Dept. Agric, May 9, 1907. 2. Alliot, H., Methode d'Essais des Produits Anticryptogamiques. Inst. Nat. du Bois Bull. Techn. 1, 1945. 3. Amadon, C. H., Recent Observations on the Relation between Penetration, Infection and Decay in Creosoted Southern Pine Poles in Line. Am. Wood Preservers' Assoc, Proc, 35, pp. 187-197, 1939. 4. Baechler, R. H., Relations Between the Chemical Constitution and Toxicity of Aliphatic Compounds. Am. Wood Preservers' Assoc, Proc, 43, pp. 94- 111, 1947. 5. Baechler, R. H., The Toxicity of Preservative Oils Before and After Artificial Aging. Am. Wood Preservers' Assoc, Proc, 45, pp. 90-95, 1949. 6. Bateman, E., Quantity and Quality of Creosote Found in Two Treated Piles After Long Service. Circ. 199, Forest Service, Forest Products Laboratory Series, U. S. Dept. Agric, May 22, 1912. 7. Bateman, E., Coal -Tar and Water-Gas Tar Creosotes: Their Properties and Methods of Testing. U. S. Dept. Agric, Bui. No. 1036, pp. 57-65, Oct. 20, 1922. 8. Bateman, E., A Theory on the Mechanism of the Protection of Wood by Preservatives. Am. Wood Preservers' Assoc, Proc, 1920, p. 251; 1921, p. 506; 1922, p. 70; 1923, p. 136; 1924, p. 33; 1925, p. 22; 1927, p. 41. 9. Bavendam, W., Die pilzwidrige Wirkung der im Holzschutz benutzten Chemi- kalen. Mitteilungen d. Reichsinst. f. Forst und Holzwirtschaft. No. 7, 20 pp. Aug., 1948. 10. Baxter, D. V., Pathology in Forest Practice. 2nd Ed. XI + 601 pp. John Wiley and Sons, New York, 1952. 11. Bertleff, V., Priifung der Fungiziden Eigenschaften Ostrauer Steinkohlenteer Impragnierole und ihrer Bestandteile. Chem. Zeitung, 63, p. 438, 1939. 12. Bescher, R. H., and others. Cooperative Creosote Tests. Am. Wood Preservers' Assoc. ,Proc, 46, pp. 68-79, 1950. 13. Bienfait, J. L., and T. Hof, Buitenproeven met geconserveerde palen. I ste mededeling. Circ. 8 Ser. Ill, Conservering en Veredeling No. 3, Central Instituut voor Materiaal Onderzoek. Delft, Holland, Dec, 1948. 14. Bland, D. E., A Study of Toxicity of Australian Vertical Retort Creosote Oils to Lentinus Lepideus Fr., Polystictus Versicolor (L) Fr., and Madison 517. Australian Council Sci. and Ind. Jl. 15, pp. 135-146, 1942. 15. Blew, J. 0., Comparison of Wood Preservatives in Stake Tests. Am. Wood Preservers' Assoc, Proc, 44, pp. 88-119, 1948. 16. Blew, J. O., C. A. Richards, and R. H. Baechler, Evaluating Wood Preserva- tives. For. Prod. Res. Society, Proc, 5, pp. 230-238, (Wood-block tests, pp. 235-236), 1951. 17. Blew, J. O., Comparison of Wood Preservatives in Mississippi Post Study (1952 Progress Report). Report No. R1757, Forest Products Laboratory, U. S. Forest Service, Madison, Wis. 18. Blew, J. O., Comparison of Wood Preservatives in Stake Tests (1952 Progress Report). Report No. D1761, Forest Products Laboratory, U. S. Forest Service, Madison, Wis. 19. Boyce, J. S., Forest Pathology. 550 p. McGraw-Hill Book Company, Inc., New York, 1948. 20. Breazzano, A., Metodo Biologico di Controllo dei Sistemi di Preservazione, dei Legnami Adottato dall' Instituto Sperimentale delle FF. SS. (English translation: Biological Method of Testing Systems of Wood Preservation Adopted by the State Railway Experiment Institute.) Tech. Rev. Italian Railways, 4, No. 5, pp. 3-8, Nov., 1913. Forest Products Laboratory, Madi- son, Wis. 500 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 21. Breazzano, A., Metodi Normali di Prova sulla Putrescibilitd dei Legnani. (English translation: Standard Methods of Testing the Putrescibility of Wood.) Extract from Report of the Ninth Meeting of the Italian Associ- ation for the Study of Building Materials; Turin; April, 1922. Forest Prod- ucts Laboratory, Madison, Wis. 22. Breazzano, A., Osservazioni sul Metodo dei Blocchetti di Legno in Usa nell Analisi Tossimetrica delle Sostanze Conservatrici del Legno. Revista Technica Delle Ferrovie Italiane. 47, No. 6, June 15, 1935. (Observations on the Wood Block Method in the Toximetric Analysis of Wood Preserva- tives. Excerpt from Technical Review of Italian Railways. English trans- lation; Forest Products Laboratory.) 23. Breazzano, A., Methods for Determining the Fungicidal Power of Wood Preservatives. International Assn. for Testing Materials, Proc, London Congress, Group C — Organic Materials, Sub. group 3, pp. 484-486, April, 1937. 24. British Standard Method of Test for the Toxicity of Wood Preservatives to Fungi. British Standards Institution. British Standard No. 838, 17 pp., April, 1939. 25. Broekhuizen, S., Onderzoekingen over de Conserverende Waarde van een Aantal Houtconserveermiddelen. Rapp. Comm. Gebruikw. inh. hout. Deel II, Bijlage II, pp. 89-122, The Hague, 1937. 26.. Cartwright, K. St. G., and W. P. K. Findlay, Decay of Timber and its Preven- tion. VI plus 294 p. London: His Majesty's Stationery Office, 1946. Re- printed 1948. 27. Colley, R. H., The Effect of Incipient Decay on the Mechanical Properties of Airplane Timber. (Abstract) Phytopathology, 11, p. 45, 1921. 28. Colley, R. H., T. R. C. Wilson, and R. F. Luxford, The Effect of Polyporus Schweinitzii and Trametes Pini on the Shock-resistance, Compression Paral- lel to Grain Strength, and Specific Gravity of Sitka Spruce. Forest Products Laboratory, Project L-243-J1, Typewritten Report, 29 p., 32 figs., 4 plates. July 3, 1925. 29. Colley, R. H., and C. H. Amadon, Relation between Penetration and Decay in Creosoted Southern Pine Poles. Bell Sys. Tech. Jl., 15, pp. 363-379, July, 1936. 30. Colley, R. H., Some Observations on the Selection and Use of Modern Wood Preservatives. Reports, Twenty-fourth Session, Communications Sec- tion, Association of American Railroads, October, 1947, pp. 17-25. 31. Colley, R. H., Wood preservation and Timber Economy. Forest Products Institute of Canada. Papers presented at the First Annual Convention, Ottawa, Oct. 30-31, 1950. 32. Curtin, L. P., B. L. Kline, and W. Thordarson, V— Weathering Tests on Treated Wood. Ind. and Eng. Chem. 10, No. 12, pp. 1340-1343, Dec, 1927. 33. DIN (Deutsche Normen) DVM 2176, Blatt 1. Prufung von Holzschutz- mitteln, Mykologische Kurzpriifung (Klotzchen Verfahren). Berlin, Aug. 1939. (New Edition DIN 52176). 34. DIN DVM 52176, Blatt 2. Prufung von Holzschutzmitteln. Bestimmung der Auslaugbarkeit. Berlin, May 1941. (Reprinted Oct., 1948). 35. Duncan, C. G., and C. A. Richards, Methods of Evaluating Wood-Preserva- tives: Weathered Impregnated Wood Blocks. Am. Wood Preservers' Assoc, Proc, 44, pp. 259-264, 1948. 36. Duncan, C. G., A comparison of Two English Creosotes Produced from Coke- oven Coal Tar and Vertical-retort Coal Tar. Mss. Office Report, Division of Forest Pathology, Bur. PI. Ind., Forest Prod. Lab., Madison, Wis., Jan. 21, 1949. 37. Duncan, C. G., and C. A. Richards, Evaluating Wood Preservatives by Soil- block Tests: 1. Effect of Carrier on Pentachlorophenol Solutions; 2. Com- parison of a Coal Tar Creosote, a Petroleum Containing Pentachlorophenol or Copper Naphthenate and Mixtures of Them. Amer. Wood Preservers Assoc, Proc, 46, pp. 131-145, 1950. 38. Duncan, C. G., and C. A. Richards, Evaluating Wood Preservatives by Soil- Block Tests: 3. The Effect of Mixing a Coal Tar Creosote and a Penta- EVALUATION OF WOOD PRESERVATIVES 501 chlorophenol Solution with a Petroleum; a Creosote with a Coke Oven Tar or Pentachlorophenol Solution. Amer. Wood Preservers' Assoc, Proc. 47, pp. 264-274, 1951. 39. Duncan, C. G., and C. A. Richards, Evaluating Wood Preservatives by Soil Block Tests: 4. Creosotes. Amer. Wood Preservers' Assoc, Proc, 47, pp 275-287, 1951. 40. Duncan, C. G., Evaluating Wood Preservatives by Soil-Block Tests: 5 Lignite-Tar and Oil -Tar Creosotes. Amer. Wood Preservers' Assoc, Proc. 48, 1952. 41. Duncan, C. G., Soil-Block and Agar-Block Techniques for Evaluation of Oil-Type Wood Preservatives: Creosote, Copper Naphthenate and Penta- chlorophenol. Division of Forest Pathology, Bur. PI. Ind., Forest Prod- ucts Lab, Madison, Wis., Special Release No. 37, Jan., 1953. 42. Eden, Johan, och Erik Rennerfelt, Undersokninar enligt klotsmetoden av nagra fraimpregneringsmedel. (Studies on Wood Preservatives, According to the Block Method) Meddelanden irkn Statens Skogsforskningsinstitut, Bd. 35, No. 10, 1946. 43. Eden, Johan, och Erik Rennerfelt, Fait och rotkammarforsok avsedda att utrona skyddsverkan hos olika fraimpregneringsmedel. (Field and Decay- chamber Experiments to Ascertain the Protective Effect of Various Wood Preservatives.) Meddelanden fr&n Statens Skogsforskningsinstitut. Bd. 38, No. 4, 1949. 44. Falck, R., Die wichtigsten reinen Holzschutzmittel, die Methoden ihre Wertzahlen, Eingenschaften und Anwendung. Hausschwammforschungen, 8, pp. 18-20, 1927. 45. Findlay, W. P. K., A Standard Laboratory Test for Wood Preservatives. British Wood Preserving Assoc, Jl., 5, pp. 89-93, 1935. 46. Finholt, R. W., Improved Toximetric Agar-dish Test for Evaluation of Wood Preservatives. Anal. Chem., 23, No. 7, pp. 1039-1039, July, 1951. 47. Finholt, R. W., M. Weeks, and C. Hathaway, New Theory on Wood Preserva- tion. Ind. and Eng. Chem. 44, No. 1, pp. 101-105, Jan., 1952. 48. Flerov, B. C, und C. A. Popov., Methode zur Untersuchung der Wirkung von antiseptische Mitteln auf holzzerstorende Pilze. Angew. Bot. 15, pp. 386- 406, 1933. 49. Frosch, C. J., V — The Correlation of Distillation Range with Viscosity of Creosote. Physics, 6, pp. 165-170, May, 1935. 50. Gillander, H. E., C. G. King, E. O. Rhodes, and J. N. Roche, The Weathering of Creosote. Ind. and Eng. Chem., 26, No. 2, pp. 175-183, Feb., 1934. 51. Harrow, K. M., Toxicity of Water-Sol uble Wood-Preservatives to Wood- Destroying Fungi. New Zealand Jl., Sec B., 31, No. 5, pp. 14H9, Mar., 1950. 52. Harrow, K. M., Note on the Soil Moisture Content Used with the Leutritz Technique for Testing Toxicity of Wood Preservatives Against Fungi. New Zealand Jl. Sci. and Tech., 4, pp. 39-40, Jan., 1951. 53. Hatfield, I., Information on Pentachlorophenol as a Wood Preserving Chemi- cal. Am. Wood Preservers' Assoc, Proc, 40, pp. 47-65, 1944. 54. Holzschutzmittel Prufung und Forschung III. Wissenschaftliche Abhand- lungenderDeutschen Materialpriif ungsanstalten : Berlin-Dahlem. II Folge, Heft 7, 132 p. Springer-Verlag, Berlin, Gottingen, Heidelberg, 1950. 1. Schulze, B., G. Theden u K. Starfinger, Ergebnisse einer vergleichenden Prufung der Pilzwidrigen Wirksamkeit von Holzschutzmitteln. pp. 1-40. 2. Becker, G., Ergebnisse einer vergleichenden Prufung der insektotenden Wirkung von Holzschutzmitteln. II Teil. pp. 40-62. 3. Becker, G., Prufung der **Tropeneignung" von Holzschutzmitteln gegen Termiten. pp. 62-76. 4. Becker, G. u B. Schulze, Laboratoriumspriifung von Holzschutzmitteln gegen Meerwasserschadlinge. pp. 76-83. 55. Hopkins, C. Y., and B. B. Coldwell, Surface Coatings for Rotproofing Wood. Canadian Chemistry and Process Industries. N. R. C. No. 1256, Dec, 1944. 56. Howe, P. J., Weathering and Field Tests on Treated Wood. Amer. Wood Preservers' Assoc, Proc, 1928, p. 192. 502 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 57. Hubert, E. E., A study of Laboratory Methods Used in Testing the Relative Resistance of Wood to Decay. Univ. of Idaho Bulletin, 34, No. 15, July, 1929. 58. Hubert, E, E., An Outline of Forest Pathology. 543 pp. John Wiley and Sons, New York, 1931. 59. Hudson, M. S., and R. H. Baechler, The Oxidation of Creosote— Its Signifi- cance in Timber Treating Operations. Amer. Wood Preservers' Assoc, Proc, 36, pp. 74^112, 1940. 60. Hudson, M. S., Poria radiculosa, a Creosote Tolerant Organism. For. Prod. Res. Soc, Jl., 2, No. 2, pp. 73-74, June, 1952. 61. Humphrey, C. J., and R. M. Flemming, The Toxicity to Fungi of Various Oils and Salts, Particularly Those Used in Wood Preservation. Bui. No. 227, U. S. Dept. Agric, Aug. 23, 1915. 62. Hunt, G. M., and G. A. Garratt, Wood Preservation. VIII, 457 p. McGraw- Hill, New York, 1938. (In course of revision.) 63. Hunt, G. M., and T. E. Snyder, An International Termite Exposure Test — Twenty-First Progress Report. Amer. Wood Preservers' Assoc, Proc, 48, 1952. 64. Jacquiot, C, Controle de I'Efficacit^ des Fongicides Utilises pour I'lmpregna- tion des Bois. fitude Critique de la Technique Standard Anglaise et de la Norme Allemande DIN DVM 2176. Principes pour I'Etablissement d'une Norme Frangaise. Extr. d'Ann. I'Ecole Eaux For^ts, 8, pp. 185-206, 1942. 65. Kaufert, F. H., A Survey of Laboratory Methods Used in the Evaluation of Wood Preservatives. Report of Conunittee P-6, Appendix A. Amer. Wood Preservers' Assoc, Proc, 45, pp. 55-59, 1949. 66. Leutritz, J., Jr., The Toxic Action of Various Compounds on The Fungus Lentinus Lepideus Fr.). Unpublished Thesis, Columbia University, Nov., 1933. 67. Leutritz, J., Jr., Laboratory Tests of Wood Preservatives. Bell Lab. Record, 16, No. 9, pp. 324-328, May, 1938. 68. Leutritz, J., Jr., Acceleration of Toximetric Tests of Wood Preservatives by the Use of Soil as a Medium. Phytopathology, 39, No. 10, pp. 901-903, Oct., 1939. 69. Leutritz, J., Jr., Outdoor Tests of Wood Preservatives. Bell Lab. Record, 22, No. 4, pp. 179-182, Dec, 1943. 70. Leutritz, J., Jr., A Wood-Soil Contact Culture Technique for Laboratory Study of Wood-Destroying Fungi, Wood Decay and Wood Preservation. Bell Sys. Tech. Jl., 25, No. 1, pp. 102-135, Jan., 1946. 71. Liese, Nowak, Peters, Rabanus, Krieg, and Pflug., Toximetrische Bestimmung von Holzkonservierungsmitteln. Angew. Botanik, pp. 484-488, Nov .-Dec, 1935. 72. Liese, J. et al., Toximetrische Bestimmung von Holzkonservierungsmitteln. Angew. Chemie, 48, Beihefte 11, 1935. 73. Loseby, P. J. A., and P. M. D, Krog, The Persistence and Termite Resistance of Creosote and Its Constituent Fractions. Jour. South African Forestry Assoc, Jl., No. 11, pp. 26-32, June, 1944. 74. Lumsden, G. Q., and A. H. Hearn, Greensalt Treatment of Poles. Amer. Wood Preservers' Assoc, Proc, 38, pp. 349-361, 1942. 75. Lumsden, G. Q., Proving Grounds for Telephone Poles. Bell Lab. Record, 22, pp. 12-14, Sept., 1943. 76. Lumsden, G. Q., A Quarter Century of Evaluation of Wood Preservatives in Poles and Posts at the Gulfport Test Plot. Amer. Wood Preservers' Assoc, Proc, 48, 1952. 77. Lutz, M. L., M^thodes Permittant de Determiner la Resistivity des Bois Bruts ou Immunises Soumis a I'Attaque par les Champignons Lignicoles. Ann. I'Ecole Nat. Eaux For^ts 5, pp. 317-327, 1935. 78. Mahlke-Troschel-Liese, Holzkonservierung (Wood Preservation), 3rd Ed. XII -f 571 p. Springerverlag, Berlin/Gottingen/Heidelberg, 1950. 79. McMahon, W., C. M. Hill, and F. C. Koch, Greensalt — A New Preservative for Wood. Amer. Wood Preservers' Assoc, Proc, 38, pp. 334-348, 1942. 80. Martin, S. W., Characterization of Creosote Oils. Amer. Wood Preservers' Assoc, Proc, 45, pp. 100-130, 1949. EVALUATION OF WOOD PRESERVATIVES 503 81. Mayfield, P. B., The Toxic Elements of High Temperature Coal Tar Creosote. Amer. Wood Preservers' Assoc, Proc, 47, pp. 62-85, 1951. 82. Narayanamurti, D., V. Ranganathan, Ragbir Singh, T. R. Chandrasekhar, and A. Banerjee, Studies on Coal Tar Creosote as a Wood Preservative, Part II. Indian Forest Bulletin, No. 144, 1948, 7 + 43 pp., Jl. of India Press, Calcutta, 1950. 83. Peters, F., W. Krieg and H. Pflug, Toximetrische Priif ung von Steinkohlen- teerol. Chem. Zeit., 61, pp. 275-285, 1937. (English Edition, Pub. Int. Adv. Off. Wood Pres. The Hague. 1937.) 84. Preservative Treatment of Poles. (Condensed from report by American Telephone and Telegraph Company of Aug. 3, 1931.) Am. Wood Preservers' Assoc, Proc, p. 237, 1932. 85. Preservatives Committee; Report of Committee P-6. Am. Wood Preservers' Assoc, Proc, 48, 1952. 86. Rabanus, Ad., Die Toximetrische Priifung von Holzkonservierungsmitteln. Angew. Bot. 13, p. 352-371, 1931. (Partial translation in English. Am. Wood Preservers' Assoc, Proc, pp. 34-43, 1933.) 87. Reeve, C. S., Comment on Creosote-Permanence Toxicity Relationships. Am. Wood Preservers' Assoc, Proc, p. 78-79, 1934. 88. Rennerfelt, Erik, och Bo Starkenberg., Traskyddskomittens fait- och rot- kammarforsok. (Field and Decay-Chamber Experiments with Wood Pre- servatives.) Meddelanden fr§,n Statens Skogsforskningsinstitut, Bd., 40, No. 4, 1951. 89. Rhodes, E. O., J. N. Roche, and H. E. Gillander, Creosote Permanence- Toxicity Relationships. Am. Wood Preservers' Assoc, Proc, pp. 65-78, 1934. 90. Rhodes, E. O., History of Changes in Chemical Composition of Creosote. Am. Wood Preservers' Assoc, Proc, 47, pp. 40-61, 1951. 91. Rhodes, F. H,, and F. T. Gardner, Comparative Efficiencies of the Com- ponents of Creosote Oil as Preservatives for Timber. Ind. and Eng. Chem., 22, No. 2, pp. 167-171, Feb., 1930. 92. Rhodes, F. H., and I. Erickson, Efficiencies of Tar Oil Components as Pre- servative for Timber. Ind. and Eng, Chem. 25, pp. 989-991, Sept., 1933. 93. Richards, A. P., Cooperative Creosote Program; Preliminary Progress Report on Marine Exposure Panels. Am. Wood Preservers' Assoc, Proc, 48, 1952. 94. Richards, C. A., Methods of Testing Relative Toxicity of Wood Preserva- tives. Am. Wood Preservers' Assoc, Proc, 19, pp. 127-135, 1923. 95. Richards, C. A., and R. M. Addoms, Laboratory Methods for Evaluating Wood Preservatives: Preliminary Comparison of Agar and Soil Culture Techniques Using Impregnated Wood Blocks. Am. Wood Preservers' Assoc, Proc, 43, pp. 41-56, 1947. 96. Richards, C. A., Laboratory Decay Resistance Tests — Soil-block Method. (In Cooperative Creosote Tests by R. H. Bescher et al.) Am. Wood Pre- servers' Assoc, Proc, 46, pp. 71-76, 1950. 97. Scheffer, T. C, Progressive Effects of Polyporus Versicolor on the Physical and Chemical Properties of Red Gum Sapwood. U. S. Dept. Agric, Tech. Bui., No. 527, Sept., 1936. 98. Scheffer, T. C, T. R. C. Wilson, R. F. Luxford, and Carl Hartley, The Effect of Certain Heart Rot Fungi on the Specifid Gravity and Strength of Sitka Spruce and Douglas-Fir. U. S. Dept. Agric, Tech. Bui., No. 779, 24 pp., May, 1941. 99. Schmitz, H., Laboratory Methods of Testing the Toxicity of Wood Preserva- servatives. Ind. and Eng. Chem., Anal. Ed., 1, No. 7, pp. 76-79, April, 1929. 100. Schmitz, H., and Others, A Suggested Toximetric Method for Wood Preserva- tives. Ind. and Eng. Chem., Anal. Ed., 2, p. 361, 1930. 101. Schmitz, H., and S. J. Buckman, Toxic Action of Coal-Tar Creosote. Ind. and Eng. Chem., 24, No. 7, pp. 772-777, 1932. 102. Schmitz, H., W. J. Buckman and H. von Schrenk, Studies of the Biological Environment in Treated Wood in Relation to Service Life. Changes in the Character and Amount of 60/40 Creosote-Coal Tar Solution and Coal Tar 504 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 and Decay Resistance of the Wood of Red Oak Crossties after Five Years Service. Am. Wood Preservers' Assoc, Proc, 37, pp. 248-297, 1941. 103. Schmitz, H., H. von Schrenk, and A. L. Kammerer, Studies of the Biological Environment in Treated Wood in Relation to Service Life, III. Am. Wood Preservers' Assoc, Proc, 41, pp. 153-179, 1945. 104. Schulze, B., und G. Becker, Untersuchungen iiber die pilzwidrige und insek- tentotende Wirkung von Fraktionen und Einzelstoffen des Steinkohlen- teerols. Holzforschung., 2, No. 4, pp. 95-127, 1948. 105. Sedziak, H. P., The Wood-Block Soil Method of Accelerated Testing of Wood Preservatives. Report of Committee P-6, Appendix B, Am. Wood Pre- servers' Assoc, Proc, 46, pp. 55-58, 1950. 106. Sedziak, H. P., The Evaluation of Two Modern Wood Preservatives. For. Prod. Res. Soc, Proc, 1952. 107. Snell, W. H., The Use of Wood Discs as a Substrate in Toxicity Tests of Wood Preservatives. Am. Wood Preservers' Assoc, Proc, 25, pp. 126-129, 1929. 108. Snell, W. H., and L, B. Shipley, Creosotes — Their Toxicity, Permanence and Permanence of Toxicity. Am. Wood Preservers' Assoc, Proc, 32, pp. 32- 114, 1936. 109. Standard of N. W. M. A., Method for Testing the Preservative Property of Oil-soluble Wood-Preservatives by Using Wood Specimens Uniformly Impregnated. Nat. Wood Mfg. Assoc, M-1-51, April 27, 1951. 110. STAS 650-49, Incercarea Toxicitatii Substantelor de Impregnat Contra Ciupercilor. Comisiunea de Standardizare (Rumania) April 1, 1950. 111. Suolahti, Osmo, tJber Eine das Wachstum von Faulnispilzen Beschleunigende Chemischen Fernwirkung von Holz. Statens Tekniska Forskningsanstalt. 95 p. Helsinki, Finland, 1951. 112. Tamura, T., New Methods of Test on the Toxicity and Preservative Value of Wood Preservatives. Phytopathologische Zeitschrift, 3, No. 4, pp. 421- 437, 1931. 113. Teesdale, C. H., Volatilization of Various Fractions of Creosote After Their Injection into Wood. Circ 188, Forest Service, Forest Products Laboratory Series, U. S. Dept. Agric, Oct. 17, 1911. 114. Tippo, O., J. M. Walter, S. J. Smucker, and W. Spackman, Jr., The Effective- ness of Certain Wood Preservatives in Preventing the Spread of Decay in Wooden Ships. Lloydia, 10, pp. 175-208, Sept., 1947. 115. Trendelenburg, R., Uber die Abkurzung der Zeitdauer von Pilzversuchen an Holz mit Hilfe der Schlagbiegepriifung. Holz als Roh- und Werkstoff. 3. No. 12, s. 397-407, Dec, 1940. 116. van den Berge, J. Beoordeeling van de Waarde van Fungicide Stoffen voor Houtconserveering. 183 p. N. V. Technische Boekhandel, J. Waltman, Delft, Holland, 1934. 117. van Groenou, H. Broese, Weatheringsproeven met Houtconserveermiddelen, (Weathering Tests with Wood Preservatives). Materiaalenkennis, 7, No. 10, pp. 63-65, Oct., 1940. 118. van Groenou, H. Broese, H. W. L. Rischen and J. van den Berge, Wood Pre- servation During the Last 50 Years. XII -}- 318 p. A. W. Sijthoff, Lei- den, Holland, 1951. 119. von Pechmann, H., u. 0. Schaile, Uber die Anderung der dynamischen Festig- keit und der Chemischen Zusammensetzung des Holzes durch den Angriff Holzzerstorender Pilze. Forstwissenschaftliches Zentralblatt, 69, No. 8, s. 441-465, 1950. 120. Verrall, A. F., Progress Report on Tests of Soak and Brush Preservative Treatments for Use on Wood off the Ground. Southern Lumberman, Aug. 1, 1946. 121. Verrall, A. F., Decay Protection for Exterior Woodwork. Southern Lumber- man, June 15, 1949. 122. Warner, R. W., and R. L. Krause, Agar-Block and Soil-Block Methods for Testing Wood Preservatives. Ind. and Eng. Chem., 43, No. 5, pp. 1102-1107, May, 1951. 123. Waterman, R. E., Forecasting the Behavior of Wood Preservatives. Bell Lab. Record, 11, No. 3, pp. 67-72, Nov., 1932. EVALUATION OF WOOD PRESERVATIVES 505 124. Waterman, R. E., F. C. Koch, and W. McMahon, Chemical Studies of Wood Preservation. III. Analysis of Preserved Timber. Ind. and Eng. Chem.. Anal. Ed., 6, No. 6, pp. 409-413, Nov. 15, 1934. 125. Waterman, R. E. and R. R. Williams, Chemical Studies of Wood Preserva- tion. IV. Small Sapling Method of Evaluating Wood Preservatives. Ind. and Eng. Chem., Anal. Ed., 6, No. 6, pp. 413-418, Nov. 15, 1934. 126. Waterman, R. E., J. Leutritz, and C. M. Hill, A Laboratory Evaluation of Wood Preservatives. Bell System Tech. Jl., 16, pp. 194-211, April, 1937. 127. Waterman, R. E., J. Leutritz and C. M. Hill, Chemical Studies of Wood Preservation: The Wood-block Method of Toxicity Assay. Ind. and Eng. Chem., Anal. Ed., 10, No. 6, pp. 306-314, June, 1938. 128. Weiss, J. M., The Antiseptic Effect of Creosote Oil and Other Oils Used for Preserving Timber. Soc. Chem. Ind. Jl., 30, No. 23, Dec. 15, 1911. 129. Williams, R. R., Chemical Studies of Wood Preservation. I. The Problem and Plan of Attack. Ind. and Eng. Chem., Anal. Ed., 6, No. 5, pp. 308-310, Sept. 15. 1934. Abstracts of Bell System Technical Papers* Not Published in this Journal Principles and Applications of Converters for High-Frequency Measure- ments. D. A. Alsberqi. I.R.E., Proc, 40, pp. 1195-1203, Oct., 1952. (Monograph 2030). The heterodyne method permits measurements over wide frequency bands with the standards operating at a fixed frequency. The accuracy of such measure- ments depends upon the performance of heterodyne conversion transducers or converters. Design principles are derived to maximize hnearity and dynamic range and minimize zero corrections. These principles have been applied to point- to-point and sweep measurements of delay, phase, transmission, and impedance. Ferroelectric Storage Elements for Digital Computers and Switching Sys- tems. J. R. Anderson^. Elec. Engg., 71, pp. 916-922, Oct., 1952. (Mono- graph 2014). These ferroelectric storage devices, although still comparatively new, show great promise. They can store up to 2,500 bits of information per square inch in a surface only a few thousandths of an inch thick with pulses less than a micro- second long. Transistors in Switching Circuits. A. E. Anderson^ I.R.E., Proc.^ 40, pp. 1541-1558, Nov., 1952. Corrections to Figs. 17, 18 and 19 giving synopses published in December issue, pp. 1732 and 1733. The general transistor properties of small size and weight, low power and voltage, and potential long hfe suggest extensive apphcation of transistors to pulse- or switching-type systems of computer or computer-like nature. It is possible to devise simple regenerative circuits which perform the normally employed functions of waveform generation, level restoration, delay, storage (registry or memory), and counting. The discussion is hmited to point-contact type transistors in which the alpha or current gain is in excess of unity and to a particular feedback configuration. Such circuits, which are of the so-called trigger type, are postulated to involve negative resistance. On this basis an analysis, which approximates the negative- resistance characteristic by three intersecting broken lines, is developed. Con- * Certain of these papers are available as Bell System Monographs and may be obtained on reauest to the Publication Department, Bell Telephone Labora- tories, Inc., 463 West Street, New York 14, N. Y. For papers available in this form, the monograph number is given in parentheses following the date of pub- lication, and this number should be given in all requests. * Bell Telephone Laboratories. 506 ABSTRACTS OF TECHNICAL ARTICLES 507 elusions which are useful to circuit and device design are reached. The analysis is deemed sufficiently accurate for first-order equiUbrium calculations. Transistors having properties specifically intended for pulse service in the cir- cuits described have been developed. Their properties, limitations, and parame- ter characterizations are discussed at some length. Mobility of Electrons in Germanium. P. P. Debye^ and Esther M. ConwellI. Letter to the Editor. Phys. Rev., 87, pp. 1131-1132, Sept. 15, 1952. The Telephone Industry in National Defense. C. A. Armstrong^. Te- lephony, 143, pp. 44-46, 114, Oct. 25, 1952. Infrared Absorption in High Purity Germanium. H. B. Briggs^ Letter to the Editor. Jl. Opt. Soc. Am., 42, pp. 686-687, Sept., 1952. New Infrared Absorption Bands in p-Type Germanium. H. B. Briggs^ and R. C. Fletcher^ Letter to the Editor. Phys.'Rev., 87, pp. 1130-1131, Sept. 15, 1952. Automatic Switching for Nation-Wide Telephone Service. A. B. Clark^ and H. S. Osborne^. A.I.E.E., Trans. Commun. and Electronics Sect., 2, pp. 245-248, Sept., 1952. (Monograph 2015). Western Electric' s Service with Standards. K. B. Clarke^ Standardi- zation, 23, pp. 332-338, Oct., 1952. Properties of Silicon and Germanium. Esther M. Conwell^ I.R.E., Proc, 40, pp. 1327-1337, Nov., 1952. This article provides the latest experimental information on those fundamental properties of germanium and silicon which are of device interest, currently or potentially. Electrical properties, especially carrier density and mobility, have been treated in greatest detail. Descriptive material has been provided to the extent necessary to give physical background. Effects of Space-Charge Layer Widening in Junction Transistors. J. M. Earlyi. LR.E., Proc, 40, pp. 1401-1406, Nov., 1952. Some effects of the dependence of collector barrier (space-charge layer) thick- ness on collector voltage are analyzed. Transistor base thickness is shown to decrease as collector voltage is increased, resulting in an increase of the current- gain factor (a) and a decrease in the emitter potential required to maintain any ^ Bell Telephone Laboratories. 2 American Telephone and Telegraph Company. ^ Western Electric Company. 508 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 fixed emitter current. These effects are shown to lead to two new elements in the theoretical small-signal equivalent circuit. One, the collector conductance (g'c), is proportional to emitter current and varies inversely with collector voltage. This term is the dominant component of collector conductance in high-quality junction transistors. The other element, the voltage feedback factor (fiec), is independent of emitter current, but varies inversely with collector voltage. The latter element is shown to modify the elements of the conventional equivalent tee network. Four-Terminal p-n-p-n Transistors. J. J. Ebers^ I.R.E., Proc, 40, pp. 1361-1364, Nov., 1952. The equivalent circuit of a p-n-p-n transistor is obtained. It is demonstrated that a p-n-p transistor and an n-p-n transistor can be connected so that the com- bination has the same equivalent circuit as the p-n-p-n structure. A simplified circuit is obtained which can be used when the p-n-p-n transistor is connected as a hook-collector transistor. A method of adjusting the current gain of p-n-p-n transistors by external means is given as w^ell as experimental results. Dynamics of Transistor Negative-Resistance Circuits. B. G. Farley^ LR.E., Proc, 40, pp. 1497-1508, Nov., 1952. A general method is presented for calculating approximately the behavior of many nonlinear circuits by dividing the region of operation into subregions, within each of which the circuit may be considered linear to a good approxima- tion. The method is appHed to a high-speed transistor switching circuit as an il- lustrative example. Regenerative Amplifier for Digital Computer Applications. J. H. FelkerI. i^r^e., Proc, 40, pp. 1584-1596, Nov., 1952. A description of the negative-resistance properties of the point-contact transis- tor is presented as an introduction to the description of a regenerative amphfier. The choice of circuit parameters for the amplifier is discussed and a sample de- sign presented. The illustrative amplifier regenerates digital information at a megacycle rate and develops pulses with rise times of less than 0.05 /xsec. It op- erates from supply voltages of +6 and —8 volts, with a battery drain of less than 0.05 watt. A complete set of computer building blocks has been designed around the amphfier. Their use is illustrated in two computer applications. Evidence for Domain Structure in Anti-ferromagnetic CoO From Elas- ticity Measurements. M. E. Fine^ Letter to the Editor. Phys. Rev., 87, p. 1143, Sept. 15, 1952. Optical Position Encoder and Digit Register. H. G. Follingstad^ J. N. Shive^ and R. E. Yaeger^ LR.E., Proc, 40, pp. 1573-1583, Nov., 1952. The usefulness of transistors in systems has been given a feasibility proof through the construction and operation of a six-digit position encoder and serial- 1 Bell Telephone Laboratories. ABSTRACTS OF TECHNICAL ARTICLES 509 output digit register. This system performs the functions of photoelectric en- coding, pulse regeneration, digit storage, reflected-to-natural binary translation, and digit shifting by means of circuits using transistors and other semi-conductor devices. The model occupies a volume of about \i cubic foot, weighs seven pounds, and consumes 16 watts of power. Comparison of Recording Processes. J. G. Frayne^, I.R.E., Trans., PGA-7, pp. 5-8, May, 1952. S.M.P.T.E., JL, 59, pp. 313-318, Oct., 1952. The three common forms of sound recording may be classed as mechanical (disk), photographic and magnetic. All three methods are in common use today and each is employed in a field for which it appears to be pecuharly fitted. The purpose of this article is to examine briefly the factors which determine the fidelity of each method. By fideUty we mean how true the tonal range can be reproduced, the amount and nature of h'armonic distortion present, the signal- to-noise ratio possible with each method, and the amount of wow or flutter that may be expected under average conditions of reproduction for each recording process. Transistor Shift Register and Serial Adder. J. R. Harris^ I.R.E., Proc, 40, pp. 1597-1602, Nov., 1952. A small set of basic functions, such as binary memory and elementary binary logic, can be remarkably versatile; such functions are important in switching and computing. This paper describes a piece of computing equipment which can store a pair of binary numbers and add them, producing the sum a digit at a time. The equipment is built from a basic set of functional blocks, all of which are de- signed around transistors. This set of building blocks consists of a binary cell, a pulse amplifier, a pulse amplifier w^ith delay, and logic circuits. The binary cell is a flip-flop; amplifiers are monostable circuits, and logic is performed in diode gates. Some interesting special features arise from the use of transistors. These features are discussed and the designs are evaluated. Charge Transfer and the Mobility of Rare Gas Ions. J. A. Hornbeck^ Jl. Phys. Chem., 56, pp. 829-831, Oct., 1952. Ion-atom collisions in the rare gases between an atomic ion and a parent gas atom, such as Ne+ and Ne, involve quantum mechanical symmetry effects which though rigorousty inseparable have been fisted as (a) a force of resonance attrac- tion, (b) a force of resonance repulsion, and (c) charge exchange. Drift velocity measurements at high fields show that this compficated interaction may be rep- resented to a good approximation by the hard sphere model of kinetic theory in which the collision cross section is several times the viscosity cross section of the atoms themselves. Broad Band Matching with a Directional Coupler. W. C. Jakes^ I.R.E., Proc, 40, pp. 1216-1218, Oct., 1952. (Monograph 2033). ^ Bell Telephone Laboratories. ^ Westrex Corporation. 510 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 This paper presents the results of a theoretical and experimental study of a waveguide matching technique which allows a directional coupler to be located any distance away from the discontinuity causing the original mismatch and a broad-band match to still be obtained. Design curves are included which give the required coupling coefficient of the directional coupler and the power loss for a given initial mismatch and desired vswr reduction. Experimental confirmation of the theory is also presented. New General-Purpose Relay for Telephone Switching Systems. A. C. KellerI. Elec. Engg., 71, pp. 1007-1012, Nov., 1952. Monograph 2034). This new general-purpose electromagnetic relay, called the AF type is a wire spring relay. With variations providing slow release or marginal characteristics, it is known as the AG and AJ relay, respectively. It provides improved perform- ance at lower cost. Spherical Model of a Ferromagnet. H. W. Lewis^ and G. H. Wannier^ Letter to the Editor. Phys. Rev., 88, pp. 682-683, Nov. 1, 1952. In order to interpret the properties of the tetragonal crystal ND4D2PO4 (deuterated ADP) a thermo-dynamic treatment has been developed which re- lates the observed crystal structure change and the dielectric constant change at the transition temperature to the appearance of spontaneous polarization. For an antiferroelectric crystal, the average spontaneous polarization is zero, being oppositely directed for adjacent layers, but the square of the spontaneous polarization is large. This results in quadratic strain components which cause a change in the crystal structure below the transition temperature. It is shown that the change observed is consistent with an antiferroelectric arrangement with one of the a axes being the antiferroelectric axis. The dielectric constants in all three directions suffer a large drop below the transition temperature. Piezoelectric^ Dielectric, and Elastic Properties of NDiD^POi {Deuter- ated ADP). W. P. Masoni and B. T. Matthias^. Phys. Rev.y 88, pp. 477^79, Nov. 1, 1952. (Monograph 2036). Transistors in Our Civilian Economy. J. W. McRae^ I.R.E.j Proc, 40, pp. 1285-1286, Nov., 1952. /. R. E. Editor's Note: At relatively long intervals there appear on the tech- nical and industrial horizons devices of such broad scope and major significance that they profoundly affect the fields of their use. One of these epochal develop- ments is the transistor, which bids fair to take its place beside the electron tube as one of the foundation stones of future communications and electronics. It is accordingly timely and suitable that certain of the probable future in- dustrial uses and effects of the transistor should be here analyzed in a guest edi- torial by an engineer especially qualified for this task, and who is a Fellow and Director of the Institute, and a Vice President of Bell Telephone Laboratories. * Bell Telephone Laboratories. ABSTRACTS OF TECHNICAL ARTICLES 511 Domain Properties in BaTiOz. W. J. Merz^ Letter to the Editor. Phys. Reu., 88, pp. 421-422, Oct. 15, 1952. Notes on Methods of Transmitting the Circular Electric Wave Around Bends. E. S. Miller^. LR.E., Proc, 40, pp. 1104-1113, Sept., 1952. (Monograph 2037). The tendency for energy to be converted out of the circular electric wave in bent round pipe may be avoided by one of three general approaches: (1) by re- moving the degeneracy between TEoi and TMn, (2) by converting to a normal mode of the bent guide at both ends of the bend, and (3) by utilizing dissipation in the unwanted modes to prevent power transfer to them. All three approaches are discussed. Normal attenuation in round pipe should be effective in moderat- ing straightness requirements. Elliptical guide and special waveguide structures may be used to negotiate intentional bends; bending radii in the range one to 1,000 feet appear acceptable at 50,000 mc for waveguides %-uich to 2 inches in diameter, respectively. Multi-Element Directional Couplers. S. E. Miller^ and W. W. Mum- FORD^ I.R.E., Proc, 40, pp. 1071-1078, Sept., 1952. (Monograph 2038). It is shown that the backward wave in a directional coupler is related to the shape of the function describing the coupling between transmission lines by the Fourier transform. This facihtates the design of directional couplers for arbitrary directivities over any prescribed frequency band. Tightly coupled directional couplers are analyzed in simple terms, and it is shown that any desired loss ratio, including complete power transfer between lines, may be achieved. The theories are verified using waveguide models operating at 4,000, 24,000 and 48,000 mc, and it is indicated that the work is appUcable to many types of electrical and acoustic transmission lines. Transistor Noise in Circuit Applications. H. C. Montgomery^. I.R.E.y Proc, 40, pp. 1461-1471, Nov., 1952. Linear circuit problems involving multiple noise sources can be handled by famihar methods with the aid of certain noise spectrum functions, which are described. Several theorems of general interest dealing with noise spectra and noise correlation are derived. The noise behavior of transistors can be described by giving the spectrum functions for simple but arbitrary configurations of equivalent noise generators. From these, the noise figure can be calculated for any desired external circuit. Transistor Noise in Circuit Applications. H. C. Montgomery^ I.R.E.y Proc, 40, pp. 1314-1326, Nov., 1952. The invention of the transistor provided a simple, apparently rugged device that could amplify — an abihty which the vacuum tube had long monopoHzed. As with most new electron devices, however, a number of extremely practical hmitations had to be overcome before the transistor could be regarded as a ^ Bell Telephone Laboratories. 512 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 practical circuit element. In particular, the reproducibility of units was poor — units intended to be alike were not interchangeable in circuits; the rehability was poor — in an uncomfortably large fraction of units made, the characteristics changed suddenly and inexplicably; and the ''designability" was poor — it was difficult to make devices to the wide range of desirable characteristics needed in modern communications functions. This paper describes the progress that has been made in reducing these limitations and extending the range of performance and usefulness of transistors in communications systems. The conclusion is drawn that for some system functions, particularly those requiring extreme miniaturization in space and power as well as reHabiHty with respect to life and ruggedness, transistors promise important advantages. In Search of the Missing 6 Dh. W. A. Munson^ and F. M. Wiener^ Acoustical Soc, Am., JL, 24, pp. 498-501, Sept., 1952. (Monograph 2019). The unexplained difference in sound pressure in the ear canal which appears to exist when equally loud low frequency tones are presented alternately from an earphone and from a loudspeaker has bedeviled acousticians for many years and, unfortunately, still continues to do so. There are presented here the results of some of the measurements carried out at the Bell Telephone Laboratories which show the magnitude of the effect and various attempts at explaining it. While no satisfactory explanation has been found, it is hoped that publication of these results will stimulate interest in the problem. Nation-Wide Numbering Plan. W. H. Nunn^. A.I.E.E., Trans., Com- mun. and Electronics Sect., 2, pp. 257-260, Sept., 1952. Elec. Engg., 71, pp. 884-888, Oct., 1952. (Monograph 2015). At the present time a great variation in the types of telephone numbers exists. This is because of the number of telephones in communities of different sizes. With the advent of local dialing and now nation-wide dialing, a uniform num- bering system has become necessary. How to Detect the Type of an Assignable Cause. P. S. Olmstead^. Ind. Quality Control, 9, pp. 32-34, 36, Nov., 1952. Silicon p-n Junction Alloy Diodes. G. L. Pearson^ and B. Sawyer^ I.R.E., Proc, 40, pp. 1348-1351, Nov., 1952. A new type of p-n junction silicon diode has been prepared by alloying ac- ceptor or donor impurities with n- or p-type silicon. The unique features of this diode are: (a) reverse currents as low as 10~^° amperes, (b) rectification ratios as high as 10* at 1 volt, (c) a Zener characteristic in which d(\og I) rf(log V) may be as high as 1,500 over several decades of current, (d) a stable Zener voltage ^ Bell Telephone Laboratories. 2 American Telephone and Telegraph Company. ABSTRACTS OF TECHNICAL ARTICLES 513 which may be fixed in the production process at values between 3 and 1,000 volts and (e) ability to operate at ambient temperatures as high as 300°C. Hard Rubber. H. Peters^ Ind. and Eng. Chem., 44, pp. 2344-2345, Oct., 1952. As judged by the literature, the general trend during the past year on the subject of hard rubber has been toward de-emphasis of fundamental research and more emphasis on use. Plastics, through substitution, continue to make gains in the field of hard rubber. A renewed interest is again shown in the use of latex ebonite for industrial applications. The patent situation appears to be unusually active and the interest in sjTithetic hard rubbers continues to increase. Application of Information Theory to Research in Experimental Pho- netics. G. E. Peterson^ Jl. Speech and Hearing Disorders^ 11 j pp. 175- 188, June, 1952. Principles of Zone-Melting. W. G. Pfann^ Jl. of Metals, 4, pp. 747- 753, July, 1952. A.I.M.E. Trans., 194, pp. 747-753, 1952. (Monograph 2000). In zone-melting, a small molten zone of zones traverse a long charge of alloy or impure metal. Consequences of this manner of freezing are examined with respect to solute distribution in the ingot, with particular reference to purifica- tion and to prevention of segregation. Results are expressed in terms of the number, size, and direction of travel of the zones, the initial solute distribution, and the distribution coefficient. Nonsynchronous Time Division with Holding and with Random Sampl- ing. J. R. Pierce^ and A. L. Hopper^. I.R.E., Proc, 40, pp. 1079-1088, Sept., 1952. (Monograph 2041). There is a general tj-pe of system in which an indefinitely large number of transmitters can have access to anj^ of an indefinitely large number of receivers over a medium of limited bandwidth. In these systems, signal-to-noise ratio goes down as more transmitters are used simultaneously. This paper describes a particular system which sends samples by means of coded pulse groups sent at random times. The signal-to-noise ratio is good in the absence of interference and the effect of interference is minimized by holding the previous sample if a sample is lost. An experimental sj'stem worked satisfactorily and gave close to the predicted signal-to-noise ratio. Such a system might be used to provide com- munication and automatic switching in rural telephony, or for other applica- tions. Fundamental Plans for Toll Telephone Plant. J. J. Pilliod^. A.I.E.E., Trans., Commun. & Electronics Sect., 2, pp. 248-256, Sept., 1952. (Mono- graph 2015). ^ Bell Telephone Laboratories. 2 American Telephone and Telegraph Company. 514 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Organization of the Engineering Profession. D. A. Quarles^ Elec. Engg., 71, pp. 963, 964, Nov., 1952. Since the organization of the American Society of Civil Engineers 100 years ago, professional engineering has assumed a major role in American life. The goal now to be attained is the closer organization of the entire engineering pro- fession. We begin a New Institute Year. D. A. Quarles^. Elec. Eng., 71, pp. 867-868, Oct., 1952. In an address before the recent Pacific General Meeting in Phoenix, Mr. Quarles, President of the Institute, evaluates the evolution in A.I.E.E. organiza- tion and poHcy as the Institute enters a new administrative year. Mean Free Paths of Electrons in Evaporated Metal Films. F. W. Reynolds^ and G. R. Stilwell^ Letter to the Editor. Phys. Rev., 88, pp. 418-419, Oct. 15, 1952. Single-Sidehand System for Overseas Telephony. N. F. Schlaak^ Elec- tronics, 25, pp. 146-149, Nov., 1952. Single-sideband transmitter furnishes four voice channels for overseas tele- phone service. Pushbutton tuning permits rapid frequency shifts and load- control circuit minimizes interchannel crosstalk and out-of-band radiation. Copper-oxide and germanium varistors replace modulator tubes. Automatic Toll Switching Systems. F. F. Shipley^ A.I.E.E., Trans., Commun. and Electronics Sect., 2, pp. 261-269, pp. 889-897, Oct., 1952. (Monograph 2015). The new system was designed to implement the nation-wide switching plan which integrates the telephone switching network of the entire nation into a single unit. Requiring a high order of mechanical intelUgence, this system is one of the most comprehensive ever devised. Properties of M-1740 p-n Junction Photocells. J. N. Shive^ I.R.E., Proc, 40, pp. 1410-1413, Nov., 1952. The p-n junction photocell has a sensitivity of 30 ma per lumen for light of 2,400 degrees K color temperature, corresponding to a quantum yield approxi- mately unity in the spectral range from visible to the long wave cutoff at 1.8 microns. Dark currents of a few microamperes are observed at room temperature, with a temperature coefficient of about -|-10 per cent per degree C. Both dark and light currents exhibit saturation in the range from 1 to 90 volts applied. The frequency response is flat into the 100-kc region. Short-circuit noise currents are observed around 20 Ai^a in a 1-cps band at 1,000 cps. The photocell element is encapsulated in a plastic housing ^i X Me X % inch in dimensions. 1 Bell Telephone Laboratories. ' Sandia Corporation. ABSTRACTS OF TECHNICAL ARTICLES 515 Interpretation of e/m Values for Electrons in Crystals. W. Shockley^ Letter to the Editor. Phys. Rev., 88, p. 953, Nov. 15, 1952. Transistor Electronics: Imperfections, Unipolar and Analog Transistors. W. Shockley^ I.R.E., Proc, 40, pp. 1289-1313, Nov., 1952. The electronic mechanisms that are of chief interest in transistor electronics are discussed from the point of view of solid-state physics. The important con- cepts of holes, electrons, donors, acceptors, and deathnium (recombination center for holes and electrons) are treated from a unified viewpoint as imperfections in a nearly perfect crystal. The behavior of an excess electron as a negative particle moving with random thermal motion and drifting in an electric field is described in detail. A hole is similar to an electron in all regards save sign of charge. Some fundamental experiments have been performed with transistor techniques and exhibit clearly the behavior of holes and electrons. The interactions of holes, electrons, donors, acceptors, and deathnium give rise to the properties of p-n junctions, p-n junction transistors, and Zener diodes. Point-contact transistors are not understood as well from a fundamental viewpoint. A new class of unipolar transistors is discussed. Of these, the analog transistor is described in terms of analogy to a vacuum tube. Unipolar ''Field-Effect" Transistor. W. Shockley^ I.R.E., Proc, 40, pp. 1365-1376, Nov., 1952. The theory for a new form of transistor is presented. This transistor is of the "field-effect" type in which the conductivity of a layer of semiconductor is modu- lated by a transverse electric field. Since the amplifying action involves currents carried predominantly by one kind of carrier, the name ''unipolar" is proposed to distinguish these transistors from point-contact and junction tj^jes, which are "bipolar" in this sense. Regarded as an analog for vacuum-tube triode, the unipolar field-effect traijs- sistor may have a m^ of 10 or more, high output resistance, and a frequency response higher than bipolar transistors of comparable dimensions. Control of Frequency Response and Stability of Point-Contact Transistors. B. N. SladeI. I.R.E., Proc, 40, pp. 1382-1384, Nov., 1952. The frequency response and stability of point-contact transistors are deter- mined to a large degree by control of the point-contact spacing and germanium resistivity. Stabihty is particularly important in amplifiers in which the im- pedances of the emitter and collector circuits are very small in the frequency range in which the transistor is designed to operate. Satisfactory stability has been obtained with developmental transistors having a frequency cutoff (3-db drop in the current amplification factor, alpha) ranging from 10 to 30 mc. These transistors operate under approximately the same dc bias conditions used with lower-frequency transistors, and have an average power gain of approximately 20 db. By means of the methods outlined, transistors which oscillate at frequen- cies as high as 300 mc have been made. ^ Bell Telephone Laboratories. 516 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Junction Transistor. M. Sparks^. Set. Am., 187, pp. 29-32, July, 1952. It is one of two forms of the remarkable device that amphfies electricity by the flow of electrons in a crystal. An account of its underlying principles and present state of development. Telephone Answering Services. L. R. Stang^ Telephony, 143, pp. 53- 55, 123-124, Oct. 25, 1952. Traffic Engineering Design of Dial Telephone Exchanges. J. A. Stewart^ Telephony, 143, pp. 13-16, 41-42, Oct. 18, 1952. Low-Drain Transistor Audio Oscillator. D. E. Thomas^. I.R.E., Proc, 40, pp. 1385-1395, Nov., 1952. A nine-element transistor audio oscillator is described. This oscillator operates with relatively low drain from a single 6-volt battery. The oscillator gives re- liable performance with an output uniform to approximately dbl db with sub- stantially all type 1768 point-contact transistors and without any circuit element adjustment required for variation in transistor parameters from unit to unit or with transistor ambient temperature. Trandstor Amplifier — Cutoff Frequency. D. E. Thomas^ I.R.E., Proc, 40, pp. 1481-1483, Nov., 1952. The effect of positive feedback through the internal base resistance of a transis- tor on circuit cutoff frequency is considered. Transistor Reversible Binary Counter. R. L. Trent^ I.R.E., Proc, 40, pp. 1562-1572, Nov., 1952. The feasibility of performing a fairly complex switching function using a few elementary transistor circuits is illustrated and experimentally verified. The specific function discussed is reversible vinary counting. The mechanism used to achieve reversibility and the circuitry within each building block is described. Operating margins and suggestions for design improvements for systems applica- tion are given. Effect of Electrode Spacing on the Equivalent Base Resistance of Point- Contact Transistors. L. B. Valdes^. I.R.E., Proc, 40, pp. 1429-1434, Nov., 1952. A theoretical expression for the equivalent base resistance rb of point-contact transistors is derived here. This expression is shown to check experimental values reasonably well if the severity of some assumptions made for purposes of analysis is considered. Electrode spacing, germanium-slice thickness, and re- sistivity of the semiconductor are shown to be the properties that affect rj, primarily. * Bell Telephone Laboratories. • Illinois Bell Telephone Company. ABSTRACTS OF TECHNICAL ARTICLES 517 Measurement of Minority Carrier Lifetime in Germanium. L. B. Valdes^ LR.E,, Proc., 40, pp. 1420-1423, Nov., 1952. A method for measuring the lifetime of minority carriers in germanium is described. Basically, it consists of liberating the carriers optically on a flat face of a crystal and measuring the concentration of minority carriers as a function of distance from the point of Uberation. The mathematical model is analyzed and experimental results are presented here. Drift Velocities of Ions in Krypton and Xenon. R. N. Varney^ Phys. Rev., 88, pp. 362-364, Oct. 15, 1952. (Monograph 2028). Drift velocities and mobihties of ions of Kr and Xe in their respective parent gases have been measured over a wide range of values of E/po , the ratio of elec- tric field strength to normahzed gas pressure. Two ions appear in each gas identified as Kr+ and Kr2+ in Kr and Xe+ and Xe2 in Xe. The relation that drift velocity varies as (E/po)^ at high E/po has been found to hold for the atomic ions and has been used to determine the equivalent hard sphere cross sections at high fields. The cross sections are 157 X 10"^^ cm^ for Kr and 192 X 10"^^ cm^ for Xe. The Langevin theory of mobilities gives excellent agreement with experimental results extrapolated to zero field strength provided that, in the theory, the hard sphere cross section is taken as large for the atomic ions and very small for the molecular ions. The range of the polarization forces is such as to render them insignificant in atomic ion colHsons and of primary importance in molecular ion colhsions. Junction Transistor Tetrode for High-Frequency Use. R. L. Wallace^ L. G. ScHiMPFi and E. Dickten^. I.R.E., Proc, 40, pp. 1395-1400, Nov., 1952. If a fourth electrode is added to a conventional junction transistor and biased in a suitable way, the base resistance of the transistor is reduced by a very sub- stantial factor. This reduction in r^, permits the transistor to be used at frequen- cies ten times or more higher than would other\\'ise be possible. Tetrodes of this sort have been used in sine- wave oscillators up to a frequency of 130 mc and have produced substantial gain as tuned amplifiers at frequencies of 50 mc and higher. Nature of Solids. G. H. Wannier^. Sci. Am., 187, p. 39 Dec, 1952. The theory that explains their various properties is a comparatively recent development of physics. From it practical benefits already begin to flow. Magnetic Double Refraction at Microwave Frequencies. M. T. Weiss^ and A. G. Foxi. Letter to the Editor, Phys. Rev., 88, pp. 146-147, Oct. 1, 1952. Stress Relaxation in Plastics and Insulating Materials. E. E. Wright^ A.S.T.M. Bull, 184, pp. 47-49, Sept., 1952. (Monograph 2024). 1 Bell Telephone Laboratories. 518 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 Organic materials are in ever-increasing use for mechanical and electrical devices where satisfactory performance is required over long periods.of time and under a wide assortment of atmospheric influences. Although the dimensional stabiUty of plastics and electrical insulating materials under no-load conditions has been estabUshed reasonably well, there is an important gap in existing knowledge with regard to a material's abihty to maintain adequate counter- stresses under compressive loading. A.S.T.M. Method of Test D 621 - 5P uses a constant load system and meas- ures the material's resistance to gross deformation. However, this fails to simu- late the usual appHcation where the material is subjected to constant deflection (such as fastening devices, inserts, etc.) and is required to maintain adequate counter-stresses for the foreseeable life of the part. Therefore, it has been neces- sary to integrate data from Method D 621 with long practical experience in order to extrapolate between two dissimilar systems. This paper describes a constant-deflection procedure for direct measurement of stress relaxation or change thereof, thereby permitting evaluation in terms of a material's abihty to maintain a tight assembly under conditions simulating actual use. Apparatus for carrying out the test is described and typical data illustrating its usefulness are included. i Contributors to this Issue Charles Clos, C.E., New York University, 1927; New York Tele- phone Company, plant extension engineering, valuation and depreciation matters, intercompany settlements and tandem and toll fundamental plans, 1927-47. Pratt Institute, Evening School, Mathematics Instruc- tor, 1946-49. Bell Telephone Laboratories, studies on development plan- ning for local and toll switching systems and research in switching probabiUty, 1947-. Member of A.I.E.E., New York Electrical Society, Mathematical Association of America, A.A.A.S., American Statistical Association, Iota Alpha, and Tau Beta Pi. R. H. CoLLEY, A.B., Dartmouth College, 1909; A.M., Harvard Uni- versity, 1912; Ph.D., George Washington University, 1918; Austin Teaching Fellow in Botany, Harvard University, 1910-12; Instructor in Botany, Dartmouth College, 1909-10 and 1912-16; Pathologist, Division of Forest Pathology, Bureau of Plant Industry, U. S. Department of Agriculture, 1916-28. Bell Telephone Laboratories, 1928-52. Dr. Col- ley was chairman of Conamittee 05 — Wood Poles, of the American Standards Association for nearly twenty years. He was president of the American Wood-Preservers' Association 1943-44. During his years with the Laboratories he worked particularly on development and research problems connected with material and preservative treatment specifica- tions for poles and other timber products used in outside plant. His more recent activities were directed toward improvement of laboratory tech- niques for evaluating wood preservatives, and toward the development of a coordinated plan for fundamental research on oil preservatives. He was Timber Products Engineer for the Laboratories from 1940 to 1950, and Timber Products Consultant from 1950 to 1952. His article in this issue of the Journal was prepared before his retirement on May 31, 1952. Karl K. Darrow, B.S., University of Chicago, 1911. He studied at the Universities of Paris and Berlin in 1911 and 1912, speciaUzing in physics and mathematics; Ph.D., University of Chicago, 1917. He then joined the staff of Bell Telephone Laboratories, at that time known as the Engineering Department of Western Electric Company. Here his 519 520 THE BELL SYSTEM TECHNICAL JOURNAL, MARCH 1953 work has included the study, correlation, and representation of scientific information for his colleagues, keeping them informed of current ad- vances made by workers in fields related to their own activities. As a corollary to his work. Dr. Darrow appears from time to time before scientific and lay audiences to lecture on current topics in physics and the related sciences. He has taken an active interest in education, teach- ing physics during summer and other sessions at Stanford, Chicago, and Columbia Universities and at Smith College. From 1944 to 1946, he served as consultant to the Metallurgical Laboratory in Chicago. Dr. Darrow is the author of Introduction to Contemporary Physics (1939), Electrical Phenomena in Gases (1932), Renaissance of Physics (1936), Atomic Energy (1948). He is a member of the American Physical Society, which he has served as secretary since 1941, the Physical Society of London, Soci^t^ Francaise de Physique, the American Philosophical Society, of which he was a counsellor for four years, and the International Union of Pure and Applied Physics, of which he was vice president from 1947 to 1951. In 1949 he received an honorary doctorate from the Uni- versity de Lyon. George R. Frost, Bell Telephone Laboratories, 1943-. Mr. Frost has had over twenty-three years of service with the Bell System. Since 1943 he has been an instructor at the Laboratories' School for War Training and in the Communications Development Training Program, until recently when he became a member of the Publication Depart- ment's group responsible for displays. From 1941-43 he taught communi- cations at Fenn College, and from 1946-47 mathematics at Pratt Institute. William Keister, B.S. in E.E., Alabama Polytechnic Institute, 1930; Bell Telephone Laboratories, 1930-32 and 1936-. As a member of the switching department, he is currently preparing text material and teaching general switching circuit theory and telephone switching sys- tems to members of the technical staff. Co-author of The Design of Switching Circuits, with S. H. Washburn and A. E. Ritchie (Van No- strand, 1951). Member of A.I.E.E., Eta Kappa Nu, Tau Beta Pi, and Phi Kappa Phi. Alistair E. Ritchie, A.B., Dartmouth College, 1935; M.A., Dart- mouth College, 1937. Bell Telephone Laboratories, 1937-. As a member of the switching development group, Mr. Ritchie tested panel and cross- bar circuits and made noise studies on panel and crossbar systems until CONTRIBUTORS TO THIS ISSUE 521 1942. He then became an instructor in the Laboratories' School for War Training. From 1945-51, he taught switching circuit design. Since 1951, in the switching engineering group, he has been working out new tech- niques for measuring telephone traffic in central offices. Co-author of The Design of Switching Circuits with W. Keister and S. H. Washburn (Van Nostrand, 1951). Member of the A.I.E.E. H. A. Stone, Jr., B.S., Yale University, 1933. Bell Telephone Labora- tories, 1936-. A member of the Transmission Development Department, Mr. Stone is in charge of a group engaged in the development of induc- tors and loading coils and cases. He previously assisted in the develop- ment of inductors and networks for use in military radio and telephone projects, and the design of radar pulse generators. Member of the A.I.E.E. Roger I. Wilkinson, B.S. in E.E., 1924, Professional Engineer (hon- orary), 1950, Iowa State College; Northwestern Bell Telephone Company, 1920-21; American Telephone and Telegraph Company, 1924-34; Bell Telephone Laboratories, 1934-43 and 1946-. U. S. War Department, Washington and South Pacific, 1943-45. Mr. Wilkinson has been en- gaged in applications of the mathematical theory of probability to telephone problems. Medal for Merit, 1946. Member of A.S.E.E. ; A.S.A. ; Institute of ^lathematical Statistics; American Society for Quality Control; Fellow, Operations Research Society of America; Associate Member of A.I.E.E.; and Member of Eta Kappa Nu; Tau Beta Pi; Phi Kappa Phi; and Pi Mu Epsilon. HE BELL SYSTEM nical ournal EVOTED TO THE SC I E N T I FIC^^^ AND ENGINEERINC SPECTS OF ELECTRICAL COMMUNICATION y^ ^U^^mmm j GLUME XXXII MAY 1953?^)^^^"^ ^NUMBER JUU^-^^" Solderless Wrapped Connections Introduction J. w. mcrae 523 Part I — Structure and Tools R. f. mallina 5%5 Part n — Necessary Conditions for ^Obtaining a Permanent Connection W. p. MASON AND T. F. OSMER 557 Part III — Evaluation and Performance Tests R. H. VAN HORN 591 An Improved Circuit for the Telephone Set a. f. bennett 611 Automatic Line Insulation Test Equipment for Local Crossbar Systems R. W. BURNS AND J. W. DEHN 627 Theory of Magnetic Effects on the Noise in a Germanium Filament harry suhl 647 DC Field Distribution in a ''Swept Intrinsic'^ Semi-Conductor Configuration R. C. PRIM 665 Transmission Properties of Laminated Clogston Type Conductors E. F. VAAGE 695 A Coupled Resonator Reflex Klystron e. d. reed 715 Abstracts of Bell System Papers not Published in this Journal 767 Contributors to this Issue "^"^^ COPYRIGHT 1953 AMERICAN TELEPHONE AND TELEGRAPH COMPANY THE BELL SYSTEM TECHNICAL JOURNAL ADVISORY BOARD S. BRACKEN, President, Western Electric Company F. R. K A P P E L, Vice President, American Telephone and Telegraph Company^ M. J. KELLY, President, Bell Telephone Laboratories EDITORIAL COMMITTEE E. I. GREEN, Chairman A. J. B U S C H F. R. L A C K W. H. DOHERTY J. W. MCRAE G. D. EDWARDS W. H. NUNN J. B. FISK H. I. ROMNES R. K. HONAMAN H.V.SCHMIDT EDITORIAL STAFF J. D. T E B O, Editor M. E. STRIEBY, Managing Editor R. L. SHEPHERD, Production Editor THE BELL SYSTEM TECHNICAL JOURNAL is published six times a year by the American Telephone and Telegraph Company, 195 Broadway, New York 7, N. Y. Cleo F. Craig, President; S. Whitney Landon, Secretary; Alexander L. Stott, Treasurer. Subscriptions are accepted at $3.00 per year. Single copies are 75 cents each. The foreign postage is 65 cents per year or 11 cents per copy. Printed in U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XXXII MAY 1953 nuubbr 3 Copyright, 1952, American Telephone and Telegraph Company Solderless Wrapped Connections Introduction By J. W. McRAE (Manuscript received February 9, 1953) In the telephone plant during the course of a single year, the operation of connecting a wire to a metal terminal is carried out approximately one billion times. Many of these connections are made in the factory. Others are made during the installation of equipment and a substantial number are made in the course of normal operation of the telephone plant. Successful functioning of the plant depends on trouble-free per- formance of each of these connections, most of which are now soldered in accordance with long-standing practice. Recently, a new technique for joining wires to terminals has been developed which will have im- portant technical and economic advantages in the Bell System and which should have similar advantages in other fields. The immediate need for a new connection arose with the develop- ment of the new wire spring general purpose relay.* In this relay the terminals appear in the form of closely spaced wires and the standard methods of applying connections were not satisfactory. Since the pro- duction schedule for these new relays required something like fifty million connections a year on relay terminals alone, an intensive effort was made to devise a satisfactory method for wiring. The first result was the development of a tool which could wrap a few turns of wire around the terminals of the relay, and do this efficiently on the closely- * Keller, A. C, A New General Purpose Relay for Telephone Switching Sj^stems. Bell System Tech. J., 31, pp. 1023-1067, Nov., 1952. 523 524 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 spaced wire terminals.t It was soon found that similar tools could be used to advantage on existing types of terminals, and the Western Electric Company is now making extensive use of wrapping tools. Con- nections made with these tools are being soldered after wrapping. In the meantime, further development work indicated the possibility that by wrapping the wire under tension around a properly shaped terminal, the need for soldering might be eliminated. Major economies appeared possible if such a solderless wrapped connection were applied extensively in wiring communication equipment. In addition, there would be freedom from trouble due to solder splashes in equipment and there would be an appreciable reduction in the consumption of tin for use in solder. Three papers in this issue of the Journal describe the present status of the program undertaken to exploit these possibilities. The paper on design indicates that practical tools for making solderless wrapped connections can be designed, built and used; the paper on analysis describes the basis for belief that the method is fundamentally sound; and the paper on evaluation indicates that the resulting connections are satisfactory for use in the telephone plant. Other types of tools capable of cutting, skinning and wrapping the wire in one operation are under development, and the problems pre- sented in adapting the basic techniques to other conductors, such as aluminum wire and stranded copper wire are being studied. In fact, a whole new area of development effort has been opened up. Thus the work reported in the following papers is of interest from two points of view. On the one hand it is a record of progress in the continual quest for less expensive and more reliable equipment. On the other hand, it is an example of the broad effects which are often the result of development aimed at a specific problem. The search for a solution to the problem of making connections to a new type terminal has led to a new approach to the whole problem of wire-terminal con- nections. t Miloche, H. A., Mechanically Wrapped Connectors. Bell Labs. Record, 29, pp. 307-311, July, 1951. Solderless Wrapped Connections PART I — STRUCTURE AND TOOLS By R. F. MALLINA (Manuscript received February 17, 1953) In the search for a better way of connecting wires to apparatus terminals a new joining method has been discovered. The new method not only elimi- nates soldering and its hazards but also reduces cost, improves quality and conserves space. In contrast to the solder joint which depends largely on human judgement and skill, the new connection is made with a calibrated tool. A degree of uniformity has been obtained which virtually eliminates the need for product inspection. The trend toward smaller apparatus and automation may now be further intensified due to the use of this new method of making electrical connections. INTRODUCTION Methods of joining wires to apparatus terminals for the purpose of electrical conduction can be broadly divided into two groups: solder connections and pressure connections. There are others such as welded and brazed connections; however, they are relatively few in number. The annual production of solder connections in the Bell System is esti- mated to be one billion. In television and radio manufacture the num- ber of connections made per year is in the order of ten billion. Because of the high cost of manual soldering, the pressure connection is of great importance to the comjnunication industry. One form of pressure con- nection — the solderless wrapped connection — will be described in this article. In order to determine the technical and economic value of a new type of pressure connection it is necessary to compare it with those now ac- cepted as good connections in the communication industry. A large por- tion of this article will, therefore, be devoted to the analysis of pressure connections some of which have been in use since the early development of the telephone. 525 526 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 WHAT IS A PRESSURE CONNECTION? The chart Fig. 1 shows six typical pressure connections classified in terms of seven requirements. In this classification the screw connection, for example, meets the following requirements: large contact area, high contact force, great mechanical stability, long life, easy to disconnect. The space, however, which the screw connection occupies is large and its cost is high. In the history of electricity it is probably the oldest and best pressure connection. In the second column of the table in Fig. 1 is the plug connection. It is small in size, easy to disconnect, but has no large contact area, no high contact force, no long life, no mechanical stabiUty and is not low in cost. As will be shown later the solderless wrapped connection in Colimin 6 of Fig. 1 is indicated as meeting all seven requirements. Its main advan- tage over the screw connection is that it is low in cost and small in size. CONTACT AREA The effective contact area relative to the cross sectional area of the wire is of great importance since it controls the resistance of the con- nection. It must remain uniform in size, metallically bright and not be affected by temperature changes, vibration and handling. Contact area is not easily defined. For example two flat metal sur- faces having an area of one square centimeter each and brought into contact do not necessarily have a contact area of one square centi- meter. If the force holding them together is small, only the high spots REQUIREMENTS 1. LARGE CONTACT AREA 2. HIGH CONTACT FORCE 3. LONG LIFE 4. SMALL SIZE 5. MECHANICALLY STABLE 6. EASILY DISCONNECTED 7. LOW COST FAHNESTOCK CLIP PLUG SOLDERLESS SCREW WRAPPED \^ \^ \^ v^ w' %^ »^ v^ %^ V-- v^ V- ^ »^ %^ Fig. 1 — Classification of pressure connections. SOLDERLESS WRAPPED CONNECTIONS — PART I 527 of the surfaces touch and large currents passing through such a connec- tion may develop heat and melt the metal at the high spots. CONTACT FORCE To make the above mentioned area of one square centimeter effective for electrical conduction it is necessary to press the two metal parts together with a force so high that essentially all particles of the area are intimately interlocked and free from insulating impurities. If the pres- sure is high enough, the film which appears in the form of oxide on the terminal surface is crushed. In general it is assumed that in a good con- nection the contact force should be such that the contact area produced is equal to or greater than the cross sectional area of the wire. In screw connections, crimped connections and wrapped connections the contact area is normally a multiple of the wire cross sectional area. In plug con- nections, such as on vacuum tube sockets, the contact area is very small. In a Fahnestock clip for example, the contact area is about one quarter of the cross sectional area of the wire. LIFE If the electrical resistance of a pressure joint is to remain constant with time, it is the contact area which must remain substantially con- stant, but not necessarily the contact force. Once the metal particles are tightly interlocked a subsequent reduction in contact force within rela- tively wide limits does not change the electrical resistance. The resist- ance will increase only when the force is reduced to such a low value that vibration and handling cause partial separation of the contact area. In such a case two changes may take place : 1. The atmosphere may enter through the fringe of the contact area and a process of corrosion may begin. 2. The effective contact area may be reduced through dislodging some of the contacting particles. In both cases the resistance is increased. Therefore, to produce a durable connection it is important to have a firm joint and one such that the atmosphere cannot enter the contact area. The term commonly used for such a joint is ''gas tight." ELASTIC RESERVE The question now arises how much reduction in contact force can be tolerated before a joint loses its gas tightness? In all types of pressure 528 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 F F SPRING ANALOGUE Fig. 2 — Screw tightened and wire compressed (with accompanying "Spring Analogue")- connections, the forces which hold the wire and terminal together are provided by the springiness or elasticity of the materials. The elasticity in a Fahnestock clip is quite apparent because of the long spring mem- ber. On the other hand, a screw connection, such as shown in Fig. 2, does not appear to have spring members of any kind. Analysis, however, shows that there is considerable elastic deformation. Most of this elas- tic deformation is in the elongation of the screw shank and there is also some bending in the screw head and some compression in the screw threads. When the screw is tightened, the wire which is interposed be- tween screw head and nut is also elastically deformed. Since in most electrical connections the wire is a soft material such as copper or alu- minum, it is nearly always compressed beyond the yield point and only the recovery of the overstressed material can be considered as elastic reserve. To determine the usefulness of a connection and compare pressure connections of different kinds, the elastic reserve in the deformed wire and the deformed terminal must be measured or computed. Elastic re- serve might be expressed either in terms of stiffness or the potential energy stored in the system. Stiffness S is defined as the ratio of the ''applied force F to the elastic return Z)"; potential or elastic energy E is "one half of the product of the force F and the elastic return D." {E = H™.) Example: A wire is placed under a screw (Fig. 2) and compressed by the screw head to a thickness Di . The screw is then loosened so that it just touches the wire. The wire now has expanded to a certain extent and its new thickness is D2. (Fig. 3.) The difference (D2 — Di = Dw) is the elastic return and the ratio F/Dw = Sw is the useful stiffness of the wire. A preferred way of expressing elastic reserve is in terms of stored energy. (Strictly speaking the equation E = J^FD holds only for springs with constant stiffness. A round wire compressed by a screw head becomes stiffer as the compression increases). The distance, Ds = Di — Dir , is the elongation of the screw (see Fig. 3). The total energy SOLDERLESS WRAPPED CONNECTIONS PART I 529 stored up is therefore the sum of the energy in the screw and wire (E = Es-{- Ew). Screws in terminal blocks are normally made of hard materials such as brass or phosphor bronze. Wires used for the interconnection of com- ponents are nearly always of a soft material and have a tendency to creep. If creep takes place in the wire during the many years a screw connection is in use, it is advantageous to have the loss of potential energy in the wire compensated for by the energy stored in the screw. F=o F=o SPRING ANALOGUE Dw= D2-D, Ds = D3-Dw Fig. 3 — Screw loosened and wire not compressed. The recovery of the wire is denoted Dw. The recovery of the screw is Dz — Dw (with accompanying "Spring Analogue"). A screw, for example, made of soft copper would not be expected to make a lasting connection. If on the other hand the screw is made of a material which has little creep and much elasticity, such as brass or steel, it would act as a spring member and tend to keep the connection tight. Several typical screw connections were measured to determine the elastic reserve. It was found that on an average the potential energy stored in the screw is about equal to that stored in the wire. Plastic flow of the wire creates an effective bearing area comparable to the area of the screw shank. THE SOLDERLESS WRAPPED CONNECTION The detailed analysis of the screw connection as an introduction to the solderless wrapped connection was necessary not only because the screw has such wide use as an electrical pressure connection but chiefly because of its proven value as a (durable connection. When new types of pressure connections are put into large scale production, the question invariably arises. What is their life? While considerable analytical work has been done on the cold flow of metals under stress* and while certain See Part II. 530 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 theoretical predictions can be made on the durabiUty of new connec- tions, it affords additional satisfaction to be able to show that the solder- less wrapped connection is in many respects similar in structure and performance to the conventional screw connection. If then this fact is supported by parallel analytical work, there should be little doubt that the solderless wrapped connection is a durable pressure connection. THE RECTANGULAR TERMINAL Generally speaking the terminal best suited for a wrapped connection is a terminal of rectangular cross section. It is an inexpensive terminal since it can be blanked from sheet stock or coined from round wire. It is ideally suited for a pressure connection because the edges produce a concentrated high pressure on the wire. The stress distribution in the wire produced by the terminal edges is shown diagrammatically in Figs. 4 and 5. If the wire is wound with high tension around the rectangular terminal, the terminal edges dig into the soft copper wire, crush and shear the oxide on both the wire and the terminal and form a large, intimate and metallically clean ''gas tight" contact area. An indication of the high pressure is the crushing of the hard nickel silver terminal edge by the soft copper wire. A pattern of contact areas on both wire and terminal is shown in Fig. 6. Several turns of wire are required to preserve the high contact force. In general it is assumed that the first and last two edges around which the wire is wrapped do not contribute much to the joint as contact areas. A seven-turn wrapped connection on a rectangular terminal thus has six effective turns. Each turn contacts four edges or a total of twenty-four contact areas for six effective turns. /MEDIUM TENSION MAXIMUM TENSION "" TERMINAL MAXIMUM/ COMPRESSION TERMINAL Fig. 4 — Stress distribution along one-quarter turn of wire. Fig. 5 — Cross section through terminal edge showing stress distribu- tion in the wire. SOLDERLESS WRAPPED CONNECTIONS — PART I 531 Pattekx of Contact Are Contact Areas Enlarged. The Soft Copper Wire Crushes the Hard Nickel Silver Terminal Edge IN the Wrapping Process. Fig. 6 — Contact areas. 532 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 WHAT IS A GOOD CONNECTION? The quality of a connection depends fundamentally on two factors: the contact area and the contact pressure. As long as there is sufficient pressure and the atmosphere cannot enter the joint, the connection is considered a good one. If, however, the elastic energy which holds the two surfaces together is small, various disturbances may cause a partial separation of the interlocking metal particles and thus effect a change in resistance. For normal telephone applications a good connection may, therefore, be defined as one which not only has sufficient contact area and contact pressure but which also has sufficient elastic reserve to maintain contact area and contact pressure throughout the desired life, which may be forty years or more. The mechanical disturbances to which a connection may be subjected are: handling, vibration, temperature changes and cold flow. HANDLING AND VIBRATION The solderless wrapped connection is well protected from the point of view of handling and vibration. The locking effect on the rectangular terminal or a terminal having well defined edges does not permit loosen- ing of the center turns from the terminal. In vibration tests where con- ventional soldered connections were compared with solderless wrapped connections, it was found that solderless wrapped connections outlast soldered connections. This is due to the fact that a sudden change in cross section from wire to solder lump localizes the stresses at a very small area. (See Figs. 7 and 8.) In the screw connection a similar con- dition exists where the wire emerges from under the screw head. In the CLIPPING 0.125' Fig. 7 — Standard solder connec- tion of U-relay. TAPERED STIFFNESS Fig. 8 — Solderless wrapped con- nection of modified U-relay terminal. SOLDERLESS WRAPPED CONNECTIONS — PART I 533 solderless wrapped connection there is no sudden change in cross section and therefore no locahzation of stresses. The term commonly used to indicate the gradual change in rigidity of the wire as it approaches the anchoring point is "tapered stiffness." (See Fig. 8.) HEAT AND COLD FLOW When a pressure connection is subjected to high temperatures, which may be due to large current or to heat transfer from adjacent com- ponents, the pressure at the joint is relaxed. This is true in the solder- less wrapped connection as well as in the screw connection. The same process of relaxation takes place in normal temperature with time. The relaxation of pressure with temperature and time will be shown in an- other part of this paper. Under ordinary conditions the relaxation of pressure in a solderless wrapped connection is not sufficiently large to indicate any change in resistance during a forty-year life. Furthermore, as Mason and Osmer point out in their paper, solid state diffusion takes place as time goes on. This process strengthens the joint mechanically and improves it electrically. QUANTITATIVE EVALUATION OF ELASTIC RESERVE Because of the above mentioned disturbances to which a pressure connection may be subjected, it is important to know how much elastic reserve is stored in a connection. If no potential energy were stored in the wire and in the terminal, no contact pressure would be produced. If little potential energy were stored in a connection, a slight change in temperature due to differential expansion of the metals would loosen the connection. The same would be true with vibration and handUng. A rough comparison with other pressure connections will serve to illustrate how much elastic reserve a solderless wrapped connection has to have in order to withstand the disturbances to which it may be sub- jected. The best known pressure connection, and the most universally used, is the screw connection. On a No. 4 screw (0.112"), the force ex- erted in clamping the No. 24 gauge (0.020") wire is about 135 lbs. The elastic energy is stored by compressing the wire and by elongating the screw shank. Similarly, in a solderless wrapped connection as shown in Fig. 9 a total force of 90 lbs is exerted on the edges of the terminal (24 corners) . Here the greater part of the energy is stored in the terminal which receives torsional as well as compressional stress from the tension in the wrapped wire. (See Figs. 10(a) and 10(b)). 534 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig. 9 shows in diagrammatic form how the stored energy in a screw connection and a solderless wrapped connection compare. A typical solderless wrapped connection — seven turns of 20-mil copper wire wound with 1300 grams applied force on a 0.0148" x 0.062" nickel silver terminal — has approximately 3 mil pounds of stored energy E. (2.4 mil pounds are stored in the terminal and 0.6 mil pounds in the wire). The screw connection — No. 4-40 screw (0.112") tensioned to 135 lbs on 20-mil copper wire — has approximately 2.7 mil pounds of stored energy E. (Approximately half in the screw and half in the wire). In the screw connection the energy is about equally divided between the screw and the wire whereas in the solderless wrapped connection a larger part of the energy is stored in the terminal. This is advantageous since the hard materials of which terminals are generally made have less cold flow than copper. In a solderless wrapped connection, if the NO. 4-40 (0.112") BRASS SCREW N0.24GA (0.020") COPPER WIRE CONTACT FORCE 135 LBS (a) SCREW CONNECTION 0.0148" X 0.062" NICKEL SILVER TERMINAL 7 TURNS N0.24GA (0.020") COPPER WIRE (1300GR AF) CONTACT FORCE 90 LBS (24 CORNERS) (b) SOLDERLESS WRAPPED CONNECTION 2.4 SCREW 0.6 WIRE TENSION COMPRESSION 18,750 PSI 18,250 PSI E = 1,4 + 1.3 =2.7 TORSION AND TENSION COMPRESSION 8500 PSI E = 2.4 + 0.6 = 3 ( C I TOTAL ENERGY E (IN lO'^ INCH LBS) Fig. 9 — Elastic energy stored in screw connection and in solderless wrapped connection. SOLDERLESS WRAPPED CONNECTIONS PART I 535 terminal size is changed to 0.020'' x 0.062'' there is considerably less energy stored in the terminal and slightly more in the wire. Inasmuch as a screw connection in most cases depends on the human element, that is the amount of torque applied by the operator, it can be expected that some screw connections will be made with a force that may vary from 75 lbs to 150 lbs. The wrapped connection on the other hand, being made wil h a calibrated tool, can be expected to give substantially the same contact force at all times. In order to understand more clearly how the wire and the terminal interact when they are under mutual stress and exposed to heat, the elastic deformation of the wire and the terminal must be analyzed. It has been sho^Mi in Fig. 4 that the wrapped wire on the four sides of the rectangle is under tension. This tension causes the terminal to twist. If instead of a helix the terminal were surrounded by a series of hoops, NO CONTACT FORCE -^ CONCENTRATED _^ CONTACT FORCE STIFF TERMINAL INDENTATIONS AND CONCENTRATED CONTACT FORCE COMPRESSION RELATIVELY SOFT WIRE TURNS LONGITUDINAL TENSION IN A WIRE CAUSES TERMINAL TO TWIST (b) MODEL MADE WITH A RUBBER TERMINAL WRAPPED WITH RUBBER TUBING EMPHASIZES THE TWIST PRODUCED BY THE LONGITUDINAL TENSION IN THE TUBING Fig. 10 — (a) Longitudinal tension in wire causes terminal to twist, (b) Model made with a rubber terminal wrapped with rubber tubing emphasizes the twist produced by the longitudinal tension in the tubing. 536 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig. 11 — Twist of rectangular terminal (100 turns of wire). the terminal would be compressed at the edges but the terminal would not twist. The terminal twist in the wrapped connection is therefore due to the fact that the terminal is surrounded by a helix and not by hoops. Figs. 10(a) and 10(b) show that a left-hand helix produces a right-hand twist in the terminal. As will be shown later, this visible deformation of the terminal is being used to determine the tension in the wire. The twist in the terminal of a wrapped connection with many turns can readily be seen in Fig. 11. For example an initial twist of 46° is produced in a nickel silver terminal 0.0148" X 0.062" wrapped with 100 turns of No. 24 (0.020" dia.) copper wire with an applied force of 1300 grams. One way to visualize the behavior of the wire and the terminal when wrapped under tension, exposed to time and heat and then unwrapped, is to represent the wire and the terminal by linear springs. This is shown schematically in Fig. 12. Position 1 represents both wire and terminal before wrapping. Position 2 represents the wire wrapped on the ter- minal. Position 3 is the same as Position 2 except that the wrapped terminal has been exposed at room temperature (20°C) for eight days. This causes the terminal twist to relax from 46° to 39°. Positions 2 and 3 are analogous to the wire under tension and the terminal under tor- ROOM TEMPERATURE V/////////////////////////////////. WIRE TINNED COPPER 24 GA (0.020") 100 TURNS APPLIED FORCE OF 1300 GRAMS TERMINAL SET TERMINAL NICKEL SILVER 0.0148" X 0.062" TENSIONED P0S.4 RELEASED V/////////////////////////A PCS. 5 PCS. 6 POS.7 TENSIONED AFTER RELEASED HEAT (AFTER HEAT) Fig. 12 — Energy in unheated connection proportional to 30°. Energy in heated connection proportional to 14°. SOLDERLESS WRAPPED CONNECTIONS — PART I 537 sion. The force WF is the tension in the wrapped wire. This force can be determined by dividing the torque necessary to twist the terminal by the effective moment arm. Since the elongation of the wire cannot readily be measured, the terminal twist was chosen to determine the force exerted at the terminal edge. The 39° terminal twist shown in Posi- 80 0.0148" X 0.062" NICKEL SILVER TERMINAL WRAPPED WITH 100 TURNS OF 24 GA (0.020") TINNED COPPER WIRE 600 800 1000 1200 1400 1600 1800 APPLIED WRAPPING TENSION, AF, IN GRAMS 2000 2200 2400 Fig. 13 — Angle of twist in terms of applied wrapping tension. tion 3, however, cannot be used for determining the force since the ter- minal may be overstressed as is shown in Position 4. Instead of return- ing 39° the unwrapped terminal returned only 30°. In other words the terminal has taken a set of 9°. Fig. 12 illustrates the deformation of wire and terminal only for one value of applied tension, namely 1300 grams. If the angle of twist is measured for applied tension ranging from 100 to 2400 grams, a set of curves is obtained as shown in Fig. 13. Curve A shows the angle of twist immediately after the terminal is wrapped. Curve B shows the relaxation after eight days aging at room tempera- ture. Curve C represents the terminal set. The value between Curves B and C is the elastic reserve. For 1300 grams applied tension the elastic reserve is expressed as 30° reserve twist. Using the before mentioned ratio of torque and moment arm, the force WF can now be determined. The torque required to twist the terminal 30° is 37.2 inch grams. (See Fig. 14.) The effective moment 538 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 40 50 60 TORQUE IN INCH-GRAMS Fig. 14 — Torque required to twist the unwrapped terminal. 100 0.020 250 500 750 1000 1250 1500 1750 2000 APPLIED WRAPPING TENSION, AF, IN GRAMS Fig. 15 — Moment arm in terms of tension. 2250 2500 SOLDERLESS WRAPPED CONNECTIONS PART I 539 arm A^ , which decreases a slight amount as the wrapping tension AF increases (see Fig. 15), is equal to 2ah/{a + 6)*. Therefore, Ae = 2 (0.042'' X 0.0242'0/(0.042'' + 0.0242'0 = 0.0307 in. Thus the tension in the wire WF = T/Ae = 37.2/0.0307 = 1210 grams. Using the re- covery angle of the terminal as a measure of force, the tension in the wire can now be plotted in terms of angular twist. This is shown in Fig. 16. It should be noted that the tension in the wire WF (wrapped 2600 2400 2200 2000 1800 0.0148"X 0.062'' NICKEL SILVER TERMINAL WRAPPED WITH 100 TURNS OF 24 GA (.0.020") TINNED COPPER WIRE _ ^ ' AT 1800 GRAMS AF THE WIRE BEGINS TO BREAK WHEN UNWRAPPING ^ ^ y^ 1710 J y ^ 1600 < a. O 1400 z 1 ^^ ky^ ^ 1 1 1 > \y 1300 r ^ ^ -^ ^ 1200 cc o A > 121C ) / stl<>^ ^ 1 1 800 600 400 200 0 / / 1 1 1 CF CONTACT FORCE WF TENSION IN WRAPPED WIRE AF TENSION APPLIED WHEN WRAPPING / '/ y 1 1 ^y / / 1 1 1 > / 1 I37- ^^ 1 1 25 30 35 40 45 ANGLE OF TWIST IN DEGREES Fig. 16 — Forces in terms of angular twist. force) is not directly proportional to the wrapping tension AF (appUed force). The reason for this is that at low applied wrapping tension the bending of the wire around the corner of the terminal produces an addi- tional increment of tension. For example at an angle of 15° the wrapped tension WF is nearly twice as high as the applied tension AF. The wrapped tension WF and the applied tension AF are about equal when the angle of twist is 33°. At higher values of applied tension, the wrapped tension increases at a much lower rate. This is caused by the terminal taking a set. (See Fig. 13.) At 1300 grams of applied tension, which is the reconamended wrapping tension for No. 24 copper wire, the wrapped tension is 1210 grams. See Appendix I. 540 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 50 L46*'WHEN wrapped 40 7° RELAXATION AT20OC <0 LU \° 39°" 1 1 UJ Z < in \ ^175°C ^^--^ 30° (100%) """"**—- 1 -23° ^ A ' ELASTICI 1 RESERVE ^14°— 470/0 J 1 L_ _ 1 _ J. I L T_ 5 0 ~r -} f |105 i 40 60 80 100 TIME IN MINUTES Fig. 17 — Relaxation of wrapped wire after heating for three hours at 175°C. The reserve twist of 14° is an indication of the tension left in the wire which amounts to 47 per cent of the original wrapped tension. A severe test for a wrapped connection is heating to 175°C for three hours. This relaxes, about half the stress and is considered the equiva- lent of a 40-year life at 135°F. Position 5 in Fig. 12 is the same as Posi- tion 3, that is, the wire has just been wrapped onto the terminal and its tension is 1210 grams. If this connection is now heated to 175°C and the angle of twist noted every fifteen minutes, a curve is obtained as shown in Fig. 17. It should be noted that at 105 minutes the curve is for all practical purposes asymptotic at an angle of 23°. If the heated connection is cooled and unwrapped and the set in the terminal meas- ured, it is found to be 9°. The 14° difference is a measure of the elastic reserve. This is 47 per cent of the wire tension before heating. The cor- responding tension WF is then 570 grams. This process is illustrated in positions 6 and 7 of Fig. 12. Position 7 shows that the terminal set of 9° was the same as before heating. A similar experiment was made with formex insulated wire wrapped on a nickel silver terminal. Instead of subjecting the connection to the heat of an oven, a high current was passed through the wire. Essentially the same curve as shown in Fig. 17 was obtained. To further check the behavior of springs with complex elastic defor- mation such as in a wrapped connection on a terminal having edges, measurements were also made with simple helical springs tensioned SOLDERLESS WRAPPED CONNECTIONS — PART I 541 within the yield point. Two springs, one of nickel silver wire and the other of copper wire having a stiffness ratio of 5 to 6, were coupled in series (Fig. 18), tensioned to 30 grams and then heated to 173°C for two hours. The tension left after heating was 13 grams or 43 per cent of the original tension. The tension decay curve was similar to that shown for the wrapped connection. (See Fig. 17.) This test shows that in spite of the complex deformation of the wire at the corners of the terminal there is substantial agreement in results of the measurements ob- tained — namely 43 per cent remaining stress in the case of the helical spring and 47 per cent for the wrapped connection. STRESSES IN THE FINISHED CONNECTION Having determined the interacting forces in the solderless wrapped connection, the next questions of primary interest are — what are the stresses in various parts of the connection and what will happen to these stresses in forty years? Most of the elastic energy stored in the wire is in the portion marked ''Medium Tension" (Fig. 4). Here the stress is about 8,500 P.S.I. This assumes that a 20-mil wire is wrapped with 1300 grams applied force. The wrapped force or the useful force obtained from the elastic reserve is then 1210 grams. The stresses at the corners are not easily determined because they are not uniform. This is shown in Fig. 5. As can be seen, the point of highest A 'A B NICKEL SILVER COPPER _-A,— >)^--ai— >K li > |<— 100%— 1— ' T, = 30 GM untensioned'I BEFORE yy , HEATING TENSIONED xk TENSIONED NTENSIONED AFTER HEATING |wvvwvw\4aaa/^«/vv\a/VvV|aavwvv\/^ D ^VWVWWWVJAf I JMAMAjWWWvVWp " __^ |<— ELASTIC RESERVE 43% Fig. 18 — Tension Ti in coupled springs after heating for two hours at 173°C is about 43 per cent of the original tension Tx- (Tension Tx approximately 10 per cent below the yield point). 542 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 12 xlO^ 1 < 10 ^10,500 PSI ^'^4,,,^ 1 ^^ 20° C 8 " 't i I 6 4 2 ■ 0 8500 PSI 175° C 4000 PS y\ 2 3 4 5 6 7 IME IN DAYS AT 20 DEGREES C 8 0 0.5 1.0 1.5 2.0 2.5 3.0 TIME IN HOURS AT 175 DEGREES C Fig. 19 — Stress relaxation in wire. concentration is in the center of the contacting area. From this point to the periphery there is a pressure gradient which is similar to that of a circular compressed thin film of viscous material. At the boundary line the pressure is zero. The average pressure within the contact area is about 29,000 P.S.I. but the maximum stress in the center of the contact area may be as high as 100,000 P.S.I. The relaxation, that takes place in eight days at room temperature (Fig. 13), is assumed to be due to the very high initial stress in the center of the contact area which seeks equalization. 1.20 1.00 S 0.80 u a. uu a ^ 0.60 111 a. 0.40 u 0.20 1 4,72C PSI — — " "■ / ELECTROLYTIC TOUGH PITCH COPPER ANNEALED-0.040 MM 80° F ' ^ — •10,6 90 P W / 835 0 PS 1 -— ^ — 1 1 1 >--4270 PSI 14 16 18 20 TIME IN MONTHS Z2 24 26 28 30 32 34 Fig. 20 — Creep curves of annealed copper for various stresses. (Courtesy of Chase Brass and Copper Co., Waterbury, Conn.). I SOLDERLESS WRAPPED CONNECTIONS PART I 543 Summarizing the stresses in the connection, one may therefore say that in the portion of the wire where most of the elastic energy resides, the stress after eight days is about 8,500 P.S.I, and at the points of con- tact 29,000 P.S.I. After forty years these stresses will be approximately 4,000 and 13,500 P.S.I., respectively or 47 per cent of the original stress. (See Fig. 19). As may be seen in the creep curves shown in Fig. 20, a stress of 8350 P.S.I, reaches a creep value of about 0.07 per cent in three years and from then on for all practical purposes ceases to creep. 50 20 ^ ^_ A :57°C 100 TURNS OF NO. 24 GAUGE (0.020" ) BARE TINNED COPPER WIRE \ ^ \. ^ WRAPPED ON 1/16" WIDE TERMINAL WITH APPLIED FORCE OF 1300 GRAMS \ s \ N \1 w:75o"c V > \ \ -!^ \. \ 'V,''' 's -"^ ^"^>^ \ K^ c j 175" C 1 1 200° C v^ 0.014 3" THICK NICKEL^ ! < SILVE R TERMINAL TWIST 0.0124" THICK SPRING 1^ MEASUREMENTS 1 STEEL TERMINAL j j COMPUTED BY MASON & OSMER FROM MEASUREMENTS OF DASHED CURVES 1 1 3 HOU 1 1 1 40 YEARS — H 1 i 1 MIN 10 MIN I HR 10 HR 1 DAY 10 DAYS 100 DAYS I YR 10 YRS lOOYRS TIME Fig. 21 — Stress relaxation in wire plotted on a logarithmic time scale. The effect of time and temperature on the longitudinal stress in the wire can better be seen by curves plotted on a logarithmic time scale (Fig. 21). The initial stress after transient relief of three days is con- sidered as 100 per cent. Curve A shows that at a temperature of 57°C (135°F) — which is the maximum temperature that solderless wrapped connections will be subjected to — the longitudinal stress relaxes ap- proximately 50 per cent in 40 years. To reach the 50 per cent value at a temperature of 175°C takes approximately three hours (Curve C). Curves B and C show that the relaxation in the wire is essentially the same for either nickel silver or spring steel terminals. METHOD OF WRAPPING In nearly all soldered connections where a wire is to be joined to a terminal the procedure is as follows: The operator takes the skinned 544 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig, 22 — Air-driven wiring tool. end of an insulated wire, hooks or threads it onto the terminal and ap- plies solder. The hooking or threading is important because it is cus- tomary in production to attach several wires at first and then solder. The question now is: If a similar procedure is to be followed with a wrapped connection where the wire must surround the terminal wdth a high pressure, how do we produce that pressure? In a screw connection a force is obtained by a high lever ratio. In a crimped connection the force is applied by heavy and powerful com- pression tools. These tools are not suitable for connecting wires to closely spaced terminals such as shown in Fig. 22. To produce high tension in the wire while it is being wrapped onto the terminal, a new method of tensioning had to be devised. In the manufacture of helical springs it is customary to anchor the end of the wire in a hole in the arbor and tension the wire with a friction pad. By rotating the arbor a helical spring is produced. For closely spaced terminals this method is not practical as the wire cannot be fed tangentially to the terminal and the terminal cannot be rotated. A new SOLDERLESS WRAPPED CONNECTIONS PART I 545 wire connecting concept was proposed whereby a rotating spindle housed a stationary terminal in an axial opening in the spindle and was pro- vided with a second opening radially separated from the axial opening and arranged to accommodate a wire. When the spindle was rotated the wire was caused to form a spiral about the stationary terminal. One method involved anchoring the wire in the second opening and feeding the wire tangentially to the terminal as the spindle was rotated. Due to certain limitations inherent in tangential feed onto a stationary ter- minal an improved method was finally chosen. This is the axial feed method which is particularly adapted to wrapping closely spaced ter- minals of all cross sections. The operation of loading the wire and wrap- ping the connection is shown in Fig. 23. Position A shows the tool tip, Position B the bare wire 2 inserted into the feed slot 4, Position C the anchoring of the wire by bending it into the notch 5, Position D the ter- minal insertion and Position E the wrapping of the wire 2 by rotating the spindle 1 around the terminal 3. Position F is the finished connec- tion. A more detailed drawing of the tool tip is shown in Fig. 24. WRAPPING TENSION The tension in the wire is produced by rotating the spindle 1 around the terminal 3 (Fig. 24) thus pulling the short skinner wire 2 out of the feed slot 4. In the process of pulling the wire out of the slot and wrap- ping it around the terminal each increment of the skinner wire length undergoes several bending operations. The first bending occurs at the edge R of feed slot 4 where the wire is bent through an angle of less than 90°. The second bending is the straightening out operation of the bent wire. The third bending takes place as the wire is wrapped around the terminal. All three bending processes contribute to the tension with which the wire is wrapped. The dimensions which control the tension and are therefore of engineering importance are the radius R at the tool tip (See Figs. 25 and 25(a)) and the wall thickness W (Fig. 24). The bend- ing forces are inversely proportional to the respective bending curva- tures and the frictional forces in turn are proportional to the bending forces. The tension imparted into the wire as it is wrapped around the terminal, however, is not only due to the friction alone but to the com- bined effect of friction and bending effort. If the wire were completely elastic and the friction zero, no tension could be produced. But there would be tension in the wire if the friction were zero and the wire only partly elastic such as copper wire. There also would be tension if a com- pletely elastic wire would be pulled around an edge having friction. 546 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 .^^^^ (b) (C) lihiiii^ir'rfifiiffi (f) Fig. 23 — The wrapping process. A — Tool Tip. B — Wire Inserted. C — Wire Anchored. D — Terminal Insertion. E — Wire Wrapped. F — Finished Connec- tion. SOLDERLESS WRAPPED CONNECTIONS PART I 547 1 ^6 STATIONARY SLEEVE (ROTATING SPINDLE) Fig. 24 — Method of wrapping the skinner wire on a rectangular terminal. CLOSELY SPACED TERMINALS When the tenninals are closely spaced the stationary sleeve 6 and the anchoring notch 5 are used in order to anchor the first turn of wire to the terminal. (See Fig. 26.) However, when the terminals are not closely spaced the sleeve and notch are desirable but are not necessary since the insulated portion of the wire can be held by some means external of the tool at an angle of approximately 90° with respect to the tool spindle. The high acceleration of the wrapping motor produces a mass reaction of the wire leading up to the terminal. This counterforce cou- pled with a slight tension of the supply wire applied by the external RADIUS, R »► Fig. 25 — Wrapping tension is controlled by edge radius. 548 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 WRAPPED TERMINAL ^ '0.050" ■0.090 Fig. 26 — Space occupied by wrapping tool between closely-spaced terminals. wire guiding means is sufficient to insure wrapping of the first turn. The following turns need no further anchor as the first turn locks the wire to the terminal. The trend toward making circuit components smaller is now marked in all branches of communication engineering. With a tool tip such as Fig. 27 — Forty-four point terminal block, 132 connection capacity, only 48 connections of 26 gauge (0.0159*) wire shown. Occupies Ij^i" x %* x %6* or 3^ cu in of space. SOLDERLESS WRAPPED CONNECTIONS PART I 549 Fig. 28 — Double connection. sho^\^l in Fig. 24, it is possible to wire apparatus having a terminal spacing as close as 2J^ times the terminal width. (See Fig. 26.) A terminal block 13-^" by ^" by ^g" having 44 terminals is shown in Fig. 27. The cables shown contain forty-eight No. 26 (0.0159" dia.) wires all wrapped on the terminals. Each terminal is capable of accommodating three wires or a total of 132 connections may be made in an area of less than one square inch. An enlarged view of a double connection is shown in Fig. 28. REMOVAL OF CONNECTION The solderless wrapped connection may be removed from its terminal by two methods. The most convenient method is by stripping. Two types of tools may be used for this purpose. The specially formed jaws of a pair of pliers are hooked in the back of the connection as shown in Fig, 29. By applying a force the connection may be stripped off. The other tool for stripping is shown in Fig. 30. The stripping force varies with the tightness of the wrapping and is plotted in teims of applied 1 1 1 \ --^v?""'^^ ^ g _,^ 2 ^^^^^^^-^ F wraj: [G. 29 — Stripping of solderless ped connection. Fig. 30 — Stripping a solderless wrapped connection from a terminal. 550 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 wrapping tension in Fig. 31. Another method of removing a connection is by unwinding the helix. This may be done by using a pair of pliers as shown in Fig. 32. Either end of the wire may be used for unwrapping. A terminal is not seriously damaged by stripping off a wrapped wire, however, the re-use of the stripped off wire is not recommended. A wire may be reconnected by skinning to the proper length and wrapping. When the wire is not sufficiently long to provide the necessary number of turns to insure a good connection, one or two turns may be wrapped and then soldered. CONNECTION OF LARGE AND SMALL WIRES There is no upper limit to the size of wire wrapped on adequately proportioned terminals. Connections have been made with both alu- minum and copper wire over 200 mils in diameter with satisfactory re- sults. The torque necessary to wrap large wire is considerable, since it increases with the third power of the diameter. A 20-mil wire requires a winding torque of 100 inch grams whereas, a 200-mil wire requires 100,000 inch grams (18 foot pounds). Wires as small as No. 39 (0.0035'' dia.) may also be wrapped, however, the design of the wrapping tool must be changed slightly in order to facihtate the loading of the fine wire into the tool. DIMENSIONAL RELATIONS The data given in this paper refer only to No. 24 copper wire 20 mils in diameter. The terminal width ♦most frequently used in conjunction with this size wire is about one-sixteenth inch or three times the wire diameter. The terminal width may also be twice the wire diameter or slightly less, however, the one-sixteenth inch size has been chosen for APPLIED WRAPPING TENSION, AF »• Fig. 31 — Stripping force in terms of applied wrapping tension. Fig. 32 — Unwinding wrapped con- nection with pliers. SOLDERLESS WRAPPED CONNECTIONS PART I 551 (a) SQUARE TERMINAL (d) COINED AND FLATTENED WIRE TERMINAL (b) U TERMINAL (e) COINED WIRE TERMINAL (C) V TERMINAL (f) TWIN WIRE TERMINAL Fig. 33 — Various terminals. good visibility. When smaller wires are used, the tendency is to make the terminal width greater than three times the wire diameter for better visibility. The terminal thickness depends to a great extent on the shape of the terminal. For a rectangular terminal the thickness may vary from three times to one-half the wire diameter. When the terminal thickness is less than one-half the wire diameter, the terminal may twist too much during the wrapping operation. TYPES OF WRAPPED CONNECTIONS The rectangular terminal is not the only terminal which lends itself to a good solderless wrapped connection. Any terminal offering one or more contacting edges substantially crosswise to the axis of the wrapped wire will make a good connection. Since rectangular terminals of very thin material may twist exces- sively during the wrapping process the preferred shape is a U or V as shown in Fig. 33. These terminals are capable of storing even more elastic energy than a rectangular terminal of equal cross sectional area. The U and V terminals are particularly suited for vacuum tube sockets and thin relay springs. Flattened or coined single wires as well as coined twin wires may be used as terminals for solderless wrapped connections. These are shown in Fig. 33. Stranded wire connections have been made by laying the strands 552 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig. 34 — Stranded wire connection. along the serrated edge of the terminal and then wrapping both the terminal and the strands with solid wire. This is shown in Fig. 34. A stranded wire may also be wrapped in the same manner as a solid wire. However, the strands of the wire to be wrapped must be dipped in pure tin in order to bind the strands into the equivalent of a solid wire. A test was made on a No. 24 (0.020'' dia.) wire having seven strands and three twists per inch. The preliminary resistance and aging tests have shown connections of this type to be good. Enameled wire also has been wrapped. The tool for this purpose has a combination edge at the wrapping tip. Part of the edge has an arc for producing the wrapping tension and the other part has a scraping edge for removing the enamel from the underside of the wire as it is being wrapped onto the terminal. 1 2 FAHNE STOCK PLUG CLIP SCREW SOLDERLESS WRAPPED 0.0148" N0.15 0.040"DIA. 0.120" DIA. 0.350"DIA. NO.4-40 x 0.062" PIN X0.112"L6. X 0.550" LG. (0.112") TERMINAL CONTACT FORCE IN POUNDS CONTACT AREA IN SQUARE INCHES CONTACT PRESSURE IN PSI SPACE 10* CUBIC MILS ELASTIC ENERGY IN MIL-POUNDS 1.4 2.2 22 UNKNOWN 135 90 0.000079 UNKNOWN UNKNOWN UNKNOWN 0.0074 0.0031 15,000 UNKNOWN UNKNOWN UNKNOWN 18,250 29,000 41 8.78 1.75 52.8 15.6 1.53 21 UNKNOWN UNKNOWN UNKNOWN 2.77 3.05 Fig. 35 — Comparison of pressure connections for No. 24 (0.020") wire. J SOLDERLESS WRAPPED CONNECTIONS PART I 553 EVALUATION The solderless wrapped connection has been compared with other pressure connections. (See Fig. 35.) However, when compared with a soldered connection its advantages are as follows: 1. A substantial reduction in wiring defects in manufacture and in service because of: — a. Greater uniformity obtained with a calibrated tool. b. Less breakage of wires due to handling and vibration. c. No solder splashes. d. No clippings. e. No cold joints. f. No rosin joints. 2. Less expensive connection. 3. More compact connection. 4. More clearance between current carrying parts. 5. Easy to disconnect. 6. Saving of tin — a critical material. 7. No contact contamination from soldering fumes. 8. No damage to heat sensitive materials in circuit components. 9. No hazard from hot soldering iron. SUMMARY A good pressure connection depends on the amount of elastic energy which can be stored in the mutually stressed members, namely the wire and terminal. If the ratio of elastic energy to the size of the connecting members is favorable, and the contacting areas are sufficiently large, then the connection can be termed good. The solderless wrapped con- nection when properly proportioned not only meets these requirements, but is uniform in quality and low in cost. Appendix I EFFECTIVE MOMENT ARM FOR TORSION OF RECTANGULAR TERMINAL In this appendix the relationship between the wrapped tension in the wire and the twist of the terminal will be analyzed. The structure of the solderless wrapped connection is equivalent to a terminal having springs attached between its edges as shown in Fig. 36. The springs are arranged in such a way as to form a helix of pitch p. Now let: Su = torsional stiffness of a unit length terminal WF = wrapped tension 554 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 2a and 25 be the sides of the rectangular cross section of the terminal. A = lead of wire helix on long side (2a). 4 = lead of wire helix on short side (26). N = number of turns of wire The classical formula for Torsion is Torque X Length e Su (1) For an equivalent "spring" attached to the long side the axial length of the terminal is ^i , the torque is WFb* and hence the deflection WFMi/Su ; for the short side the length is 4 , the torque is WFa* and the deflection WFa^2/Su . For a complete turn of the helix the length of the terminal is therefore, 2(^i + 4) = p, and the torsional deflection is 5 = ?W (^^_ ^ ^^^^ Su It is logical to assume that (2) (3) Fig. 36 — Equivalent structure of wrapped connection (only half turn shown), * This is approximate and holds closely for A «C 2a and (i <^ 2b SOLDERLESS WRAPPED CONNECTIONS PART I 555 Substituting (3) into (2), the deflection per turn becomes 2ab WF (4) a + 5 ^ Su Since all effective turns are similar the total deflection A for A^ turns is A = WF h 2ah (pN) Su (5) pN is the total length of the terminal. By equation (1) the effective torque must then be Torque = WF T^^l and hence the effective moment ann, A, A. _ r 2ab 1 (6) (7) Solderless Wrapped Connections PART II — NECESSARY CONDITIONS FOR OB- TAINING A PERMANENT CONNECTION By W. P. MASON and T. F. OSMER (Manuscript received February 9, 1953) In order to study the stresses and strains occurring in a solderless wrapped connection, a photoelastic technique using photoelastic hakelite and a photo- plastic technique using polyethylene have been used. Polyethylene has a stress strain curve similar to a metal and can he used to investigate strains in the plastic region. Using these techniques, it is shown that the connec- tion is held together hy the hoop stress in the wrapping wire. In order to lock this in, a dissymmetry from a circular form has to occur. This may he in the direction of an oval shape or a square or rectangular shape. Sharp corners are preferred since a more definite contact area results. A num- ber of rules are derived for constructing the most satisfactory solderless wrapped connection. It is shown that the connection between the wire and terminal is intimate enough to permit solid state diffusion, but the strains are not high enough to cause cold welding of the connection. The life of the joint depends on the twin processes of s'ress relaxation and self diffusion. Stress relaxation occurs at a rate such that half the hoop stress is relaxed in 2500 years at room temperature. This loss of stress is compensated by the diffusion of one part of the joint into the other. Since the activation energies for stress relaxation and self diffusion are approximately equal for most metals, the two effects complement each other and produce a connection which should remain unchanged for times in excess of forty years under any likely ambient conditions. INTRODUCTION The solderless wrapped connection described in the paper by R. F. Mallina (see page 525) provides a very satisfactory and economical method for making connections with apparatus terminals when such 557 558 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 connections are properly made. Not all methods of wrapping or all types of terminals are equally satisfactory and it is the purpose of this paper to describe investigations that have been made to determine the necessary conditions for the best wrapped terminal. These investiga- tions include a photoelastic investigation of the stresses in the terminal and a photoplastic investigation of the strains in the outside wrapping wire. A new photoplastic material, polyethylene, has been used which has a stress strain curve similar to a metal and a birefringence propor- tional to the strain. The use of this material makes possible the evalua- tion of strains in the plastic region and may find applications in other plastic flow problems such as the extrusion of metals. Even after such terminals have been satisfactorily made, there re- mains the question of whether they will have sufficient life to satisfy the requirements of the telephone plant. A design objective for most relays and other switching apparatus of the telephone plant is an unin- terrupted trouble-free life of forty years. Hence, unless the connections are to be the limiting factor in the maintenance of the equipment, they also should have a minimum life of forty years under the conditions for which the apparatus is designed. In order to investigate the probable length of life of such connections, theoretical and experimental work on stress relaxation in metals has served as the basis for calculations and t€sts. These have been extended to the materials and conditions of the wrapped solderless connection and the results indicate that the life should be adequate even under very severe ambient conditions. PHOTOELASTIC ANALYSIS OF STRAINS IN TERMINALS OF THE SOLDERLESS WRAPPED CONNECTION In studying the conditions necessary to insure a good solderless wrapped connection, it is desirable to know what strains occur in the terminals and in the wrapping wires and how these vary with the ter- minal shape, the winding force and other variables entering into the con- struction of the connection. While some of these strains can be surmised from the winding conditions and the shape of the terminals, it is difficult to obtain any quantitative results by calculations on account of the fact that the desirable terminal shapes are rather complicated and because a good many of the strains are in the plastic region. To remedy this difficulty, use has been made of a photoelastic and a photoplastic technique. For the inside terminal, all the strains, except at the corners where the wires make contact with the terminals, are elastic and can be approximated with an ordinary photoelastic tech- SOLDERLESS WRAPPED CONNECTIONS PART II 559 Fig. 1 — Photograph of photooLastic modol and solderless wrapped connection. nique using photoelastic bakelite. Fig. 1 shows one of the photoelastic models as compared with the metal solderless wrapped connection that it simulates. While the ratios of the wire diameter to the terminal di- mensions are different in the model from those in the connection, consid- erable information can be obtained about strains in the terminals and wires from the photoelastic model. The wrapping gun described in the previous paper puts a tension on the wire. To simulate the tension, the photoelastic model is placed in a chuck, one end of the wire is anchored to the chuck and an appropriate weight is applied to the other end of the wire. The specimen is then rotated by the chuck and a definite number of turns of copper wire are wound around the specimen. The extra wire is then clipped off and it is found that the wire tightly adheres to the terminal in the manner of a metal solderless wrapped connection. The specimen is then polished on the two ends up to the end wires and is put into the polariscope of a photoelastic analyzer. For obtaining the isochromatic lines, i.e., the lines occurring when the ordinary and extraordinary rays differ in path length by a half wavelength or some multiple of a half wavelength, the elements of the polariscope, as shown by Fig. 2, contain a quarter- wave plate before and after the specimen. These have the effect of making the plane wave from the polarizer circularly polarized and elimi- nate the isoclinic lines which mark the directions of the slow and fast axes of the material. Fig. 3 shows the isochromatic lines for a square 560 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 specimen 0.4 inches square wound with nineteen turns of 0.050-inch copper wire under a constant weight of twenty-eight pounds, which is a stress of 14,300 pounds per square inch. It is evident from the sharp- ness and number of the Hues that can be seen that we are deaUng with a case of plane stress that can be analyzed by the method discussed in the Appendix. Since the stress strain curve of copper wire has the form shown by Fig. 4 with a yield stress* of about 26,000 pounds per square inch and a breaking stress of about 34,000 pounds per square inch, the appHed winding stress is about 42 per cent of the breaking stress. Fig. 4 shows also the recovery measured for copper wire. The recovery curves are quite accurately parallel to each other, but the larger the strain the smaller the percentage recovery. Using the isochnic lines shown by Fig. 24 of the Appendix and the method of analysis discussed there, the stresses across and perpendicular to the line of ''eyes" of Fig. 3 are shown by Fig. 5. The stress perpendicular to the line of eyes meas- ures the total compressive stress put on the terminal by the hoop stress in the wire and from this measurement the average hoop stress remain- ing in the wire can be calculated as follows. The cross section of which QUARTER-WAVE ^'- PLATES --^ POLARIZER SAMPLE Fig. 2 — Elements of polariscope ANALYZER PHOTOGRAPHIC PLATE this force is applied is the width 0.4 inches by the length of the specimen 0.95 inches and hence the force applied by all the turns is F = 0.95 X 0.40 X 2000 = 760 pounds. (1) Since there were nineteen turns of wire wound around the specimen and each turn has two sides exerting a tension on the bakelite, the aver- age tension remaining in the wire, required to balance the compressive stress, is T ^ 760 2 X 19 = 20 pounds. (2) * In this paper tho. yield st ress is taken jis the point of greatest curvature of the stress-strain curve. SOLDERLESS WRAPPED CONNECTIONS PART II 561 Fig. 3 — Isochromatic lines for a square model 0.4 inches on a side wrapped with nineteen turns of 0.050 inch copper wire with a constant load of 28 pounds. 2 4 6 8 10 12 14 16 18 STRAIN IN PER CENT Fig. 4 — Stress-strain recovery curves for copper wire. 20 22 562 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 2200 2000 1800 1600 1400 1200 1000 800 600 400 200 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 DISTANCE ACROSS PICTURE IN CENTIMETERS Fig. 5 — Stresses along X and Y directions for a square terminal. or 71 per cent of the winding stress applied to the wire. The rest of the stress is lost in the contraction that occurs when the wire causes plastic flow to occur at the corners of the specimen. This plastic flow causes the terminal (both bakelite and metal) to flow around the wire and pro- vides an air tight joint which is an essential requirement for a good solderless wrapped connection. That the stress in the terminal is high enough to do this can be seen directly from the photoelastic picture, Fig. 3. Counting the lines as far as they can be distinguished by a mi- croscope, there are 72 lines from the eyes of the picture (which are iso- tropic points) up to 0.0525 inches from the geometrical corner lines. Using the stress constant of photoelastic bakelite which is 88 pounds per square inch per fringe per inch length along the optic path, this cor- responds to a stress per unit area of ^> A 1^- c r^ i^^ . ^, ^-, L'-' / >^- 1 s. / N / \ STRESS PERPENDICULAR TO LINE OF EYES (T2) / t \ / \ r \l \ / y STRESS ALONG \line of EYES(Ti) / \ 1 \ / \ \ / \ !r = 72 X 88 .95 = 6700 pounds per sq in. (3) Since the total force put on by the wire is supported by successively smaller cross sections, as one approaches the corners the yield stress of bakehte of 15,000 pounds per square inch will be attained at a radius of SOLDERLESS WRAPPED CONNECTIONS PART II 563 ( 9:9^\ X 6700 = 15,000 pounds/sq in. or a; = 0.0235 in. (4) Hence, plastic flow should occur for about 23 mil inches into the plastic. Unwrapping the wires from the terminal, it is found that depressions of this order are cut in the terminals. Since one of the requirements of a better connection is that an air tight bond shall be formed between the wire and terminal, it is obvious that the terminal should have a low enough yield stress so that a sizable groove can be cut in it by the hoop stress of the wire. This rules out such terminals as hardened steel in the most satisfactory connections. It has been found that copper, brass, aluminum, soft iron and nickel silver are soft enough to meet this requirement. Any material with a plastic flow limit in compression, much lower than photoelastic bakelite, would probably have such a deep groove that it would be difficult to maintain the desired hoop stress. Using the photoelastic technique as a tool, considerable data has been obtained on desirable shapes for the terminal and limiting winding stresses that can be used. One of the most used terminals is the rectan- gular terminal and Fig. 6 shows a photoelastic picture of a terminal 0.8 inches by 0.4 inches wound with nineteen turns of 0.050-inch copper wire with a winding load of 28 pounds. This figure is particularly easy to analyze for stress across a line half way down the long edge since the stress along this in the direction of the line varies only a little. Hence, Fig. 6 — Isochromatic lines for a rectangular model 0.4 inches by 0.8 wrapped with nineteen turns of 0,050 inch copper wire with a constant load of 28 pounds. 564 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 one can count the number of fringes from the "eye" (which is an iso- tropic point) to the center of the specimen which in this case is eighteen fringes. Using the stress constant, 88 pounds per square inch per fringe per inch along the optic path, the compressive stress normal to the mid line is 18 X 88 0.95 1670 pounds/sq in. and the hoop residual stress per wire is 1670 X 0.4 X 0.95 2 X 19 = 16.6 pounds (5) (6) which is 60 per cent of the winding stress. Hence, the change in shape has not made any appreciable difference in the residual hoop stress. A specimen 0.4" x 1.6" was also tried and this had a residual stress of 56 per cent of the winding stress. In order to determine the number of turns required to make a satis- factory joint, measurements were made of the residual stresses as a function of the number of turns with the results shown by Fig. 7. All of these experiments were made with the same weight, 28 pounds, which results in a stress of 14,300 pounds/square inch. Down to five turns, about 50 per cent of the winding stress is maintained. The results are consistent with assuming that the wire unwinds to the extent of two corners on each end while 60 per cent of the winding stress is main- tained in all the other turns. As seen from Fig. 8, this is what one might expect, for when the constant tension is released, recovery will cause the l.U o = 0.4 "x 0.8" BAKELITE 0.8 0.6 0.4 ^ = -S-''^ t" BAKELITE 8 4 y" ^ -^T-^ I / 0.2 0 c / Y / / / / / 14 16 18 0 2 4 6 8 10 12 NUMBER OF TURNS Fig. 7 — Residual hoop stress as a function of number of turns. SOLDERLESS WRAPPED CONNECTIONS PART II 565 first corner to bend out and lose contact with the terminal. When the second corner attempts to unwind, it pulls the first corner up against the sample and no umvinding beyond the second corner can occur. In order to obtain the most satisfactory connections, at least five or six turns should be used. The fact that the first two corners on each end do not make close contact to the inside terminal produces a very beneficial result when the connection is subject to vibration due to the handling and operation of Fig. 8 — Locking in effect in a rectangular terminal which allows un- wrapping at only two corners on each ■WIRE end. a relay. In a soldered connection, large bending strains are caused by vibrations at the point of contact between the wire and the solder and in standard vibration tests when large 60-cycle vibrations are impressed upon the wires, the wires fatigue and break off at the point where the wire enters the solder in times in the order of fifty hours. Similar tests have been carried out for solderless wrapped connections and up to times of 2000 hours and longer no breaks have occurred. This is due to the fact that a bending strain is not enhanced by a sharp discontinuity as it is in the soldered connection and strains for a given vibration am- plitude should be less than half as large as those for a soldered connec- tion. Since the relation between fatigue and strain is such that a reduc- tion of strain of two to one or more caused an increase in the number of cycles before breakage of factors of 1000 or more, the increased life under vibration for the solderless wrapped connection is not surprising. Next a series of measurements were made on the value of the winding stress necessary to preserve a hoop stress in the wire. This was meas- ured by winding wires with different weights on photoelastic samples and measuring the residual hoop stress by the technique described above. This was done for both copper and aluminum wires with the re- 566 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 0^ 0.3 0.4 0.5 WINDING STRESS, W YIELD STRESS, Y Fig. 9 — Relation between remanent hoop stress, ratio of wire curvature to wire diameter and the applied winding stress for copper and aluminum wire. suits shown by the dashed lines of Fig. 9. To obtain a hoop stress greater than zero, the ratio of winding stress to yield stress must be greater than 0.1. To obtain a hoop stress that is 0.2 of the yield stress, a winding stress of 35 per cent of the yield stress has to be employed. This is the value recommended to give the most stable terminal. The winding stresses were carried up to 0.8 of the yield stress. At this stress the copper wire at the corners tends to draw down to a low value and may break when it unwinds and hence this is probably the upper limit for winding stresses. The radius of curvature of the middle of the wire, as it is bent around a corner was also measured from photographs similar to that shown on Fig. 6 and the ratio of the wire diameter to the mean radius of curvature is shown by the solid lines of Fig. 9. This is an alternate way of specifying the necessary winding force which may be useful for other shapes of terminals. While square and rectangular terminals are very satisfactory shapes for the inside terminal, they are not the only ones that can be used. A number of coined and U shaped terminals are in general use as discussed by Mallina. In order to investigate the necessary requirements for such shapes, a number of experiments have been made on circular and ellip- tical terminals. When a wire under tension is wound around a rod of circular cross section, there are two sets of opposing stresses, one of which tends to make the helix smaller and the other to make it larger. As shown by Fig. 10, the tension in the wire tends to make the helix hug the cylinder while the bending strains introduced by the wrapping of the wire around the cylinder tend to make the helix open up when the constant stress is released. A number of experiments were made on SOLDERLESS WRAPPED CONNECTIONS PART II 567 wrapping copper wire 20 mil inches in diameter on a steel cylinder hav- ing a diameter of 0.124 inches. In all cases even up to stresses of 80 per cent of the yield stress, the helix failed to grip the cylinder. Some fur- ther measurements were made with smaller sized inner cylinders down to 20 mil inches in diameter. At this small radius the wire barely gripped the cylinder and it took about 70 grams stripping force to pull the wrapped mre off the cylinder. As seen from Fig. 11, the normal force per unit length against the cylinder is balanced by the hoop force in the wire according to the equation BFur = 2Fn sin or rFs = Fh (7) The stripping force SF, when the wire does not dig into the terminal, should be equal to SF = 2wmFj,f = 27rn/Fj (8) where n is the number of turns, / the coefficient of friction which is about 0.15 to 0.2 between metals. For a stripping force of 70 grams for 6 turns, the remaining hoop stress is equal to Fn = 9.3 grams. (9) Since the wire was wound with a 700-gram force, it is evident that only about 1 per cent is maintained in the wire, which is entirely inadequate. In order to obtain a good wrapped connection with high hoop stress in the wires, some means has to be employed to eliminate the unwrap- ping effect of the strain due to bending. This can be accomplished by changing the shape of the terminal from a circular cylinder to a dis- symmetrical shape. For then, as shown by Fig. 8, the tendency to un- -WIRE Fig. 10 — Strains in a wire wrapped around a circular ter- minal. Fig. 11 — Relation between hoop stress and stripping force for terminal not indented. 568 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 wind is opposed by the locking in effect of the diss3Tnmetry which in the most preferred types of terminals, as in Fig. 8, comprises abrupt changes in direction around the periphery of the terminal cross section. Some experiments were made with a cyhnder whose cross section, as shown by Fig. 12, consisted of parallel sides with a semi-circle on each end. When the length A was twice the width, the top curve of Fig. 12 shows the stripping force as a function of the winding force. Assuming that most of the grip occurs on the two semi-circular surfaces, the hoop 12.2 YIELD STRESS IN PER CENT 24.4 36.6 48.8 18 17 16 15 14 13 12 Q i" O Q. lU ^ 9 O u. Q. |7 6 5 4 3 2 1 0 ^y y^ y y ^ y Y ,^ / 0.125X0.063/ / f /\ > / / * / 0.079 X 0.063^ 1 A ^ ! ,y y' ! $ 1 / / / MANDREL CROSS-SECTIOr 0.063"— •1 H h A / f / / / 1 / 3 4 LOAD IN POUNDS Fig. 12 — Stripping force as a function of winding load for an elliptical type terminal. SOLDERLESS WRAPPED CONNECTIONS PART II 569 Fig. 13 — Photoelastic picture of a terminal when the outside wire diameter approaches the terminal diameter. stresses indicated are in the order of 50 to 60 per cent of the winding stresses in agreement with the photoelastic experiments. Ratios of 5 to 1 on the length- width ratio were also tried with substantially the same result. To determine how much dissymmetry is necessary to lock in the bending stresses, a ratio of 1.25 to 1 of the length to width was tried with the result shown by the lower curve of Fig. 12. The stripping force for this ratio is about half that for the larger ratios. The conclusion is that the stripping force decreases as the symmetry increases but that a small deviation from circular symmetry is sufficient to lock in the bending stress. However, for the more satisfactory connections, other factors such as adequate intimate gas tight contact areas indicate that the required dissjTnmetry involves abrupt surface changes in the nature of edges having appreciable penetrating power with respect to the wire. Therefore, while a terminal like that of Fig. 12 may have the ability to lock in the bending stresses, it may not, for other reasons, be the better terminal to use. All of the photoelastic and other types of stress measurements were made for terminals that have a large stiffness in torsion and as seen by the photoelastic pictures, the stresses are nearly plane stresses, i.e., they are all tensions, compressions or shears in a plane perpendicular to the axis of the connection. If, however, the torsional stiffness becomes low, another type of deformation can take place, namely, a twist of the whole terminal due to the torque put on by the helical form of the winding. Fig. 13 shows a photoelastic picture of the strain in the terminal when 570 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 the size of the outside wire is comparable to the size of the terminal and it is obvious that a twist in the terminal is occurring. This is a case of three dimensional stress rather than plane stress and cannot easily be analyzed from the photograph. The photograph does, however, show a twist of the terminal. The twisting strain can be most easily analyzed by taking a long section of terminal of low torsional stiffness, winding 100 or more turns on the terminal and measuring the angle of twist as discussed in the paper by Mallina. Calculations by Love^ show that a twist in an ellipti- cal section with its length along the Z axis introduces shearing strains in the X, Z and F, Z planes, i.e., ezx — S^ and eyz = 84, shearing strains equal to where 2a is the diameter of the ellipse along the X direction and 26 the diameter of the ellipse along the Y direction and r the angle of twist in radians per centimeter. For the terminal with 100 turns of 0.020 mil copper wire discussed by Mallina whose data are given by Fig. 13, 45° angle of twist occurs in 2 inches giving a value of r = 0.157. This causes a shearing strain of about 2 per cent in the worst case which is enough to cause a considerable permanent set. While this twist is useful in studying stress relaxation in the wire, it is undesirable for a solderless wrapped connection to have too much twist since it may cause the ter- minal to twist off in the winding process. According to the data of Fig. 13 of MalUna's paper, no terminal set occurs for nickel silver if the twist in radians per centimeter is less than 0.09 which corresponds to a maxi- mum shearing strain in the X, Z plane of 1.1 per cent. Hence in order to avoid excessive permanent set and twisting off of the inner terminal, the size and shape of the terminal should be controlled so that shearing strains due to twisting should be less than 1 per cent. For standard shapes such as rectangles and ellipses formulae are available to relate the maximum strain to the dimensions of the terminals and the moment due to the winding stress. For a given wrapping tension, this moment can be calculated from Appendix I of Mallina's paper. Summarizing the results of this section, the necessary conditions that the terminal should meet are: 1. The wrapped connection is held together by the hoop stress in the outside wrapping wire. This can be locked in if the terminal has a dis- synmietrical shape in which the length-width ratio is 1.5 or greater, 1 Love, Theory of Elasticity. Chap. XIV, p. 310, 4th Edition, Cambridge Uni- versity Press. SOLDERLESS WRAPPED CONNECTIONS — PART II 571 or if some regular shape such as a square, rectangle or rhombus is used. Sharp corners are helpful since a sharp bend in the wire occurs around them. 2. The material of the terminal must be strong enough so that the wire will not deform or cut through the terminal but must be plastic enough so that an appreciable groove can be cut in it by the hoop stress of the wire, in order that an air tight connection shall be made. The most satisfactory metals are brass, copper, soft iron, nickel silver and aluminum. 3. In the most satisfactory solderless wrapped connection, the wrap- ping wire unwinds to the extent of half a turn on each end and about six turns or more are required to make a good connection. 4. To maintain sufficient hoop stress, the constant wrapping stress should be from 0.2 to 0.7 of the breaking stress of the wire. 5. Shearing strains in planes parallel to the axis of the terminal should not exceed 1 per cent in order to eliminate terminal set. PROTOPLASTIC ANALYSIS OF STRAINS IN WIRES OF A WRAPPED SOLDER- LESS CONNECTION All of the strains in the inner block or terminal are elastic except at the corners. Hence, for the interior terminal, photoelastic bakelite is a satisfactory material for strain investigations. However, the outer wire is necessarily stressed beyond its elastic limit and ordinary photoelastic techniques cannot be applied. In order to see if the wire strains could be studied with photoelastic bakelite, some time was spent in heating rods to a temperature for which they become elastic, winding them under a stress and cooling under the applied stress. Although the wind- ing process was carried out successfully several times, the bakelite rod always broke on cooling. This appeared to be due to the fact that bake- hte is nearly linear up to the breaking point, i.e., it suffers from brittle fracture and does not simulate a metal in this respect. Some measurements had previously been made at the Bell Labora- tories and in England^ on polyethylene which indicated that it had properties similar to a metal in the plastic range. Stress-strain curves up to 15 per cent strain are shown by Fig. 14, and it is evident that on the ascending part, the curve is very similar to that for copper or soft iron. On the relief from stress, however, a considerably larger recovery is obtained than for a metal. This material is fairly transparent and the lower curve marked R/t shows the relative retardation for a 5461 A° 2 Miss S. M. Crawford and Dr. H. Kolsky, Stress Birefringence in Polyethylene. Proc. Phys. Soc, Section B, London, 6, Part 2, pp. 119-125, Feb. 1, 1951. 572 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 1000 6 7 8 9 10 11 STRAIN IN PER CENT 14 15 16 Fig. 14 — Stress and birefringence strain characteristic of quenched poly- ethylene. mercury line. This relative retardation R/i is related to the number of fringes A^ and the wavelength X by the equation t (11) and hence for a 1 per cent strain there are 7.3 fringes for a 1-cm path length in the optic direction. It will be observed that for both increasing and decreasing strains, the birefringence is directly proportional to the strain. Hence, by using polyethylene it appeared possible to measure the strain even in the plastic region. The first experiment tried was to wrap a square metal rod with a polyethylene ''wire" one-sixteenth inch in diameter with a wrapping stress about half the yield stress. It was found, however, that the wire sprang off the metal rod when the constant stress was released. This is due to the fact that the polyethylene has considerably more re- covery than the metal wire and the dissymmetry is not sufficient to lock in the bending stress. This shows that one of the requirements of the wire is that the recovery shall not be too large. If we are to use polyethylene as a photoplastic material, it is neces- sary to simulate the unloading curve as well as the loading curve. This can be done by heating up the polyethylene when it is wound under a SOLDERLESS WRAPPED CONNECTIONS — PART H 573 load, cooling under a load, and then removing the load at room tem- perature. Fig. 15 shows the loading and unloading curves for poly- ethylene as a function of temperature. The stress required to produce a given strain decreases very rapidly as the temperature increases, al- though the recovery remains about the same irrespective of the tem- perature. Suppose now that we apply a load and cool the polyethylene down to room temperature maintaining the strain. When the weight is taken off, the unloading curve will parallel that of the 20°C curve, and 42 per cent recovery will be obtained at 50°C and 12 per cent for 90°C. In order to see if a wrapped joint would be simulated by this means, a one-quarter inch rod of polyethylene was heated up to 97°C, was wound with a one pound winding weight and was cooled to room tem- perature with the weight attached. This was sufficient to prevent the ''wire" from unwrapping and when the weight was removed, the poly- ethylene "wire" gripped the metal closely and formed a bond similar to the wrapped connection. In order to analyze the strain, one has to cut a section through the center of the wire and put the section in the polariscope. First, a number of unstrained polyethylene samples were cut by various techniques and it was found that if they were cut with a 0.02 0.04 0.12 0.06 0.08 0.10 STRAIN IN PER CENT Fig. 15 — Stress strain recovery curves for quenched polyethylene. 0.14 574 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig. 16 — Photoelastic picture of a polyethylene "wire" wrapped at 97°C around a rectangular terminal and cooled off under the applied load. jewelers saw with a good deal of set to the teeth, no strains were intro- duced in the sawing process. Using a jig with two parallel guides a sample 0.040 inches thick was cut through the polyethylene wire. Tak- ing one in the center of the wrap, a photoelastic picture was taken with the result shown by Fig. 16 with an enlargement of one corner shown by Fig. 17. The easiest parts to analyze are the strains in the two legs of the sample since the strains are similar to those for a bent section. The long leg which was twice as long as the short leg has its zero order fringe along the inside edge. There are twelve lines to the outer edge which corresponds to a tensile strain of 16 per cent with an average tensile strain of about 8 per cent which is about the strain caused by the load- ing weight. In the short leg the zero order fringe is the inside oval and the strain is about 11 per cent compressive at the inner edge of the segment and about 38 per cent tensile at the outer edge. These values are consistent with the radius of curvature that the wire is bent around for it can be shown that the strain in a wire of diameter d bent around a cylinder of diameter D without tension is equal to S = 2p D + d (12) where p is the radial distance measured from the center of the wire outward, D the diameter of the cylinder and d the diameter of the wire. If p is positive or the point is outside the center line, the strain is posi- SOLDERLESS WRAPPED CONNECTIONS — PART II 575 tive or tensile, while if p is inside the center line the strain is negative or compressive. From measurement of Fig. 16, it appears that the equivalent inner cylinder that corresponds to the radius of the center of the wire is about 3.0 times the wire diameter and hence the strain should be 25 per cent tensile at the outer edge and 25 per cent compres- sive at the inner edge. The addition of a tension of 13 per cent due to the winding force makes the outer strain 38 per cent and the inner one 12 per cent compressive which are close to the values found. This indi- cates that the tensile strain is somewhat higher in the short leg than in the long one. The same tensile strain occurs around the outside periph- ery of the wire opposite the corners of the terminal, but a very high Fig. 17 — Enlargement of one corner of the photograph of Fig. 16. 576 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 point of compression develops just below the point of contact between the wire and terminal. The distribution of compressive and tensile strains in the wire is shown pictorially by Fig. 18. The distribution of strains in a metal wire can be considered to be quite similar except that the tensile strain due to winding should be only 2 to 3 per cent as seen from Fig. 4, while the strains due to bending may be even higher, for as seen from Fig. 9 the ratio of B/d may be in the order of 1 and the bending strains may be as high as 50 per cent. In order to specify the properties that the wire must have in making a good solderless connection, experiments have been made on how much recovery can be tolerated in the wire. Since it is difficult to get a series of metal wdres having different amounts of recovery, the technique was resorted to of heating polyethylene to a definite temperature, winding under a load equal to half the yield stress at the winding temperature, and cooUng to room temperature under the load. As shown by Fig. 15, known recoveries can be obtained in this way. It was found that the largest amount of recovery that could be tolerated to make a joint at all was 20 per cent while a reasonable hoop stress was not obtained until the recovery was less than 10 per cent. In summary, the necessary conditions that the wire should fulfill are: 1. Since strains of 50 per cent may be encountered in bending wires around sharp corners, wires should be used which have a large difference between the yield strain and the breaking strain. Copper, aluminum and soft iron are materials of this class while phosphor bronze and music wire are not as satisfactory. WIRE - TENSION - COMPRESSION Fig. 18 — Distribution of compressional and tensile stress in the wire of a wrapped solderless connection. SOLDERLESS WRAPPED CONNECTIONS PART II 577 2. The amount of recovery from strains of 20 to 50 per cent should not be greater than 20 per cent. PERMANENCE OF WRAPPED SOLDERLESS CONNECTIONS By the photoelastic, photoplastic and strain analysis of the previous two sections, it has been demonstrated that the wrapped solderless connection is held together by the hoop stress in the outside wire whose value is determined by the winding stress and the locking in effect dependent on a dissymmetry of the terminal. The high stresses cause plastic flow in the wire and terminal in such a manner that the two materials flow together and produce an intimate air tight joint. The inti- mate nature^ of this contact has been demonstrated by dip coating nickel silver terminals with pure tin and wrapping them with cleaned bare copper wire. The wrapped terminals were then placed in a glass tube, evacuated, sealed off and heated for 400 hours to 180°C (37°C below the melting point of tin). The samples were then removed, mounted vertically, polished, etched and examined microscopically for distinguish- ing constituents. It is believed that if such a constituent appeared on the originally bare copper wire after such treatment, that the contact was sufficiently intimate to permit solid state diffusion. A section of the wire in contact with the corner is shown by Fig. 19. The copper is seen COPPER NICKEL BRASS Fig. 19 — Solid state diffusion of tin into copper in a wrapped solderless con- nection. ' This experiment was conducted by G. S. Phipps. 578 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 to have a heavy layer of tin constituent at the contact surface. From the test, it can be concluded that wrapped connections are sufficiently tight to allow solid state diffusion and are therefore good electrical contacts. On the other hand the contacts are not welded contacts such as occur when two pieces of aluminum are cold pressed together with strains in excess of 75 per cent. This is shown experimentally by the simple process of imwinding the wire from the terminal which takes place with no excess force when the wire is removed from a terminal corner. Since the strains at the points of contact do not exceed 30 to 40 per cent, one would not expect cold welding. It is possible, however, that in tinned terminals, some long time diffusion takes place at room temperature in the manner demonstrated at higher temperatures by the data of Fig. 19. This would occur very slowly and cannot be relied upon solely to maintain the contact. Hence, it appears that long life in the connection depends on main- taining sufficient hoop stress in the wrapping wire to keep the elements of the connection sufficiently tightly pressed together so that no corro- sion can occur in the connection in such a manner as to interrupt the electrical continuity. This is a problem in stress relaxation rather than creep. Stress relaxation is intrinsically a simpler phenomenon since the major fraction of stress that can be relaxed will be relieved through viscous flow in previously formed slip bands or along grain boundaries, and no generation of new slip bands is required. However, under ordinary creep conditions, an increase in stress is presumably attended by the generation of new slip bands. It appears likely then that stress relaxa- tion phenomena even at quite high stresses should more nearly follow the conditions that have been established for low stresses than would be the case for creep phenomena. A good deal of work has been done on stress relaxation at low stresses, particularly by Zener'^ and his coworkers, and this will be briefly re- viewed. According to these studies, stress relaxation can be caused by several mechanisms including stress induced migration of impurities in the metal, viscous behavior of slip band material and the viscous be- havior of grain boundaries. At- the common junction between the two metal grains, there is an amorphous layer of material which acts as a viscous medium, i.e., if there is a shearing stress applied across it, the two grains will move with respect to each other with a velocity * Zener, C, Elasticity and Anelasticity of Metals. University of Chicago Press, 1948. T'ing-Su K6, Experimental Evidence of the Viscous Behavior of Grain Bound- aries in Metals. Phys. Rev., 71, No. 8, pp. 533-546, April 15, 1947. T'ing-Su Kd, Anelastic Properties of Iron. Tech. Publication No, 3370, Metals Technology, Jqne, 1948, SOLDERLESS WRAPPED CONNECTIONS — PART II 579 V = vT/D, (13) where D is the thickness of the layer, 77 the coefficient of viscosity and T the shearing stress. Hence no matter how small the shearing stress, one grain will move with respect to the other in a finite time. The amount that the grains can move is limited by the necessity of making the grain boundaries fit. According to Zener, the situation is analogous to the case of a jigsaw puzzle in which the overall configuration possesses rigidity in spite of the fact that no shearing stress exists between adja- cent pieces. Zener has calculated that the ratio of the relaxed stress to 1.2 ,o^.o 0.8 < I- 5 0.4 0.2 -CM HEAT OF ACTIVATION = 34,500 CAL/MOL CL, r- -o^ «>^ WJ, so, y ' ^=^^<^ ■Ov >n TEMPERATURE OF MEASUREMENT A tSCC (TIME DIVIDED BY 74) o 175°C (TIME DIVIDED BY 7.7) X 200° C D 225°C(TIME MULTIPLIED BY 6.2) xtI^^ ^■oxo. b-> -> o «Ox^ -x-o^ -6- .X —j- ■ 0.06 0.1 0.2 0.4 1.0 2 4 6 10 20 40 60 100 200 400 TIME IN SECONDS (REFERRING TO 200° C ) 1000 2000 4000 10,000 Fig. 20 — Stress relaxation in aluminum at three temperatures for strains less than 10-4 (after Ke). the initial stress is equal to i(7 5(7) 7+ (T 5(r2 (14) where a is the value of Poisson's ratio. For values of Poisson ratio of from 0.25 to 0.5 this ratio Ues betw^een 0.595 and 0.76. Fig. 20 shows measurements of stress relaxation plotted against time for aluminum for three different temperatures. These were obtained^ by twisting an alimiinum wire through a definite angle and observing the force required to hold it at this angle as a function of time and the temperature of the wire. All the curves can be made to coincide by multiplying the times by different factors. If we define the relaxation time r as the time required to relax half of the variable component of stress, i.e. H(l — 0.67) of the stress, this relaxation time fits an equa- tion of the form r = Ke^^«^ (15) where K is a constant, H an activation energy, T the absolute tempera- 580 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 ture in degrees Kelvin and R the Boltzman constant for one gram mole of the material. R is closely equal to 2 calories per degree K. Hence, if H is expressed in calories per gram mole and the value of K is obtained to fit equation (15), we have 34,500 T = 9.2 X 10"'' X e"2^ . The constant K is close to that given by the Langmuir-Dushman' theory J. _hN _ 6.62 X 10-^^ X 6.06 X 10^^ _ ,5 .... ^"H 34,500 X 4.187 X 10^ " '^ ^ ^ ' ^^^^ where h = Planck's constant equal to 6.62 X 10"^^ ergs, A^" is Avogadro's number equal to 6.06 X 10 and H is the activation energy expressed in ergs. Su K6 has shown that the activation energy for grain boundary slip is essentially the same as for self diffusion and for creep. Similar results have been found for a-brass and a-iron. These have activation energies shown by equation (17) a-brass 41 kilocalories per mole, (17) a-iron 85 kilocalories per mole. Although measurements have not been made for copper, the activation energy of self diffusion is about 57.2 kilocalories^ per mole, but 39.9 kilocalories for the principle impurity silicon. All of these measurements were made for strains under 10~*, and the question arises as to whether these concepts are valid for the much higher strains experienced in the wrapped solderless joint. From the photoelastic pictures. Figs. 16 and 17, it is obvious that the greatest stress inhomogeneity occurs in the neighborhood of the corners and flow will take place in such a way as to relieve the high stress concen- tration. This will have the effect of making the terminal and wire mate even closer and may result in a slight transient lowering of the hoop stress. After the initial formation, however, it will be the long time re- laxation of the hoop stress in the wire that determines the lasting quality of the joint. As discussed previously, the twist that the terminal takes is deter- mined by the mean value of the hoop stress in the wire, and any relaxa- tion in this hoop stress can be studied by observing the angle of twist as a function of time and temperature. By using a long terminal wound » S. Dushman and I. Langmuir, Phys. Rev. 20, (1922) p. 113, 1922. • Zenner, Elasticity and Anelasticity of Metals. Table 12, p. 98, Chicago Uni- versity Press. SOLDERLESS WRAPPED CONNECTIONS — PART II 581 \\dth 100 turns or more of copper wire, a twist of 50° or more can be obtained which is sufficient to measure. In order to test first the relaxation in the copper wire alone, the inner terminal was made of clock spring steel 0.0124 inches by 0.062 inches. This was wound with 100 turns of 0.020 inch copper wire tensioned at 2.87 pounds (9,300 pounds per square inch). A twist of 25° was obtained which is sufficiently large to measure. If one observed the angle after transient creep has occurred, the angle decreased on the average about 17 per cent in the first month as shown by the circles of Fig. 22. The values agree with the solid curve which was estabUshed by relaxation measurements as a function of temperature as discussed in the next paragraph. At room temperature, further decreases in the angle of about 10 per cent were observed out to times in the order of a year. If, however, the wrapped connection was heated up, a faster relaxa- tion of the hoop stress occurred. Fig. 21 shows the ratio of angle meas- ured to the initial angle as a function of time when the connection is subjected to a temperature of 200°C. As shown by the dashed line, which is a plot of the exponential equation R = e~"' where a = 2 X 10~^ (18) this is not a single relaxation of the type found for grain boundary motion but is a sum of effects occurring -with different activation ener- 1.1 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 D 200' 'c A^ "^ s "O v \ - O UO-C ALL VALUti Uh- TIME DIVIDED BY 9.5 - A 150° C ALL VALUES OF TIME DIVIDED BY 120 ^. ^ N N k_ •s s. sj N„ ^ X k \ s \ \ u. 4 ^ 4 \^^ SINGLE \RELAXATION TIME ^N ^ ^ 4 \ \ \ ^, sA r ^ i.5^ > ^^^ ^^ ..^ s ^N V > 2 468 2 468 2 468 2 10 102 103 10^ TIME 'N SECONDS AT 200 DEGREES CENTIGRADE 4 6 8 105 Fig. 21 — Ratio of angles of twist at time shown to initial angles. 582 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 gies. Furthermore, the change is not Hmited to two-thirds as in the case of grain boundary motion, but is much more tending toward a value of 0.2 for very long times. It appears that several processes are involved in addition to grain boundary motion. These are probably connected with slips along slip planes which occur with a lower activation energy when the stress is high. As the slip increases, strain hardening occurs with a resultant increase in activation energy until the activation energy of grain boundary movements is reached. For 0.3 relaxation and lower, the activation energy remains constant and equal approximately to 40 kilocalories per mole — the self -diffusion value — as can be calculated from the 150°C, 175°C and 200°C relaxation curves of Fig. 21. These curves show that the activation energy varies from 12.5 kilocalories at 0.9 relaxation, to 40 kilocalories for long time effects. Similar measurements have been made on nickel silver terminals which are the terminals actually used and as seen from Fig. 21 of Malhna's paper, these agree quite well wdth those measured for spring steel terminals. The nickel silver terminals had the dimensions 0.0148 inches by 0.062 inches. A twist of 46° was obtained for 100 turns of 0.020 inch copper wire with 2.87 pounds winding stress applied. At this angle of twist, a permanent set of 19° occurred when the outside wire was un- wound. If we subtract that value from the initial twist, the time-angle curve is very similar to that for spring steel and indicates that no addi- tional relaxation occurs in the nickel silver terminal. The conclusion from these experiments is that stress relaxation for the value of strain used in the wrapped solderless connections follows a simi- lar activation energy pattern to that followed for smaller strains except that instead of a single process with a single activation energy we are dealing with many separate processes having an activation energy range from 12.5 kilocalories to 40 kilocalories. For each stage of the process a different activation energy is effective. For example, for a ratio of re- laxed stress to initial stress of 0.9 the curves of Fig. 21 indicate an activa- tion energy of about 12.5 kilocalories. With this value of activation energy the time required to relax this amount of stress at room tem- perature of 25°C (77°F) is 2.98 X 10^ seconds or 0.0095 years as shown by Fig. 22. For a ratio of relaxed to initial stress of 0.8, the activation energy is 15.3 kilocalories and the time at room temperature is 0.126 years. Similar values can be calculated for the other relaxation ratios and the complete relaxation ratio and time curves are shown by Fig. 22 for temperatures of 77°F and 135°F. To reach a value of 0.5 of the initial hoop stress requires 2500 years at 77°F and about forty years at 135°F. The circles show the measured values at 77°F carried out in a tempera- SOLDERLESS WRAPPED CONNECTIONS — PART II 583 1.0 0.9 I 0.7 lU 0.5 I- (0 0.4 o O 0.1 a 0 ^ 3 h o OBSERVED VALUES OF RELAXATION AT 25° C ^ n ---« ^ ^ ■^ ^ ^ S-- \ ^ -- ^^cCtt^f) "^ ^ ^^ ■^ se-T^cdas^pT^ ^ ^^ ' "H ' ■ ^ 10" 10" 2 5 2 1 10 TIME IN YEARS 5 2 5 2 5 102 103 10" Fig. 22 — Aging at room temperature and at 135°F plotted as ratio of angle to initial angle as a function of time. ture controlled air conditioned room and these agree well with the cal- culated values. 135°F is in general the maximum temperature that wrapped solderless connections will be subjected to. For cases of very high temperature it is planned to use copper covered soft iron wires for the wrapping wires since, as shown by Equation (17), iron has a much higher activation energy than copper and can be expected to maintain its hoop stress for forty years even under an ambient temperature of 200°C. The question arises as to whether the hoop stress of a small part of the initial value, that has been shown to continue for a long time by the data of Fig. 21, is sufficient to maintain a good contact. This ques- tion may be important if aluminum is to be used as a wrapping wire since with an activation energy of only 34.5 kilocalories, 0.5 of the hoop stress will be maintained for forty years only for temperatures lower than 100°F. The problem then is whether corrosion can occur between the wrapping wire and the inside terminal when the relaxable com- ponent of hoop stress has been relaxed. A very sensitive test for this question is obtained by winding aluminum wire on an aluminum terminal since if any break occurs in the contact between the wire and the termi- nal, oxidation of the aluminum surface takes place very rapidly and should affect the resistance of the solderless connection. Accordingly, a number of aluminum-aluminum solderless connections were wound up and their resistances were measured. They were then put in an oven and heated to a temperature of 200°C for twelve hours, which was suffi- 584 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 cient to relax all the stress that can be relaxed. On remeasuring the resistance, it was found that there was no change within the experi- mental error of 1 per cent, which corresponds to a resistance of 3 X 10~^ ohms. A similar result is found by studying the corrosion of surfaces of copper and nickel silver in solderless wrapped connections when they are fully relaxed and subjected to a corrosive atmosphere as discussed in the paper by Van Horn. The stripping force of the aluminum-aluminum connection subjected to a temperature of 200°C for 12 hours has actually been found to in- crease by a factor of about 2 which suggests that the two parts have diffused into each other and formed a permanent connection. This has been confirmed by the increase in force required to unwrap the wire. Since the activation energy of self diffusion is about the same as the activation energy for stress relaxation, then as the hoop stress is relaxed at high temperatures, solid state diffusion takes place and a diffusion joint is formed in aluminum. The same process, both for stress relaxa- tion and diffusion, should take place at a much lower rate at lower tem- peratures and as the hoop stress is relaxed, a diffusion connection between the two parts is formed so that no decrease in the conductivity of the connection occurs and an actual increase in the strength of the connec- tion results. The same process should result between any two materials provided the energy of seK diffusion from one into the other is less or equal to the activation energy of stress relaxation for the weakest component. It is planned to tin plate both terminals and wires for all of the wi'apped solderless connections used in the telephone system. In order to find how the two processes of stress relaxation and self diffusion, which progress as a function of time and temperature, affect the two fundamental properties electrical conductivity and mechanical strength of the connections, a large number of connections were wound up and subjected to temperatures of 200°C for times corresponding to 0.9, 0.8 etc., of the initial stress as determined from Fig. 21. The electrical re- sistances of the connections were determined before and after the treat- ment and the stripping force was also measured. Within the experimental error the resistances of the connections remained the same while the average stripping force for twenty connections for each point are shown by Fig. 23. No significant change in the stripping force occurred out to values of stress relaxation less than 0.2 times the initial hoop stress. These experiments show that as the hoop stress is relaxed by time and temperature, self diffusion occurs between the two parts of the connec- SOLDERLESS WRAPPED CONNECTIONS PART II 585 tion in such a way that the mechanical strength and conductivity are maintained unchanged with time. In summary it has been shown that 1. The connections are sufficiently intimate to permit solid state diffusion but the strains are not high enough to cause cold welding of the connection. 2. The hoop stress relaxes as a function of time and temperature according to well known activation energy equations with an activation energy for copper wires on nickel silver connections varying from 12.5 to 24 20' 12 STRIPPING FORCE ▲ MAXIMUM • AVERAGE , ■ MINIMUM L k A i i ' i 1 ( » m 1 , ' , ■ ' ' ' ■ 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 RATIO OF RELAXED HOOP STRESS TO INITIAL HOOP STRESS Fig. 23 — Stripping force plotted as a function of stress relaxation. 40 kilocalories. It requires about 2500 years to relax half the hoop stress at 77°F (25°C) and 40 years at 135°F. 3. The twin processes of stress relaxation and self diffusion occur in such a way as to maintain or increase the strength of the connection and to leave the resistivity of the connection unchanged with time. 4. The conclusion from all of these tests is that the life span of the solderless wrapped connection appears to be ample for meeting any of the likely requirements. This conclusion is reinforced by life tests of a large number of wrapped solderless connections in service that have been carried out over two years without a failure and by the long and satis- factory tests of the screw connection whose success also depends on the stress relaxation, diffusion processes. Such connections have been shown to be satisfactory over periods of time in excess of twenty years. 586 the bell system technical journal, may 1953 Appendix method for analyzing stresses and strains from photoelastic pictures The methods for analyzmg photoelastic pictures are given in detail in a number of books and other publications^ and only a brief summary of the method used here will be given. When polarized light is sent normal to a plane of photoelastic material that is strained in the plane, the light is broken up into an ordinary and an extraordinary ray that travel with different velocities. It is shown^ that the birefringence, which is defined as the difference between the two indices of refraction m and 112 (i.e. the index of refraction is the ratio between the velocity of light for one of the rays in a vacuum to the velocity in the medium) is given by B = Ml - M2 = CV(Ti - T2y + 4.TI , (19) where C is a constant called the relative stress optical constant, Ti is the tensional stress along the X axis, T2 the tensional stress along the Y axis and Te the shearing stress in the XY plane. If we change the direction that we call the X axis until we reach the direction of maxi- mum stress, the relations between this stress and the tensional stress at right angles to it are given by the equations T[ = Ti cos'^ + 2 sin (9 cos OT^ + T2 sin' 6, (20) T2 = Ti sin'^ - 2 sin ^ cos STq + T2 cos' 6, where d is the angle between the X axis and the axis of maximum tension. [f 0 is chosen so that Ti is a maximum, we find tan 29 = ^.^^ (21) and r; = ^'' t ^^ - iVcr, - T^f + ATI (22) ' Coker and Filon, Photoclasticity, Cambridge University Press, 1931. M. Hetenyi, Handbook of Kxporimental Stress Analysis, John Wiley and Sons, Chap. 17, 1950. R. D. Mindlin, J. Applied l^hysics, April 1939, pp. 222-241 and May 1939, pp. 273-294. W. P. Mason, Eleotrooptic and Photoelectric Effects in Crystals, Bell System Tech. J., 29, pp. 161-188, April, 1950. SOLDERLESS WRAPPED CONNECTIONS — PART II 587 and hence T[ - T2 = V{Ti - T^y + 4T|. (23) Hence, the birefringence is directly proportional to the difference be- tween the principal stresses. The retardation R is the difference between the path length for the fast and slow rays when they are transmitted through the thickness of the specimen. If h is the thickness of the specimen h = V2t, h - R = vit (24) (25) and hence R V2 h - Vi V2 and c R _ vi h /V2 — \ V2 vi\c J Vl [c _ c ^ Hence, the retardation is given by R h t;) = C{T',- T[), If Xis the wavelength of Ught used for the measurement, N\ R h h = C{T'2 - T[), (26) (27) or the number of fringes is related to the difference of the principal stress by AT = ^ (T2 - T[). (28) A o For photoelastic bakelite for the green mercury line 5461 A, the fringe constant \/C is 88 pounds per square inch/inch/fringe. Hence the difference between the principal stresses is given by (T2 - T[) = ^ (88) pounds/sq in. (29) The isochromatic lines such as shown on Fig. 3 are taken with quarter- wave plates in addition to the crossed polaroids. These lines give directly the multiple number of times the stress is greater than the value (88//i) pounds per square inch. In order to know the exact multiple, one has to know the starting point or the points of zero stress difference, called isotropic points. These are determined by using white light and locating the gray fringes. The zero order fringe is gray because for this case, all the wavelengths of light are delayed the same amount and hence no 588 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 color effects appear. A first order fringe will have red (absence of violet) nearest the zero order fringe and violet (absence of red) furthest from the zero order fringe. High order fringes will not appear at all in white light since they are the resultant of a number of colors which add up to white light. Having located the zero order fringe, a simple count will give the number of times the factor (S8/h) has to be multiplied by to obtain the number of pounds per square inch. This stress will, however, only be a stress difference and in order to resolve this into stresses along X and Y axes and a shearing stress in the XY plane, other information is necessary. One part of the information is obtained when the isoclinic directions are obtained. These directions are the directions of the principal stress axes and these are obtained by taking out the quarter wave plates and rotating the axes of the polaroids (keeping them crossed) until the polarization axes coincide with the principal stress axes at any point. When this occurs, the picture will be black because if no model were there, the polarized light passed by the polarizer would be cancelled by the analyzer. If the principal axis of the stress eUipsoid coincides with the direction of polarized light from the polarizer, only one ray will be generated whose plane of polarization coincides with that of the polarized Ught and hence this will be cancelled in the analyzer. The isoclinic lines show up much better if a white light source is used. Hence, the isoclinic Hnes locate the direction of the principal axes of the stress ellipsoid. From equations (20) and (23) we have r-^)" sin 2(9, Ti - T2= {T[ - T2) cos 26. (30) Hence, if we know 6 (the direction of the principal stress axes with re- spect to the axes for which the stresses are to be analyzed) the shearing stress T« and the difference between Ti and T2 can be obtained from equation (30). The other necessary relation can be obtained from the equilibrium stress relations that have to be satisfied by any stationary body, namely '^ + fi =0; ^^ + '^ = 0. (31) dx dy dy dx Integrating these equations SOLDERLESS WRAPPED CONNECTIONS PART II 589 For example, the isoclinic lines for the square plate of Fig. 3 are shown by Fig. 24. All the isoclinic lines converge on the positions of the *'eyes" which is a characteristic to be expected of an isotropic point {Ti — T^) — 0. For the Une through the center of the "eyes", which we designate as the X direction, the isoclinic line Hes along the X axis and hence there is no shearing stress along this axis and Ti — T2 = T'l — T2 . At either edge the stress in the X direction is zero and hence at this point the stress T2 (which is a compression) is obtained by counting the number of lines from the isotropic point to the edge (in this case 20) so that the stress at the edge is ^ 20 X 88 1850 pounds/square inch (33) To determine the shearing stresses at any point, one needs to fit the isoclinic lines over the isochromatic lines. For example, for any point along the isoclinic line marked 15, the direction of the principal axis is 0" ( ^^^^ 90" c ^;s.^"^vr^ "^^^^° Fig. 24 — Isoclinic lines for a square terminal whose isochromatic lines are shown by Fig. 3. 590 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 15° from the X axis, since it was obtained by setting the analyzer and polarizer at 15° from the direction X. Counting the number of isochro- matic Hues from the "eye", the principal stress difference T{ — T2 is known and using equation (30), T^ is at once calculated. To obtain dT^/dy, it is sufficient to divide Tq by the distance y that the point is above the X axis. Proceeding this way, values of dT^/dy are plotted as a function of X. Then from the first of equations (32) with Ti^ = 0 at the edge, one can obtain Ti by integrating dT^/dy dx over the X axis. The result is the curve marked Ti of Fig. 5. Having Ti , T2 can be ob- tained from the isochromatic lines which determine T2 — Ti and this is plotted on the upper part of Fig. 5. Solderless Wrapped Connections PART III — EVALUATION AND PERFORMANCE TESTS By R. H. VAN HORN (Manuscript received February 9, 1953) In the development of solderless wrapped connections the basic require- ments of electrical and mechanical stability have been translated into test requirements on laboratory samples of these connections and on the manu- factured product. These tests have shown that the connections can with- stand satisfactorily the effects of corrosion, humidity, vibration, and relaxa- tion. The effects of terminal dimensions, materials, corner sharpness, wrapping tool construction, etc. are noted. INTRODUCTION The previous tv^^o articles have described the fundamental considera- tions involved in solderless wrapped connections. A description of these connections together with a rather detailed explanation of the forces which maintain them has been presented. This third article discusses the results of a number of tests where the actual fabrication and use of these connections have been simulated. From these results it will be seen that reliable performance can be expected over the central office Hfe of these connections, and that the variations permissible in their fabrication will privide sufficient margin to make that process easy to control. GENERAL REQUIREMENTS The minimum physical requirements which a solderless wrapped con- nection must meet are: 1. Intimate contact between wire and terminal. 2. The points of contact should be gas-tight to withstand corrosion. 3. The minimum dimension of the gas-tight area should be great enough so that it does not decrease appreciably during the expected 591 592 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 life (forty years) because of corrosion or because of relaxation of the internal stresses in the wire or terminal. 4. The sum of the areas of intimate contact should be equal to or larger than the cross-section of the wire to prevent local heating. 5. The connection should be mechanically stable so that forces applied to the connection during shipment, installation and subsequent main- tenance activities will not dislodge the wire and break the points of intimate contact. 6. The wire should not be embrittled during the wrapping operation so that it will subsequently break due to vibration, handling, or un- wrapping. TRANSLATION OF GENERAL REQUIREMENTS INTO TEST REQUIREMENTS In order to evaluate the feasibility of solderless wrapped connections, extensive development studies were necessary so that a good estimate could be obtained as to whether these connections would meet these general requirements under a wide variety of conditions and with suffi- cient margin to provide for ease of manufacture. It was necessary to translate these requirements first into a set of development test require- ments and second into a set of shop inspection requirements. These two translations of requirements will not necessarily be the same although there will be a large degree of overlap. In either translation they can be broken down into two distinct areas. These two areas of tests cover those tests which are related to evaluating (1) life and electrical stability of the connection and (2) mechanical stability. A great deal of engineering judgment was used in the translation of the physical requirements into inspection requirements and this judg- ment took into account the special nature of conditions to be encountered in telephone offices. Consideration was given to the methods of handling the wire, the manipulations of installation and maintenance men when working on central office equipment, the effect of vibration, the effect of variation in tool dimensions and the like. Furthermore, a good deal of knowledge of the corrosion and relaxation process had to be developed before it could be judged on the basis of accelerated tests that the con- nections might be expected to have a satisfactory field life. There may be applications where the translation of the physical re- quirements into test requirements may be different, perhaps quite different, from the translation made for the telephone apparatus which was in mind during this investigation. The use of other kinds and sizes of wire or terminals may require an evaluation quite different from the one presented here. Nevertheless, the test requirements herein estab- SOLDERLESS WRAPPED CONNECTIONS — PART III 593 lished should have a wide apphcation in many areas. In telephone prac- tice they provide a reasonable latitude for variations in the process of making the connection, including tolerances in the parts, and at the same time guarantee a good product. Most of the product tests which have been made so far apply specifi- cally to connections which use Standard No. 24 tinned solid copper wire such as is used in 95 per cent of the telephone switching plant, and terminals having a rectangular cross-section punched from sheet nickel silver, brass or bronze and whose dimensions are one-sixteenth inch wide by the thickness of the stock (0.013" to 0.062''). These terminals are typical of those which are or could be used on relays, switches, resistors, capacitors, terminal strips, etc. Studies with #20 and #22 wire have been made with results similar to those with # 24 wire. A similar investigation is under way on connections to terminals which are made from round silicon copper and nickel silver wire such as are used on the wire spring relay, and where the wire terminal had been prepared for connection by various treatments such as flattening, coining, serrating, annealing, etc. AGING OF WRAPPED CONNECTIONS Assuming that a connection can be wrapped with sufficient mechan- ical strength to withstand handling, vibration, etc., there appear to be two factors which might cause the connection to fail after a period of time. These factors are (1) relaxation of the internal stresses in the metals, and (2) corrosion of the metal surfaces. As Mr. Mason points out in his paper, it now appears that solid state diffusion of metal across the boundary between the wire and the terminal improves the con- nection as much or more than relaxation degrades it. Tests have been designed to relax the connections in a short time to the same degree that will occur in the normal forty-year life which is required. These tests will be described later. Several investigators have studied the rate of surface corrosion in indoor atmospheres of metals such as are used in these connections. Studies of corrosion* where oxidation is the primary factor indicate for example that the corrosion in zinc varies linearly with time and on copper it varies as the square root of the time exposed. The corrosion products are ZnO and CU2O. The corrosion rate for brasses fall between * British Non-ferrous Metals Research Association, Investigation on the Atmospheric Corrosion of Non-ferrous Metals, First and Second Experimental Reports to the Atmospheric Corrosion Resistance Committee, May, 1924, and May, 1927, W. H. J. Vernon. 594 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Table I — Thickness of Material which Corrodes in Forty Years Zinc 0.00023" Copper 0.000105'' Tin Negligible copper and zinc. There are very few data available on tin but its cor- rosion should be less than either copper or zinc. From the data presented the depth of metal which will corrode during a forty-year period in a central office for several metals is estimated* to be as shown in Table I. Samples of copper bus-bar taken from telephone exchanges where they have been exposed for periods up to forty years show that the tarnish is primarily CU2O and the actual rate of corrosion may be ap- preciably less than that estimated by the above figures. Thus the depth of metal corroded is small enough to neglect when the metals are subject to indoor atmospheric exposure. When dissimilar metals are joined in a connection there is the possibility of electrolytic corrosion in addition to atmospheric corrosion. The particular metals involved here, however, are relatively close to each other in the electro- motive force series of metals so that it is expected that this effect will be negUgible especially as there is no condensation on these connections. It is therefore expected that the most important factor in the aging of these connections is the relaxation of stresses internal to the wire and the terminal rather than corrosion. The test procedures and results in the following sections reflect that view. development test procedure — electrical requirements General Requirements 1 through 4 are considered together. A set of tests has been designed to evaluate the degree to which these require- ments can be met. Since the tests which have been devised are destructive it is necessary to check connections in production on a sampling basis. In determining whether a connection meets the General Require- ments 1 through 4, Test Procedure I as follows was devised. Test Procedure I for Solderless Wrapped Connections 1. Check connection for insulating barrier film between wire and terminal. 2. Measure the variation of the resistance of connection while pro- ducing movement between the terminal and connecting wire. * Unpublished memoranda, D. H. Gleason, Bell Telephone Laboratories. SOLDERLESS WRAPPED CONNECTIONS PART III 595 3. Chill for two hours at 0°F. 4. Heat for three hous at a temperature (175°C) which will relax the stresses as much as will occur during the expected life at the normal central office operating temperature. 5. Expose the connection to a suitable agent which will discolor all the non gas-tight area. 6. Remeasure the resistance variation as in 2. 7. Unwrap the wire and estimate the size of the gas-tight areas. Items 1 and 2 are intended to show whether initially there is the inti- mate contact between wire and terminal demanded by the General Requirement No. 1. At present, the multiple terminal banks on step-by- step switches are connected together with a clinched solderless con- nection. Based on experience with these connections, together with an estimate of the noise produced (See appendix A) a resistance variation in excess of 0.002 ohms is considered to indicate a poor connection. Items 5 and 7 will show the existence of the gas-tight areas demanded by General Requirement No, 2, and the size estimate from Item 7, the area of contact demanded by General Requirement No. 4 can be evalu- ated. Items 3 and 4 are intended to simulate most of the aging which will take place during the life of the connection. It is believed, as de- scribed earlier in this paper, that relaxation of internal stress is the chief factor in the aging of these connections. The chill at 0°F puts the maximum initial stress in the wire by shrinking it at extreme operating MILLIAMMETER FULL SCALE 100 MA MILLIVOLTMETER/^ FULL SCALE V 7.5 MV V^ Fig. 1 — Test set for measuring resistance of solderless connection. Test Procedure (1) With Ri set at "a" and switch SW closed, adjust R2 so that the millivolt- meter reads 25 microvolts with Rx open. (2) If millivoltmeter returns to zero with test connection across Rx , no barrier film is indicated and therefore connection is closed. (3) All resistance measurements are made with R^ set to give 100 milliamperes through Rx . Then Rx = millivolts/0.100. (4) A variation of not more than 2 milliohms as the connection is moved or disturbed indicates a stable connection. 596 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 temperatures. The heating for three hours at 175°C produces the degree of relaxation expected over a 40-year period. By examining and esti- mating the size of the gas-tight areas after this process, the necessity for maintaining the intimate minimum gas-tight area demanded by General Requirement 3 is considered to be met. Item 6, the remeasure- ment of resistance is further confirmation of proper performance after relaxation. For the measurement of resistance, and barrier films a circuit is used as sho^vn on Fig. 1. With Ri set at "a," R2 is adjusted so that the voltage of about 25 microvolts is applied across R^ before the connection to be measured is inserted. This voltage is low enough to insure the absence of a film and at the same time gives a convenient reading on the test set. If the millivoltmeter drops to approximately 0 when the connection is inserted it indicates that no barrier film exists at the connection. The current is then increased to 100 milliamperes by means of the potenti- ometer, Ri, and the resistance is determined from the millivoltmeter reading. Since most of the measured resistance is in the wire rather than in the connection the important criterion of quality is the variation in resistance as the wire is moved relative to the terminal. If the variation in resistance of the connection does not exceed two milliohms when the wire is moved back and forth the connection is considered to be —TERMINAL STRIPPER Fig. 2 — Stripping force test for solderless connection. (1) The connection shall consist of six turns minimum of which at least four are close wound. (2) The connection shall be capable of withstanding a stripping force, F, of at least 30()0 grams applied as shown. (3) The wire shall be capable of being unwrapped from a terminal without breaking. SOLDERLESS WRAPPED CONNECTIONS — PART III 597 stable and the requirement for intimate contact between wire and ter- minal is considered met. In a typical local talking circuit this amount of resistance variation would correspond to a noise level of approximately — 8 db where 0 db equals 10~^^ watt and where anything less than +26 db would give no noise transmission impairment (See Appendix A). DEVELOPMENT TEST PROCEDURE — MECHANICAL REQUIREMENTS In determining whether a connection meets the General Require- ments 5 and 6, Test Procedure II as follows was devised: 1. The connection shall be capable of withstanding a stripping force of 3000 grams applied as shown in Fig. 2. 2. The wire shall be capable of being unwrapped from a terminal without breaking. These items are related directly to General Requirements 5 and 6. Experience has shown that under ordinary conditions of handling of connections as cables and frames are wired in the shop, or when a second connection is being made on an already wired terminal, a resistance to stripping in excess of about 3000 grams is required if the demands of General Requirement 5 is to be met. To insure adequate mechanical strength and current carrying capacity, a minimum of six turns is re- quired. If the turns are close wound the strip-off force is readily met. If they are open wound they may strip off with a much lower force. When the edge radius R (Fig. 3) of the wrapping tool is too small very high tension can be developed in the wire while wrapping. At very small radii the tension can be sufficient to break the wire. Although wrapping with high tensions in the wire produces a connection which will sustain very high stripping force, the wire is so embrittled that WALL THICKNESS Fig. 3 WRAPPING SPINDLE TERMINAL HOLE Solderless connection wrapping tool. 598 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 9 cc O Ah o m O o O W o O O P Pi 02 s o CO pq t5 1 Number of Connections Where Resistance Exceeded 0.002 Ohms 0)0 0 010) fl fl G a fl OOOOO o o o oooooooo £ ^ Months at 85°F Humidity -St 1 05000 1— 1 l-H 1—1 »o OiO50^(M(M(M(N(N Hours In Corro- sive Agent t5 :^^:^;:^:^:i^ :^:^::^:^^ \N ooo^:^:^:^^::^^::^ 1 ' ' WKWWK 1 ^ "^0 X ^ ^^ SHARP CORNERS (25 SAMPLES)^ 0^ ^ r ^ ^ A^ J^ - ^ u ^ ^ ROUNDED CORNERS 6 TO 8 MILSv RADIUS (25 SAMF*LES) A 0.1 12 5 10 20 30 40 50 60 70 80 90 95 98 99 PER CENT OF TERMINALS BELOW INDICATED STRIPPING FORCE Fig. 5 — Effect of terminal edge sharpness on stripping force. 10 < 3 >^ Z8 O O 7 S: TINNED COPPER WIRE ON UNTINNED BRASS TERMINALS 0.03l"x 0.062" ^^ .'^ 0 ^ ^ ANNEALED (25 SAMPLES) ^ - rr >^ A A A ^ ^^ ^ -!? ^ ,^ ^ ' ^ HARD (25 SAMPLES) „^^^^L 'a ^ ^ A A — 0.1 12 5 10 20 30 40 50 60 70 80 90 95 98 99 PER CENT OF TERMINALS BELOW INDICATED STRIPPING FORCE Fig. 6 — Effect of terminal hardness on stripping force. 99.9 SOLDERLESS WRAPPED CONNECTIONS — PART III 603 tests are shown on Figs. 5 to 9 inclusive. These charts show a plot on arithmetic probability paper of stripping force against the cumulative per cent of the sample below the indicated stripping force. The straight lines shown there are actually drawn through the average and average minus Sa points where a is the standard deviation of the observed read- ings. In the cases shown, and in fact in practically all the tests made to date the results have come out so that the experimental values fit very well on the straight hne drawn through these points, thereby indicating good normal statistical distributions of the data. 8 (/) < o o _J Z6 TINNED COPPER WIRE ON NICKEL SILVER TERMINALS 0.013" x 0.062" .-^ ^^ ^ r^ ^ TINNED (100 SAMPLES) r^ ^ >^ 12 5 10 20 30 40 50 60 70 80 90 95 ?8 99 PER CENT 0«= TERMINALS BELOW INDICATED STRIPPING FORCE Fig. 8 — Effect of wire plating on stripping force. 99.9 curves run higher than the curves for the harder nickel silver in Fig. 5. Fig. 6 shows that the presence of tin plate on a 0.013" by 0.062" terminal increased the minimum expected stripping force by more than 1,000 grams. Figure 8 shows that on a 0.031" y 0.062" terminal that while the effect of the tin plate on the wire was to raise the stripping force as before, the increase was not as great as that in Fig. 7. It is also clear that the thicker terminal of Fig. 8 results in a higher stripping force than that for the corresponding thinner terminal of Fig. 7. In analyzing the effect of tool variations it has been convenient to take a series of curves like those in Figs. 5 to 8, and from each curve select two points, namely one for the average stripping force and one for the average minus 3o- where o- is the standard deviation of the ob- served values of stripping force. Using these two points for each of several sets of curves a plot may be obtained of the average and minimum expected stripping force as one or another of the parameters of the tool design are changed. In this manner curves of the type shown on Fig. 9 were obtained. By preparing curves of this type for any given appli- cation, the permissible range, Z, of variation of tool radius, /^, can be obtained. This range includes those values of R which are above that value which results in wire breakage and below that for which the mini- mum expected stripping force falls below the stripping force limit. Table II summarizes the results of many of the tests which have been run on a large number of samples in which the materials and dimensions of the terminals, wires, tightness of wrap, etc., were varied. All of sample SOLDERLESS WRAPPED CONNECTIONS PART III 605 connections originally met the initial test requirements outlined above for resistance variation. They were then subjected to the indicated ex- posures including being kept in the Humidity Room at 85°F and 95 per cent relative humidity for periods up to two and one-half years, fol- lo\ving which they were remeasured. In no case was any barrier film found which did not break down at 25 microvolts nor was any resistance variation in excess of 0.002 ohms observed. Some of these terminals are sho^vn in the photographs of Figs. 10 and 11. Upon unwrapping some of these connections there were still considerable areas which were bright (See Fig. 12), indicating that they were still gas-tight. This would seem to indicate a life expectancy of many years. Since connections 2 300 i:?375 0C225 (/) IL O200 H Z 111 O 175 cc UJ a. z 150 ^ 100 75 50 25 ^ RADIUS BELOW WHICH r— WIRE J^ AVAILABLE RANGE BREAKS WHEN UNWRAPPING ^i^ FOR VARIATION IN RADIUS K i \ \ i \ \ \ j < j >v AVERAGE M 'N ^ \ h N v^ 1 N ^ \ 1 1 \ \> ^--^ ^ 1 1 MINIMUM ^"^ (AVERAGE -3^)^ ^ 1 1 ^ >^ 1 ~i 1 t 1 c iTRIPPIN FORCE LIMIT 5 1 1 i 1 10 20 80 90 30 40 50 60 70 PER CENT OF MAXIMUM RADIUS Fig. 9 — Stripping force as a function of radius of wrapping tool. 100 606 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Fig. 10 — Terminal at left exposed to H2S for 14 hours and then 30 months at 85°F 90%RH. The connections did not develop resistance variation in excess of the 0.002 ohm limit. Corresponding new terminal is shown at right. which meet the 3,000-gram strip-ofif requirement appear satisfactory on the other tests, the strip-off test used on a sampling basis has been stand- ardized as the shop control of the wrapping tool and indirectly of the quaUty of the connections themselves. FIELD TRIALS A number of equipment units using these connections is currently on field trial. At Tonawanda, N. Y., one trunk unit using wire spring relays of an early design and 300 solderless wrapped connections has been in service successfully since 1951. At Elmhurst, L. I., a sender frame com- prising five senders with about 7,500 connections has been in use about a year and a half with no troubles reported to date. At Boston, Mass., an outgoing sender frame for the No. 4 crossbar office comprising three SOLDERLESS WRAPPED CONNECTIONS -- PART III 607 Fig. 11 — Terminals on left exposed to H2S for one-half hour and 85°F 90 per cent RH for seven and one-half months without development of excessive resist- ance variation. Corresponding new connection is shown on right. BRIGHT SPOTS Fig. 12 — Corroded nickel silver terminal after unwrapping connection. Note the bright spots where the wire was in contact with the terminal. 608 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 senders wdth 6,000 solderless connections has been in service about nine months, again vdth no troubles reported. In order to test these connections in a transmission circuit one channel of the K-1 carrier system between Newark and Atlanta was selected. Rather than ^^dre a regular equipment unit with solderless wrapped connections, four units were built consisting of two groups of 320 con- nections in series. Each unit was exposed to 15 weeks at 85°F 90 per cent relative humidity and then inserted in the system. One unit was located at each of the following places: New York, Philadelphia, Balti- more, and Richmond. No troubles or adverse service reactions have been received to date, more than six months after installation. SUMMARY The tests described indicate that solderless wrapped connections are practical when wrapping No. 24 solid tinned copper wire on fiat punched terminals of brass or nickel silver where the mdth is one-sixteenth inch and the thickness varies from 0.010" up to one-sixteenth inch. Similar tests with heavier wire, (No. 20 to No. 23) such as would be used in distributing frames and tests with No. 24 wire on flattened, coined, or otherwise treated wire spring relay terminals have also been successful. These connections are mechanically stable, and have less tendency to break due to handling and vibration than solder connections, and will e = f L(R,) 2B NOISE 600n<' MEASURING SET son vw (a) MEASURING CONDITION 60on. e /- — \ LINE r^ \receiver / Vr A > > > 1 Vb 60on -3DB SUBSCRIBER SET -3 DECIBELS — 1 1 1 ^TRANSMITTER Vr= 2VB-4' (b) subscriber CONDITION Fig. 13 — Comparison of noise measuring circuit with subscriber circuit. SOLDERLESS WRAPPED CONNECTIONS PART III 609 probably have a central office life of more than fifty years without de- veloping objectionable resistance variation. Tests on other terminals and using wires heavier than No. 20 and on wires smaller than No. 24 are to be made. ACKNOWLEDGEMENT The author acknowledges that this paper reflects largely work done under the supervision of D. H. Gleason whose recent retirement pre- vented his authorship. Appendix A NOISE LEVEL ARISING FROM VARIABLE CONTACT RESISTANCE The measurement of contact resistance noise in a telephone central office has been standardized, and the estimate of the effect of a given resistance variation has been made in accordance with the standard technique. In this measurement, a 2-B noise measuring set is connected as shown in Fig. 13 (a) where the voltage, e, is that due to a variation in contact or connection resistance. If a dc current i is flowing and the resistance variation is (A/^)o sin w^, then ee„= ^J-(A/i!) (1) and F^ = I e. (2) In the subscriber condition (Fig. 13 (b)) there is a 3 db loss in the subscriber's loop and a 3 db loss at the subset so that the voltage at the receiver and It therefore follows that A 2e ^ 1 ^8 Vr 3*4^ 3 610 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 and 20 log Tj:^ =9 dh (to the nearest decibel). Extensive tests have shown that a level of 17 db at the receiver where 0 db = 10~^^ watts will produce no noise impairment of transmission. The corresponding level at the input to the noise meter in the measuring condition is 17 + 9 or +26 db. Currents in talking circuits may vary from 0.025 amperes in many central office circuits to approximately 0.150 amperes in short sub- scriber's loops. At a current of 0.025 and assuming AR = 0.002 ohms the noise voltage will be e = ^ X 0.025 X 0.002 = 35.4 X 10"' volts and the voltage at the input to the noise measuring set will be % e = 23.5 X 10~* volts. The power into the set will be 23.5 X 10"' 600 or 1 X 10"^^ watts or 0 db. The noise measuring set is equipped with a weighting network which takes into account the relative interfering effect of different noise frequencies on received speech. For random noise, the effect of this network is to reduce the measured noise level by about 8 db as compared with the reading that would be observed without the network. Accordingly the noise produced by the above 23.5 X 10~^ volts would measure —8 db. An Improved Circuit for the Telephone Set By A. F. BENNETT (Manuscript received August 16, 1952) A telephone set known as the 500 type has been in prodiiction for Bell System use for some time. The successful outcome of an intensive study has made it possible to simplify the circuit and some of the components of this set, and thereby to increase dependability and life and significantly to reduce the manufacturing cost. This change now has been completed and telephone seos embodying the necessary modifications are in production. This paper discusses some of the problems involved in this work and outlines the improvements which have been effected. Presented also is information concerning the superior performance of the 500 set over the preceding telephone set when used in noisy locations. INTRODUCTION One of the outstanding characteristics of the 500 type set is a 10 db increase in combined transmitting and receiving performance on long loops. This gain is equally divided between receiving and transmitting. This improvement has resulted largely from the use of a transmitter and a receiver which are not only more efficient, but also have better frequency response characteristics. To take full advantage of these trans- mission gains, two new elements were introduced in the original 500 set design. One of these elements was a better sidetone balancing circuit to offset the more sensitive transmitter and receiver, and maintain sidetone at a level no greater than it was with the previous design of set (known as the 302 type). The other was a tungsten filament and thermistor element to control automatically the transmitting and receiving levels so that the desired gains occur on a graduated basis as the loop length increases. This combination of filament and thermistor bead was called the transmission equalizer. While the transmission objectives were met with telphone sets having these elements, this additional component increased the manufacturing cost of the set appreciably, and therefore more economical means of attaining the desired results were sought. Such means have been found in the form of an arrangement in which a 611 612 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 pair of silicon carbide varistors serves for both the sidetone balancing and equalizing functions, and with certain novel modifications in the circuit of the set, yields essentially the same overall transmission per- formance as the original design at lower cost. The advantage is retained of a telephone set which from a transmission standpoint is universal in its application in the telephone plant. This universality avoids additional codes of set which require effort and expense to administer. Telephone sets having these features are designated as 500C (manual) and 500D (dial) types. The original design is known as the 500 A (manual) and 500B (dial types). GENERAL When the sidetone, which is the sound level in the receiver caused by voltage developed in the local transmitter, becomes too high, the sub- scriber subconsciously lowers his talking level thereby reducing the level of outgoing transmission.^' ^ To avoid this loss and make the higher efficiency provided by the new transmitter and receiver effective, a reduction in sidetone must be made. When the room noise level at the station is high, another undesirable effect of high sidetone occurs. This is the loss in effective receiving which results from the masking of in- coming speech by the high level of sidetone noise in the receiver. This will be discussed in more detail later. In developing the original design of 500-type set, several different equalizing means were considered, and the type involving a filament in series with the transmitter, and a thermistor bead in shunt with the receiver was selected as the most suitable with the means available at that time. Reduction in sidetone was realized by the use of a special balancing network which required an auto transformer. The set now being produced is based on a new circuit arrangement, utilizing a pair of semiconductors in the form of silicon carbide varistors, one of which has a resistance in a range which required development for this specific purpose. CHARACTERISTICS OF SILICON CARBIDE VARISTORS Fig. 1 shows the relation between the dc voltage and current for the two varistors used in the 500C and D sets. These varistors were coded 312D and 312E, but for brevity in this paper, they are referred to as Vi and V2 respectively. Also shown in Fig. 1 is the dc resistance of these varistors as determined by the ratio of applied voltage to current. It is IMPROVED CIRCUIT FOR TELEPHONE SET 613 100 60 40 20 10 6 to •3 4 o > u Q 2 1.0 0.6 0.4 0.2 0.04 \ V2 X V- s . ""^^v^ ^N ^s •* ^\l ^ \ \ V ■v \ \ S^ X s ^c" ^^ ^S J^'j>j^ ' V ><' ^ S>V ^\, — — - g ^ ^ *^'* \ S ■^ n;; 1 N% »^: ^ - N -> ^ '-.^ ""< ^V2- ^'^ "^ — ^Oc "*' "^-^ -.»^ " — — ^'^. •--.^ -.^c — set. 0.03 0.04 0.05 0.06 0.07 0 08 0.09 0.10 0.11 0.12 0.13 0.14 0.15 016 LOOP CURRENT IN AMPERES Fig. 4 — Characteristics of Vi and V2 varistors in circuit of 5(X)D telephone J IMPROVED CIRCUIT FOR TELEPHONE SET 617 >!5 UJZ OR /NO EQUALIZATION / Sv ■""■ ■*""■ -t ^ ^ % V AC LOSS . \ K V k WITH EQUALIZATION -- X ^^Oc /-o^3 i -^J 0.02 0.04 0.16 0.18 0.06 0.08 0.10 0.12 0.1 LOOP CURRENT IN AMPERES Fig. 5 — Level of 500D telephone set at various loop currents with and with- out equalization. amount for long loops to a maximum for short loops. It is important to note that the greatest portion of this loss is the ac shunting loss intro- duced by the varistors. SIDETONE BALANCE Before considering the matter of sidetone balance, it will be helpful to examine some of the features of the telephone set circuit which are required for good receiving and transmitting performance. For sim- plicity, the induction coil is shown in a hybrid arrangement in Fig. 3. The speech currents received from the loop pass through the windings A and B of the induction coil which are so connected that they produce additive voltages in winding C which is connected to the receiver. These additive voltages would cause a resultant current to flow in the network resistance (68 ohms), were they not opposed by an approxi- mately equal voltage 180° out of phase which results from the receiving current in winding B. Therefore, there is little or no voltage drop across the network resistance. Maximum receiving levels are thus obtained without appreciable power loss in the network resistance. The transmitter of the set is made low in resistance because it is in series with the loop and influences the permissible loop range of the set from the standpoint of supervision. The impedance Zi looking toward 618 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 the loop is relatively high. To obtain acceptable transmitting levels, the low impedance of the transmitter must be approximately matched to the high impedance loop by providing a proper ratio of turns between mndings A and B. Therefore, the impedance Zi looking toward the balancing network must be made low and since an important element of this network is the V2 varistor, this had to have an unusually low resistance. This is the reason why it was necessary to undertake a special development to make available a silicon carbide varistor having suitable characteristics. Sidetone would be eliminated if the vectorial sum of the voltages in the mesh which includes the receiver, and which arise as a result of speech and noise picked up by the transmitter, were zero. The complete elimination of sidetone is not desirable, but the objective is to keep it at a low level by a balance of opposing voltages. To achieve this result any voltage developed in the local transmitter is divided in the windings A and B so that the voltages induced in winding C are opposing. Further- more, the voltage across the network resistance arising from current flowing in winding B is arranged to oppose this resultant of voltages induced in winding C. The overall effect of this balance is that the current in the receiver as a result of voltages developed in the transmitter is small. Also, this result gives maximum transmitting levels because there is little power loss in the receiver. However, to maintain the balance which gives low receiver currents the impedances Zi and Z2, as affected by the transformation ratio of the coil, must be comparable. This is the key to good sidetone balance — that the impedances Zi and Zi be effectively matched both in magnitude and phase. Now in the telephone plant, the impedance looking from the station toward the loop varies widely from one installation to another, and even from one call to another. The loop may be long or short, of small gauge or large gauge, and composed of cable or open wire or combina- tions of the two. Furthermore, the loop may be loaded or non-loaded and it may be connected through central office circuits of different char- acteristics to a trunk or distant loop and telephone set which also may cover a range in impedance. It is quite obvious that under these con- ditions the impedance looking toward the loop will not only vary over a wide range in magnitude, but may be inductive, or capacitive or es- sentially resistive. If we assume that the V2 varistor were not present, the impedance 7j2 looking toward the balancing network is not influenced by loop current. This is the situation that has obtained in balancing sidetone in preceding designs of telephone set. One of the impedances to be matched IMPROVED CIRCUIT FOR TELEPHONE SET 619 was fixed and the other varied over a wide range. The sidetone balance under such conditions represented the best compromise that could be made among a large number of uncontrolled factors. Let us examine next how the varistors affect this balancing problem and consider the influence of long and short loops. From what has already been shown, the varistor is a variable impedance element — that is, both its dc and ac resistance depend on the direct current. If the loop is long and composed of open wire the impedance looking toward it is high, perhaps 1200 ohms, and the direct current is low. The dc resistance of the Vi varistor is then so high (approximately 10,000 ohms) that it has no appreciable effect on the impedance Zi. Even the ac resistance of about 4,000 ohms does not have an appreciable shunting effect on Z\. On short loops and local connections where the current is large and the impedance looking toward the loop may be of the order of 900 ohms, the dc resistance of 1,800 ohms has little effect. However, the ac resistance of 600 ohms does have an appreciable effect on Zi, bringing about a better impedance match. The impedance looking toward the balancing network and receiver without V2 connected in the circuit is approximately 75 ohms, and is not affected by the loop current. If we go through the same process of examining the shunting effect of V2 on the impedance Z2, we see that at small loop currents both the dc and ac resistance of V2 are too high to have any appreciable effect. At large loop currents the dc resistance of V2 is under 150 ohms, but the ac resistance has dropped to well below 50 ohms and has a decided effect on the impedance. Since the varistors are essentially pure resistances they tend to make the impedances Z\ and Z2 presented to the coil more resistive than they would be if the varistors were not present. This has the effect of reducing the difference between loop impedances which are capacitative, inductive, or resistive from the standpoint of phase angle and thereby improves the impedance match. The overall effect of the automatic changing in resistance of the Vi and V2 varistors is to provide better matching and thereby reduced sidetone. This is shown in Fig. 6, where the loudness level of sidetone of the 500D set is plotted against loop current. It will be noted that mth Vi and V2 open circuited, there is a continuing increase in sidetone loudness level with increasing loop current. When both Vi and V2 are connected in the circuit a decrease in sidetone level of 10 db or more is obtained at medium and large loop currents. The decrease is less at small loop currents. The close interrelation between the varistors and the circuit is well illustrated by the fact that if either the Vi or V2 620 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 108 106 104 g,02 ID Q z 100 -I UJ 92 90 y'' -/^ i*^ V, OPEN^^' ■/ V, 8.V2 OPEN /^ ^^^' / -/ ^-'^ "^M^ OPEN y^ ^ /J ;^ /^ /^ / /7 / / ^ V, & V2 CONNECTED ^ --^ 0.02 0.04 0.06 0.08 0.10 0.12 0.14 LOOP CURRENT IN AMPERES 0.16 0.18 0.20 Fig. 6. — Effect of Vi and V2 varistors on sidetone loudness levels of 500D telephone set for average speech. varistor is opened, almost all the effects of sidetone balancing are lost. It requires the presence of both Vi and V2 to obtain the impedance matching necessary for good balance. From all that has been said, it should not be thought that the side- tone balance of the 500D set is perfect under all plant conditions. The impedance looking toward the loop varies so greatly in magnitude and phase that the best overall balance involves compromise. However, the balance provided is as good as with the original design of 500 set, which was considerably better than that attained by any previous design of commercial telephone set. OTHER MODIFICATIONS The introduction of the new equalization and sidetone controls in the telephone set has required a number of changes in other components of the set to realize all possible economies. Fig. 7 is a photograph of the transmission network (425A) which was employed in the earlier 500 A and B type sets. In the new arrangement shown in Figure 8 IMPROVED CIRCUIT FOR TELEPHONE SET 621 Fig. 7 - 125A Network used with early design of sut the oUUA and li typo. the Vi and V2 varistors have been included in the network (425B), thereby providing mechanical protection for these elements. It has been necessary to design a new terminal block which is the plastic member to which the components of the network are mounted. OVERALL RESULTS By means of the Vi and V2 silicon carbide varistors and the rearrange- ments made in the circuit, the overall transmission performance of the modified set has been made substantially the same as that provided by the original design of set. The savings in manufacturing cost resulting from this modification are significant. Figure 9 shows on the left the Fig. 8 — ■ 425B Network used in present production set — the 500C and D type, showing the Vi and V2 varistors. 622 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 O Fig. 9 — The 311A equalizer and autotransformer on the left are replaced in the 500C and D set by the Vi and V2 varistors on the right, parts which have been ehminated — the 31 lA equalizer and auto-trans- former. On the right are shown the Vi and V2 varistors which provide essentially the same performance as the replaced parts. Another advantage of the change in design, though a difficult one to evaluate, is that of dependability. The silicon carbide element has an extremely long life and in this respect is inherently better than the type of equaUzer which involves the use of a heated tungsten filament. OPERATION IN NOISY LOCATIONS Since the previous paper on the 500 set was published,^ a study has been completed on the comparative performance of the 500-type set and its predecessor, the 302 type, in locations where room noise is severe. This is of particular interest, because the sidetone characteristics which have been discussed play an important part in transmission per- formance under such conditions. For many years the securing of improved performance of the tele- phone set under noisy conditions has been a serious problem. It has been established that receiving impairment in a noisy location is far more limiting than the transmitting impairment resulting from the noise picked up at the noisy location and delivered to the distant party. This is due to the fact that the signal to noise ratio in the transmitting direction is improved by the natural tendency of the speaker to raise IMPROVED CIRCUIT FOR TELEPHONE SET 623 his voice when the ambient noise level is high. A major factor in the impairment of received speech in the noisy location is the masking of received speech by noise picked up by the transmitter and delivered via the sidetone path to the receiver. Therefore, a telephone set which minimizes the masking sidetone noise during the listening interval should operate more satisfactorily in noisy locations. In many previous in- stances where the* noise conditions were extreme, a "push-to-listen" switch was provided to cut off the local transmitter while listening and thus prevent the introduction of noise to the receiver via the sidetone path. With its higher efficiency and better sidetone balance, the 500- type set approaches the good performance of such a * Variable" set under noisy conditions and provides this adequate performance to the user more conveniently and with the expenditure of less effort. Laboratory tests have been made which permit appraisal of the merits of various telephone sets or modifications of them under high noise level conditions. These include talking and listening tests between two telephone stations under typical loop and central office conditions with an adjustable amount of typical room noise at the listening end. During the tests the length of trunk between the two stations was in- creased until the received speech was just sufficiently intelligible to per- mit carrying on a conversation. This threshold of intelligibility expressed in db of trunk loss was taken as a criterion of the performance of a given set. Fig. 10 shows the results of the tests made with the 500-type set compared with the 302-type set, and with variations of the 500 type set. 500C- and D-type sets having silicon carbide varistors were used in this comparison, but in this respect the performance is no different than for the earlier design (500A and B) . The curves shown are for the average of eight observers. The abscissae of the curves are the noise levels at the listening end expressed in db above 10~ watts per sq. cm. For reference purposes it should be kept in mind that the average noise in a fairly large business ofl&ce is, on this scale, approximately 65 db. For further orientation as to the significance of the noise levels shown, it should be noted that when a level of about 100 db or higher is reached, it is extremely difficult to carry on a face-to-face conversation where no telephone is involved. The ordinates show the relative trunk loss in db over which it is possible, under the stated conditions, to just carry on a telephone conversation. Therefore, the larger the trunk loss, the better the performance. It is seen from Figure 10 that the 500-type set is from 6 to 8 db better than the 302 set over the indicated range of ambient noise conditions. 624 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 80 82 84 t06 86 88 90 92 94 96 98 100 102 104 NOISE LEVEL AT LISTENING END IN DECIBELS (above 10-'6 WATTS PER SQ Cm) Fig. 10 — Effect of noise at listening end on trunk loss. This improvement is attributed to the following: 1 . The better sidetone balance of the 500-set circuit reduces the level of sidetone noise picked up by the transmitter and reproduced by the receiver, thereby providing a gain in signal to noise ratio. 2. The acoustic impedance looking from the ear canal into the Ul receiver of the 500-type set is lower than the acoustic impedance of the HAl receiver of the 302 set and, therefore, the acoustic noise pressure built up in the ear canal by leakage around the receiver cap is lower than when the HAl receiver is used. 3. The output from the transmitter of the 500-type set under noise conditions contains less extraneous distortion products, giving it a char- acter which is less disturbing than with the 302 set. Included in the laboratory tests were experiments in which the out- put of the transmitter was intentionally lowered. It is known that a speaker in a noisy location instinctively raises his talking level. Since it is desirable to keep the sidetone level low to improve the signal to noise ratio, the output level of the transmitter at the listening end was reduced about 10 db by shunting a resistance of approximately 40 ohms across the transmitter. The sensitivity of the transmitter is reduced IMPROVED CIRCUIT FOR TELEPHONE SET 625 and sidetone noise is directly lowered. The level of the received speech is unaffected. Therefore, the signal to noise ratio in the receiving direc- tion is improved. The transmitting level is not too adversely affected, because the tendency of the subscriber to increase his talking level when the room noise level is high largely offsets the lower eflficiency of the transmitter to speech sounds. These effects are borne out in the applicable curve of Fig. 10 which shows that an improvement of approximately 5 db in trunk results from the shunting of the transmitter. Carrying this experiment further and acoustically shielding the local transmitter completely from noise, the lower curve of Fig. 10 was ob- tained. This is equivalent to shorting the transmitter as is done in the earlier "push-to-listen" types of telephone set, except that it does not introduce any electrical effects in the circuit. This shielding provides about 4 db additional discrimination against noise. This, then, indicates the limit to which we can go in reducing the effects of room noise on received speech by operating on the transmitter element alone. From the data presented in Fig. 10 it is evident that the 500-type telephone set provides a significant improvement when the subscriber is carrying on a telephone conversation under noisy room conditions. While it has been indicated above that the greatest gain of the 500-type set over the 302 type under high room noise conditions is with respect to received transmission, it is also better in transmitting from noisy locations. This is because of the less disturbing character of the trans- mitted noise and because the signal to noise ratio is improved by having the transmitter closer to the subscriber's Ups, which is a feature of the new handset. Where noise conditions in the field are severe, the 500-type set will provide material improvement. Typical of such conditions are noisy business locations, telephone stations which are located in power plants and near loading ramps at airports. Where the noise is particularly severe, the provision of a 40-ohm shunt resistance on the transmitter of the 500-type set offers still further improvement and is recommended for application by the local telephone people. CONCLUSION The simplification of the 500 set made possible by the use of silicon carbide varistors, with its attendant reduction in manufacturing and maintenance costs and increase in life represents a significant step forward in station set design. A valuable characteristic of the 500-type set is the substantial improvement in transmission performance under conditions of severe ambient noise. 626 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 REFERENCES 1. Iiiglis, A. H., and W. L. Tuffriell, An Improved Telephone Set. Bell System Tech. J. 30, pp. 239-270, April, 1951. 2. Gibbon, C. O., The Common Battery Anti-Sidetone Subscriber Set. Bell Sys- tem Tech. J. 17, pp. 245-257, April, 1938. 3. Mott, E. E., and R. C. Miner, The Ring Armature Telephone Receiver. Bell System Tech. J. 30, pp. 110-140, Jan., 1951. 4. Inglis, A. H., Transmission Features of the New Telephone Sets. Bell System Tech. J. V. 17, pp. 358-380, July, 1938. Automatic Line Insulation Test Equipment for Local Crossbar Systems By R. W. BURNS and J. W. DEHN (Manuscript received February 9, 1953) Moisture entering faults in the insulation of subscriber lines provides so-called "leakage^^ paths which reduce the insulation resistance. Testing the insulation resistance of all lines under the environmental conditions which tend to produce these leakages is a maintenance technique, relatively new, for detecting the insulation defects. The faults can then be corrected before they become serious enough to affect the customers^ service. Subscriber reports are thereby reduced and the correction of the faults on a preventive maintenance basis tends toward a more uniform work load for the repair personnel. Rapid testing of the lines is necessary, otherwise the environ- mental conditions may change and the leakages will disappear without de- tection. Rapid line insulation testing is practiced quite generally in all the switching systems throughout the Bell System, but the testing arrangements used are wholly or partially manually controlled in the testing and recording operations. While the benefits derived from rapid line insulation testing apply to all systems alike, this article is confined to a discussion of the entirely automatic testing and recording arrangements which are now being introduced in the No. 1 and No. 5 crossbar systems. INTRODUCTION The insulation resistance of subscriber lines is an important considera- tion in the design and operation of central office switching circuits. If the insulation resistance between the two conductors of the line, or the insulation resistance from the "ring" or '^battery side" to ground, be- comes low enough, the "leakage" current flowing produces the same effect as lifting the handset and failing to dial or to pass a number to the opera- tor. This condition is called a permanent signal and the line is said to be "permanent." As long as the condition persists, the line is out of service both to incoming and outgoing traffic. Insulation resistance of a slightly 627 628 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 higher value mil not cause a permanent signal, but its presence may- interfere with other circuit functions, for example, dial pulses are dis- torted and a Avrong number may be reached, the ringing signal may be tripped before the called party answers, or the switching circuits may fail to restore on hang-up of the receiver. If the insulation resistance is at least as high as the design limit, failures of the kind described above will not occur. The design limit for some switching systems used in the Bell System is 10,000 ohms; for others 15,000 ohms. WHERE LINE LEAKAGES OCCUR The outside plant distribution system for exchange lines usually con- sists of some underground cable, many miles of aerial cable to distribute the line conductors throughout the area and a small amount of open ^vire on the fringes. The insulation resistance of the cable conductors is normally quite high. If, however, a break in the cable sheath occurs moisture may enter and be absorbed by the paper insulation of the conductors near the break. This reduces the insulation resistance of the affected cable pairs. Sheath breaks may exist for a considerable length of time without reducing the insulation resistance sufficiently to cause any reaction on customer service. Eventually these sheath breaks will, if not detected and repaired, permit the entrance of sufficient water during a rain to reduce the insulation resistance to the point where permanent signals occur on several pairs. Then repairs must be made on an emergency basis to guard against a complete failure of all line cir- cuits in the cable. One of the common causes of sheath breaks in some residential areas is gnawing by squirrels — squirrel bites. Cable terminals are located on the poles or on the cable for making drop wire connections to the customers' premises. Binding posts mounted in a face plate within the terminal are connected to some of the cable pairs through a terminal cable stub joined to the aerial cable. Each cable pair is thus connected to binding posts in about four or five ter- minals on the average. When water or condensation wets the face plate, leakage currents will flow between the binding posts. If dirt and dust have accumulated on the face plates, the combined resistance of all leakage paths in parallel across the pair or to ground may become relatively low. While open wire makes up only a small part of the outside plant circuits most areas have some lines containing from a few spans up to several miles of open-wire conductors. It is difficult to keep the open-wire plant in a condition so that it will be free from leakages under wet INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 629 weather conditions. Trees grow up into the wires and during rainy- weather the branches often drop across the wires. When this oc- curs at many points on the open-wire run the combined leakage along the pair may become very low. In some localities, moss growing on the \\ires, salt spray or heavy fog conditions causing leakage at insulators are additional causes of low insulation resistance on open-wire lines. Drop wire used to make the connection from the cable terminal to the customer's premises is subject to damage by abrasion and the insulation deteriorates from the effects of the weather. The old style of drop wire, a large amount of which is still in use in the plant, is in- sulated with a rubber compound and covered with a water proof cotton braid. When this braid protection is lost due to abrasion or effects of the weather after many years in service cracks form in the rubber in- sulation. Moisture enters these cracks in wet weather causing low re- sistance leakages. It is expected that the latest type of drop wire with the tougher neoprene covering will withstand the hazards for a longer period of time than does the old drop wire but undoubtedly the end of its useful life wdll ultimately be determinable by measurement of the insulation resistance under wet weather conditions. Inside wiring on the customer's premises often remains in service for a long period of time and the insulation deteriorates over the years. If the insulation is in poor condition, inside wiring will develop leakages during periods of prolonged high humidity indoors which occur fre- quently during the summer months. EFFECTS OF WEATHER ON LINE INSULATION RESISTANCE When weather conditions are favorable — clear weather with low humidity — the insulation resistance of the line conductors in the out- side plant is quite high compared to the design limit for central office switching circuits. When the plant becomes thoroughly wet from a hard rain the insulation resistance drops very considerably because of leakages which are in parallel all along the line. Fig. 1 shows this very clearly. The data for the curves in this figure were collected in a special study conducted in 1931 to determine the insulation resistance distribution of exchange lines in underground and aerial cable and open-wire plant under different weather conditions — dry, humid and wet. About 6,000 dial lines selected at random in twelve large cities in different parts of the United States were tested under different weather conditions and no repairs were made throughout the study period except to correct un- satisfactory service conditions. The tests were made with the voltmeter 630 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 „ o Z D OZ 0.8 ^E 0.6 a. §0.5 pz 0.4 dI- 0.3 U UJ -J 0.2 z I 0.1 / WET WEATHER y TESTS / / DRY WEATHER TESTS / / / / / / / - / / - / f / / / / y / y ^ / / 1 1 1 10-* 106 INSULATION RESISTANCE IN OHMS Fig. 1 — Wet weather versus dry weather tests — all types of Outside Plant combined. test circuit of the local test desk which is used for testing subscriber Unes reported in trouble. The speed at which lines can be tested from the test desk is necessarily slow because each number must be dialed or called individually therefore, only a small sample of hues was selected from any one office. Rapid line insulation test equipment was not in use in any area when this study was made. The upper curve of Fig. 1 shows that about 0.3 per cent of the Hues were below 11,000 ohms in wet weather while in dry weather only about 0.3 per cent of these same lines were below 140,000 ohms. Similarly, in wet weather two and one-quarter per cent of the lines were below 47,000 ohms but in dry weather only two and one-quarter per cent were below 900,000 ohms. The same wet weather test data summarized by types of outside plant is shown in the curves of Fig. 2. Lines in underground cable only, show the highest line insulation resistance. Those with aerial cable are next and those with some open wire have the lowest insulation resistance. Fig. 3 shows the wet weather line insulation resistance distribution of 11,600 lines in an eastern city where the exchange outside plant had been thoroughly reconditioned prior to conversion from manual to cross- bar dial operation in 1940. The upper curve represents the approximate distribution before reconditioning and the lower curve represents the INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 631 distribution after the reconditioning was completed. The lower curve can be considered as representative of an outside plant in good condition as of the year 1940 when rapid hne insulation testing was not yet used. Since that time, a considerable number of improvements leading to better insulation resistance characteristics have been made in outside plant items, such as drop wire and cable terminals, and a higher av- erage insulation resistance would currently obtain. However, during wet weather, an undesirably large number of lines would still be closer than desired to the design limit. MAINTENANCE WITHOUT RAPID LINE INSULATION TESTING It may appear at first sight that the per cent of lines near or below the design limit is so small as to be unworthy of particular notice. How- ever, an examination of this from a maintenance standpoint ^vill prove otherwise. The testing of lines reported in trouble and dispatching of craftsmen on the outside to make repairs are handled from a local test center which on the average serves about 50,000 stations. There will be about six local test desk positions manned to do the testing and dis- o 2 < (OUJ UJoC z CL Z uj m Q.> >z 5^. 8 6 5 4 3 2 1.0 0.8 0.6 0.5 0.4 0.3 0,2 0,1 - r / OPEN WIRE - / / / . f / AERIAL CABLE / / / J / y / / / / / f / / UNDERGR CABL OUND E - O / / / / / / yy / y J f / > / 1 1 _l_ 10' 8 1d8 Fig. 2 2 34568 ^0^ 2 3456 INSULATION RESISTANCE IN OHMS Wet weather tests — insulation resistance by types of Outside Plant. 632 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 patching work. Such a test center would handle at the current trouble rate about 120 reports on subscriber lines per day. If the outside plant is in a condition represented by the lower curve of Fig. 3 where 0.2 per cent of the lines are below 10,000 ohm resistance, this represents 100 additional stations in the region where service reactions may be expected from wet weather conditions. Consequently where these plant conditions obtain, there is usually a noticeable increase in subscriber's reports in wet weather and the repair load rises sharply. Without line insulation test equipment to test the lines rapidly while the low insulation re- sistance obtains, there is no way to pick out the worst lines from an insulation resistance standpoint after the weather has cleared. Visual inspections are costly and superficial indications do not always give reliable evidence of low insulation resistance. Consequently the large percentage of repairs to correct insulation defects are made as a result of subscriber reports and only a small per cent by routine preventive maintenance where rapid line-insulation testing equipment is not used. MAINTENANCE WITH RAPID LINE INSULATION TESTING The only experience to date with the fully automatic line insulation test equipment for crossbar offices is in the Media, Pa., No. 5 crossbar office. This test equipment has been in use for about one year and low insulation resistance cases recorded on each test have been investigated and repairs made. The condition of the outside plant in the exchange area is represented in Fig. 4. The curves in this figure are based upon the results of tests of the 4200 working lines in the summer of 1952 under very wet conditions of the outside plant. The small percentage of lines below 40,000 ohms indicates that the poor insulation cases are being corrected well before reaching the stage where the customers' service would be affected. This is done with a minimum of maintenance effort as the test equipment spots the line automatically. DETECTING SHEATH BREAKS While sheath breaks in aerial cable are brought to light by rainy weather it is inadvisable to wait for rain to disclose them because of the possible serious effects on service and the need then for repairs on an emergency basis. Rapid line-insulation testing techniques have been very successful in disclosing sheath breaks in aerial cable. During the night the cable sheath cools and as the pressure inside decreases, outside air with a high moisture content enters the sheath break. The paper INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 633 uict: ILP 2 cr z LUai a.> UJ' >z o z I 1.0 0.8 0.6 0.5 0.4 0.3 0.2 0.1 - - _^^^ • ^^'before reconditioning y / ^-■' - - /after reconditioning / y / / / 1 _L 1 1 10^ 5 6 2 34568 10^ 2 3 insulation resistance in ohms Fig. 3 — Wet weather tests — reconditioned Outside Plant. 8 ,o6 insulation of one or more pairs near the break absorbs the moisture and the insulation resistance of the affected pairs is lowered. Tests made from about 3 A.M. to 5 A.M. are effective in detecting these leaks and by applying the latest fault locating methods, many of the sheath breaks can be located. The test equipment is so arranged that, for the most part, lower leak conditions on the line which may be present outside of the aerial cable will not register on these tests for cable defects. BENEFITS DERIVED FROM RAPID LINE-INSULATION TESTING With the aid of rapid line-insulation test equipment the maintenance personnel can correct the greater part of insulation defects on a pre- ventive maintenance basis which results in a reduction of subscriber reports. Service to the customer is thereby improved, maintenance effort is reduced and repairs on an emergency basis often involving the ex- penditure of overtime are required less frequently. Rapid line-insulation testing is also of great value in rapidly determining the extent of storm or flood damage and makes it possible to direct immediate efforts where the greatest benefits will be derived. 634 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 UJ Z p<0.2 - -^ — - -1— ] - / f / / > 1 1 r 1 1 1 1 Fig. 4 used. INSULATION RESISTANCE IN OHMS — Wet weather tests — automatic line insulation testing equipment CROSSBAR TEST EQUIPMENT — TYPES OF TESTS AND SENSITIVITIES The crossbar line-insulation test equipment is arranged to make three different kinds of tests from the standpoint of the way in which the resistance measuring circuit (line-insulation test circuit) is connected to the line and to battery or ground. Each kind of test may be made in three different resistance ranges, hence there are nine different tests. When the test equipment is performing one of these tests a record is made of all lines which have an insulation resistance below the top limit of the particular test. Each test range is divided into three bands of resistance and the record indicates whether the insulation resistance lies within the first quarter (low band), the second quarter (medium band) or the upper half (high band) so that preference can be given to the worst cases in clearing the trouble. The test numbers, ranges and resistance bands are shown in the Table I. SHORT AND RING GROUND TEST The first kind of test is called "short and ring ground test." This test measures the leak in the way in which it affects the line circuits, pulsing circuits and supervisory circuits of the central office switching system. The hne-insulation test circuit (LIT circuit) which is described later on is connected to central office battery potential and to the fine in the manner shown in Fig. 5 (a). As indicated in this illustration, leakage resistance from ring to tip and from ring to ground are measured INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 635 Table I — Crossbar Line-Insulation Tests Types of Tests and Test Numbers Range Resistance Bands* (kilo-ohms) Short and Ring Ground Tip and Ring Ground Foreign E.M.F. Low Medium High 1 2 3 Not used 4 5 6 Not used Not used 8 9 A B C D 0-40 0-160 0-640 0-1250 40-80 160-320 640-1250 1250-2500 80-160 320-640 1250-2500 2500-5000 * The equipment is arranged so that by cross connection changes the bands of resistances in each range may be halved or doubled, if necessary, to meet local conditions. in parallel. If the line-insulation resistance is more than the top limit of the test range, the test equipment proceeds to the next Une. If, however, the resistance measured is less than the top limit the test equipment immediately retests the line with the central office ground removed as shown in Fig. 5 (b). This re-test measures the leakage from ring to tip. If leakage from ring to tip only is indicated the drop wire is probably the cause. The No. 1 test is run under wet weather conditions to detect all lines which are below the top limit of this test (normally 160,000 ohms). If the maintenance force keeps these cases cleared out by running tests during every rain the wire plant can be kept in good condition. During long periods of dry weather all line insulation test indications pre- viously recorded may have been investigated, then if a Ught rain occurs it may be desirable to run the No. 2 test to provide a satisfactory number of failure indications for subsequent investigation. The No. 2 test can also be used to detect leakages in inside wiring which occur frequently during the summer when houses are unheated and the inside humidity r TIP H'l IJ-400/XF LINE INSULATION TEST CIRCUIT TIP + -400//.F 1 LINE INSULATION TEST CIRCUIT RING 1 CONTACT OPENED DURING TEST (a) FIRST TEST CONDITION MEASURES COMBINED SHORT AND RING TO GROUND RESISTANCE (b) RETEST AFTER FAILURE ON COMBINED TEST MEASURES SHORT ONLY Fig. 5 — Short and ring ground test. 636 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 is very high. Tests to disclose insulation defects in inside wiring are made when the weather is clear with low outside humidity at which time leakages in other parts of the exchange plant will rarely be found. TIP AND RING GROUND TEST The "tip and ring ground" test can be made in three ranges of sensi- tivity as shown in Table I. The arrangement of the test circuit for this type of test is shown in Figure 6. The first test condition, Fig. 6 (a), measures the combined leakage resistance from tip and ring to ground. The tip and ring are connected together therefore leakage across the pair is not indicated. This eliminates practically all of the drop wire TIP CONTACT OPENED DURING TEST TIP RING LINE INSULATION TEST CIRCUIT -400 < < I > L (b) FIRST TEST CONDITION MEASURES COMBINED TIP TO GROUND AND RING TO GROUND RESISTANCE RETEST AFTER FAILURE ON COMBINED TEST MEASURES RING TO GROUND RESISTANCE Fig. 6 — Tip and ring ground test. leakages. If a failure is indicated on the first test condition, a re-test is made under the condition shown in Fig. 6 (b) to check only the resistance from ring to ground. If the ring tests clear it is known that the leakage is from tip to ground. This kind of test is of particular value in checking terminal face plate leakages which are predominately leakages from tip to ground. This type of test can also be used to check that tip con- ductors on party lines in message rate areas are free of low resistance grounds which might result in wrong party identification on calls made by the ring party. FEMF TEST The FEMF (foreign e.m.f.) test is used to measure leakages in cable to detect sheath breaks. These leakages are high resistance compared to other leaks which may be present across the pair connected for test or from the pair to ground. To make the latter ineffective so as not to INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 637 mask the higher resistance cable leakages the test circuit is grounded and connected to the line with the tip and ring short circuited as shown in Fig. 7. The test circuit measures the leakage current flo^\dng from the pair connected for test to the ring conductors of other subscriber lines which are connected to battery potential in the central office. Leakages outside the cable will not cause leakage currents to flow through the line insulation test circuit. This test is called FEMF because it measures leak in the same way as does the test desk voltmeter when the FEMF test key is operated to connect ground instead of test battery to the voltmeter. Leakages as high as about 2 megohms can be successfully located after the cable has been identified by analysis of pairs affected by leaks. r LINE INSULATION TEST CIRCUIT LINE UNDER TEST XH' LINE RELAYS LEAKS IN CABLE TO OTHER RING CONDUCTORS CONNECTED TO CENTRAL OFFICE BATTERY (FEMF) Fig. 7 — Foreign E.M.F. test. TEST AND RECORDING CIRCUITS The equipment for automatically measuring and recording line-in- sulation resistance consists of three principal parts. First, the device for measuring the low currents involved; second, the means for connect- ing this measuring device to each of the lines in rapid succession; and third, the apparatus required for recording the faults discovered. The first of these, kno\vn as the line-insulation test circuit, is capable of detecting insulation resistance as high as ten megohms, with fast enough response, so that a satisfactory rate of line testing (about 10,000 to 12,000 lines per hour) is attainable. It is provided wdth a filter to attenuate both 25- and 60-cycle interference, so that leakage faults may be separated from other types such as induction from power lines. As shown in Table I, its sensitivity may be easily changed, so as to be compatible with atmospheric conditions, the kind of test to be made, and the condition of the outside plant. When a particular sensitivity or 638 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 "range" is chosen, the insulation fault detector will report all lines having lower insulation resistance than the chosen value, and as already stated it will further identify those lines having insulation resistance lower than one-half and one-quarter of this resistance. This test circuit is connected to each of the lines in rapid succession by the line-insulation test control circuit. As shown in the block diagram of Fig. 8, the control circuit appropriates the control and testing wires between one of the markers (usually marker No. 2) and each of the line link frames. The arrangement shown is for a No. 5 crossbar office, but the No. 1 crossbar arrangement is similar in principle. By means of these appropriated connections the control circuit is able to select the LINE CONNECTED I FOR TEST LINE INSULATION TO MARKERS Fig. 8 — Block diagram of line insulation LOCAL TEST CENTER test control circuit connections. ^ INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 639 lines and if the selected line is not busy its tip and ring conductors are connected to the insulation resistance measuring circuit (line-insulation test circuit). To establish this connection, the control circuit chooses an idle line link to connect the line vertical to the no test vertical and the test path is completed through the no test connector to the line-insulation test frame. The operation of the control circuit may be started in the central office by operation of keys at the test frame, but it is more commonly started by remote control from the local test center. This permits in- sulation tests to be made even though the central office is unattended. When the control circuit is started, it must also be directed to make one of the three types of tests, and to choose one of the three test sensitivities for each of these types as shown in Table I. This is accomphshed by operating one of nine keys at the test frame or by selecting the test trunk at the test center and dialing one of nine codes and then operating a key at the test desk. Either action chooses the type of test and the sensitivity, and causes the test circuit to start. A tenth key or a tenth code is used to stop the test before the end of a complete cycle, when this is required, so that the type of test or the sensitivity may be changed. A pre-arranged regular pattern is followed in testing the lines . A hne link frame is selected, a particular five lines are tested, and then the frame is released, and the next frame is selected, and another five lines are tested, and so on. Lines found busy, dial PBX trunks and line link frame terminations of intertoll trunks and similar circuits which would cause false operation of the test circuit are passed by without test. The line link frames are selected in order, beginning with the lowest numbered frames and continuing to the highest. When one cycle through the frames has been completed, another cycle will be started. However, on the next cycle, different groups of five lines will be tested. On the first cycle, the five lines in vertical group 0, horizontal group 0 will be tested, on the second cycle, the five lines of vertical group 1, horizontal group 0 will be tested, in each line link frame and so on until all lines in horizontal group 0 of all line link frames have been tested. On subsequent cycles the lines of other horizontal groups are tested in regular order. Testing only five lines for each selection of a line link frame minimizes interference with traffic. Since one to two seconds is required for testing five lines, there will be only a slight delay to calls which require access to the frame. In addition, a heavy traffic load will cause the control circuit to stop insulation testing and restore to service the marker whose cabHng it has been using. When the traffic has been reduced sufficiently, the test will be restarted automatically. 640 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 When a line fault is discovered, the control circuit makes use of either of two types of mechanism to record (a) the location of the line on the line link switches, (b) the type of test being made, and (c) a rough approxi- mation of the insulation resistance. Since a substantial time is required to make this record, the line link frame is restored to service during this interval, to reduce interference with service. One of the above two devices is the trouble recorder, used only in the No. 5 crossbar offices. The control circuit connects itself to this machine in much the same way that a marker does, when it needs to record a trouble. Having made this connection, a card is perforated showing the Une location and other data pertinent to the trouble indicated. The other device consists of teletype- writer equipment at the central office which transmits the required data to a teletypewriter page printer located at the local test center. The equipment at several central offices has common use of the same page printer at the test center and several test circuits in one building use the same teletypewriter transmitting equipment. This second recording arrangement is the more convenient of the two, since it produces the record at the place from which the activities of the outside plant repair force are directed. This arrangement is available for both No. 1 and No. 5 crossbar offices. A typical teletype record is shown in Fig. 9. LINE-INSULATION TEST CIRCUIT The insulation resistance measuring device is required to respond, not only to the very low current (five micro-amperes) obtainable with insulation resistance up to ten megohms, but it must also give good discrimination between resistance values in the order of 20,000 ohms. This is accomplished by desensitizing the measuring device by means of series and shunt resistors when a test using less than the maximum range is desired. A second requirement is that the measuring device be low in resistance so that the time constant of this resistance in combina- tion with the line and filter capacity will be low enough to attain high testing speeds. These leakage current amplitudes are so small that amphfication is required in order to actuate the recording and control mechanisms of the measuring system. These small direct currents could be amplified by means of a dc amplifier. However, since it is easier to design and construct an ac amplifier of suitable stability, it is desirable to use a measuring device which has an alternating voltage output which varies with the direct current input. A type of magnetic modulator, called a magnettor, which has these INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 641 desirable characteristics, is used as the current measuring instrument. It has a low impedance input circuit, in which the low amplitude dc leakage current flows. An alternating current of constant amplitude and frequency is supplied to separate windings of the magnettor, so that, as explained below, its output circuit delivers an alternating voltage which varies with the dc input. Fig. 10 shows the basic circuit, which operates as follows. Two identical mndings, a and a', and two other identical windings, b and b', are wound on identical permalloy cores. Windings LINE LINK FRAME BAND TEST NUMBER OFFICE u 0-82 0-82 0-82 0-82 0-82 0-80 0-82 0-82 0-80 0-82 -VERTICAL GROUP -HORIZONTAL GROUP -VERTICAL FILE 0801-01 0704-10 0705-22 0007-21 0205-32 •0406-63 •0800-74 ■0601-70 •0806-73 ■0100- S3 Fig. 9 — Teletype record of failures. a and a' are connected in series and supplied with an alternating current. The magnettor cores have a characteristic as sho^vn in Fig. 11 (b), and the amplitude of the input voltage is great enough so that the core is driven to saturation on each half cycle. Fig. 11 (c) shows the resulting flat topped flux versus time curve*. Since the voltage induced in winding b is proportional to the rate of change of flux, it will have a wave form as sho^vn in Fig. 11 (d). The voltage peaks occur when the flux rate of change is maximum, and during the "flat" intervals the induced voltage is small. Since the b and b' windings are connected so that their output voltages are in opposition (see Fig. 11 (d)) the net output with no dc * The wave shapes of Figs. 11 (c), 11 (d), 11 (e) and 11 (f) have been exagger- ated to illustrate the action involved. 642 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 input will be zero. Manufacturing tolerances which produce dissimilari- ties in the mndings and cores may cause a small output with zero dc input. When the dc leakage current flows through windings b and b', a bias flux is established in each of the cores. This causes the shape of the flux versus time curve to be changed as illustrated in Fig. 11 (e). Since the ac input will saturate the cores, both half cycles of the wave will be flat topped, but the flat portion of one-haK cycle will be increased and the other decreased. Also the steeper parts of the curve will be brought closer together on one-half cycle and further apart on the other. As illustrated in Fig. 11 (e) the dc bias will have a different effect upon the flux in each of the two cores. In one core the bias aids the positive half cycle of the input current, and in the other the negative half cycle. The result is that the core reaches saturation more quickly and, therefore the flux curve is flatter on the half cycle which is aided by the bias. This change in the shapes of the flux time curves produces correspond- ing changes in the shapes of the voltage-time curves of the output windings b and b'. These are shown in Fig. 11 (f). The peaks of the voltage curves occur at the points of maximum rate of change of the flux curves, and of course the flat portions of the voltage curves (near zero) are lengthened or shortened depending upon the flatness of the flux curve tops. This skewing of the two voltage curves prevents the output voltage cancellation which was obtained with no dc, and gives MAGNETTOR DC ■^ DC 4/i.F /S'c 2000 -CYCLE BAND PASS FILTER AMPLIFIER [> DETECTOR CONNECTED TO LINE ^X rH5!5- AC INTERFERENCE FILTER I I I I I ■* ADJUSTED FOR DIFFERENT TEST RANGES BY TEST CONTROL CIRCUIT > ..X-J Fig. 10 — Block (liagnim of line insulation test circuit. INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 643 a resultant output which contains even harmonics of the input voltage. The second harmonic is selected by means of a filter for use as an in- dicator of the presence of leakage current. The second harmonic is amplified by a three stage negative feedback vacuum tube amplifier, whose output is applied to the grids of three thyratrons. Adjustments are made, as described in the following, so that one of the thyratrons conducts when the insulation resistance is less than the value selected for the test, an additional thyratron conducts if the insulation resistance is one-half of this value or less, and all three conduct if the resistance is one-fourth or less of the selected value. Three relays, whose windings are connected in the anode circuits of the thyra- trons, are actuated when the associated thyratrons conduct and cause a record of the trouble to be made, or cause the control circuit to select another line if the insulation resistance is above the range of interest. In order to calibrate the detecting circuit, a test resistor of 160,000 ohms is connected to the magnettor by operation of a key. The amplifier gain is then adjusted so as to just cause conduction of the thyratron (b) (c) (e) VIAGNETTOR CORE CORE FLUX VS TIME CORE FLUX VS TIME CHARACTERISTICS (NO BIAS) (DC BIAS) FLUX RESULTANT OUTPUT OF b & b' EVEN HARMONICS ONLY Fig, 11 — Graphical representation of voltage and flux in the magnettor. 644 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 LINE INSULATION TEST CIRCUIT Fig. 12 — Front view of line insulation test frame. INSULATION TEST EQUIPMENT FOR LOCAL CROSSBAR SYSTEMS 645 which responds to the lowest insulation resistance value. Test resistors of 320,000 ohms and 640,000 ohms are then substituted and the grid bias of the other thyratrons adjusted so that they just conduct on the proper value of resistance. This procedure cahbrates the device for one range, using a shunt and series resistor which corresponds to this range. Facihties are provided for substituting suitable test resistors for check- ing the cahbration of other ranges, with the shunt and series resistors for the range connected to the magnettor. EQUIPMENT FEATURES The apparatus components of the test control circuit and the line insulation test circuit are assembled and wired in an 11 -foot bay, 23 inches wide. Fig. 12 is a front view of the test frame. Fig. 13 is an enlarged view of the line insulation test circuit equipment and the control panel which includes features for checking the accuracy of the test circuit and for calibrating it. When the teletype method of recording failures MAGNETTOR Fig. 13 — Front view of control and test panel (at bottom) and line insula- tion test unit (center) for No. 5 crossbar. 646 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 is used, the teletype transmitting equipment common to all line insula- tion test frames in a building is mounted in a separate 23-inch bay and occupies about one-third of the vertical space. CONCLUSION The initial installation of the automatic hne insulation test equipment in a No. 5 crossbar office has been in operation for more than a year and the maintenance advantages of remote control of starting and automatic testing and recording have been fully demonstrated. It is expected that any future developments of line insulation test equipment undertaken for other switching systems will follow this same general pattern. ACKNOWLEDGEMENT The authors gratefully acknowledge the assistance of R. C. Avery, F. E. Blount and D. H. MacPherson in the preparation of the technical descriptions and analyses presented in this paper. Theory of Magnetic Effects on the Noise in a Germanium Filament By HARRY SUHL (Manuscript received October 10, 1952) A magnetic field will influence the current noise in a germanium fila- ment. This fact hears out the hypothesis that at least part of the noise arises from minority carriers emitted in random hursts and recomhining at the surfaces. A quantitative theory of this effect is given. INTRODUCTION In a series of fundamental experiments, H. C. Montgomery^ has es- tablished that minority carriers play an important part in the current- noise associated with semiconductors. He found that on the one hand, the noise voltage is usually proportional to the biasing current, suggest- ing fluctuations in the conductivity, and hence the carrier concentration. On the other hand the spectrum of the noise suggested a rather coarse- grained time variation, not likely to be caused by fluctuations in the normal carrier density. One might conclude, therefore, that the noise is caused by a distribution of sources emitting or absorbing minority carriers in random bursts. Such carriers would be subject to the same laws of motion and of recombination as intentionally injected carriers. Montgomery was, in fact, able to verify that the noise along a filament showed marked correlation over a distance roughly equal to that through which minority carriers could drift in the biasing field before recom- bination. W. Shockley has pointed out another corollary of this theory: A mag- netic field transverse to the filament should have a pronounced effect on the noise. This conclusion, too, Montgomery was able to verify experimentally.^ His results are in good qualitative agreement with theory. Complete quantitative agreement was perhaps not to be ex- pected, since technical difficulties prevented attainment of the idealized conditions assumed by the theory. This paper gives an account of that 647 648 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 theory. On it are based the computed curves in Montgomery's paper showing the change in noise power with magnetic field. To see how such a change comes about, we imagine the magnetic field applied normal to one pair of the long faces of a rectangular fila- ment. This field, and the longitudinal drift current used to measure the noise, 3deld a sidewise thrust on the carriers, directed at right angles to the other pair of long sides. As a result the density distribution over the cross section is distorted, the minority carriers tend to accumulate near one of those sides, while the neighborhood of the opposite side is depleted. But for the usual conditions the recombination of carriers occurs mainly near the surface, and is proportional to their density there. Hence the magnetic field will change their lifetime.^* ^ Clearly the amount of noise is dependent on the length of time carriers are able to contribute to the change in conductivity, that is, dependent on their lifetime. Therefore, the magnetic field should change the noise power. In simple extreme cases one can even make a semiquantitative argument for the maximum variation to be expected on the basis of such considerations. FORMULATION OF THE PROBLEM In order to make an exact calculation, we require a few preliminaries: The conductivity g is supposed to undergo a small time-dependent fluctuation Ag{t) about its mean value. The fluctuation arises from certain sources each of which, for macro- scopic purposes, may be considered to emit a noise-current J(t) of minor- ity carriers. Thus in a small time-interval dt^ near i' the excess charge injected is J{t') dt' . This charge decays by recombination. Let r{t — t') de- note the fraction of carriers left over at time t{>t'). Then at time t there remains a charge r{t — t') J{t') dt' of the original injection. Now provided the excess density is small compared with the mean density, Ag{t) is proportional to the excess charge at time t, due to all the previous emissions added together. Therefore Ag(0 oc f r{t- tViO dt'. (1) In practice we do not literally plot Ag{t) as a function of <, but rather its frequency component Agr(/) in a narrow range df of frequencies near/. In other words, we single out for observation* the contribution to Agr from that part J{J) of the injected current J{t') which varies as e~^'*-^' Suppose now that 1// is large compared with the time over which r{t) MAGNETIC EFFECTS ON NOISE IN A GERMANIUM FILAMENT 649 is appreciably different from zero (that is, let 1// be much greater than the lifetime). Then, in the integral (i), r{t — t') will have gone from unity to zero long before J(f)e~^''^^^ has changed appreciably from its value at t' = t. Therefore, for purposes of observation at frequencies much smaller than the reciprocal lifetime, we can rewrite (1) as ^g{t) oc J{t) [ r{t - t') dt' J— 00 (2)* = J(t) / r{t) dt. Jo The integral in (2) can be interpreted as the average lifetime of car- riers. For, by definition, the rate of recombination at time t is — dr(t)/dtj so that — {dr/dt)dt is the number of carriers recombining between time t, t + dt. Hence the average lifetime is r = -f^ ^ ^ ^^ = -[^KO]^ + 1^ r(t) dt = f r{t)dt since tr{t) -^ 0 as ^ -^ co . If 1// is not large compared with r one cannot simplify the integral (1) in this way. One then has to consider separately each frequency component Ag{f)e~^'''^' due to the current J{f)e'^'''^\ Then J— CO = J{f)e-'"" f r{f) = r\ If the emission processes are stationary in time, is time in- dependent : = r. Now r can be written as I r{t - t') dt', which is simply the total concentration at the present time t due to a constant injection from time — oo to the present. Therefore the problem is reduced to finding the total carrier concen- tration in the filament due to a distribution of sources of constant strength V<«/o> • Let w{x^ y, z; Xi , yi , Zi) denote the carrier concentration at x, y, z due to a steady unit source at o^i , yi , Zi . Then the total carrier con- centration is r{xi ,yi,Zi) = j w(x, y, z; Xi , yi , Zi) dx dy dz. The reason for the dependence on xi , yi , Zi , is that the recombination process takes place largely on the surface. Therefore a source near the surface will yield a smaller concentration than one well inside the fila- ment. (Volume recombination will be neglected throughout this paper.) The mean square conductivity modulation due to many statistically independent sources at Xr , yr , Zr (r = 1, 2 • • • ) is then < {Agf> = i:T\Xr , yr , Zr) . The behavior of w is governed by the diffusion equation, subject to the boundary conditions expressing the recombination process, and sub- ject to a suitable singularity at Xr , yr , Zr , expressing the injection of a unit current. J^ut in two and three dimensions the solution is not avail- able in closed form, or at any rate not in terms of the elementary trans- cendental functions. The infinite series for the solution is not easy to MAGNETIC EFFECTS ON NOISE IN A GERMANIUM FILAMENT 651 handle computationally. It is therefore desirable to simplify the experi- mental conditions to a point where the problem becomes almost one- dimensional. A solution for w can then be found in closed form. Consider a very long uniform rectangular filament with one pair of sides very much wider than the other pair. Suppose that the y and z directions are respectively parallel to the wide sides and to the length of the filament (Fig. 1). Consider sources located anywhere on a plane x = |, which is parallel to the wide sides of the filament. If the recombination properties of the filament are uniform in the y — z directions, the Hfetime due to a unit source anywhere in that plane is independent of the location of the source on that plane, and depends only on ^. Hence the conductivity modulation due to sources of strength Ji^, Vr , Zr) (r = 1,2...) in that plane is simply T(?)EA?,2/r,2r) I ^2 (REAR PLANE) (FRONT PLANE) Fig. 1 — Geometry of the filament, and disposition of the fields. 652 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 and is the same as that due to an infinite source of strength Z) «^(^, Vr , Zr) r uniformly distributed over the plane x-^. But the density w due to such a source will be a function of x and ^ only, and will be the solution of the one-dimensional diffusion equation. Hence for a geometry approaching that of figure 1 sufficiently closely, the problem is one-dimensional. The Evaluation of r We now have to write down the one-dimensional diffusion equation in the presence of a magnetic field along Oy , which combines with the drift velocity of the carriers so as to force them towards one of the surfaces x = dz a.li Fx is the effective field arising in this manner, and D is the diffusion coefficient, the equation is which expresses the fact that the diffusion current —Dq dw/dx plus the drift current qtiFxW must be constant since the carrier density cannot build up indefinitely, n is the mobility of the minority carriers: At„ for electrons, /Up for holes. As is shown elsewhere^ the effective field F^ is given by Fx = (Bn + Bp) Ez = BE^ , where Ez is the biasing field causing the drift current, Bn , Bp are the Hall angles for electrons and holes, respectively. If /x„ , Hp are the electronic and hole-mobilities, and if the magnetic field is not too large, B = Bn + Bp= 10-\fjLn + tJip)H, where ff is in oersteds, and the mobilities are in cm^/ volt-second, and 6 is in radians. (Strictly speaking, the diffusion current is not in the direc- tion of the density gradient when a magnetic field is present.'* As the result mixed derivatives dxdy occur in the diffusion equation. But in the reduction to one dimension these terms integrate out. All that remains is a small correction to 7), MAGNETIC EFFECTS ON NOISE IN A GERMANIUM FILAMENT 653 negligible for ordinary values of H.) It is convenient to specify a dimen- sionless parameter in the same notation as H. C. Montgomery. 2ayLF^ ^ 2aBE^ By the Einstein Relation Z)//i = kT jq this may be written ^ 2oBE^ ~ kT/q where q is the absolute value of the electronic charge. $ is the ratio of the voltage corresponding to the transverse field to the thermal voltage kT/q. In terms of $, equation (3) can be rewritten dw _ ^ dw _ ^ , . dx^ 2a dx The integral of this equation has the form w = Ae^^'"""^ + B (5) where A and B are two constants. Because of the existence of a singu- larity at X = Xo , say, the constants A, B take on different values for X Xo . To see what these values are, we first write the solu- tion (5) in the form w, = Aie*^^-^°^^'" + B, x> xo, w, = A 26*^"-^°^^'" + B2 X 1 - «2 - (1 - e-") «1 + "26 2o^ I>V'i «! + die* Similarly, when only — o is emitting, we get 1 2 «i THi-a) ^ 2a Z)^2 a\e * + 0:2 - 1 But in an actual experiment, both faces will be emitting, with mean square strengths (Ji), (J2) say. The quantity that is then measured •? 0 §-. 8 5 • Z 5-2 -» ,/" "\ ■ — ■-X L \ K.- X / ^ % \« N ^ '0 y z / \ \, ^ ^ ^^ / ^ 0 ^^v * X ^v. - ^ ^ ^^ ■20 ■le •12 16 20 -e -4 0 4 6 MAGNETIC FIELD PARAMETER, 4> Fig. 3 — Contribution to the noise change from a unit source at the plane X — +0. MAGNETIC EFFECTS ON NOISE IN A GERMANIUM FILAMENT 657 is the ratio -. _ ^gl {JDrli+a) + {Jl)Tl{-a) ly H — ~" {Jiy„=o i+a) + {Jl)rUi-a)' (11) ASf/y=0 Therefore we also need the lifetimes at zero H (that is, zero $). But at $ = 0, the r's are indeterminate, and we therefore have to take limits. Expansion in powers of ^ shows that It T„(+a) = ^ It Th{-0) = — — (-^) 1 + Y- + -V l + 2>- Thus we finally get Nj, = ^9l {Ji} V^ L Oil + 0:26* {jVi Oil {^-^) + 1 q;i6 + «2 (12) (Jl) F(-^)^f('n^) There remains one small difficulty in the way of comparing experiment with theory: We do not know (Jl), (Jl)- As suggested by H. C. Mont- gomery, we are able to overcome this difficulty as follows : We first draw a number of curves of rU+a) th (-a) versus , rli+a) rK-a) for various sets of parameters (^i , ^2.) (See figures 3, 4). Then we contrive to match a linear superposition Ci Tlr(+a) . . rli-a) rl{+a) + C2 To i-a) where Ci + C2 = 1 to the experimental curve. This will be possible only for one particular set of values (^1 , ^2). From Ci and C2 = 1 — Ci and 658 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 16 12 -I ui S » UJ a z z < X o UI -• -12 -16 -20 V/i=o.) -^ ?A^ ^^0 / / / ^ 50 ^"^ / / .^ 1.0 50 /^ l//\=^° ^2 ^20 i M //I / // A 1 //// 1 y \v/ ^.^ vy .X ^^/ / ' ^ = 0 and varies as — /6 there. Hence the initial variation is as 1 + $2 rYQ-|(-.i))- Hence the curve rises or falls initially according as ^»4('+J)- The noise therefore increases initially if 12 2 xp and falls initially if 511 — ->— or ^>6.9 approximately lA < 6.9. 2. Volume Generation. At first sight it may seem that if volume generation is considered, so should volume recombination (detailed balancing). This is not neces- sarily so, since we are not dealing with an equilibrium situation here. 660 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Therefore the possibility of volume generation and surface recombina- tion cannot be discarded. This case is somewhat more difficult. Assuming all sources to be un- correlated and uniformly distributed throughout the interval { — a, +a), we have to square expression (10) and integrate it from Xq = —atoxo = -ha in order to find (Agl). (We suppose that all the sources are of equal strength («/*)). Substituting the values of the A and B from 9 and 10, we get, after some obvious cancellations th rixo) = 1^ [^0 + Se-^''^''^' + T] where S = 2a (-Kk+s)) «ie ^'^ + ^26 */2 T = — $ aie -*/2 0:26 ,+*/2 ciia2 sinh $ L2 aie-*/2 + 0:26+*/' aie-*/^ -f a:2e*/2 - 1 Hence 4a' {Agl)= (J) f rUx„)dx, J— a **Z)2 -a -f 2r a + -— — sinh - 6 $ 2 (14) cosh a> Before proceeding with this general case, we first consider the limiting case ^, = ^2 = «>, when ai = -1, ag = +1. Then T = -acoth * and 1^ tJ(xo) = 8a^ 5 tt sinh I i+coth'l coth $ $ + ^ MAGNETIC EFFECTS ON NOISE IN A GERMANIUM FILAMENT 661 To find Nh = (^oD/i^gl^) we need the limit of /!« r^ as $ ^ 0. After some tedious algebra, we find this limit to be 2 ^a Th=o — 15Z)2 so that / (A^7/)^1.^2=* ~ ~ /'• 30 $2 1^ ..* *''^°*'^l^8- _ + eoth--— ^+- (15) In the general case we can again take the limit of (14) as $ — > 0 in order to determine A^^ , but this would be too tedious. Instead, we solve the diffusion equation directly when = 0. The equation is then simply = 0 and the solution subject to the correct boundary conditions and allowing for a steady unit injection at Xq is Wi = Ai(x — Xq) -\- B W2 = A2{x — Xq) + B X > Xq X < Xq where Ai = -^ lAi [l + "^2 ('l + ^) / ih + 4^2 + 2M2), A2 = ^Ml + h (l - ^° j / (tAi + h + 2iAi , this quantity tends to 4aVl5D^, as before in our limiting process.) When ^1 = ^2 = ^, 'S and T also simplify somewhat, and the result is iy H — -. — 2 — r _ + 2T^ + - (^- - r j smh 2 - ^ cosh ^ + -^ smh $ •■[(^O'-a where now s- ^ An alternative form is 4 45 + ^ ...y Ut^ + coth ^ + 7 - 2 2 2 coth - (16) where A = l+-coth-, a result which correctly tends to (15) sls xp -> ^ . 10 log Nh for various ^ ^ is shown in Fig. 5. ACKNOWLEDGMENTS : The author gratefully acknowledges the help given by Dr. Shockley, I who suggested the conceptual picture underlying this theory and who 664 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 independently calculated some special cases. Thanks are also due to H. C. Montgomery for many valuable discussions and to Mrs. C. A. Lambert for computing the curves. REFERENCES 1. Montgomery, H. C, Electrical Noise in semiconductors. Bell System Tech. J., 31, pp. 950-975, Sept., 1952 bockley, 2. Shockley, W., Electrons and Holes in Semiconductors, Chapter 3, Section 2, Van Nostrand. 3. Suhl, H., and W. Shockley, Phys. Rev., 75, pp. 1617-1618, 1949. 4. Shockley, W., Electrons and Holes in Semiconductors, p. 299, Van Nostrand. DC Field Distribution in a '""Swept Intrinsic" Semiconductor Configuration By R. C. PRIM (Manuscript received January 15, 1953) This paper contains an analysis of the dc field intensity distribution in an idealized one-dimensional n-intrinsic-p semi-conductor configuration biased in reverse. It gives some quantitative insight into the progressive penetration of the field into the intrinsic region as the magnitude of the bias voltage is increased. INTRODUCTION Possible applications have been suggested for semi-conductor con- figurations involving intrinsic regions adjacent simultaneously to n- and p-type extrinsic regions. The basic idea behind some of these pro- posals is that a suitably large reverse bias voltage (n-regions positive with respect to p- regions) will set up a substantial electric field in the interior of the intrinsic region. This field would sweep most of the mobile carriers out of the intrinsic material, producing a region of material ("swept intrinsic") supporting a large field and having a high resistivity. This paper contains a dc analysis of an idealized one-dimensional swept intrinsic structure with abrupt transitions from strongly n-type to highly intrinsic to strongly p-type material. It gives some quantita- tive insight into the penetration of the electric field into the intrinsic region as the bias voltage is progressively increased. FORMULATION OF PROBLEM A one-dimensional structure will be considered having the distri- bution of excess of donor conceptration over acceptor concentration (Nd-Na) shown in Fig. 1. It will be supposed that N/ni and P/ui are ^ 1 and that a reverse bias voltage (Fig. 2) is applied between the bodies of the n- and p-type regions, (n^ denotes the thermal equilibrium con- 665 666 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 n-TYPE N INTRINSIC p-TYPE t « z 1 JO -p L . Fig. 1 — Assumed distribution of excess of donor concentration over acceptor concentration. centration of mobile electrons — or of holes — in intrinsic material at the given temperature.) The set of equations which (together with boundary conditions to be specified later) determines the electric field in the intrinsic region 0 < y < Lis: dE a , n), dy kT '^ ixJtT ' — = -±nF -t- ^'" dy kT "^ ^,„kT ' Ty--^' ^ = 9(!/ - r), du dy = -qig - r), Inhere E: electric field intensity, volts/m. q : electronic charge magnitude, coulombs. K : absolute dielectric constant, farads/m. p : hole concentration, m~'. n : electron concentration, m~'^ k :Boltzmann'8 constant, joules/°K. (1) (2) (3) (4) (5) (6) DC FIELD IN A "SWEPT INTRINSIC SEMICONDUCTOR 667 T : temperature, °K. ip : electric potential, volts. ip : hole current density, amps/m^. in : electron current density, amps/m^. lip', hole mobility constant, m^ /volt-sec. Mni electron mobility constant, m^/volt-sec. g : rate of generation of hole-electron pairs, m~^ sec""\ r : rate of recombination of hole-electron pairs, m"'^ sec~\ An order-of-magnitude comparison of the terms in (2) or (3) reveals that the currents probably have Uttle influence on the field distribution. For example for 5 amps/m^, amp-m' 10~m. finkT = 1.44 X 10"'' ' N = 2X 10'' m~', and L / -^dy^ S'WL ^ 3-10" m Jo IJinkT while L dn T- dy 0 dy n(0) - n(L) ^N = 2.10''m- On this basis, the current terms in (2) and (3) can be omitted without serious error. No use then has to be made of (5) and (6), so the govern- ing equations for the intrinsic region become: dE q , dy K n), (10 1 n-TYPE INTRINSIC j p-TYPE ^\[ Fig. 2 — Qualitative picture of potential distribution in reverse-biased n-m- trinsic-p structure. 668 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 . ^ = -E. (40 dy Equations (2')-(4') will also be used within the n-type extrinsic region {y < 0) and the p-type extrinsic region (y > L). However, for the n-type region (1') is replaced by ^ = 2(N + p-n), (I'a) dy K where N denotes the excess concentration of ionized donor over ac- ceptor centers for y < 0. Similarly, for the p-type region (1') is replaced by ^ = ?(_P + p-n), (I'b) dy K where P denotes the excess concentration of ionized acceptor over donor centers for y > L. In order to solve the equation set (l')-(40 for the intrinsic region 0 < y < L, it will be necessary partially to solve the sets (I'a) (2')- (4') and (I'b) (2')-(4') governing the two extrinsic regions because only in this way can a sufficient number of appropriate boundary conditions be imposed. Deep inside the extrinsic regions the electric field intensity will be negligible and the mobile carrier concentrations will have their equi- librium values. This leads to the conditions E =^ 0, np = n], n — p = Nsity=— ^ (7) and E = 0, np = n% p - n = P at ?/ = + 00 . (8) It will further be supposed that there is no infinite charge concentration at the extrinsic-intrinsic interfaces so that the electric field intensity is continuous at the interfaces. The concentration of the (local) majority carrier will also be assumed continuous at an interface. In short, E and n are continuous at ?/ = 0 (9) DC FIELD IN A ''sWEPT INTRINSIC" SEMICONDUCTOR 669 and E and p are continous at ?/ = L. (10) Finally, we choose the reference level for the electric potential in the intrinsic region so that ^ = 0 f or n = p (11) and regard the potential at the interface ?/ = 0 as a prescribed parameter, ;/, = !^ . [/ at 2/ = 0. (12) The two conditions (11) and (12) apply directly to the solutions for the intrinsic region. The conditions (7)-(10) indirectly imply the two additional restraints necessary to determine a unique solution of (1')- (40 inO < y < L. NORMALIZED VARIABLES AND EQUATIONS It is convenient to introduce dimensionless normalized variables before proceeding further with the mathematical analysis. As reference voltage it is natural to adopt the Boltzmann voltage rl^B^—, (13) Q the voltage equivalent of the mean kinetic energy of an electron at temperature T. (At room temperature the Boltzmann voltage is about 1/40 of a volt.) As reference quantity for carrier concentrations we choose the geometric mean of the majority carrier excess concentrations for the two extrinsic regions, i.e., reference concentration = (NP)^^l (14) The reference voltage and carrier concentration having been so chosen, it is natural to select as reference length the mean Dehye length £ = kT/q 2 2 (Arp)i/2 1/2 (15) K This mean Debye length is related by £ = {£n£py" (16) to the n-region and p-region Debye lengths defined respectively by 670 THE HELL SYSTEM TECHNICAL JOURNAL, MAY 1953 and £n = <£p = 'kT/q' K 'kT/q 2Sp 1/2 -11/2 (17) (18) We now use the reference quantities defined in (13)-(15) to introduce the normalized distance y = £' the normalized thickness of the intrinsic layer (19) (20) the normalized concentrations of positive and negative mobile carriers _ P fjVPV/2 the normalized potential and 1-t- *=^.' n = n (NP) 1/2 (21, 22) (23) and the normalized electric field intensity E tl^ ^b/£ (24) In terms of these normalized variables, the governing equations ((l')-(4')l for the intrinsic region become: dtS 1,. ., <4 55 = -S. (25) (26) (27) (28) For the n-type region, (25) should be replaced by rlP ^ = i(A + p-n) (25a) where 1/2 ..©■" (29) For the p-type region, (25) should be replaced by ^ = i(-A-' + i>-n). (25b) FORMAL SOLUTION OF EQUATIONS FOR INTRINSIC REGION Division of (25)-(27) by (28) yields, after evident rearrangements of factors, (30) (31) (32) (33) (34) where A and Ax are constants of integration. The condition (11) that ^ = 0 f or p = n implies that Ai = A, Substitution of (33) and (34) into (30) yields % = ''^'-^-'' (35) = 2 A sinh ^. dfl' 4 = n — p. dhip d^ = -1, dlnn 4 = 1. From (31) and (32) follow p = -Ae-* and n = - aJ, 672 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Integration of (35) now leads to S = [2A (cosh ^ + B)f" (36) where B is another integration constant. Substitution of (36) into (28) in the form yields after another integration where C is the fourth integration constant. In order to express in terms of tabulated functions the relationship between ^ and y defined by (37) we shall consider two cases: — 1 < B < I and B > 1. (It is not necessary to consider B < —I because A is essentially positive [see (33)] so that B < —1 would imply an imagi- nary field strength [see (36)] at the plane in the intrinsic region where The changes of variable of integration s = 2 sinh"' cot X for - 1 < B < 1, s =. 2 sinh"^ tan ^ for B > 1 permit the carrying out of the integration indicated on the left side of (37). This gives .,.(4(L-)-]_,[(i^«J»,.,„-.^.|]) = i2Ar\C - y) (38a) or — {(^7'4(^T for -1 < ^ < I and - A"\C - J)) - A"\C y)) (38b) (grrTP^Ciri)" ■ ^'"" ^"^"^l = (2^)"'^^ - ^-'^ (39.) DC FIELD IN A "sWEPT INTRINSIC" SEMICONDUCTOR 673 or 1 — sn for B > I. Note: ((ii-;)'(^')^--"') F[k, ^] s f Jo Vl - k^sin^d is the Elliptic Integral of First Kind, usually tabu- lated for O<0. ^.^_ 1 ... "^ ^" - — »«.,_ U = ?no L=4000 ^■^"^ -. ""n r \ 20 ' L-^ .;;;;;- k. \ < " N.N \ ^v ^ \ S \ \ A \ \\ \ \ ll \ \ \ \u=. i \ 1 \ \ \ \ \ \ \ \ \ \ , \ \ \ 1 \ 1 V \ \ 1 \ 1 1 1 0.2 0.3 0.5 y/L=y/L Fig. 3 — (Potential at point in intrinsic layer/Potential of n-intrinsic junction) versus (Distance from n-intrinsic junction/Intrinsic layer thickness) for L = (Intrinsic layer thickness/Mean Debye length) = 4,000 and several values of U = (Potential of n-intrinsic junction/Boltzmann voltage). 076 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 1.0 8 s — ki>«^ -i'V-.-. '"•=■-"■»« ... 20O0 L — xn nnn \ ^-»**gj rf^i- -"" i V «^^ ^ -»^^ ^200 ^*^ 20^ \ \ ^^^ V \. \ U=7 \\ m-2 \ \ \ , \! 1 in-3 ! 1! II li 10-* i ii 10"* II i • • ii II 2 II io-« A 1, 10-^ 0.1 0.2 A 0-3 L = y/L 0.5 Fig. 4 — Same as Fig. 3 except for L = 40,000. DC FIELD IN A "sWEPT INTRINSIC" SEMICONDUCTOR 677 ^•"n r"=''-<5^_. I I ' 0 r^ r'*'^«.<\,^ ^ 4 '-■^ k-w \ ''^J ^v. L io-^ "^^ \, \ e V s a V \ ^N \ "-X \ 10-2 > ^ , V N V v_ 1 N \ \ 10-3 \ t. = 400,000 \ \ \ "^ . " 9 U= 20 U= 200 U= 2000 O U = 20,000 \ -« 1 5 10-7 I li 0.2 0.5 Fig. 5 — Same as Fig. 3 except for L = 400,000. 678 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 10' I (VJl liJ' 10" II 1 L = 4000 \\ « \\ V- ._ U = 200 _-. — '^*- ■ — ^. L 20 \ v \ \ \ 1 0.3 Fig. 6— (Field intensity at point in intrinsic layer/ Average field intensity over intrinsic layer) versus (Distance from n-intrinsic junction/Intrinsic layer thickness) for £ = (Intrinsic layer thickness/Mean Debye length) = 4,000 and several values of [/ « (Potential of n-intrinsic junction/Boltzmann voltage). DC FIELD IN A "sWEPT INTRINSIC'* SEMICONDUCTOR 679 1 L = 40,000 3 in2 ^"^ ll I ^U = 2000 |<- i ^v. 1.0 ^ U=200 1 \ 1 \ V 1 \ \ 1 \ \ I \ \ \ \ \ \ ^ \U=200 \ r\ K, \ \ \ \ ^20 \ \, s. \ 7 \ \ \ \ 1 \ > \ \ 1 \ \ \ \ v\ \ \ 0 V ^ 0 0.1 0.2 0.3 0.4 0.5 Fig, 9 — Same as Fig, 3 except with non-logarithmic potential scale. 682 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 1.0 0.9 0.8 0.7 0.6 ^(0) 0.5 0.3 0.2 ■ U = 20 1 1 I l\ y V \ \ \ \ \ \ \ \ \ s. \ \ L = 4000 s \, \ \ \ \40,000 \ \ \ V \ \ \ V \ N, \ s. \l ^00,000 \ A ^v ^ \ \ 0.4 0.5 0 01 0.2 03 y/L Fig. 10 — Similar to Fig. 9 except for U = (Potential of n-intrinsic junction/ Boltzmann voltage) = 20 and several values of L = (Intrinsic layer thick- ness/Mean Debye length). Because of the extreme values of the parameters encountered in these computations, it was necessary to make extensive use of the following expansions to supplement available tables of elliptic functions: for k" « 1 k^ snlk, v] = sin 2; — - cos v{v — sin v cos v) + k^ cn[ky v] = cos ?; + — sin v{v — sin !; cos v) + for k' ^ (1 - kY' « 1 Klk] -'4-^H-')t-^ sn[k, v] = tanh v -\- — sech ?;(sinh v cosh v — v) + •••• cn[k, v] = sech v — — tanh v sech z;(sinh v cosh v — v) + .... Also useful were; for (f> « 1 for » 1 fnl±^ = tn 2 1-* K^^^ l-<^ !- 2 2 In the determination of B from (43) the problem arises of solving the transcendental equation k'K[(l - k'')] = a « 1 where a is a given positive quantity. This equation must be solved by iteration or plotting and a reasonably good estimate of the root saves a great deal of labor. Making use of the approximation vahd for /b' < 1, implying substantial uniformity of field through- out the intrinsic region. Subject to the assumptions (40) relations will now be derived among the apphed voltage, the intrinsic layer thickness, the asymmetry parameter A, and the field penetration parameter r]. The minimum field intensity occurs where ^ = 0. Therefore, from (36) and (45) ^ l2AiB + 1)]"- 2L~\U -/nA) Ehminating A from (46) by the use of (41) and L by the use of (43) leads to 2"\B + iY'MB) V /^^ U - Ink C8G THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 II \ 11 \ \ \ J in 4 s dlld r \ \ \ 4 \ 1 \ "6\ \ \ \ \ \ '-i \ \ 1 S, \ 11 \ \ 1 >■ t- «0 z lU l- 2 o w u. i 1 2 > »- in 2 U z u. UJ \ \ \ \ \ s, \ \ \ 1 \ \ 1 \ , \ II \ \^ \ \ \ — > v\ \\ \ \ \\ ^ 0) .s a O c3 o tf a o o rg o CM fVJ ?2 I ° vui-n DC FIELD IN A "sWEPT INTRINSIC" SEMICONDUCTOR 687 or Then substitution of (47) into (43) yields L « 2e"%{B) exp [l (^^)"' *(B)] . (48) For fixed r/, (47) and (48) are parametric equations of a function Ua(L). These equations were used to compute the curves of Fig. 11 in which U - /nA is plotted against L for 77 = 0.05, 0.1, 0.5, 0.9 and 0.95. This figure gives a quantitative picture of the dependence of the field pene- tration parameter rj on impressed voltage and intrinsic region thickness. [The ordinate U — InK is 3^ the total voltage drop across the intrinsic layer in (kT/q) units.]. The foregoing analysis clarifies the progressive elimination of the low field region near the center of the intrinsic material as the applied volt- age is increased for fixed L. Now the high field regions near y = 0 and y = L will be described. Making use of (41), together with cosh U w\eU and 2AB « A, A~^ implied by (40), in (36) leads to m)^e-"\"' (49) and to — (0) ;^ -i e-'A. (50) di) Hence a length characterizing the ''space charge layer thickness" at ?/ = 0 is ^(0) _, o 1/2 A -1/2 ^ 2e"'K-"\ dy Similarly, for the p-intrinsic interface at y = L, we have E{L) ^ e-"'K-"\ (52) ^ (L) ^ -he-'K-\ (53) dy I 688 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 and -fit) '''' dy If it is noted that the n-region and p-region Debye lengths are related to the mean Debye length £ and the asymmetry parameter A by £n = A-^^^£ and it is seen that (49)-(54) can be written: MO)^e-'''r— perccj, (49') ECD^e-^'^T— perjeJ, (520 ^ (0) « -he-' [— per £„ per £ J , (500 dy 19 J ^ (L) ^ -ie"' [— per £p per £ J , (530 dy LQ. J -§^;^2e^^^ [times £„], , . ^^^^ ^2e^/' [times £,^ Equations (490-(540 show that the maximum field intensity and the "space charge layer thickness" at either extrinsic junction is dependent only on the Debye length of the adjacent extrinsic material. Similarly, the concentrations of the majority carrier at either inter- face is found to be substantially independent of U and L, and deter- mined by the neighboring extrinsic material: ^«e- (55) ^«e-'. (50) DC FIELD IN A "sWEPT INTRINSIC" SEMICONDUCTOR 689 The minority carrier concentration at the interfaces depends too, on the neighboring extrinsic material, but is also U dependent: ^ « e-'e-^", (57) ^ « e-'e-'". (58) At the point of minimum field intensity, ^ = 0 and J = |«e-e- (59) The extremely low carrier concentrations given by (57)-(59) are not really meaningful, of course, because the analysis has neglected the carrier concentrations due to thermally generated hole-electron pairs and to saturation currents injected through the biased junctions. While these latter concentrations are neghgible compared to (55) and (56), they are undoubtedly large compared to (57) and (58). Therefore, al- though they can be neglected in determining the electric field intensity distribution, they are of principal importance in determining the small residue carrier concentrations in the "swept" region. A computation of these concentrations can be made by regarding the fields determined in the present analysis as impressed and studying the resulting motion of the generated and injected carriers. Appendix I EVALUATION OF INTEGRATION CONSTANTS A, B, AND C. In this section the conditions described in (7)-(12) will be used to evaluate the integration constants J., B, and C in terms of the pre- scribed parameters L, U, and A. First a partial integration of the differential equations for the extrinsic regions will be performed to obtain from (7)-(10) relations between JS and n at ^ = 0 and J§ and p aty = L. Division of (25a) by (27) gives (for the n-region ^ < 0) dn ' n = 1-?-^. (00) Addition of n times (26) to p times (27) yields ^ ipn) = 0, dy 690 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 whence ^^T (61) where v is a constant. Condition (72) then implies , = A . (62) "^ NP Substituting (61) into (60) and integrating we obtain ^2 ^ n - A In 71 + ^ + D, where D is a constant of integration. Now by (73), for y =^ _ 00 ^ = 0 and -O+i) Or, since An nN n « A for ^ = 0. Therefore giving Dj^-A + AlnA-^, or, for (1 — n/A) positive but not <^ 1, ^'^n-VK\n-. [for^ = 0]. (63) en A directly parallel computation for the p-region {y > t) leads to A-^ ^^ « p + A-^ In K [for ^ = I]. (64) '^ ep DC FIELD IN A SWEPT INTRINSIC" SEMICONDUCTOR 691 Upon substituting (34) and (36) into (63) and setting f = U (12), we obtain Ae'"" + 2AB = A In j^ , (65) or, because both terms on the left are negligible under the assumptions (40)*, A ^ Ae'^^'K (66) Similarly, substitution of (35) and (36) into (63) and setting of ^ = — V yields Ae'"" + 2AB = A"' In ^T+i , (67) or, by virtue of (40), A^A'^e'^^'K (68) Combining (66) and (68) we obtain F ^ C7 - 2 hi A. (69) This formula gives the normalized potential magnitude at the 2>-intrinsic interface in terms of that (U) at the n-intrinsic interface and the asym- metry parameter A. Now the condition ^ = U ior y = 0 requires (from 38a and 39a) that for -1 < B < 1, or (mf^Kl^y'' «in^>tanh^] = A-C (70b) forB > 1. In addition, the condition \j/ = —V ior y = L requires 4(^r]-4(4-T •-■~'a= "■•' A''\C - I) for -1 l. Fortunately, because of the assumptions (40a, b) the formidable relations (70a-71b) can be simplified to HB) ^ A"'C + 2e-''" and *(B) « A"\L -C)+ 2e-'"\ where *(B)^ '4(^)"'] for -1 < 5 < 1 f or B = 1 [{^T^Umi '- > 1 -F/2 Subtracting (73) from (72) we obtain 0 = 2A"'C - A"'L + 26-""' - 2e , whence, substituting (66) and (69), Finally, substitution of (66) and (74) into (72) gives or, by virtue of (40c, d), (72) (73) (74) (75) Equations (66), (69), (74), and (75) are the desired expressions for determining A, B, C, and V when values are assigned to A, L, and U (subject to (40)). (If nectossary, some or all of the restrictions (40) could DC FIELD IN A "sWEPT INTRINSIC" SEMICONDUCTOR 693 be eliminated, but the transcendental equations to be solved for A, B, C, and V would become quite formidable.) It should be noted that (75) permits an easy determination whether the formulae for B < I or those f or B > 1 should be used in any partic- ular case. Since $(1) = 7r/2, B^ liov A"'Le-^"-''' ^ T, (76) Appendix II CONDITIONS FOR 2ab <3C A, A~^ It has been stated without proof in the foregoing analysis that the conditions (40) imply 2AB «: A, A~^ (and hence also AB « 1). This must now be demonstrated. For B not » 1, (40a-40d) are sufficient, for (41) shows that A « 1. However, f or 5 ;::>> 1, the product AB is not necessarily small because of A <^ 1 and additional limitations are required. To establish suitable additional conditions we shall consider combinations of U, L, and A for which AB is very small and estimate the conditions under which this smallness begins to weaken. By eliminating U between (41) and (43) we can write $(B) w \LA^'\ or B"%{B) ^ \L{ABf'\ (77) Now for /c ?^ 1, Therefore, for B » 1 Substitution of (78) into (77) now yields tnBx 2-"'UABf'\ or B « exp [2-"'UABf'\ (79) 694 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 From (78) and (79) $(5) ^ ^UABY'' exp [-2-"'L{AB)]. (80) Now using the expression just obtained for ^(B) in terms of (AB), we can eliminate ^{B) from (43) and solve for C/" as a function of (AB). After some manipulations we obtain f « i2ABy- - 2iL^ - I in (2^) . (81) For 2ABA or 2ABA~' of the order of 10~^ (40c, d) insure that the right side of (81) will not change in order of magnitude if the two terms multiplied by IT^ are dropped. Therefore 2ABA and 2ABA~^ will be of order 10~^ or less if T 10 ^ ' To "^ • ^^^^ Thus for a given intrinsic layer thickness there is a ceiling on the im- pressed voltages for which AB is negligibly small. Transmission Properties of Laminated Clogston Type Conductors By E. F. VAAGE (Manuscript received December 8, 1952) The transmission properties of ideal laminated conductors of the Clogston type are discussed hy introducing the concepts of equivalent inductance, capacitance and resistance values which are analogous to tfieir corresponding counterparts in the treatment of ordinary transmission lines. From these constants the attenuation, phase constant, and speed of propagation are obtained using conventional transmission line theory, and the results com- pared with those for ordinary coaxial conductors. This paper is divided into two parts. In the first part a general discussion is given of Clogston cables and a comparison made with the conventional coaxial cable. This is illustrated with a few numerical examples, based on formulas which are developed in the second part of this paper. INTRODUCTION The discovery that deep penetration of the current can be obtained in laminated conductors, when the speed of propagation is made constant over the entire cross-section of the cable, is described in an earlier issue of this magazine/ The theoretical study of the problem was based on ,,/maxwell's field equations dealing with a stack of parallel plates of alternate conducting and insulating layers. When applied to concentric laminated tubes, this method results in a set of extremely complex equations. S. P. Morgan has given a rigorous solution for the case when the laminated layers are of infinitesimal thickness. The present paper uses a different approach which leads to simpler approximate formulas. Available theoretical results are combined with simplifying approximations and certain somewhat arbitrary assumptions ^ Clogston, A. M., Reduction of Skin Effect Losses by the use of Laminated Conductors. Bell System Tech. J., 30, pp. 491-529, July, 1951. 2 Morgan, S. P., Mathematical Theorv of Laminated Transmission Lines. Bell System Tech. J., Part I, 31, pp. 883-949, Sept., 1952, and Part 2, 31, pp. 1121- 1206, Nov., 1952. 695 696 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 in such a way that formulas for the counterparts of the usual transmission line properties of inductance, capacitance and resistance are obtained. The approximate formulas for attenuation, phase constant and speed of propagation are then derived using conventional transmission line theory. Some computations of attenuation are presented which illustrate the interesting transmission properties of Clogston cables under ideal con- ditions. Exploratory work on this type of conductor is still in the early research stage, with some very difficult problems imposed by the need for a large number of thin layers with very close tolerances ; Part I — General Discussion of Clogston Cables and a Comparison with Conventional Coaxial Cable 1. SKIN effect An alternating current transmitted over a solid conductor has the tendency to crowd toward the surface of the wire. This phenomenon is known as the skin effect and the depth of penetration of the current is usually referred to as the skin depth. The skin depth is defined as the distance, measured from the surface toward the center of the wire, where the current density is reduced to 1/e = 0.367. For a copper conductor it is given by: 2.61 where 8 = Skin depth in mils. Fmc = Frequency in megacycles. When the skin depth is a fraction of the wire radius, the ac resistance of the wire increases about as the square root of the frequency. Laminated conductors disclosed by Clogston have the property that the ac resistance will remain very nearly equal to the dc resistance over a wide band of frequencies, if the conducting and insulating layers can be made thin enough and sufficiently uniform. The dc resistance of a Clogston conductor will be higher than the dc resistance of a solid con- ductor of the same over-all dimension by a factor (w + t)/w, where w and t are the thicknesses of the conducting and insulating layers respectively. As discovered by Clogston, the depth of penetration in a laminated conductor is much greater than in a conductor of solid copper if the TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 697 speed of propagation is constant over the entire cross-section. A coaxial cable having an inner laminated conductor and an outer laminated sheath must obey the following relation to obtain the desired effect: =«■('+ 7) • /Z/€z = ;xe 1 + - , (2) where: Hi = Permeability in space between inner and outer conductor. jLt = Permeability of laminated conductors. €1 = Dielectric constant of insulation between inner and outer con- ductor. € = Dielectric constant of insulating layers in the laminated con- ductors. w = Thickness of copper layers. t = Thickness of insulating layers. In (2) the expression €(1 + w/t) is of course the mean dielectric con- stant of the laminated conductor. Since I/Vmi^i is the speed of propaga- tion in the main dielectric, equation (2) indicates that the speed of propagation is the same over the entire cross section of the cable. Equa- tion (2) must be satisfied to a high degree of accuracy, otherwise deep penetration is not possible. 2. DEFINITION OF CLOGSTON CABLES Two laminated conductors arranged as a coaxial cable are shown in Fig. 1. The inner conductor consists of a solid copper wire of diameter di, over which a large number of alternate layers of insulation and copper are arranged as concentric thin tubes. The over-all diameter of the inner conductor is Di. The outer conductor of the coaxial cable con- sists of a laminated tube of inner diameter d2 and outer diameter D2. The space between D2 and 6^2 is filled with thin concentric tubes of copper and insulation of the same thicknesses as for the inner conductor. The outside of the outer conductor is covered with a sohd copper sheath for protection, shielding and energizing purposes. This type of cable has been named Clogston I. By adding more layers to the outside of the inner conductor and more layers to the inside of the outer conductor, the space between them is completely filled when ^2 = Di. Such a cable is shown in Fig. 2, and has been named Clogston II. Clogston I may be thought of as a physical variant of the conventional 698 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 I. - Fig. 1 — Clogston I cable. L- " Fig. 2 — Clogston II cable. TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 699 coaxial cable, in that when one of the conductors carriers a current in one direction, the other will carry a current in the opposite direction. In Clogston II there is also a reversal of currents somewhere between the outermost layers and the innermost. It is therefore a kind of a two conductor cable, but the point of division between the conductors is determined by the electromagnetic field configuration of the situation. This point has been worked out by S. P. Morgan and will be referred to in the second part of this paper. 3. OPTIMUM PROPORTIONING The cross-sectional aspect of Clogston II is completely characterized by that proportion of the diameter D which is occupied by the lamina- tions. This proportion is called the Fill Factor and is defined by: 011 = (D - d)/D Clogston II. (3) The fill factor is also a useful parameter for Clogston I, though it is not sufficient to determine its geometry. It is defined by: 0j = (Di - di-^ D2 - d2)/D2 Clogston I. (4) The additional parameters which, with the outer diameter D2, will completely determine the geometry, are the ratio of the over-all thick- ness of the inner laminated conductor to the over-all thickness of the outer conductor, and a parameter which locates the inner diameter di of the inner conductor. These parameters are defined by: T = (Di - dO/iD2 - d2), U = di/D2. In a conventional coaxial cable, shown in Fig. 3, the optimum value of attenuation^ is obtained when D/d = 3.59. In Clogston cables no such optimum values exist. S. P. Morgan, however, has shown that there are useful relative optimum relations in Clogston I, which direct the choice of T and U for the cables which are illustrated in this paper. For example, for a fill factor of one-half, there is a broad optimum of attenuation when T = 1.96 and U = 0.0842. Thus the over-all thickness of the inner conductor is about twice that of the outer conductor, and the diameter di of the inner core is about one-twelfth of the outer diameter D2 . With 3 Green, E. I., F. A. Leibe and H. E. Curtis, The Proportioning of Shielded Circuits for Minimum High-Frequency Attenuation, Bell System Tech. J., 15, pp. 248-283, April, 1936. 700 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 these dimensions the cross sectional area of the outer conductor is nearly twice that of the inner conductor, so that the optimum arises from other causes than matching the conductivity of the two conductors. The problem of determining the relative optimum values of Clogston I, in genera), is complicated, but numerical studies indicate that for values of the fill factor other than one-half, the values for T and U are not greatly different. Another factor in Clogston cables which can be optimized is the ratio of the layer thicknesses of the conducting and the insulating layers. Clogston* has shown that in the frequency range where the attenuation is substantially fiat with frequency, this optimum ratio is equal to: w/t = 2. (6) For this condition, the dc resistance of the laminated conductor is increased by {w + t)/w or 3/2 over a solid conductor. The dielectric constant of the main insulation, according to (2) above must equal €i = 3c, w^hich reduces the speed of propagation by \/3, assuming Mi = M- At frequencies where the attenuation begins to increase, other op- timimi values of w/t can be obtained, and the ratio will depend upon what top frequency is considered. In a practical case a fill factor of unity will probably not be used. A little space in the center will be made available for a solid conductor for energizing or other purposes. Fig. 3 — Conventional coaxial cable. * Loo. cit. TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 701 4. ATTENUATION The attenuation of a transmission circuit at high frequencies, where oiL^ R and o)C ^ G, is usually given in the following form: i/1+li/i' (7> where R is the total ac resistance of both conductors, L, C and G the inductance, capacitance and leakance of the circuit. It will be assumed that the insulation consists of polyethylene or some other material having a very low^ leakance. Thus as a first approximation the second term in (7) can be neglected. In a conventional coaxial cable, in the frequency range considered, R will increase in proportion to the square root of frequency, so that neglecting leakance, (7) may be written as follows: a=^VF^r (8) where Ki is a constant depending upon the dielectric constant of the insulating material and the resistivity of the conductors. D is the inside diameter of the sheath, and F^ac the frequency in megacycles. In the second part of this paper it is sho^vn that the attenuation of Clogston I or II cables can be written in the following form: a. = ^' + K,wV,^, (9) where D = Over-all diameter of laminated cable. w = Copper layer thickness. K2 and Kz are constants, different for Clogston I and Clogston II, which depend upon the geometry of the cables, the dielectric constant of the insulating material and the resistivity of the conducting layers. The first term in (9) gives a constant loss independent of frequency. The second term contributes little to the attenuation provided w is small enough. In fact, the attenuation mil remain constant within p % of the first term provided: w = lor i^3 dfL' ^ ^ This equation (10) determines the copper layer thickness, which will result in a "flat" attenuation within p per cent up to a frequency F, t c • 702 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 By neglecting the second tenn in (9) and comparing the result with (8) it can be seen that the attenuation of a conventional coaxial cable and a Clogston cable are equal at a frequency given by: (11) The numerical examples given later in this paper indicate that a Clogston cable will have higher attenuation than a conventional coaxial cable at frequencies below F^c , and less attenuation at higher frequencies. At frequencies sufficiently higher than F^c , the attenuation of a Clogston cable will increase rapidly which is evident from the second term in (9) . It is in the region between Fmc and frequencies where the second term in (9) becomes important that Clogston cables can theoretically pro- vide less attenuation than a conventional coaxial cable. 5. IMPEDANCE, PHASE CONSTANT AND SPEED OF PROPAGATION The equivalent impedances of Clogston I and Clogston II cables are developed in the second part of this paper and are equal to: (12) ^" "" VL C Clogston II. In these expressions Lin, Lex and Cex are the internal and external inductances and capacitances respectively. For conventional coaxial cables they are discussed by S. A. Schelkunoff in * 'Electromagnetic Waves" (Van Nostrand, 1943). For the Clogston analogy. Part II of the present paper gives the reasoning adopted in defining them. In a Clogston II cable the external inductance goes to zero and the external capacitance to infinity, but the product LexCex nevertheless remains constant. The impedance of a conventional coaxial cable, in the frequency range considered, is given by: =/ I (.8) where L and C are the external inductance and capacitance of the con- ventional coaxial cable. The equivalent impedance of a Clogston cable is lower than that of conventional coaxial cable of the same outer diameter. Numerical eval- TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 703 uations using the above formulas indicate an impedance of about 23 ohms for a half -filled Clogston I cable, and about llj^ ohms for a completely filled Clogston H cable, assuming polyethylene insulation with € equal to 2.3 and copper conducting layers having a thickness twice that of the insulating layers {wit = 2). These values compare with about 76 ohms for a conventional coaxial cable having air dielectric and the same outer diameter and 51 ohms for a corresponding coaxial with soHd polyethylene dielectric. The phase constants of both Clogston cables in the frequency range where the attenuation is nearly flat, are equal and independent of the geometry of the cables. They are given by: iSi = ^11 = WL^C^\ (14) assuming uniform layer thicknesses. For a conventional coaxial cable the phase constant, neglecting leak- ance, is given approximately by: ^ = coVLc[l-i(2-^)]. (15) The computed speed of propagation, which is equal to co//?, is about 71,000 mi/sec for Clogston cables with polyethylene insulating layers. This compares with 123,000 mi/sec for a conventional coaxial cable with polyethylene insulation and 186,000 mi/sec for a coaxial with pure air dielectric. 6. COMPARISON WITH A CONVENTIONAL COAXIAL CABLE To illustrate the effect of the various parameters involved in the attenuation of a Clogston cable, and to compare the result with a convential coaxial cable, a few numerical examples have been evaluated. A one-half filled Clogston I and a completely filled Clogston II have been selected arbitrarily for comparison purposes. Fig. 4 shows the attenuation characteristics of these cables for several values of copper layer thicknesses. In each case, polyethylene insulation (e = 2.3) is assumed, with w/t = 2, i.e., the insulating layers have one-half the thick- ness of the conducting layers. In the same figure, the attenuation charac- teristics of two conventional coaxial cables of the same outer diameter, one with air dielectric and one with polyethylene insulation, are shown also. The regions where Clogston cables have in theory less attenuation than conventional coaxial cables of the same outer diameter can be seen in this figure. 704 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 s, rr n 1 \ ^ ^ is T~ V — V ^1 N N n 7 / > ^ S s \ \ 1 r n "> V, \ 4s ^v \ / ,^\ N. V s °°^\\ / CLOGSTON I: OUTER DIAMETER OF SHEATH = 375 MILS INNER DIAMETER OF SHEATH= 312 MILS OUTER DIAMETER OF INNER CONDUCTOR = 156 MILS INNER DIAMETER OF INNER CONDUCTOR = 31.6 MILS CLOGSTON n : OUTER DIAMETER = 375 MILS INTERIOR COMPLETELY FILLED - % > \ \ \v / - >\ \ v V V X % ^ V ^ \, "^ \ / V L } \ \ V ( w v^ \ ^ \ s A \ ^ \ > \ ■J y ^A ,- \ \ A. \ ' \ - i > \ ^ > \ \\ \ - (VJ *s ^ / \ Nv\ 1 (0 ^ > 1 g -1 o ^^^ V \ \ ft \\ -1 ^ ^ / v ^^. \ s % > r^ s: \^ \ \ 5 u i > / > \ \ \ / \ \ - / '' - • rvi 1 > 1 / \ Y^ / \ \ \ V / / \ \ \ \ X / / § N ^ \ i J. / i 3 o J- \ \ 1 \ -do (asddOD onos uod) siii^ ni nidsa nivs qnv 3111^ ti3d sisgoaa ni noiivonbhv TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 705 The dotted curve shown on Fig. 4 gives the skin depth in solid copper. It will be noted that the copper layer thicknesses become smaller and smaller fractions of the skin depth as the frequency increases. This is also evident from (1) and (10) above, which show that the copper layer thicknesses are inversely proportional to the frequency, while the skin depth is inversely proportional to the square root of the frequency. The effect of fill is illustrated in Fig. 5 for a 375-mil Clogston I cable, for fill factors of one-eighth, one-quarter, and one-half. It will be noted that the attenuation increases rapidly with decrease in fill, accompanied by an increase in the frequency band over which the attenuation is flat. The attenuation of a completely filled Clogston II cable is also shown for comparison. The above estimates are based on ideal conditions; that is, it is as- sumed that the laminated structures are perfectly uniform. The effects of departures from ideal are not shown by the approximate methods used in this paper, but it has been shown by Morgan^ that even small departures from ideal conditions will result in increases and irregularities in attenuation, and a decrease in the band over which the attenuation is approximately uniform. 100 80 60 J 40 — J ;30 — uj20 Q. CD 10 8 6 LU 8 Q D Z o I.O 0.8 - II 1 1 1 1 1 DIAMETER RATIOS 1/8 FILLED 1/4 FILLED 1/2 FILLED — d, = 0.0210 D2 0.0421 D2 0.0842 D2 D, = 0.1037 D2 O.2075D2 0.4150 D2 d2= 0.9577 D2 O.9155D2 O.8309D2 TWICE TOTAL THICKNESS =D,-di + D2-d2 OR 0.125 D2 0.250 D2 0.500 D2 D2= 375 MILS d2/Di = 9.25 4.42 2.00 / / / / i / / // / / / / V - / 7 - / ^ CLOGSTON I 1/8 FILLED 31 LAYERS -f^ // ~ r- ~~ ' / * CLOGST ONI 1/i 4F II LEC ) e 2 LAYEF S -y i / CLOGST ONI 1/2 ?F LL .ED 125 LAYERS, ^ / / 1 M 1 1 1 ^ eBS CLOGSTON I ORE FILLED 250 uA' 1 1 111.1,1 20 40 60 100 200 400 600 1000 2000 4000 FREQUENCY IN KILOCYCLES PER SECOND Fig. 5 — ■ Attenuation of Clogston I. 10,000 20,000 ^ Loc. cit. Part II, pages 1161-1201, 706 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Part II — Derivation and Evaluation of Formulas In this part of the paper, approximate formulas for the counter- parts of the usual primary constants used in transmission line computa- tions are derived for Clogston I and II type cables. From these formulas the various constants entering into the expressions for attenuation, im- pedance and phase as given in Part I are evaluated in terms of the di- mensions of the cables and the frequency. In deriving the formulas, certain simplifying assumptions have been made as explained in the test. The effects of these assumptions are examined by comparing the results with those obtained by more rigorous methods. 1. RESISTANCE IN GENERAL A Clogston I cable consists of a laminated inner conductor and a laminated outer conductor as shown in Fig. 1. Each of these may be represented by the laminated conductor shown in Fig. 6. At present, we have no exact formula for the ac resistance of such conductors over a wide range of frequencies. However, S. P. Morgan has shown that for a stack of parallel plates, the resistance per unit cross-sectional area is equal to: fl, = R^ !g±l Fi + ('^^ - y5)«>- FL +....], (16) w L 412 J where i^dc is the dc resistance of the stack, when completely filled with — D-' Fig. 6 — Laminated conductor. TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 707 copper. The other parameters are: w = Thickness of copper layers in mils. t = Thickness of insulating layers in mils. n = Total number of layers. Fmc = Frequency in megacycles. With curvature disregarded, equation (16) also gives the ac resistance of the laminated conductor shown in Fig. 6. For only one copper layer, n = 1, and t = 0, (16) reduces to: R^. = R..[l + ^+ ....'\, (17) which is exactly the expression for a copper tube at frequencies where the ac resistance begins to depart from the dc resistance and 3^(D — d) = w. For a single layer, the effect of curvature is very small^ and can be dis- regarded. It will be assumed that (16) also holds to a fair degree of ac- curacy for a laminated conductor made up of a large number of layers. Since n is large, the small fraction one-fifth can be neglected. For the optimum condition (minimum attenuation) Clogston^ has shown that: w = 2t. (18) The total number of layers can be obtained from the following expression : D-d D-d, _^ ,^^, n = -^, — -r = —5 for w = 2/. (19) With (18) and (19) substituted in (16) the ac resistance is given by: ^^\ _ 82080 r w\Ti-drFl^ . '° S^^^L 3710 +••••!' (20) where the diameters and the copper layer thickness are given in mils and the frequency in megacycles. The resistivity of copper is taken to be 1.724 X 10"' ohms/cm'. 2. CLOGSTON I CABLE 2.1 Resistance The ac resistances of the inner and outer laminated conductors of ClogstoD I cable, shown in Fig. 1, can be obtained from (20) above by substituting Di and di for the inner conductor and D2 and ^2 for the ^ Schelkunoff, S. A., The Electromagnetic Theory of Coaxial Transmission Lines and Cylindrical Shields. Bell System Tech. J., 13, pp. 532-579, Oct., 1934. BSTJ, October, 1934, by S. A. Schelkunoff. ;■ ^ Loc. cit. 708 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 outer conductor and adding the result. Thus: „ , ,3 S20S0Ai [ . , Biw'DIfL , 1 1. / • /o1^ /Ci + I^ac2 = —^^2 — [1 + — 37lO~ ■ ■ ■ J ^^"^^/"^^' (21) where: Dl . d! „ 2(D,Dt - d,(k){Di - dx)(J>2 - ^2) .„„, ^' - {Dl -dl + Dl- dl)Dl ■ ^^^> From (4) and (5) in the first part of this paper, it is possible to ex- press A i and Bi wholly in terms of <^i , T and U. Thus, Ai and Bi are independent of D2 , and it follows that the first term in (21) is inversely proportional to Dl , while the second term, when multiplied out, is in- dependent of D2 , assuming fixed values of 0i , T and U. 2.2 Impedance J Inductance and Capacitance In a coaxial cable the flux in the space between the two conductors gives the external inductance. The internal inductance is obtained from considering the flux within the walls of the conductors themselves but not that in the space between them. The effective inductance is then the sum of the two. Analogous considerations apply to the external and internal capacitance and the effective over-all capacitance is the value of the two acting in series. Similarly in the frequency range where o)L^R and o)C ^ G, but where the ac resistance is nearly equal to the dc resistance, the internal in- ductance and capacitance of the laminated conductors must be taken into account. Since they are in series with the external components they tend to increase the total inductance and decrease the total capacitance. The impedance of the circuit can therefore be expressed as follows: Z, = i/| = 7^^^+/"^-, ohms, ./ r CinCex 1 "' (24) (I) (#3 Lex = ^ ^n ( ;i ) henries/cm, (25) 2ir€/ ^« = 7T\ farads/cm, x^^v ^^fd,\ (26) TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 709 where: ei = e(l + i^/O = Dielectric constant of insulation between inner and outer conductors. At the present time no exact formulas for the internal components of either L or C are available. The internal inductance, however, must be nearly equal to that of a solid wire for the inner laminated conductor and to that of a solid sheath for the outer laminated conductor, and will be assumed so in this paper. It should be remembered that deep penetration of the currents will be obtained when Clogston condition of constant speed of propagation over the entire cross section of the cable is satisfied. The speed of propagation over the laminated conductors is equal to iVLindn and in the main dielectric between the two conductors equal to IVLexCex • Thus to obtain the Clogston condition the following relation must hold: LinCin = Les^Cex • (27 ) By solving for the unknown quantity Cin and substituting the result in (24), the impedance of Clogston I cable is found to be: z,= ^in /Z/ex Led V Ce. 1 + i^ i/^ ohms. (28) Formulas for Lin, Lex and Cex in units convenient for numerical evalu- ation are given later in the section summarizing the formulas. 2.3 Attenuation and Phase The attenuation of Clogston I cable neglecting leakance is obtained from the following expression: The phase constant, at the frequencies considered, and where the ac resistance of the conductors does not depart appreciably from the dc value, is equal to: /3l = CO VLC = ^ /j/ (Lin + LeJ cJYCe. ' (30) which with (27) above reduces to: i8i = CO VL^iC;; . (31) 710 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 3. CLOGSTON II CABLE 3.1 Resistance A Clogston II cable may be looked upon as having an "inner" and "outer" conductor which are separated by an infinitesimal amount in which /?ac given by (20) above is an expression for the parallel connection of R^ci and /?ac2 of a Clogston II cable having an outer diameter of D and an inner diameter of d. That is: R&cl + R&c2 It will now be assumed that (1) the currents flowing in one direction through the "inner" conductor and in the opposite direction in the "outer" conductor, will separate in such a way that R^d is equal to R^c2 y and (2) that the currents are uniformly distributed over the cross sections of the conductors. With these assumptions the respective cross sections would then be equal, and the reversal of the current would take place in a completely filled {d = 0) Clogston II cable at a radius equal to: r = — ^ = 0.3535A (33) By substituting (32) and (33) in (20) it can be shown that: ^ , ^ 328320 r, ^ B2w'D^FL ^ 1 u / • /o.^ R^i -f R^2 = ^, _ ^, I 1 + 3y^Q + • • • J ohms/mi, (34) where The above assumptions relating to division and distribution of the current are not exact. S. P. Morgan has shown that the current distribu- tion is not uniform and that the reversal of the current takes place at a radius equal to 0.3138D. As shown in a later section of this paper, the error resulting from these simplifying assumptions is not large. 3.2 Impedance, Attenuation and Phase In a Clogston II cabU;, tlie main dielectric insulation between "inner" and "outer" conductor has vanished. Thus the external inductance approaches zero and the external capacitance becomes infinitely large. B2= 1 + TRANSMISSION PROPERTIES OF CLOGSTON TYPE CONDUCTORS 711 From (25) and (26) above it is evident, however, that the product of Le^Cex remains constant, since it is independent of the diameter ratio d2/Di , which in a Clogston II cable approaches unity. With (27) in- serted in (24) and with L^ = 0 and Cex = <», the impedance of a Clogston II cable may be written: The internal inductance of a Clogston II cable is not known, but will be taken equal to 0.1609 X 10"^ Henries/mi, which is the internal inductance of a pair of wires at low frequency. Thus: 17.35 , Zn = —7^ ohms, (37) where e is the dielectric constant of the insulating layers. The attenuation is obtained by dividing i^aci + i^ac2 from (34) by 2Zii , where leakance is disregarded. The phase constant, in the frequency range considered, is equal to the phase constant of a Clogston I cable since L^C^x is a constant value. 4. Comparison of Results with V allies Obtained from Rigorous Formulas S. P. Morgan^ has developed rigorous formulas for the attenuation of Clogston cables, assuming infinitesimal thickness of the layers but re- taining a fixed ratio of copper to insulating layer thicknesses. A correc- tion term gives the increase in attenuation with frequency for layers of finite thickness. The attenuation of a one-half filled Clogston I cable computed by the approximate formulas given in the present paper was 1.1 per cent higher than the value computed using Morgan's rigorous formulas. Similar computations on a completely filled Clogston II cable gave values 8.6 per cent higher. This decrease in accuracy with increase in fill is in line with the expectation that uniform distribution of the current is more closely approximated with low percentages of fill. 5. SUMMARY OF FORMULAS The formulas developed in the second part of this paper and those for which the derivation has been indicated are summarized below. Con- ductuig layers of copper, and insulating layers with a dielectric constant 8 Loc. cit. 712 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 of € and a thickness half that of the conducting layers {wji = 2) are assumed throughout. 5.1 Clogskm I Cable /e, = ^^[l + ?^^+...]ohn.sMile, (38) /.= 0.741 X 10- ^[^. log. g -. (39) - 0.2172 henries/mile, v-3 1 C?2 Lex = 0.741 X 10"' logio ~ henries/mile, ^ 0.0388 X 10-'€x . , / ., Cex = " -^ farads/mile, €i = t(l + W/t), Pi = wVLexCex radians/mile, IT «i = 772 + Kiw^Fl^c nepers/mile, where: _ 41040 r D\ Dl 1 Zi lD\-dyDi-diy ^ 22.124 D1D2 - did2 Zx (Di + di)(Z)2 + ^2) * 5.2 ClogaUm II Cable _ 328320 r B2u;^Z)^FL , 1 u / m 17.35 ^ Zu = — 7=" ohms, /3ii = wVLexCex radians/mile, TRANSMISSION PROPERTIES OP CLOGSTON TYPE CONDUCTORS 713 k' «n = 7)1 + KaW^Fl, nepers/mile, (51) 164160 £)' ^^= Z„ D'-d" (52) J.' 44.248 D' + d' -{D + d)Vi(D^ + d^) ^ 2(£)i£>2 - did2)iDi - di){D, - di) ^' " {Dl -d! + Dl- dl)Dl ' (55) ''—+s^('+5)'/iF&)- (56) The parameters in (38) to (56) are defined as follows : D2 = Outside diameter of outer laminated conductor in a Clogston I cable, in mils. d2 = Inside diameter of outer laminated conductor in a Clogston I cable, in mils. Dl = Outside diameter of inner laminated conductor in a Clogston I cable, in mils. dl = Inside diameter of inner laminated conductor in a Clogston I cable, in mils. D = Outside diameter of a laminated Clogston II cable, in mils. d = Inner diameter of a laminated Clogston II cable, in mils. w = Thickness of copper layers in mils, ^mc = Frequency in megacycles, ei = Dielectric constant of insulation between inner and outer lami- nated conductors in a Clogston I cable, € = Dielectric constant of insulating layers. AC KNO WLEDGMENT The author wishes to express his appreciation to H. S. Black, C. W. Carter, Jr., J. T. Dixon and F. B. Llewellyn for valuable assistance and advice in preparation of this paper. A Coupled Resonator Reflex Klystron* By E. D. REED (Manuscript received March 5, 1953) The theory of a coupled resonator reflex klystron is developed and its reduction to practice described. This tube differs from the conventional reflex klystron in that its performance characteristic is derived from the interaction between the electronic admittance due to a bunched electron stream and the input admittance of two synchronously tuned, coupled resonators. As a result: (1) power output can be made to be substantially flat over the greater part of the electronic tuning range; (2) the half-power electronic tuning range of the coupled resonator reflex klystron is more than twice that of a klystron using the same electron optical system but inter- acting with a single resonator; and (3) modulation linearity may be ob- tained over a greatly increased frequency swing. A reduction in power output of about 3db occurs for a secondary resonator Q and coupling coefficient adjustment designed to yield a maximum flat band or maximum electronic tuning while a much smaller reduction in output power will provide a substantial improvement in modulation linearity. TABLE OF CONTENTS 1.0 Introduction 716 2.0 Small Signal Reflex Klystron Theory 718 2.1 Electronic Admittance 719 2.2 Passive Circuit Admittance of Single Resonator 722 2.3 Equivalent Circuit of Single Resonator Reflex Klystron .... 723 2.4 Performance Analysis Based on Complex Admittance Plane Representation 723 3.0 Theory of Coupled Resonator Reflex Klystron 727 3.1 Driving Point Properties of Two Coupled Resonators Hav- ing Equal O's 729 3.1.1 Variation of Input Impedance with Frequency 735 3.1.2 Input Admittance Plot in g-b Plane 736 * Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy, in the Faculty of Pure Science of Columbia University. 715 716 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 3.1.3 Variation of Input Phase Angle with Frequency 737 3.2 Mode Shapes Resulting from Interaction Between Elec- tronic Admittance and Input Admittance of Two Coupled Resonators of Equal Q's 739 3.3 Driving Point Properties of Two Coupled Resonators Hav- ing Unequal Q's 742 l^ 3.4 Mode Shapes Obtainable With Two Coupled Resonators of Unequal Q's 746 4.0 An Experimental Coupled-Resonator Reflex Klystron 746 4.1 Constructional Features of Experimental Tube and Circuit . 748 4.2 Qualitative Verification of Theory 750 4.3 Quantitative Verification of Theory 750 4.3.1 Determination of Primary Q 752 4.3.2 Calibration of Secondary Resonator 754 4.3.3 Comparison of Experimental and Theoretical Mode Shapes 757 4.4 Performance Data 760 5.0 Applications of the Coupled Resonator Reflex Klystron 762 6.0 Conclusions 764 References 765 1.0 INTRODUCTION The conventional reflex klystron derives its performance characteris- tics from the interaction between the electronic admittance due to a bunched electron stream and the input admittance of a resonant cavity. As is well known, this interaction results in mode shapes which are closely related to certain input properties of the passive resonant circuit. Thus, the dependence of power output upon frquency, which results from variations in repeller voltage about its mid-mode value, bears close resemblance to the input-impedance-versus-frequency plot of a parallel resonant circuit. Similarly, the curve relating frequency to repeller voltage has the same general shape as that relating frequency to the input phase angle of the resonator. Recognition of these relation- ships has resulted in the consideration of different and, perhaps, more useful mode shapes which might be obtained if the electronic admittance were made to interact with impedance or admittance functions of passive circuits other than that due to a single resonator. What do we mean by "more useful** mode shapes? The answer, of course, depends on the application, although an "ideal" mode shape could probably be defined as one having a flat top, i.e., power output inde- A COUPLED RESONATOR REFLEX KLYSTRON 717 pendent of repeller voltage, with frequency linearly related to the latter. Moreover, these conditions should preferably obtain over the widest possible frequency range. A tube possessing such characteristics would prove exceedingly useful in a large number of applications. To list a few: (a) Electronically swept signal generator, (b) FM deviator, (c) Transmitting oscillator in radio relay systems emplo5dng fre- quency modulation, and (d) Local oscillator in microwave receivers with wide range AFC applied to the repeller. Inability to reaUze this ideal mode shape in a practical tube might make a compromise solution appear acceptable, one consisting of a reflex klystron having a variable mode shape, i.e., a characteristic which could be adjusted to fit a particular need. Thus, appUcation (a) requires constant power output with minor emphasis on frequency-repeller-volt- age linearity, whereas appUcations (b) and (c) demand a high degree of modulation linearity with constancy of power output of no great im- portance. In application (d) the emphasis is on wide electronic tuning mth both variation in power output and non-linearity in the frequency characteristic permissible. A method to obtain this variable mode shape was achieved by the use of coupled cavities. Instead of having the bunched electron stream inter- act with the electric field of a single resonator, as is done in the conven- tional reflex klystron, we can present to it the input admittance of two coupled cavities. The resultant mode shape may then be expected to resemble the input impedance of two coupled resonant circuits just as the conventional klystron mode resembles that of a single resonator. Moreover, the input impedance of two coupled cavities can assume a large variety of contours depending on the ratio of Q's of the primary and secondary resonators and on the tightness of coupling between the two. If we were now to construct such a double cavity reflex klystron with provision to vary the secondary Q as well as the coupling coefficient continuously, we would have the means of producing a large variety of mode shapes within a single tube. Depending on the appUcation, the characteristics could then be adjusted to give either a range of flat power output or optimum modulation hnearity or wide electronic tuning. As might be expected at this point, a price must be paid for the ad- vantages gained in the coupled cavity approach. It consists of the power expended in supplying the losses due to the secondary resonator. This power subtracts directly from the available useful power and, there- 718 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 fore, though resulting in broadband operation, greatly improved modula- tion linearity and a number of other useful properties, leads to a definite reduction in output level. Whether this can be tolerated will again de- pend on the appUcation. In many practical cases the performance flexi- bihty inherent in the coupled resonator reflex klystron will more than outweigh this advantage. It is the purpose of this paper to present the theory underlying the operation of the coupled resonator reflex klystron, as well as its experi- mental verification. For the sake of completeness, but also in order to emphasize methods of analysis to be used in later sections and not readily found in the hterature, a review of reflex klystron theory will precede the exposition of the coupled resonator problem. In both the single and coupled resonator case, performance analysis will be based on a separate and independent study of the electronic and passive circuit admittance developed across the interaction gap and upon the graphical combination of the two in the complex admittance plane. As a by-product of this investigation, a number of driving point properties of two coupled resonant circuits will be developed which may be found of general net- works interest. Following the theory of the coupled resonator reflex klystron, an experimental tube of this type will be described and a qualitative as well as quantitative verification of the theory given. Oscillograms will be presented showing the advantages of this device when used as a sweep generator, both in the microwave band and at lower frequencies. Additional applications will be indicated in the hope that others may try them. 2.0 SMALL SIGNAL REFLEX KLYSTRON THEORY This section will be devoted to a brief review of the small signal reflex klystron theory. Emphasis wifl be on concepts leading to the equivalent circuit representation and to the graphical admittance-plane analysis. Both of these and particularly the graphical approach will later be used in the investigation of coupled cavity behavior. As stated before, the operation of the reflex klystron is the result of the interaction between a bunched electron stream and the varying electric field existing inside a resonant cavity. In circuit language, this amounts to an interaction between an active and a passive element. The active one due to the electron stream is termed electronic admit- tance, and the passive one is the input admittance of the resonator. Derivations for the expression describing the electronic admittance may be readily found in the literature.^ It will not be repeated here. The re- A COUPLED RESONATOR REFLEX KLYSTRON 719 suit of these derivations, however, and its physical significance will be discussed at some length. 2.1 Electronic Admittance Consider an arrangement of four plane and parallel elements consist- ing of a cathode, two ideal grids and a reflector. Assume these electrodes to be of infinite extent so that all electrons will move in straight paths perpendicular to the planes of the electrodes. Let the current densities encountered be low enough so that space charge effects can be neglected. Both grids are operated at the same dc potential, Vo , positive with respect to the cathode. As shown in Fig. 1, this might be achieved by connecting them to the secondary winding of an ideal transformer having a 1:1 turns ratio. The reflector is operated at a dc potential, Vr , negative with respect to the cathode. Next, an RF voltage of amplitude V is applied to the transformer and, hence, appears across the grids. Electrons emitted from the cathode are accelerated toward the first grid and arrive at it CATHODE RF VOLTAGE OF AMPLITUDE, V REFLECTOR DC POTENTIAL PROFILE Fig. 1 — Electrode arrangement and dc potential profile giving rise to elec- tronic admittance, Ye , described by equation (2.1). 720 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 with a velocity corresponding to Fo volts. During their traversal of the interaction gap, i.e., the space bounded by the two grids, this velocity is changed by an amount depending on the instantaneous direc- tion and intensity of the electric field and on the fraction of a cycle spent in traversing the gap. They then enter the retarding dc field of the re- peller drift space. Here, they are slowed down, brought to a standstill and returned to the interaction gap for a second transit. During this process, the velocity modulation acquired during the first transit is converted to intensity modulation such that the convection current returning through the gap does so in the form of sharp and well defined pulses mth a repetition rate equal to the frequency of the applied RF voltage. This pulsed current may now be harmonically analyzed and its fundamental component evaluated. The ratio of this fundamental component of current to the applied RF voltage is termed electronic admittance, Ye . Providing the amplitude, F, of the applied RF voltage is small compared to the dc accelerating voltage, Fo , the value of Ye is given by: y _ hff e \2Fo/ ,-((»/2)-»J fr, , ^ \2Vo) where Iq = dc beam current. (3 = beam coupling coefficient.* d = round trip transit time in repeller drift space, in radians. Fo = dc beam voltage. F = amplitude of RF gap voltage. Ji = Bessel function operator. Of all these parameters affecting Yg we shall focus our attention on two, namely F and 6. The other parameters depend on such factors as tube geometry, electron gun perveance etc. and, within the scope of this investigation, will be considered constant. Referring to equation (2.1), it is seen that the phase angle associated with Ye is a function of 6 only, whereas the amplitude of Ye depends on both e and the RF gap voltage, F. Suppose now 6 is held constant while * /S, also referred to as modulation coefficient, is given by /3- smj- ?£ 2 where 6,, the transit angle in the interaction gap, is expressed in radians. A COUPLED RESONATOR REFLEX KLYSTRON 721 V is increased from zero to some finite value, a process actually occurring in a reflex klystron during the build-up of oscillations. The amplitude of Ye will then be of the form 2Ji{x)/x and will decrease according to the Bessel function plot of Fig. 2. For an infinitesimal or zero RF gap voltage the electronic admittance, now referred to as "small-signal" electronic admittance, Yes , may be derived from equation (2.1) as. Ye\ ^ 7. 2F„ ^' (2.2) When presented in the complex admittance (i.e., g-h) plane, expression (2.2) assumes the form of a geometric spiral as shown in Fig. 4. Each point on the spiral corresponds to a particular value of d and, since 6 is a function of the repeller voltage only, to a particular value of Vr . We also note that for some values of 0 the conductance component of Yes is negative while for others it is positive. Thus, there is the possi- bility of generation of RF energy for values of Yes adjusted by means of the repeller voltage to fall on the left-hand half of the admittance spiral while energy is absorbed for values of Yes having a positive conductance component. As the RF gap voltage, V, builds up from zero to its final value the magnitude of the electronic admittance shrinks along a radius 1.0 0.9 0.8 ^ ^ ^ X^ — ^ N \^ 0.6 X 0.5 > \ N S^ 0.4 N \ 0.3 0.2 0.1 0 D C .2 0 .4 C .6 0 .8 1 .0 1 .2 1 .4 1 6 1 .8 2 .0 2 .2 ^ .4 2 .6 2 Fig. 2. Bessel function plot showing the relative variation of the amplitude of the electronic admittance (ordinate) as a function of RF gap voltage (abscissa) i with repeller drift angle held constant. 722 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 vector from Ye* to Ye according to the Bessel function plot of Fig. 2. For 6 = 2t{N + ^i) radians where iV = 0, 1, 2, 3 . . . the electronic admittance becomes a pure negative conductance as may be seen by putting this relation into equations (2.1) or (2.2). Summarizing, we have seen that the presence of an electron stream bunched in accordance with the arrangement of Fig. 1 gives rise to an admittance appearing across the grids bounding the interaction space. The phase angle of this admittance is a function only of the repeller drift angle, d, while its magnitude depends on both 6 and the RF gap voltage, F. For values of 6 in the vicinity of 2x(iV + ^) radians the conductance component is negative, a necessary condition for the production of sus- tained oscillations. 2.2 Passive Circuit Admittance of Single Resonator Any resonant cavity may be represented by a simple parallel G-C-L- combination provided the desired resonance is sufficiently far removed from adjacent ones. In some cases such as in cylindrical or waveguide cavities, to name two, it is difficult to ascribe physical significance to the lumped elements appearing in the equivalent circuit representation. This is not so in the case of a conventional reflex klystron cavity. Since the latter always consists of a re-entrant type resonator, most of the electric field is concentrated in the interaction gap, i.e., the narrow region traversed by the outgoing and returning electrons and bounded by two parallel grids, while the major portion of the magnetic flux resides in the outer cylindrical section. Thus, the effective shunt capacitance ap- pearing in the equivalent circuit is associated primarily with the above grid planes, a minor contribution originating in the fringing field close to the re-entrant post and the residual electric field in the outer cylindri- cal part of the cavity. The input admittance, F, of a high Q resonator when represented by C, L, and G connected in parallel is given by, Y = G{\ -\-j2Qb) (2.3) = G + j2CAcu, (2.4) where Q = 03qC/G and 5 = Aw/coq = A///o = (/ - /o)//o , and G represents all internal resonator losses plus the external load referred to the gap. Plotted in the complex admittance plane, the locus of the admittance vector with varying frequency is a straight line parallel to the imaginary axis and spaced a distance corresponding to G to its right. The var- iation in susceptance is directly proportional to the frequency deviation A COUPLED RESONATOR REFLEX KLYSTRON 723 from resonance. Also, the frequency deviation required to bring about a given change in susceptance is inversely proportional to both C and Q. 2.3 Equivalent Circuit of Single Resonator Reflex Klystron We saw that the presence of a bunched electron stream between two closely spaced grid planes gives rise to an electronic admittance the properties of which were discussed earlier. Suppose we now let these grids become part of a re-entrant cavity so that they form the boundaries of the interaction space. This will make them elements of significance and common to both the electron stream and the passive circuit admittance. As far as the electron stream is concerned the grids become the terminals across which the electronic admittance, Ye , is developed and in relation to the resonant circuit they constitute the major portion of the effective Yg (SEE EQ. 2.t) 2 ! 2 r u -*-" r^ ^C NODES I AND 2 CORRESPOND TO GRIDS BOUNDING INTERACTION GAP Y (SEE EOS. 2.3 AND 2.4) Fig, 3. — Equivalent circuit of single resonator reflex klystron. shunt capacitance. Based on these remarks, it should be apparent that a single-resonator reflex klystron may be represented by an equivalent circuit consisting of a parallel combination of Ye , C, G, and L as shown in Fig. 3. Furthermore, the expressions for Ye and Y as given by equa- tions 2.1 and 2.4, respectively, show that these two quantities are inde- pendent of each other. The electronic admittance, Fe , is a function of the electron optics of the tube, while the passive circuit admittance, F, is a function purely of cavity parameters, including the tightness of coupling to the external load. 2.4 Performance Analysis Based on Complex Admittance Plane Rep- resentation The condition for oscillation applying to the equivalent reflex klystron circuit requires the total admittance across nodes 1 and 2 of Fig. 3 to equal zero, i.e.. r. + F = 0, 2.5a 724 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 or Y. = -F. 2.5b Attainment of this condition is preceded by a period during which the RF gap voltage increases from its initial, near-zero value to its final steady state amplitude. Concurrently, the net conductance across nodes 1 and 2 changes from its maximum negative value to zero and the electronic admittance vector shrinks from Yen to Ye . This process of build-up of oscillations and the final steady state may be conveniently studied by the graphical representation of Fig. 4. Here, the negative of the passive circuit admittance, F, has been superimposed on the small-signal electronic-admittance spiral. Since the net con- ductance across the interaction gap (nodes 1 and 2 in Fig. 3) must be negative in order for oscillations to build up, we see at once that the (— F) plot, sometimes also referred to as "load line," divides the com- plex admittance plane into two regions: the one to its left, in which the REGION OF OSCILLATION REGION OF NO OSCILLATION ^ = (1 + 3^)277-^. (2 + 3^)277' (3 + 3^)27r. FREQUENCY INCREASING LOCUS OF SMALL SIGNAL ■-ELECTRONIC ADMITTANCE VECTOR, Yes (SEE EG. 2.2) Fig. 4 — Complex admittance plane representation showing the superposition of the negative of the circuit input admittance upon the small signal electronic admittance spiral. Each point on spiral corresponds to a particular value of re- peller transit angle, 0, and hence repeller voltage, Vr , and each division on the (— F) locus to an equal frequency increment. 1 I A COUPLED RESONATOR REFLEX KLYSTRON 725 negative electronic conductance exceeds the circuit conductance and where the build-up of oscillations is possible, and the region to its right where the condition for the build-up of oscillations is not met. Thus, for the load line in the position indicated, the tube will not oscillate in the A^ = 0 mode, but will do so in the iV = 1, 2, 3 and higher order modes. Still referring to Fig. 4, suppose 6 has been adjusted by means of the repeller voltage to ^i = (3 + %)27r radians, i.e., the center of the 3 -f % mode. The small signal-electronic-admittance vector will then be a pure negative conductance terminating on the spiral at point A and together with the passive circuit conductance yielding a net negative conductance across the grids of value {OA-OB). Oscillations, therefore, will build up until the equiUbrium condition. Ye = — F, has been satis- fied. In terms of the admittance plane representation, the electronic- admittance vector will shrink without change in its phase until it termi- nates on the load line at point B. Since the electronic admittance for the particular value of d chosen is a pure conductance, oscillations will occur at the resonant frequency of the cavity, /o , which is also the fre- quency corresponding to the intersection of the electronic-admittance vector and the load line. From equation 2.1 we see that, OB _ V2Fo/ L V2F„; J \2Vo) (2.6a) v=o where Vi is the steady state RF gap voltage corresponding to drift angle, ^i . Entering the graph of Fig. 2 with OB/OA as the ordinate, we can read off the corresponding value of (i8/2Fo)7i^i . For a particular tube structure and operating conditions, iS/2Fo will be a fixed constant so that, in effect, we have determined the value of a quantity propor- tional to the product of Fi^i . Next, consider the case where d has been changed from Si = (3 -f M)27r to 02 = K(S -f M)27r radians. The electronic admittance vector for F = 0 will now terminate on the Fe«-plot at C Again a build-up of oscillations will ensue and upon attainment of the steady state condition the vector will have shrunk to the value OD, with the 726 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 frequency of oscillation determined by the location of point D on the load line. By an argument similar to the one used before, we have OD _ ^\-2W) OC ' (^V^\ ^ ^ and once more, the plot of Fig. 2 will yield the value of (2-7-0)^^^ °'- (2T;)^^^'- It is apparent that /3/2Fo is a factor common to all these determinations and may be neglected if we are only interested in the mode shape, i.e. the relative variation of gap voltage or output power with the repeller drift angle, 6. For modes higher than A/' = 2, the electronic admittance spiral approximates a number of semi-circles with their centers close to the origin. Hence OA^OC and -M_ > £^ . OC OA Figure 2, then indicates that ViOi > V^2 or Fi^i > KV^i and since K does not vary greatly from unity V\ > V2. This is as it should be since we know that a change in repeller voltage from its mid- mode value causes a decrease in power output and, hence, in gap volt- age. The latter will decrease to zero when 6 has been adjusted to a value such that the Ye, vector terminates at the intersections of the electronic admittance spiral and the load line. Another result which becomes apparent from an inspection of Fig. 4 is this. For the condition of stable oscillation the phase angle of the electronic admittance equals that of the passive circuit except for an additive constant of 180 degrees which, however, may be disregarded in this argument. The frequency of oscillation may be determined from the input-phase-angle vs. frequency plot of the passive circuit by looking up the frequency corresponding to the particular value of phase angle to which the electronic admittance has been adjusted (by means of repeller voltage). Thus, the curve relating repeller-drift-angle to fre- quency is identical with the plot of input-phase-angle vs. frequency for the passive circuit. Moreover, if repeller voltage is linearly related A COUPLED RESONATOR REFLEX KLYSTRON 727 to the repeller-drift-angle, a condition which in most practical cases holds over a restricted repeller voltage swing about its midmode value, then the repeller voltage-frequency plot ^vill have the same shape as the input phase angle-frequency curve of the passive circuit, differing from the latter only by a constant multiplying factor. We therefore conclude that the modulation performance of the reflex klystron, at least over the central portion of the mode, may be predicted from an examination of the driving point properties of the passive circuit. The above method of graphical analysis is quite useful in the case of conventional reflex klystrons, although the same results may be ob- tained analytically by making use of the equation for electronic admit- tance in conjunction with that of the input admittance of a single resonator. In the case of coupled resonators the expression for input admittance becomes much more involved, as we shall see later, Avith the result that the graphical approach outlined above was found by far the quicker and more practical method of solution. 3.0 THEORY OF COUPLED RESONATOR REFLEX KLYSTRON It has been shown that the performance of a reflex klystron can be analyzed by considering the electronic and passive circuit admittances separately and then combining the two graphically in the complex admittance plane. The same procedure can be adopted in the deter- mination of mode shapes resulting from the interaction of the electronic admittance with any arbitrary circuit admittance which can be realized across the gap. Conversely, we can determine the admittance or im- pedance function required to produce a particular desired mode shape. In other words: given an admittance function, we can determine the resulting mode shape, and given a desired mode shape, we can predict the required admittance function. As an example, consider a mode having a flat top and vertical sides as shown in Fig. 5(a). This would be the ideal shape for a reflex oscillator to be used as an electronically swept signal source. To achieve this mode shape, we must realize an admit- tance function across the bunching grids yielding a constant excess of negative electronic admittance over a range of repeller voltages. Such a function is shown in the complex admittance plane along with the plot of Yes in Fig. 5(b) and the corresponding input impedance in 5(c). It ^vill result in a frequency range of constant RF gap voltage which in turn mil give rise to a range of flat power provided it is developed across a constant, frequency-invariant conductance. In order to clarify this rather important consideration, let us write the admittance appearing 728 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 DESIRED MODE SHAPE (FREQUENCY VARIED BY MEANS OF REPELLER VOLTAGE) NEGATIVE OF ADMITTANCE FUNCTION REQUIRED TO PRODUCE MODE SHAPE OF COLUMN 1. PLOTTED TOGETHER WITH Yes '^ g - b PLANE INPUT IMPEDANCE VS. FREQUENCY OF NETWORK WHICH IN g-b PLANE APPEARS AS SHOWN IN COLUMN 2 IDEAL MODE SHAPE FOR ELECTRON- ICALLY SWEPT SIGNAL GENERATOR R F POWER OUT (a) (b) CONSTANT EXCESS OF NEGATIVE ELECTRONIC ADMITTANCE (C) REALIZABLE MODE SHAPE BY MEANS OF COUPLED RESONATORS (d) RANGE OF CONSTANT EXCESS OF NEGATIVE ELECTRONIC ADMITTANCE (f) CONVENTIONAL SINGLE -RESONATOR REFLEX KLYSTRON (SHOWN FOR COMPARISON) m (I) Fig. 6 — Desired mode shapes and passive circuit impedance functions re quired to produce them. A COUPLED RESONATOR REFLEX KLYSTRON 729 across the interaction gap as, Y = Gr{f) + jBr(f), where Grif) and Brif) denote the total conductance and total suscep- tance both of which are functions of frequency. Over the region of constant RF gap voltage we require | F | or VGlif) + BUf) to be constant. This, however, means that the power generated (as distinct from the power deUvered to the load) is not constant since it is given by. Generated Power = J^(RF gap voltage)^ (Total Conductance) = y2v'GT{i) and Gt varies with frequency. Now, the term Gt(J) is the sum of a number of conductances, all but one of which do not change with fre- quency. The frequency-invariant conductances represent the different power losses in the primary resonator plus the external load referred to the gap while the remaining frequency-sensitive conductance is due to the coupled circuit used in shaping the admittance locus. Hence the useful power output is proportional to V^ and therefore constant if V^ is constant, whereas the variation with frequency of the total conduct- ance must be taken into account in evaluating generated power. The foregoing discussion together with the illustrations of Fig. 5 should make it clear that in order to maintain a constant RF power output level, over a specified frequency range, we must have the elec- tronic admittance interact with a circuit the input impedance of which, when referred to the gap, is also constant over the same frequency range. A circuit having such characteristics can be obtained by the use of coupled resonators as will be shown later. Suppose the emphasis is on modulation hnearity rather than flatness of power output. Attention, then, must be focused on the relation be- tween input phase angle and frequency of the passive circuit. Here, again, we shall see that the application of coupled resonators offers advantages beyond what is possible with a single cavity. 3.1 Driving Point Properties of Two Coupled Resonators Having Equal Q's Using the equivalent shunt representation as shown in Fig. 6, the exact expression for the input admittance, F, of two coupled resonators 80 (b) 0 _^ 5 ^ g .-< # f ■<^, ^ y y^ ^ ^ ^ ::^ y / /• y / //, ^ ?> ^ / / / / -5^ ^/ /^ / / / / f//^^ o%/ / t / / / z / <^^/' u / / / "T-^n i W' {/ / r / 7/ ///. / / / / UJ ?20 / ///^ 7/ ^/ 7 / / r° ;^ z^ ' / /y / / / / // V ^/ V, / / / J V '/ / / / / A ^ ^ y / / \ N.. / / •>irk \^ '^ A / V -^ /^ -20 ■0.2 0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 2Q \ w 7 \ V / A /v^KOk S/y V ( \ \ \ °7 // § vr \i A \ s \ / //// 1 "/v VM \^ \ \ \ \ ^ n V // ^\ /v V 0\ \ \ V \ / 7 k / // \\| \V ^^^ \ \ \ > \ ^ / I h / X \\ ^^ >\ \\ \ V \ \ / 7 N / \ ^^^ ^^^ k> \ \ \ \ t 03 y /, 1 r / \N s^ ^ \ \ \ N \ ^ !s J / 1 f i /o ^ ^ ^ ^ >N^ s s \, \ N, -7-1 no 1 1 1 ' / s ^ $^ d ^ s \ \ J 1 / / / ^v: ^^ >^ ^ b- J / / / ^, versus frequency deviation, (c) Rate of change of with 2Q5 versus 2Qd. 731 12m , 1 I I / 732 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 having equal Q's and equal resonant frequencies is given by, Y = Gig+jb), where (QkV , QiQ'm - 1) / , \^^/ ^^ "^ '^" [Qi^'m- DTI- L ^m J/ (3.1) \ 2 ^ ^ ^^^"^:^0(Q^m-l)t' "^ L fi^ J G = shunt conductance of primary and secondary resonators. y, 6 = total input conductance and'susceptance respectively normalized with respect to G, 12 = normalized frequency = f/fo . /o = resonant frequency of both primary and secondary resonators. k = coefficient of coupling. m = 1 - k\ This expression for input admittance, though accurate, is rather un- wieldy because of the many variables involved. A few obvious approxi- mations, however, will change equation (3.1) into a much simpler and more meaningful expression, yet sufficiently accurate for the range of Q's and bandwidths of interest here. Let, Jo Jo Jo where 6 denotes the normalized frequency deviation from resonance, Af/fo . If we further assume that k^ « 1 and the range of Q's is such that {Qk) may vary from zero to about five and that the maximum value of 5 is small enough so that its higher powers may be neglected, then equation (3.1) may be simplified to. Y G -b-T^,]*H'-rf^l « A COUPLED RESONATOR REFLEX KLYSTRON 733 The above equation* essentially contains three variables: (a) The dependent variable, F/G, i.e. input admittance normalized with respect to the shunt conductance, (b) the independent variable, (2Q5), which is a factor proportional to the frequency deviation from resonance, and (c) parameter, (Qk), a measure of the tightness of coupling between the two cavities. Compared mth the input admittance for a single resonator which was given earlier, [equation (2.3),] as, Y/G = 1 + j2Q8, it is seen that the conductance component has been changed by a factor L ^ 1 + (2Qsyj and the susceptance by L 1 + (2Q8yj • Also, by setting k = 0, i.e., completely decoupling the secondary resona- tor, equation (3.2) reduces to equation (2.3) as, indeed, it should. The information contained in equation (3.2) may be presented graph- ically in a number of ways. We can plot the magnitude of the normahzed input impedance, GJ Z |, as a function of 2Q8 with (Qk)^ as parameter as shown in Fig. 6(a).t Or we can plot the input admittance given by equation (3.2) directly in the g-h plane as in Fig. 7. Finally, we may show the variation of input phase angle with normalized frequency for differ- ent degrees of coupling as in Fig. 6(b). Each of these graphical represen- tations has an important bearing on the performance of the coupled resonator klystron. Thus, the curves of Fig. 6(a) will have the same general shape as the RF power output vs. frequency plot of the reflex oscillator, the family of curves of Fig. 7, when superimposed on the electronic admittance spiral, can be used for a detailed analysis of mode shapes, and Fig. 6(b) determines the modulation characteristics obtain- able with coupled resonators. Having briefly touched upon the significance of the families of curves given in Figs. 6(a), 7, and 6(b), we can now proceed to discuss them in greater detail and to establish further valuable results. * To check the accuracy of this equation, the curve for (Qky = 1 in Fig. 6(a) was replotted using the full expression given by equation 3.1 and assuniing Q = 100. The agreement was found to be essentially perfect as far as its use in this investigation is concerned. t For a similar presentation of the transfer characteristics of coupled tuned circuits see reference (6), 734 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6 3.8 4.0 g -* Fig. 7 — Input admittance of circuit shown in Fig. 6 plotted in g-b plane. Solid lines are loci of input admittance vectors, dashed lines connect points of equal frequencies. A COUPLED RESONATOR REFLEX KLYSTRON 735 3.1.1 Variation of Input Impedance with Frequency j Fig. 6{a), Equation 3.2 may be written as G where ^ g + jh, (3.3) and Hence \y\ _ G and = VfTv It is seen that ^^ and 6^ involve even powers of (2Q5) only, so that G\ Z \ is an even function, i.e., symmetrical about the vertical axis. For this reason, one-half the normalized impedance plot only has been given in Fig. 6(a). Inspection of this figure reveals that the effect of coupling to a secondary resonator is to broaden the frequency range over which a high impedance level can be maintained across the interaction gap. Comparing the variation of input impedance with frequency for {Qkf = 0 (i.e., single cavity) with that for {Qkf = 0.3, we see that for a frequency deviation of 2Qb = d=0.6 the former shows a drop of 14 per cent while the latter only varies by ±0.58 per cent. In terms of two cou- pled resonators having Q = 100 and operating at 4000 mc this means a variation in impedance of only ±0.58 per cent or ±.052 db over a frequency range of 24 mc. Another result clearly brought out by the family of curves of Fig. 6(a) is that the process of broadbanding by coupHng to a second resona- tor results in a reduction in absolute impedance level. This reduction in impedance level at midband is related to the tightness of coupling, (Qk) , by the expression, 736 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 which may be readily derived from eq. 3.6. Note that for {Qkf = 1, the impedance level at midband has dropped to half the value obtained mth a single resonator or (Qkf = 0. The degree of coupling corre- sponding to (QkY = 1 is noteworthy for reasons other than the one just mentioned. They will be discussed in the following section. 3.1.2 Input Admittance Plot in g-b Plane j Fig. 7. The graphical repre- sentation of input admittance in the complex admittance plane is of particular usefulness in the analysis of the coupled resonator reflex klystron. For the moment, however, we shall restrict the discussion to a consideration of the passive circuit only. Each solid line in Fig. 7 is the locus of the admittance vector for a particular tightness of coupling. For {Qkf = 0 the locus is a straight Une parallel to the susceptance axis as described earlier for the case of a single resonator. For (Qk) =0.3 the shape of the locus approximates a circle with its center at the origin; this, of course, being true only over a restricted frequency range. As the coupling to the secondary resonator is progressively tightened, the locus is seen to bulge out in the direction of increasing conductance until, for (Qk) = 1, it forms a cusp. This condition will henceforth be referred to as ''critical coupling."* Coup- ling even tighter causes the formation of loops of increasing size. For the overcoupled case, (Qk) > 1, the admittance-vector locus crosses the conductance axis three times, with the first and third crossings coincident and independent of (Qk)^ and the second crossover a function of (Qk) . The location of these intersections with the ^-axis may be determined by equating the susceptance to zero, i.e., from equation (3.5), ^'['-r^J- Hence the crossing to the extreme right occurs for 2Q8 = 0 while the first and third interesections correspond to 2Qd = ±ViQky - 1. To determine the size of the loop we substitute 2Q8 = 0 into the expres- sion for the conductance, i.e., equation (3.4) and obtain, oL.^ = 1 + my, '2Qi-^ * It should be noted that the term "critical coupling" as applied to the transfer characteristics of coupled tuned circuits, though also occuring for (Qk)^ = 1, assumes a different significance in that it describes the condition of maximum flatness in response and optimum phase linearity. (See reference 6.) In the case of the coupled resonator reflex klvstron, "critical coupling" forms the transition between stable performance and load hysteresis as will be shown later. I A COUPLED RESONATOR REFLEX KLYSTRON 737 showing that the size of the loop is a sensitive function of the degree of coupUng. The conductance value for the first and third crossover points is obtained by substituting 2Q5 = ±V{QkY - 1 into equation (3.4) and results in gr = 2, i.e., a value of g independent of the coupling coefficient. The dashed lines shown in Fig. 7 connect points of equal frequencies. It is of interest to note that these loci are straight lines crossing the conductance axis at g = 2. To prove this, eliminate (Qk)^ between equa- tions (3.4) and (3.5). This yields, whence h = i'2Q8){g - 2). (3.8) For a particular and constant value of 2Q8, equation (3.8) describes a straight line of slope equal to ( — 2Q5) intersecting the g-a,xis Sit g = 2. 3.1.3 Variation of Input Phase Angle with Frequency, Fig. 6(h). The last driving point property of interest to this study is the dependence upon frequency of the input phase angle, 0. Referring to equation (3.2), this quantity is obtained as. The graphical representation of this function is given in Fig. 6(b). It shows the gradual transition from a simple S-shaped curve for (Qk) = 0, having its only point of inflection at the origin, to the type of curve corresponding to (Qk)'^ > 1 which intersects the frequency axis three times. The special case of {Qkf = 1, considered earlier and found to result in the formation of a cusp in the complex admittance plane, now gives rise to a plot of input phase which is tangent to the horizontal axis at the origin. To investigate the condition for greatest linearity between phase angle and frequency, which, when applied to the coupled resonator reflex klystron, would be the condition for optimum modulation linearity, one could simply apply a straight edge to the curves of Fig. 6(b) and pick the best value of (Qk)^ in this manner. A much more sensitive criterion of linearity, however, is the variation with frequency of the slope of these curves. An analytical expression for this slope has been 738 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 derived from equation (3.9) as, d4> [1 - mY] + (2Q5)^[2 + 4{Qm + i2Q8Y d(2QS) [1 + {QkYf + (205)2[3 + (Qky] + (2Q5)^[3 - 2{Qky + (2Q6)« (3.10) The above expression, involving even powers of 2Q5 only, results in a family of curves symmetrical about the vertical axis, the positive half of which is shown in Fig. 6(c). From it the value of (Qk)^ for greatest linearity or most constant slope is seen to lie somewhere between 0.1 and 0.2. This is further borne out by Fig. 8, in which this region has been more fully explored. Fig. 8 constitutes a plot quite similar to that 1.08 1.04 .00 a96 °0.92 -r;o.e8 o ^0.84 •0- , 0.80 O ^ 0.76 a72 0.68 a64 a 60 ^^ -^ ^ ^ y ■ — •^ "v^ N ^ ^ • — — — ■ \ \ ^ V,, ■^ ^ "^ ^ N ^ \ — 1 \ X s \ > N s .> V \ \ \ S \, s <: s^ ^^ < \ \ \ \ s \^ s> \ \ N ^ \ \ \ ^ \\ V k \ \ \ ^ s^ \\ \ ^ \ \ V \ ^ \ \ \ ^ X \ 'C \ \ ^ \ \ \ \ \ \ \ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Fig. 8 — Rate of change of input phase angle with frequency normalized with respect to its center value for the circuit shown in Fig. 6. This quantity, the constancy of which is a measure of phase angle versus frequency linearity, is seen to stay absolutely constant over a frequency range corresponding to 2Q8 = dbO.3 for (Qk)^ = 0.12 while the single resonator case, i.e. (Qk)^ = 0, shows a change of 8% over the same frequency range. A COUPLED RESONATOR REFLEX KLYSTRON 739 of Fig. 6(c) except that the ordinate now represents the instantaneous slope normaUzed with respect to its midmode value. The curves shown are for values of (Qkf ranging from zero to (Qkf = 0.18. Let us, for example, examine the curve corresponding to (Qkf = 0.12; the slope is seen to stay absolutely constant up to a frequency deviation of 2Q8 = dbO.3, while the plot for the single resonator, i.e., (Qkf = 0, included in this figure for comparison, changes by 8.4 per cent over the same fre- quency range. Putting this in another way, suppose the maximum allowable devia- tion in slope from its mid-band value is one per cent. Fig. 8 then in- dicates that the permissible frequency deviation for coupled resonators having a value of {Qkf = 0.135 is given by 2Q8 = ±0.5, whereas it must be restricted to one-fifth this value, i.e., 2Q8 = dzO.l, for the single resonator case. In terms of a midband frequency of 4000 mc and Q = 100, the permissible frequency excursions would be ±10 mc and ±2mc, respectively. It is to be noted that the value of coupling coefficient resulting in greatest modulation linearity is considerably smaller than the value of coupling coefficient found to yield a constant impedance level. 3.2 Mode Shapes Resulting from the Interaction Between Electronic I Admittance and Input Admittance of Two Coupled Resonators of EqvM Q's Having determined all relevant properties of the passive circuit as they appear across the grids of the bunching gap, we may now proceed to investigate the results arising from their interaction with the electronic admittance. The approach to be adopted is essentially the same as the one outlined earlier for the case of the conventional single-cavity reflex klystron. It involves a graphical superposition of the negative of the passive circuit admittance upon the small-signal-electronic-admittance spiral in the g-h plane, such as shown in Fig. 9(a). The location of the load lines with respect to the spiral has been chosen, somewhat arbi- trarily, such that the ratio between the length of the Fes-vector cor- responding to ^ = (2 + %) cycles and that of the input admittance vector for {Qk^ = 0 and 5 = 0 equals two. Load lines have been drawn for five values of (Qk)^ ranging from zero to unity. The determination of mode shapes from Fig. 9(a) proceeds as illus- trated in Table I. Taking the repeller drift angle as the independent variable we can obtain corresponding values of generated power (in arbitrary units) and frequency (in terms of Q8) by going through the steps indicated. DETERMINATION OF THEORETICAL MODE SHAPES FOR 33/4 REPELLER MODE AND SECONDARY Q EQUAL TO PRIMARY Q il 111 0^ O £D a tr 0-^ 0 ^^ s Co) /^ ^0.4 ^ 4 r \ r 1 1 1 0.8 \^ ^ (c) N ^ ^^- \ y$ '>>v',. ^ -0.4 -0.2 -0.4 N \S. -0.8 \^ V L\ 0. 94 0.96 0.98 1. 00 1.02 1. 04 1 NORMALIZED REPELLER DRIFT ANGLE ■t 3 -^ (d) 4 N K ^ f^- \^ ^ ^ ^ 0.6 0.2 %^l ^//V\l^/V (e) 1 //Toy \, ^ r t '/ 1 ■1.2 0 Fig. 9 — Graphical determination of mode shapes for a coupled resonator reflex klystron having the secondary cavity Q equal to primary Q. (a) Load lines for various degrees of coupling superimposed on small signal electronic admittance spiral, (b) Power output versus repeller drift angle, (c) Frequency versus repeller drift angle, (d) Power output versus frequency, (e) Power output, normalized with resi)ect to its maximum value within the mode, shown as a function of fre- quency. 740 A COUPLED RESONATOR REFLEX KLYSTRON 741 Table I — Determination of Mode Shapes from Complex Admittance Plane Plot of Figure 9(a) (1) (2) (3) (4) (5) (6) (7) (8) (9) Kdo Repeller Drift Angle (9o = Midmode Ratio of Drift Angle to its Midmode Value Small ir tronic Admit- tance Passive Circuit Admit- tance (Kveo) from Bessel- function plot of Fig. 2 V0O, (6)/(2) {V0o)\ Quantity Propor- tional to Power Output QS Value) Measured oflF Fig. 9(a) in Arbitary Units of Length my = 0.4, 2^-mode, i .e., ^0 = (2 + ^)2ir radians = 5.57r radians 5.207r 0.947 9.45 9.45 1.000 0.00 0.000 0.00 +0.85 5.257r 0.955 9.54 8.10 0.849 1.13 1.182 1.40 +0.67 5.30x 0.967 9.63 7.35 0.764 1.44 1.493 2.23 +0.53 5.357r 0.974 9.72 6.95 0.715 1.60 1.645 2.70 +0.42 5.407r 0.983 9.82 6.80 0.693 1.67 1.700 2.89 +0.30 5.457r 0.992 9.91 6.80 0.686 1.69 1.703 2.90 +0.17 ^0 = 5.50x 1.000 10.00 7.00 0.700 1.65 1.645 2.71 0.00 5.557r 1.010 10.10 6.80 0.673 1.73 1.713 2.94 -0.17 5.607r 1.020 10.20 6.80 0.667 1.75 1.715 2.95 -0.30 5.657r 1.029 10.29 6.95 0.676 1.72 1.673 2.80 -0.42 5.707r 1.038 10.38 7.35 0.709 1.62 1.560 2.44 -0.53 5.757r 1.046 10.46 8.10 0.775 1.40 1.340 1.80 -0.67 S.SOtt 1.055 10.55 9.45 0.895 0.95 0.900 0.81 -0.85 The results of this analysis are shown in Fig. 9. This illustration, in addition to giving detailed performance characteristics for the 3 + J:^ mode, also indicates clearly the wealth of information which may be obtained from the complex admittance plane representation. Although the curves shown are self-explanatory, a few comments regarding their significance would seem to be in order. It is seen, for instance, that a coupling coefficient so adjusted that (Qkf lies between 0.2 and 0.4 will produce a frequency range of essentially constant power. In par- ticular, if we pick the curve for {Qkf = 0.4 from the family of curves of Fig. 9(d), it will be seen that the variation in power over a bandwidth of Q8 = ±0.4 is ±2 per cent, while the corresponding value for the single resonator case, i.e., (Qkf = 0, is minus 23 per cent. As the value of {Qkf is increased beyond 0.4, the depression in the center of the mode becomes increasingly pronounced until it turns into a cusp for (Qk) = 1. Mode shapes for (Qkf > 1, though of no direct interest to this in- vestigation, are indicated in Fig. 10 since they may be encountered in practice in cases of excessively tight coupUng and could then be recog- nized as such. If the mode is traversed in the direction of increasing I Vr I (or increasing frequency), such that the intersection of the electronic 742 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 admittance vector with the load line moves down along the latter as indicated by the arrows in Fig. 10(a), power changes smoothly until the F«- vector becomes tangent with the loop at 4. At this point there occurs a discontinuous jump in both power and frequency caused by the sudden shortening of the Fe- vector from 4 to 7. From here on, power and frequency again become single valued functions of Vr . The mode shape corresponding to this uni-directional sweep is shown in Fig. 10(b). If the mode is traversed in the opposite sense we again encounter this discontinuity although it will now occur on the opposite side of the loop and, therefore, at a different repeller voltage and frequency. Fig. 2^*^3^^ _A\ (0 Fig. 10 — Production of load hysteresis by overcoupled cavities, (a) Load line for overcoupled cavities, (b) Oscillographic mode representation for unidirectional (sawtooth) repeller sweep, (c) Oscillographic mode representation for sinusoidal repeller sweep. 10(c) shows the load hysteresis effect as one might expect to observe it on the oscilloscope screen with a sinusoidal sweep applied to the repeller. 3.3 Driving Point Properties of Two Coupled Resonators Having Un- equal Q^8 The presentation of the theory of the coupled resonator reflex klystron will now be concluded by a discussion of the more general and, as we shall see, more useful case of two coupled cavities of unequal Q's. Specifi- cally, we are considering the equivalent circuit shown in Fig. 11. By making approximations similar to those which led to equation 3.2, the input admittance for the case of unequal Q's may be derived as,^' ** ^ J-[ 1 + k'QQ. 1 -h (2Q.5)2 ]'-'H'-rvmJr\' ''■''' A COUPLED RESONATOR REFLEX KLYSTRON 743 3.2 3.0 2.8 2.6 - 2.4 2.2 2.0 t,., 1.2 1.0 0.8 0.6 ' 0.2 Ol_ _ 0-3^ 0.4 - O.P.^^^^ Fig. 11 — Driving point properties of two coupled, synchronously tuned reso- nators for the case of the secondary Q equal to half the primary Q. 744 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 or in more convenient fonn as, where The above expression contains four variables, namely F/G, (Qk), {2Q5) and p, so that the complex admittance plane representation will now have to be restricted to particular values of p. One such plot, for p = J^, appears in Fig. 11. It is similar to that of Fig. 7 except that only the positive half has been shown since equation (3.12) is symmetrical about the conductance axis; in addition, the parameter, (Q/c)^, has been carried to the point of critical coupling only. i.e.. the formation of a cusp, and not beyond. The frequency contours are again seen to be straight Unes crossing the horizontal axis at a value of conductance equal to the in- put conductance for the condition of critical coupling and 2Q8 = 0. Equation (3.12) shows that the susceptance term will be zero for (a) 2QS = 0, or for (b) p\Qkf = 1 + p\2QSf. (3.13) The value of conductance corresponding to condition (a) is given by g = 1 -\- piQhf and the value corresponding to condition (b) by gr = 1 + 1 /p. It is interesting to note that this latter value which determines the point of intersection of the frequency contours, as well as of the admittance plot for critical coupling, with the conductance axis, is independent of the actual values of Q and the degree of coupling and only dependent upon p, the ratio of Q's. Thus in Fig. 11, which con- stitutes a plot for p = 0.5, the value of conductance at whieh all the above named contours meet is given hy g = 1 + 1/0.5 = 3. From what has been said before we know that at critical coupling the admittance locus forms a cusp intersecting the conductance axis at ^ = 1 + \/p at the frequency, 2Qb = 0. Substituting this value of 2Q6 into the conductance term of equation 3.12 and equating to 1 + 1/p yields 1 + pmy = 1 + -- . p A COUPLED RESONATOR REFLEX KLYSTRON 745 Hence vQk = Qgk = 1 (for critical coupling). (3.14) The effect of reducing the secondary resonator Q is to broaden the frequency range over which a high input impedance or low admittance may be maintained. This is clearly illustrated in Fig. 12 where curves having the same value of conductance at resonance have been selected from Figs. 7 and 11 and superimposed to facilitate comparison. (Qk) Fig. 12 — Comparison of admittance loci for coupled resonators of equal Q's ith the case in which the secondary Q equals half the primary Q. 746 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 3.4 Mode Shapes Obtainable with Two Coupled Resonators of Unequal The complex admittance plane representation for the case of the secondary Q equal to half the primary Q is shown in Fig. 13(a) and the resulting mode shapes for the 3 + % repeller mode in Figs. 13(b), (c), (d), and (e). Again we notice that the coupUng coefficient required for best modulation linearity is considerably smaller than that which re- sults in a flat topped power curve. From Fig. 13(c) the most linear frequency-repeller voltage curve is associated with (Qfc)^ = 0.8 while from Fig. 13(d) a flat power mode may be obtained with a value of (Qk)^ somewhat less than 1.6. In cases where neither modulation linearity nor constant power output are of importance but where the application requires a wide electronic tuning range. Fig. 13(e) shows the advantage to be gained from coupled cavities. Here, power output has been nor- malized with respect to its peak value within the particular mode under consideration and plotted against Q8. The ratio of half -power band- widths for the curve corresponding to (Qk) = 2 A to the single resonator case, i.e., (Qk)^ = 0, is seen to equal 1.73. A phenomenon which may be encountered in the operation of the coupled resonator reflex klystron is illustrated by the (Qk)^ = 4.0 curve of Fig. 13(b). Here we are dealing with a split mode in which the powder, though everywhere a single valued function of repeller voltage, drops to zero over a range of repeller voltages centered about the middle of the mode. The reason for this behavior may be readily understood from an inspection of the complex admittance plane representation of Fig. 13(a). It is caused by the Fcs-locus for the 3 -|- % repeller mode crossing the appropriate load line and thereby resulting in a frequency band over which the condition for oscillation cannot be met. 4.0 AN EXPERIMENTAL COUPLED-RESONATOR REFLEX KLYSTRON The reduction to practice of the theoretical results obtained in the above study raises these requirements: (1) An arrangement must be found which allows the coupling between primary and secondary cavities as well as between primary cavity and waveguide output line to be varied continuously and independently. (2) Either primary or secondary resonators (or both) must be tunable to allow frequency adjustments for synchronous operation. (3) Secondary resonator Q should be continuously variable. (4) The secondary resonator should be detachable for independent determination of Q. A COUPLED RESONATOR REFLEX KLYSTRON 747 ETERMINATION OF THEORETICAL MODE SHAPES OR 33/^ REPELLER MODE AND SECONDARY Q QUAL TO PRIMARY Q - — g 0.94 D.96 0.98 NORMALIZED DRIFT ANGLE (34 MODE) ■_ ^ liio: ^^ oca CL ir .'^ ~N^ (d) J Y -- \ y ^ ^y ^ ^ 1.6 ^ r v^ ^ \ K 0.8 Z) o tr ^ 0.6 o 1 z /^ ^ / r ^ <^ \" / /o K y \\ V. /, /// \ \V ' \v / \ 0.6 Fig. 13 — Graphical determination of mode shapes for a coupled resonator reflex klystron having the secondary cavity Q equal to half the primary Q. The effect of lowering the secondary Q may be studied by comparing the above curves with the corresponding ones shown in Fig. 9. 748 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 (5) Frequency at which these experiments are to be performed should be in a microwave band where good waveguide components and measure- ment techniques are available. These considerations have led to the adoption of an external cavity type reflex klystron, the Sylvania 6BL6, as a vehicle for the experimental studies to be described, and operation in the 3700-4200 mc band. 4.1 Constructional Features of Experimental Tube and Circuit The 6BL6 is one of a group of Sylvania low-voltage reflex klystrons^" designed for use with external cavity resonators. Electrical connection to the interaction-gap grids is made through gold plated contact rings, formed from the disc seals which pass through the glass. The top ring is slightly smaller in diameter than the bottom ring, thereby permitting the insertion of the tube without disturbance to the associated cavity. Fig. 14 shows the external appearance of this tube and also contains a view of the major components of the passive circuit with cutouts to indicate the internal construction. SECONDARY Q TUNING ADJUSTMENT SECONDARY CAVITY RESISTANCE VANE COUPLING ADJUSTMENT BETWEEN PRIMARY AND SECONDARY CAVITY SHUTTER TO VARY . OUTPUT COUPLING IRIS OUTPUT Fig. 14 — An exporiinontal coupled resonator reflex klystron having a fixed frequency primary cavity. A COUPLED RESONATOR REFLEX KLYSTRON 749 Operation in the 4000-mc band was found to require a closely fitting primary cavity resonating in its principal mode. This unit, made of gold plated brass, is coupled to the output waveguide and secondary cavity respectively through two rectangular irises the sizes of which are independently variable by means of shutters. Toroidal contact springs located in circular grooves grip the inserted tube and complete the external circuit. The secondary cavity consists of a length of rectangular waveguide and uses a movable contacting plunger for tuning. A micro- meter driven resistance vane may be inserted into this cavity through a longitudinal slot in its top surface thereby obtaining a wide range of continuously variable Q's. The entire unit is attached to the output waveguide by means of the adapter plate shown to the right of the primary cavity in Fig. 14. A coupled resonator circuit having a tunable primary cavity is shown in the photograph of Fig. 15. The secondary cavity of this unit is identical to the one shown in the Fig. 14 but the primary resonator has been SECONDARY Q ADJUSTMENT 6 BL 6 TUNING PLUNGER SECONDARY CAVITY TUNING ADJUSTMENT Fig. 15 — An experimental coupled resonator reflex klystron with a tunable primary cavity. The mechanical tuning range of this tube extends from 3700-4200 750 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 modified to take two non-contacting tuning plungers so dimensioned that they can be moved in very close to the glass wall of the 6BL6. In this way an operating frequency range from 3700-4200 mc was obtained. 4.2 Qvalitative Verification of Theory The results of a qualitative verification of the coupled-resonator reflex klystron theory are given in the form of oscillographic displays in Fig. 16. Photograph (a) shows the (3 + %)-repeller mode of the 6BL6 with the secondary cavity completely decoupled. In photograph (b) the coupling iris and secondary Q have been adjusted for a flat topped output curve. The bandwidth of this curve for a variation in power output of ±0.1 db was found to be 58 mc, i.e., considerably greater than the half-power bandwidth of the single resonator case above. Increasing Qsk beyond this point by either enlarging the coupling iris or withdraw- ing the resistance vane from the secondary resonator or a combination of both brings about the condition of critical coupling, illustrated by photograph (c). As explained earlier, this mode-shape results from the admittance plot of the passive circuit forming a cusp in the g-h plane. Comparison between this photograph and the theoretical mode of Fig. 9(b) for (Qkf = 1 indicates good qualitative agreement. Coupling the secondary cavity still tighter, i.e., making the value of Qsk greater than unity, causes the formation of a loop in the circuit admittance plot and the consequent load hysteresis shown in photograph (d). Whereas all oscillograms discussed to this point have been obtained with a 60-cycle sinusoidal sweep applied to the repeller, the last one in this group, (e), results from a unidirectional (sawtooth) repeller sweep. The nature of this hysteresis effect has been explained earlier, and oscillograms one might expect to observe with overcoupled resonators for both sinusoidal and unidirectional sweeps were shown in Fig. 10. Again we note good qualitative agreement. 4.3 Quantitative Verifixxition of Theory As further strengthening of the theory underlying the operation of the coupled resonator reflex klystron a quantitative, experimental veri- fication was undertaken. The methods used in connection with this work and the results obtained will form the subject of the sections to follow. To anticipate some of the conclusions which will be presented in the course of this description, let it be said here that the quantitative agreement between theory and experiment was found to be of an order high enough to justify amply the approximations involved in the ex- A COUPLED RESONATOR REFLEX KLYSTRON 751 pressions for both electronic admittance and passive circuit admittance, and to establish confidence in the method of analysis proposed to predict tube performance under a variety of conditions. Since theoretical results had been worked out for the cases of the secondary Q equal to the primary Q and for the secondary Q equal to |Vr| and f INCREASING — *► (a) SECONDARY RESONATOR COM- PLETELY DECOUPLED. HALF POWER ELECTRONIC TUNING RANGE EQUAL TO 49 MC. (PEAK POWER = 100 MW) SECONDARY RESONATOR COUPLING AND Q ADJUSTED FOR FLAT TOPPED MODE SHAPE. BANDWIDTH FOR POWER VARIATION OF ± 0.1 DB IS 59 MC. CRITICAL COUPLING, LOAD HYSTERESIS DUE TO OVER- COUPLING, L.e.,QsK > 1 SAME AS (d) EXCEPT FOR UNIDIRECTIONAL (SAWTOOTH) REPELLER SWEEP Fig. 16 — Qualitative verification of theory. These oscillograms were obtained * with a 6BL6 in the circuit of Fig. 14 and the following operating conditions: Fo = 325F, Ik = 28 ma, /o = 3800 mc, 3 + ^ cycle repeller mode. 752 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 half the primary one, (see Figs. 7 and 11, respectively) it was decided to establish the same conditions in the experimental tube and to compare the results thus obtained with those predicted by theory. Prerequisites for the execution of these experiments were: (a) Knowledge of the primary Q. (b) Complete calibration of the secondary cavity with respect to both the variation of Q, with penetration of resistance vane and varia- tion of coupling coefficient with size of coupling iris. (c) ICnowledge of the location of the load lines with respect to the small signal electronic admittance plot in the complex admittance plane. 4.3.1 Determination of Primary Q. The above parameter is under- stood to denote the "operating" or loaded Q of the tube with the output iris adjusted to its final and permanent size and the secondary cavity completely decoupled. For the case of an inductively tuned (fixed gap) reflex klystron, it may be determined experimentally using the expres- sion/ Q = tt/S dVn df_ dVn - ^{N + f), (4.1) /r=J\r+3/4 where t denotes the repeller space transit time. Both dr/dVR and df/dVR are to be evaluated at the center of the mode. The functional relationship between r and Vr was obtained by placing the 6BL6 in the tunable primary cavity of Fig. 15 and determining the repeller voltage for maximum power output over the mechanical tuning range of the cavity. Since the same repeller mode was used throughout this test, r must also have been the same at each of these frequencies, namely equal to {N -f %)// seconds. The experimentally determined plot of repeller voltage vs. frequency is given in Fig. 17. It is seen to be a straight line described by the equation, / = (13.5F« + 2495) mc (4.2) and since, for the 3 + %-repeller mode, r = yi^ 10-« sec, (4.3) J(mc) the desired relation between r and Vr is obtained from the above two equation.- ;is T = (3.r)F« -f 665)"' sec, (4.4) A COUPLED RESONATOR REFLEX KLYSTRON '53 whence the numerator of equation 4.1 follows as dV, 3.6 (3.6F« + 665)^ 10 ' sec/volt. (4.5) It remained only to replace the 6BL6 in the fixed tuned primary cavity, in which all subsequent tests were performed, and to determine the midmode repeller voltage and frequency. These values were found to be 96 volts and 3800 mc respectively, thus yielding by substitution into equation (4.5), dVn 3.6 X 10" (3.6 X 96 + 665)2 = 3.52 X 10"'' sec/volt. The demoninator of equation (4.1) simply denotes the "modulation sensitivity" at the center of the mode. A simple measurement estab- 4250 4200 4150 Q Z 8 4100 / A / / / /- tn ^ 4000 >- < 3950 m 2 3900 > ^ 3850 III S 3800 a. u. 3750 3700 3650 / / / fV / / / / / / / 85 90 95 100 105 120 125 130 Fig. 17 — Experimental determination of the relation between frequency and repeller voltage. Points shown were obtained with a 6BL6 in tunable primary cavity of Fig. 14 with secondary cavity completely decoupled and operation in the Z -\- % repeller mode at a beam voltage, Fo = 325 F. Frequency was varied by means of tuning plungers and repeller voltage recorded for maximum power out- put. Relation is given by,/ = (13.5Fr + 2495) mc. 754 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 lished its value as, = 1.12 mc/volt. df dV, Substitution into equation (4.2) then gave the value of primary Q as, Q = .(3.8)^10^^ ^|Sy£- " "^^-^^^ = ^^^' The entire Q-determination, incidentally, as outlined above, was re- peated for the 2 + % mode and resulted in Q = 128.8 thus affording an excellent and independent double check. 4.3.2 Calibration of Secondary Resonator. To facilitate the establish- ment of controlled and reproducible conditions of secondary Q and coupling coefficient, two cahbration curves had to be obtained. One, relating the values of secondary Q with the readings of the micrometer controlling the depth of insertion of the resistance vane and the other, relating the coefficient of coupling, /c, with the coupling iris width. The latter could be varied by means of gold plated spring shutters as shown in Figs. 14 and 15. This calibration of the secondary cavity was carried out in three distinct phases. The first phase involved the determination of the varia- tion of Q, with micrometer setting for values of Qa ranging from 500 to 2000. The second phase, which was based on the results obtained in phase one, yielded the complete calibration of the coupling iris and the third and last phase, in turn dependent on results of phase two, yielded values of Q, down to 65. The reasons for this particular sequence of measurements will become apparent in the following more detailed description. By means of a Q-measurement technique^ based on an oscillographic display of reflected power, points on the calibration curve were obtained as indicated by the circles in Fig. 18. It is seen that the lowest value of Q, which could be determined by this method was 550. For lower values of 0, the sweep range of the signal generator became insufficient to display the required fraction of the resonance curve; in addition, the cavity proved excessively undercoupled to permit reliable measurements of bandwidths. Using the values of Q, thus determined and a particular property of coupled resonators covered earlier in this text and further elaborated below, the relation between coupling coefficient and iris width was established. It was shown earlier that an overcoupled secondary cavity gives rise to load hysteresis manifesting itself in mode shapes as sketched A COUPLED RESONATOR REFLEX KLYSTRON 755 2000 1500 1000 (t 800 O !5 Z 600 [2 500 a. > 400 a. < a Z 300 o u m . 0.04 0.06 0.08 0.10 0.12 0.14 0.16 0.18 0.20 0.22 0.24 MICROMETER READING Fig. 18 — Calibration of secondary cavity. Variation of secondary Q with reading of micrometer controlling depth of insertion of resistance vane. in Fig. 10 and shown in the form of an oscilloscope pattern in Fig. 16(d). Conditions pertaining at the crossover point were described by equation (3.13), which stated: or, since pQ = Qs v\Qkf = 1 + v\2Q^)\ (Qskf = 1 + (2^35)^ Hence the two frequencies corresponding to the crossover point are given by and their difference, A5, by Equation (4.7)*' ^ provides an experimental method of determining k since A8 can be accurately measured and Qg is known from previous tests. (4.6) (4.7) 756 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 The actual procedure adopted in calibrating the coupling iris was this: The secondary cavity was set to a fairly high and known value of Q^ , say 1500, and the coupling iris adjusted to the smallest width which still resulted in overcoupling, i.e. in Q^k > 1, and thus displayed the de- sired load hysteresis on the oscilloscope screen. By means of a high-Q absorption wavemeter the crossover frequencies were determined and, along with the known value of Q, substituted into equation (4.7) which then yielded the value of A;. As a double check the above procedure was repeated for several values of Qs and in accordance with theory the coupling coefficient, k, found to be a function only of the iris coupling width. A typical set of readings is given in Table II. They show that though Qt was varied by a ratio of nearly 3:1, the resulting values of k only differed by about one-half per cent. Measurements as shown in the table were repeated for various iris widths. They resulted in the curve of Fig. 19. To complete the calibration of the secondary cavity it was necessary to extend the relation between Qs and micrometer readings to values lower than could be obtained by the reflected power technique. This was carried out with the aid of Fig. 19 in the following manner: With the coupling iris adjusted to its maximum width of 0.5", the secondary Q was varied until the condition of critical coupling, as ob- served by the formation of a cusp on the oscilloscope pattern (see Fig. 16(c)), was reached. Since this condition is characterized by Qsk = 1 and since the value of k corresponding to the iris opening could be read off the curve of Fig. 19, the value of Qa then followed simply as the reciprocal of k. This measurement was repeated for successively smaller iris openings and the points shown as triangles in Fig. 18 were obtained. Inspection of this figure also reveals that these values of Qa , ranging from 65 to 1100, overlap and coincide with values of Q, determined Table II Coupling Iris Width Secondary Cavity Q, Q, Crossover Frequencies /land/, (mc) A/ = (/1-/0 (mc) A5 = 2A/ /. + /« From Equation (inches) V(A6)» 4- {i/QsY 0.600 0.500 0.500 1430 976 673 3749.9 3694.4 3748.4 3693.1 3746.2 3690.4 65.6 55.3 54.8 14.92 X 10-» 14.90 X 10-» 14.74 X 10-» 14.93 X 10-3 14.90 X 10-3 14.85 X 10-3 A COUPLED RESONATOR REFLEX KLYSTRON 757 16 14 12 10 1*) 2 8 X 6 4 2 0 y U y^ / A A Y / Y y V _^^^^ M 0.10 0.20 0.30 0.40 WIDTH OF COUPLING IRIS IN INCHES 0.50 Fig. 19 — Calibration of coupling iris. Variation of coupling coefficient with width of coupling iris. independently by the reflected power method. The significance of this experimental agreement is twofold. It serves to prove the correctness of the calibration curves of Figs. 18 and 19 and beyond that may be regarded as verifying much of the theory of the coupled resonator re- flex klystron. 4.3.3 Comparison of Experimental and Theoretical Mode Shapes. Knowing the primary (operating) Q and the calibration of the sec- ondary resonator, controlled operating conditions were established and the experimental mode shapes shown in the left hand column of Fig. 20 obtained. The first three families of curves in this column are the results of point by point measurements in which repeller voltage, taken as the independent variable, was varied by known amounts about its midmode value and the corresponding values of RF power output and frequency were determined by a thermistor-bridge-wattmeter and high-Q wave- meter respectively. The last family of curves in this group, (d), is derived from (c) ; in it power has been normalized with respect to its maximum value within the mode thus permitting the convenient determination of midmode percentage reduction of power and electronic tuning range to any desired power level for various values of (Qk) . Additional details of the test conditions which gave rise to the ex- Vo= 325V, l^^=26MA, Vr (MlDWODE)=-96V, fo= 3800 MC, 3% REPELLER MODE, PRIMARY Q= 130, SECONDARY Q = 65, Qk VARIED BY CHANGES IN COUPLING IRIS WIDTH OBTAINED GRAPHICALLY FROM COMPLEX ADMITTANCE PLANE REPRESENTATION. NORMALIZED FREQUENCE Q^, CONVERTED TO MC BY ASSUMING Q = 130 AND fo = 3800 MC (bb) .o;^ ^ ^ .y^ >> ^y'^\^ ^ 3 •5 0 5 A|Vr| IN VOLTS 15 60 20 0 -20 L6 IN DEGREES -40 1.0 2 lU 0.6 > (C) / \ J /^ ^*1 0 / 1.6 \ \ s aOI 1 J^ — ' \^ a2 w\ ^ iOOMC \ (cc) / A / /^ — I / \ (dd) ff. \ ' \ \ \A^ o\ \\ i 11/ % / 60-60 20 40 FREQUENCY, Af, IN MEGACYCLES PER SECOND Fig. 20 — Quantitative verification of coupled resonator reflex klystron theory. •Curves (a) and (b) were obtained experimentally by changing the repeller voltage about its midmode value and noting the corresponding values of power and frequency. Curves (c) and (d) were deduced from (a) and (b). The theoretical curves were obtained graphically from a complex admittance plane plot similar to those of Figs. 9 and 13. 768 A COUPLED RESONATOR REFLEX KLYSTRON 759 perimental curves shown are as follows: 6BL6 repeller mode = 3 -{- ^^ Resonator voltage, Vq Cathode current, h Midmode repeller voltage, Vr Midmode frequency, /o Secondary Q, Qs ^4 cycles = 325 volts = 28 ma = -96 volts = 3800 mc = 3^ prunary Q = 65, obtained by mi- crometer reading of 0.234 (see Fig. The parameter, (Qkf, was varied by adjusting the iris width, and hence k, according to Table III. A few words, next, about the method by which the theoretical curves of Fig. 20 were derived. Input admittance plots for two coupled resona- tors having Qs = J^Q were given earlier in this report (see Fig. 11). Table III (Qk)^ k (for Q = 130) Iris width from Fig. 19 in inches 0.0 0.8 1.6 2.4 0.00 6.88 X 10-3 9.73 X 10-3 11.90 X 10-3 0.000 0.285 0.350 0.405 These included graphs for the values of (Qkf listed in Table III and thus of direct use in a graphical admittance plane analysis such as described in Section 3.2. To establish correlation, however, between theory and experiment, it was necessary to determine the location of the load lines in relation to the small-signal electronic admittance plot for the 3 + % repeller mode. This was performed graphically by trial and error as follows : a number of electronic admittance loci were drawn and their interactions with the load line for (Qkf = 2.4 studied. In particu- lar, the ratio of minimum to maximum power was calculated for each trial until one was found agreeing well with the experimental value of 0.845 (see Fig. 20d). This latter trial, then, was taken as representing the proper relation between electronic and circuit admittance. Mode plots for other values of (Qkf were determined without further reference to the experimental results and allowed to fall where they may. In comparing the theoretical and experimental curves, then, the fol- lowing points should be kept in mind: the frequency scales of the the- oretical curves are normally expressed in terms of Qd. This was converted 760 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 to megacycles in Fig. 20 by using the relation, A/ = (QS) -^ mc, where Q was found to equal 130 by the independent measurement de- scribed in section 4.3.1 and /o taken as 3800 mc. The vertical power scales of the theoretical curves are not entirely independent of the experimental plots in as much as one of the latter, namely the plot for (Qkf = 2.4, was used in determining the location of the load lines within the spiral diagram. The curves presented in Fig. 20 are largely self-explaining. They in- dicate good agreement between theory and experiment. What disagree- ment there is, may be traced primarily to the assumptions involved in the small signal klystron theory not being fully met in practice. Thus, the slight discrepancy between Figs. 20(a) and (aa) may be due to the drift angle not being linearly related to repeller voltage or possibly due to the phase angle of the electronic admittance being affected by the magnitude of the gap voltage.^^ These factors, however, are eliminated in the mode plots (c) and (d), which for this reason exhibit better mode symmetry and very close agreement with theory. 4.4 Performance Data The experimental curves of Fig. 20 were obtained with particular values of (Qkf and Qs/Q for which the input admittance of two coupled resonators had been computed and plotted earlier. As pointed out before, these values were chosen merely because they facilitated comparison between theory and experiment; they were not to be regarded, however, as representing optimum conditions. Whereas these curves had been obtained for a fixed secondary Q (namely equal to half the primary one), with changes in coupling-iris width producing the desired variations in (Qk) , the demonstration of optimum performance which follows was pursued along different lines. Here, the coupUng iris was opened to its maximum width (0.500") and the secondary Q adjusted until the con- ditions shown by the oscillograms of Fig. 21 were obtained.* These oscil- lograms indicate a number of mode shapes useful in applications re- quiring power output to be essentially independent of frequency. It is * In adjusting the circuit parameters for a flat topped mode it should be borne in mind that the variation in power with frequency is a function of QJc whereas the actual bandwidth varies inversely with Q, . For a maximum flat hand, there- fore, Q, should be chosen as small as possible consistent with a value of (Qnk)^ of about 0..35. In practice the lowest value of Q, which can be used will be deter- mined by the highest value of k obtainable with a given coupling iris. A COUPLED RESONATOR REFLEX KLYSTRON 761 seen that the useable bandwidth depends on the degree of flatness. Thus, if the appUcation requires power to be absolutely constant, the useable bandwidth equals 39 mc; it increases to about 70 mc for a fluctuation in power of d=0.2 db. Fig. 21 also Hsts the values of half power electronic tuning for the oscillograms shown. They are seen to range between 107 and 113 mc. By way of comparison, the mode shape for the case of a completely decoupled secondary resonator was shown in Fig. 16(a). Its peak power was found to equal about 100 mw, or about twice the power of the flat topped modes of Fig. 21, and its half power electronic range equaled 50 SINUSOIDAL REPELLER SWEEP BEYOND EXTINCTION SWEEP REDUCED AND HOR- IZONTAL GAIN INCREASED TO ENLARGE FLAT POWER BAND f, = 3935 MC f^ = 3974 MC Af = 39 MC POWER CONSTANT OVER ABOVE BAND Af, V2- 107 MC f; — 3928 MC fa = 3978 MC Af = 50 MC POWER FLAT WITHIN ±0.05 db (APPROX) Af|/2= 110 MC f, = 3922 MC fg = 3981.4 MC Af = 59.4 MC POWER FLAT WITHIN ±0.1 db (APPROX) 2 MC Af,/,= f^ = 3917 MC fz = 3986.6 MC Af = 69.6 MC POWER FLAT WITHIN ±0.2 db (APPROX) Aft/2 = ''3-3 MC Vr— *► Fig. 21 — Oscillograms showing flat power bands obtainable with coupled resonator reflex klystron. 6BL6 operating in 3 + ^ mode, Vo = 325 F, /* = 28 ma, zero DB power level corresponds to 50 milliwatts. 762 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 mc. The price, then, we must pay for a better than two fold increase in electronic tuning and for a frequency range of constant power is a 3 db loss in the available power output. Using the tunable primary cavity shown in Fig. 15, the mode per- formance just described could be obtained about any center frequency between 3700 and 4200 mc. 5.0 APPLICATIONS OF THE COUPLED-RESONATOR REFLEX KLYSTRON A numer of applications making use of the variety of mode shapes which may be obtained with the coupled resonator reflex klystron were indicated in the introductory section of this paper. Some of these ap- plications were tried experimentally and are described below, others, USE OF COUPLED-RESONATOR REFLEX KLYSTRON IN GAIN COMPARATOR TO DISPLAY BANDPASS CHARACTERISTIC OF MICROWAVE CHANNEL FILTER. TOP TRACE REPRESENTS INCIDENT POWER, LOWER TRACE TRANS- MITTED POWER. CENTER FREQUENCY OF FILTER = 3810MC PASS BAND = 28 MC USE OF COUPLED -RESONATOR REFLEX OSCILLOSCOPE DISPLAY FOR SAME KLYSTRON AS SWEPT FREQUENCY SOURCE RESONATOR RESULTING FROM USE OF IN DETERMINATION OF Q OF UNKNOWN CONVENTIONAL KLYSTRON AS SWEPT RESONATOR BY REFLECTED POWER METHOD. FREQUENCY SOURCE TOP TRACE REPRESENTS POWER INCIDENT UPON CAVITY UNDER TEST, MIDDLE TRACE REFLECTED POWER AND LOWER TRACE ZERO POWER LEVEL. (NOTE FLATNESS OF INCIDENT POWER TRACE) Fig. 22 — Examples of application of coupled resonator reflex klystron as microwave sweeper. A COUPLED RESONATOR REFLEX KLYSTRON 763 and in particular those making use of the improved modulation linearity resulting from the use of two resonators, could not be tried since the required test apparatus^ was unavailable. The close correlation between theory and experiment, however, as demonstrated in Section 4.3.3 leaves little doubt as to the feasibility of such applications. Two examples of the use of the coupled resonator reflex klystron as a microwave sweeper are given in Fig. 22. The first example shows the band pass characteristic of a channel filter, 28 mc wide. The second example demonstrates the use of this tube in a Q-measurement scheme^ SUCCESSIVE PHOTOGRAPHS OF OSCILLOSCOPE DISPLAY SHOWING IF-POWER AS A FUNCTION OF FREQUENCY WITH SIGNAL GENERATOR TUNED TO DIFFERENT FREQUENCIES GAIN-FREQUENCY CHARACTERISTIC OF IF AMPLIFIER USING COUPLED- RESONATOR REFLEX KLYSTRON AS SWEEP SOURCE CENTER FREQUENCY OF IF-AMPL=70MC FLAT BANDWIDTH OF IF-AMPL = 22MC GAIN-FREQUENCY CHARACTERISTIC OF SAME IF AMPLIFIER AS OBTAINED BY CONVENTIONAL (SINGLE RESONATOR) REFLEX KLYSTRON. DISTORTION IN GAIN ENVELOPE DUE TO MODE SHAPE OF KLYSTRON, Fig. 23 — Example of application of coupled resonator reflex klystron as inter- mediate frequency sweeper. i 764 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 based on the oscillographic display of reflected power. The advantage of a flat-topped mode shape for this measurement may be seen by com- parison ^vith the oscillogram of Fig. 22(c) which was obtained for the same resonator under test using a conventional reflex klystron as swept frequency source. The center frequency of the flat power band obtainable with the coupled resonator reflex klystron may be shifted into the intermediate frequency range by mixing its output with a suitable local oscillator. This is shown in Fig. 23 where the first three oscillograms represent a power band flat to within =b0.1 db and 60 mc wide centered around an IF frequency of 70 mc. The frequency markers are obtained from a signal generator tuned to the frequencies indicated and coupled lightly to the output of the mixer. Fig. 23(b) shows the gain-frequency char- acteristic of a 70 mc-IF strip having a useful bandwidth of 22 mc. The last oscillogram shows the gain-frequency characteristic of the same IF amplifier with the coupled resonator reflex klystron replaced by one of conventional design. Due to the inadequate electronic tuning range of this klystron and due to the asymmetry of its mode shape, the gain envelope of the IF amplifier appears distorted. 6.0 CONCLUSIONS The theory and reduction to practice of a coupled resonator reflex klystron have been presented. This device differs from the conventional reflex klystron in that the electronic admittance interacts with the input admittance of two coupled resonators. Significant and advantageous changes in performance resulting therefrom are: (1) Power output can be made to be substantially flat over part of the electronic tuning range. The frequency range of flat power is greater than the half power electronic tuning range of a klystron having the same electron-optical system but interacting with a single resonator. (2) The half -power electronic tuning range of the coupled resonator klystron is more than twice that of the equivalent single resonator klystron. (3) Modulation linearity may be obtained over a greatly increased frequency swing. A reduction in power output of about 3 db occurs for a secondary resonator Q and coupling coefficient adjustment designed to yield a maximum flat band or maximum electronic tuning, a much smaller reduction in output power, perhaps 10 per cent, will provide a substantial improvement in modulation linearity. A COUPLED RESONATOR REFLEX KLYSTRON 765 Qualitative and quantitative experimental verifications of the theory were undertaken and good agreement obtained. Oscillograms of mode shapes having a range of constant power output and wide half-power electronic tuning were presented together with a number of applications of the coupled resonator reflex klystron both as a microwave and as an intermediate frequency sweep source. Other applications were indicated and additional ones may suggest themselves to the reader. It is believed that the addition of the second resonator makes the reflex klystron a more useful device in many of its more important applications. ACKNOWLEDGEMENTS The writer wishes to express his appreciation to J. 0. McNally for the advice and encouragement received in the execution of this work. The assistance of the thesis advisor, Prof. John B. Russell of the Elec- trical Engineering Department of Columbia University, is also grate- fully acknowledged. REFERENCES 1. Pierce, J. R., and W. G. Shepherd, Reflex Oscillators. Bell System Tech. J., 26, pp. 460-690, July, 1947. This reference constitutes a very complete treat- ment of reflex klystron theory and practice. The derivation of the expression for electronic admittance is given in Appendix III, pages 639-643; of particu- lar interest to the coupled resonator reflex klystron is Pierce and Shepherd's treatment of Frequency Sensitive Loads and Long Lines Effects, pp. 523, where it is shown that certain load conditions encountered in the operation of reflex klystrons may result in load hysteresis similar to the one described in the preceding paper as due to an overcoupled secondary resonator. 2. Martin, R. A., and R. D. Teasdale, Input Admittance Characteristics of a Tuned Coupled Circuit. I. R. E., Proc, p. 57, January, 1952, and correction to this paper published in I. R. E., Proc, p. 459, April, 1952. This paper presents a precise expression for the input admittance of two coupled resonant circuits and gives plots of specific numerical examples. The expres- sion for input admittance given in the above named correction reduces to equation (3.2) and (3.12). 3. Bleaney, B., Electronic Tuning of Reflection Klystrons. Wireless Engineer, p. 6, Jan., 1948. A valuable background paper on the mechanics and maximi- zation of electronic tuning. 4. Very High-Frequency Techniques, Vol. II, Radio Research Laboratory, Harvard University. (McGraw-Hill). Chapter 31, p. 849 contains a concise and valuable treatment of reflex klystron theory. It leads up to the appli- cation of external-cavity reflex klystrons in wide tuning range coaxial resonators. In the course of explaining some load hysteresis phenomena which are sometimes caused in these circuits by parasitic resonances or excitation of undesired modes, a simplified expression for the input ad- mittance of two coupled resonators is given. This expression is similar to equation (3.12). An experimental method is also suggested (p. 869) to meas- ure the coupling coefficient between the main resonator and the undesired resonance. Use has been made of this method in the calibration of the secondary resonator in Section 4.3.2. 766 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 5. Spangenberg, K. R., Vacuum Tubes, p. 601. (McGraw-Hill). This reference covers essentially the same material as Reference 4. 6. Reference Data for Radio Engineers, Third Edition, p. 121, Federal Telephone and Telegraph Corp. Here a universal chart of the selectivity (transfer characteristic) of two coupled tuned circuits is presented. The format and choice of variables of this presentation has inspired the selection of the equivalent quantities applying to the driving point properties of coupled tuned circuits. 7. Harmen, W. W., and J. H. Tillotson, Beam-Loading Effects in Small Klys- trons. I. R. E., Proc, p. 1419, Dec, 1949. Equation (3) of this reference is identical with equation (4.1) and has been used in the experimental de- termination of the primary operating Q. 8. Reed, E. D., A Sweep Frequency Method of Q Measurement for Single-Ended Resonators. Proc. National Electronics Conference, Vol. VII, p. 162. Also reprinted as Bell Telephone System Monograph 1953. This method of Q- measurement was used in the calibration of the secondary resonator de- scribed in Section 4.3.2. The coupled resonator reflex klystron provides a useful tool for this type of measurement as may be seen from Fig. 22(b). 9. Albersheim, W. J., Measurement Techniques for Broad-Band Long-Distance Radio Relay Systems. I.R.E., Proc, p. 548, May, 1952. Figs. 4(a) and 4(b) of this reference give the block diagram of a "Linearity-Test-Set" developed at Bell Telephone Laboratories to determine the modulation linearity of a reflex klystron used as FM-deviator. 10. Boehlke, P. G., and F. C. Breedan, External Cavity Klystron. Electronics, p. 114, July, 1947. This paper gives an account of the design considerations which led to the development of the 6BL6 reflex klystron. It also contains useful information on its performance characteristics. 11. Garrison, J. B., A Qualitative Analysis of Hysteresis in Reflex Oscillators. Radiation Laboratory Report No. 650, dated Feb. 4, 1946. In this treatment on the causes of electronic hysteresis Garrison shows (on page 5) that a dependence of the phase angle of the electronic admittance upon the magni- tude of RF gap voltage will give rise to asymmetry in the mode plot of power output vs. repeller voltage although the power vs. frequency rela- tion will be symmetrical. (Note that the experimental mode plots for QA; = 0 of Figs. 20(a) and (c) show evidence of the same phenomenon.) Abstracts of Bell System Technical Papers* Not Published in This Journal Anderson, P. W/ Concept of Spin-Lattice Relaxation in Ferromagnetic Materials, Letter to the Editor, Phys. Rev., 88, pp. 1214, Dec. 1, 1952. Balashek, S., see K. H. Davis. Band, W., see R. A. Nelson. Beach, A. L., see J. J. Lander. Bell Telephone Laboratories Transistor Teachers Summer School. Experimental Verification of the Relationship Between Diffusion Con- stant and Mobility of Electrons and Holes, Phys. Rev., 88, pp. 1368- 1369, Dec. 15, 1952. The relationship between diffusion constant and mobility, called the Einstein relationship has been experimentally verified for electrons and holes in ger- manium. This has been accomplished by measuring the rate of increase in half concentration width of a pulse of minority carriers moving in an electric field. BiDDULPH, R., see K. H. Davis. Bridgman, D. C.^ College Graduates and the Country's Telephone Industry, Jl. College Placement, 13, pp. 19-27, Oct., 1952. Bullington, K.^ Radio Transmission Beyond the Horizon in the 40- to 4,000-Mc Band, I.R.E., Proc, 41, pp. 132-135, Jan., 1953. (Monograph 2060). ReHable signals have been received at distances of several hundred miles at frequencies of 500 and 3,700 mc. The median signal levels are 50 to 90 * Certain of these papers are available as Bell System Monographs and may be obtained on request to the Publication Department, Bell Telephone Laboratories, Inc., 463 West Street, New York 14, N. Y. For papers available in this form, the monograph number is given in parentheses following the date of publication, and this number should be given in all requests. ^ Bell Telephone Laboratories, Inc. 2 American Telephone and Telegraph Company. 767 768 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 db below the free-space field, but are hundreds of db (in one case 700 db) stronger than the value predicted by the classical theory based on a smooth spherical earth with a standard atmosphere. Antenna gains and beam widths are maintained to a first approximation and no long delayed echoes have been found. Clark, A. B/ Development of Telephony in the United States, A.I.E.E., Trans., Commun. and Electronics Sect., 71, pp. 348-364, Nov., 1952 (Mono- graph 2045). The telephone was invented twenty-four years after the founding of the American Society of Civil Engineers and Architects, which society com- memorated its Centennial in September, 1952. The telephone was eight years old when the American Institute of Electrical Engineers was founded. Since Bell's invention, the telephone business has grown tremendously and this growth has been greatly dependent on developments in science and engineer- ing. This paper traces, and endeavors to give the significance of, the major developments. Brief mention is made of some developments still in the making and of some ideas as to the future potentialities of the business. CONWELL, E. M.^ Mobility in High Electric Fields, Phys. Rev., 88, pp. 1379-1380, Dec. 15, 1952. An extension of conductivity theory to high fields, subject to the usual sim- plifying assumptions, is carried out for the cases in which the change of energy of an electron in a collision can be neglected. This yields a relationship between mobility and relaxation time which is valid over a wide range of fields. Davis, K. H.,^ R. Biddulph^ and S. Balashek^ Automatic Recognition of Spoken Digits, J. Acoust. Soc. Am., 24, pp. 637-642, Nov., 1952. The recognizer discussed will automatically recognize telephone-quality digits spoken at normal speech rates by a single individual, with an accuracy varying l^etween 97 and 99 per cent. After some preliminary analysis of the speech of any individual, the circuit can be adjusted to deUver a similar ac- curacy on the speech of that individual. The circuit is not, however, in its present configuration, capable of performing equally well on the speech of a series of talkers without recourse to such adjustment. Circuitry involves division of the speech spectrum into two frequency bands, one below and the other alK)ve 900 cps. Axis-crossing counts are then individually made of both band energies to determine the frequency of the maximum syllabic rate energy within each band. Simultaneous two-dimensional frequency portrayal is found to possess recognition significance. Standards are then determined, one for each digit of the ten-digit series, and are built into the recognizer as a form of elemental memory. By means of a series of calculations performed ' Bell Telephone Laboratories, Inc. ABSTRACTS OF TECHNICAL ARTICLES 769 on the spoken input digit, a best match type comparison is made with each of the ten standard digit patterns and the digit of best match selected. DicKTEN, E., see R. L. Wallace, Jr. Felker, J. H.^ T3rpical Block Diagrams for a Transistor Digital Computer, Elec. Eng., 71, pp. 1103-1108, Dec, 1952 and A.I.E.E. Trans., 71, pp. 175-182, 1952 (Monograph 2046). The superior speed capabilities of vacuum tubes have led to their use in computer designs to replace relays. Because of their small size, low power consumption, and long hfe expectancy, it now appears that transistors will replace tubes as computer elements. Here is a study of binary computer functions in which transistors are employed. FrAYNE, J. G.,^ AND J. P. LlVADARY^ Dual Photomagnetic Intermediate Studio Recording, S.M.P.T.E., Jl., 59, pp. 388-397, Nov., 1952. Selected production magnetic tracks are transferred to a recorder which lays down colHnear 200-mil push-pull direct-positive variable-area and magnetic tracks. Magnetic stripe is on base of photosensitive emulsion on the opposite edge of film from photo track. The photo track may be used for reviewing, cutting, etc. Re-recording is done from assembled magnetic tracks. This method combines advantages of photo track for editing and provides superior quality of magnetic track. Hagstrum, H. D.^ Electron Ejection from Mo by He"^, He''"^, and He2"*", Phys. Rev., 89, pp. 244-255, Jan. 1, 1953. Total yield and kinetic energy distribution have been measured for electrons ejected from atomically clean and gas covered molybdenum by the ions He+, He++, and He2+, in the kinetic energy range 10 to 1000 ev. Evidence is presented that one electron is excited into the kinetic energy continuum for each incident. He+ ion and that the electrons so excited are partially internally reflected at the potential barrier of the metal. The slowest ions observed were found to eject 0.25, 0.72, and 0.13 electron per ion for He+, He++, and He2"*', respectively. Total electron yield is found to be nearly independent of ion kinetic energy up to 1000 ev. This observation and that of the kinetic energy maximum for slow ions indicate that the electrons are released in an Auger type process for which the energy is supplied by the potential and not the kinetic energy of the ion (potential ejection). Electrons of kinetic energy greater than the upper limit predicted by present theory are observed for faster ions and are accounted for by the shift of the energy levels of the bom- barding particle when it is near the metal surface. Some conclusions con- ^ Bell Telephone Laboratories, Inc. ^ Westrex Corporation. '' Columbia Pictures Corporation. 770 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 ceming reflection processes at a metal surface and the nature of electron ejection by the alpha-particle (He++) and the molecular hehum ion (He2+) come out of this work. Heidenreich, R. D. Methods in Electron Microscopy of Solids, Rev. Sci. Instr., 23, pp. 583-594, Nov., 1952 (Monograph 2047). Methods of replicating solid surfaces for electron microscopy are reviewed and compared. Preparation of metal surfaces for electron microscopy is dis- cussed, and the advantages of employing electron diffraction techniques in evaluating prepared surfaces are pointed out. Examples of the appUcation of replicas include steel, precipitation in alloys, such as Alnico 5, and studies of slip in aluminum. Growth spirals on the surfaces of crystals of w-paraffins are demonstrated. The use of thin metal sections and of emission electron microscopy in studying metallic structures is discussed, and examples are given. Hutchinson, A. R.^ How to Conceal Telephone Wires, Keep Desks Neat, Standardiza- tion, 23, p. 407, Dec, 1952. Jakes, W. C, Jr.^ Theoretical Study of an Antenna-Reflector Problem, I.R.E., Proc, 41, pp. 272-274, Feb., 1953. This paper gives the results of a theoretical investigation of an antenna used with a plane reflector. This finds application in microwave relay stations, where the antenna is placed at ground level facing up and the reflector is located some distance above it. The results given show that there are certain values of X, separation distance, reflector and antenna size for which the received power is greater than for the same antenna alone at the elevated location. Kaplan, E. L.^ Tensor Notation and the Sampling Cumulants of k-Statistics, Bio- metrika, 39, pp. 319-323, Dec, 1952. Now and then in the Uterature one finds results relating to multivariate distributions which are derived virtually independently of, or with con- siderable effort from, the corresponding univariate relations, whereas they are in fact only very mild generaUzations of the latter, as will be shown. Only the famiUar concepts of moments, characteristic functions, cumulants, and A^statistics and their sampling cumulants will be discussed here. It should be emphasized that these concepts are identical with those ordinarily used in multivariate situations; the only novelty lies in the concise manner of repre- senting and handling them. ' Bell Telephone Laboratories, Inc. ' Western Electric Company, Inc. ABSTRACTS OF TECHNICAL ARTICLES 771 Keller, A. C/ Economics of High-Speed Photography, S.M.P.T.E., Jl., 59, pp. 365-368, Nov., 1952 (Monograph 2052). The economics of the use of high-speed photography in research and develop- ment work are discussed. High-speed photography is a relatively new tool for engineers which can be used to measure mechanical or electrical effects or both at the same time. Examples are given which illustrate the savings in engineering manpower as well as ii materials, devices and systems. Kern, H. E., see J. J. Lander. KocK, W. E.,^ AND R. L. Miller.^ Dynamic Spectrograms of Speech, Letter to the Editor, J. Acoust. Soc. Am., 24, pp. 783-784, Nov., 1952. KocK, W. E.' Problem of Selective Voice Control, J. Acoust. Soc. Am., 24, pp. 625-628, Nov., 1952 (Monograph 2048). The development of devices which can be operated automatically from the phonetic content of speech may be viewed in terms of the more general prob- lem of the reduction of channel capacity in communications systems. Signifi- , cance has been observed in formant positions and movements as regard the identification of speech sound. The basic problems in the derivation of pho- nemes from the formant patterns are reviewed. Lander, J. J.,^ H. E. Kern^ and A. L. Beach^ Solubility and Diffusion Coefficient of Carbon in Nickel: Reaction Rates of Nickel-Carbon Alloys with Barium Oxide, J. Appl. Phys., 23, pp. 1305-1309, Dec, 1952. Experimental values for the solubility of carbon in nickel in the range 700°C to 1300°C yield the equation In ^ = 2.480 - 4,880/T, where S is the solu- bility in grams of carbon per 100 grams of nickel. Values obtained for the diffusion coefficient in the same range fit the equation In D = 0.909 — 20,200 /T, where D is in cm^ per second. These results are of some interest in the problem of the activation of thermionic oxide coated cathodes, and the experimental method used to measure the diffusion coefficients is related to phenomena occurring in vacuum tubes. To extend the usefulness of the results in this direction, rates of reaction between diffused carbon and barium oxide coatings on nickel have been measured. It was found that the rates are diffusion hmited over a wide range of conditions of interest. Lewis, H. W.' Multiple Meson Production in Nucleon-Nucleon Collisions. Revs. Modern Phys. 24, pp. 241-248, Oct. 1952 (Monograph 2049). LiVADARY, J. P., see J. G. Frayne ^ Bell Telephone Laboratories, Inc. 772 the bell system technical journal, may 1953 Luke, C. L.^ Photometric Determination of Silicon in Ferrous, Ferromagnetic, Nickel, and Copper Alloys — A Molybdenum Blue Method, Anal. Chem., 25, pp. 118-151, Jan., 1952. A simple, rapid photometric method for the determination of siHcon in fer- rous, ferromagnetic, nickel, and copper alloys has been developed. Wide applicability is its most unique and important feature. Interfering elements are removed by a carbamate-chloroform extraction and siUcon is then deter- mined by the photometric molybdenum blue method. Confirmatory data on a wide variety of samples of known silicon content are presented. LUMSDEN, G. Q.^ A Quarter Century of Evaluating Pole Preservatives. Amer. Wood Preservers' Assoc, Proc, 48, pp. 27-47, 1952 (Monograph 1999). Mac Williams, W. H., Jr.^ Computers — Past, Present, and Future, Elec. Eng., 72, pp. 116-121, Feb., 1953. This article deals with the historical development of computers. It also dis- cusses current problems and indicates future structural and functional com- puter trends which will help to free man from burdensome calculations and increase his material wealth while permitting him more time for pursuits not directly concerned with earning a living. Miller, R. L., see W. E. Koch. MUMFORD, W. W.^ Optimum Piston Position for Wide-Band Coaxial-to -Waveguide Transducers^ I.R.E., Proc, 41, pp. 256-261, Feb., 1953. A coaxial line can be matched to a waveguide by means of a probe antenna located ahead of a short-circuiting plunger. An impedance match can usually be achieved by varying any two of the following three dimensions: (a) the off-center position of the probe, (b) the probe length, (c) the piston position. This paper points out that there is, theoretically, an optimum piston position for greatest bandwidth, and presents some evidence corroborating this theory. Bandwidths greater than ±10 per cent to the 1 db swr points have been realized by fixing the piston at its optimum position and varying (a) and (b) above to obtain a match. Nelson, R. A.,* and W. Band* Vapor Pressure of He' = He* Mixtures, Letter to the Editor, Phys. Rev., eSr p. 1431, Dec. 15, 1952. * Bell Telephone Laboratories. Inc. ' American Telephone and Telegraph Company. * State College of Washington. abstracts of technical articles 773 Osborne, H. S.^ A Rose by Any Other Name, Report on Work of the Anglo-American Committee on Technical Terminology, Standardization, 24, pp. 19- 20, Jan., 1953. Peck, D. S.' Ten-Stage CoJd-Cathode Stepping Tube, Elec. Eng., 71, pp. 1136- 1139, Dec, 1952 (Monograph 2054). Developments in the art of transferring a gas discharge from one point to another in a multi-electrode tube have led to the design of a 10-stage counting tube operating up to about 2,000 pulses per second. Such a tube can be used for pulse counting, frequency division, time measurements, and similar functions. Peterson, G. E.^ Information-Bearing Elements of Speech, J. Acoust. Soc. Am., 24, pp. 629-637, Nov., 1952. This study deals with those aspects of speech which are phonetically signifi- cant. A technique has been developed with which phonetically equivalent speech samples may be obtained in different phonetic contexts and from different speakers. Data on two front vowels by different types of speakers are presented. The technique has also been appUed to the evaluation of words containing these two vowels. Pierce, J. R.^ New Method of Calculating Microwave Noise in Electron Streams, I.R.E., Proc, 40, pp. 1675-1680, Dec, 1952. The noise in a temperature-limited electron beam in a drift space is calculated by a new means. Noise maxima and minima are found. The results agree with calculations made by the Rack-Llewell^m-Peterson method. Rice, S. 0.' Statistical Fluctuations of Radio Field Strength Far Beyond the Horizon, I.R.E., Proc, 41, pp. 274-281, Feb., 1953. When a sinusoidal radio wave of extremely high frequency is sent out by a transmitter, the wave received far beyond the horizon is often observed to fluctuate. Here some of the statistical properties of this fluctuation are derived on the Booker-Gordon assumption; namely, that the received wave is the sum of many Uttle waves produced when the transmitter beam strikes "scat- terers" distributed in the troposphere. Expressions are obtained for the periods of the fluctuations in time, in space, and in frequency. These expres- sions extend closely related results obtained by Booker, Ratcliffe and others. ScHiMPF, L. G., see R. L. Wallace, Jr. 1 Bell Telephone Laboratories, Inc. 2 American Telephone and Telegraph Company. 774 the bell system technical journal, may 1953 Smith, K. D.* Properties of Junction Transistors, Tele-Tech, 12, pp. 76-78, Jan., 1953. TOWNSEND, J. R.^ What We Have Learned in 1952, Standardization, 24, pp. 16-18, Jan., 1953. ToWNSEND, J. R.^ What We Have Learned in 1952, A Report to the Joint Meeting of Standards Council and Board of Directors of ASA, A.S.T.M. Bull., No. 187, pp. 22-23, Jan., 1953. Van Roosbroeck, W.^ Large Current Amplifications in Filamentary Transistors, Letter to the Editor, J. Appl. Phys., 23, pp. 1411-1412, Dec, 1952. Wallace, R. L., Jr.,^ L. G. Schimpf^ and E. Dickten^ High-Frequency Transistor Tetrode, Electronics, 26, pp. 112-113, Jan., 1953. Sine-wave oscillators at frequencies up to 130 mc and tuned amplifiers with substantial gain at frequencies of 50 mc or higher are obtained by using junction transistors with an added connection to the base electrode biased negative at six volts. 1 Bell Telephone Laboratories, Inc. ' Sandia Corporation. Contributors to this Issue A. F. Bennett, Western Electric Company, 1914-25; Bell Telephone Laboratories, 1925-. Mr. Bennett, Director of Station Apparatus De- velopment since 1948, is in charge of the development, design, field trials, and studies of telephone instruments and sets, coin collectors, telephone booths, and station systems. During World War II, Mr. Bennett supervised the development and engineering work on a number of ordnance items. For this he was awarded a Presidential Certificate of Merit in 1946. He was a representative of the office of Scientific Re- search and Development and in that capacity served in the United Kingdom in 1943. He is a member of the A.I.E.E., the Acoustical Society of America, and the Physical Society. R. W. Burns, B.A., Indiana University, 1916; B.S. in E.E., Purdue University, 1918. U. S. Army, 1918; American Telephone and Telegraph Company, 1919-34; Bell Telephone Laboratories, 1934-. Mr. Burns has been engaged in the formulation of requirements for central office main- tenance equipment, principally that used for testing exchange lines and trunks. During World War II he was in charge of a group preparing instruction books for teletype communication equipment for the armed forces. Professional Engineer, New York State. Member of Sigma XI and Eta Kappa Nu. J. W. Dehn, E.E., Polytechnic Institute of Brooklyn, 1932. Western Electric Company, 1919-25; Bell Telephone Laboratories, 1925-. Switch- ing Systems Development Engineer, 1952. Mr. Dehn has been prin- cipally concerned with the design and development of manual and dial telephone switching systems since joining the Bell System. During World War II he developed conmiunication systems for the Signal Corps and trained military personnel in the operation and maintenance of this equipment. Since 1945 he has been engaged in No. 5 crossbar design. Professional Engineer, New York State. R. F. Mallina, M.E., Vienna Technical College, 1912. London In- stitute of City and Guilds, 1914. Wurlitzer Piano Company, Acoustical Engineer, 1925. RCA Victor Company, Head of Apparatus Develop- ment Department, 1929. Bell Telephone Laboratories, 1929-. At the 775 776 THE BELL SYSTEM TECHNICAL JOURNAL, MAY 1953 Laboratories Mr. Mallina worked initially in acoustical research where he developed the first magnetic telephone message recorder and the five-reed telephone set. From 1936 on he was engaged in fundamental development on machine switching apparatus, first on AMA, later with the wire spring relay project. In connection with the latter, he developed the solderless. wrapped connection. Mr. Mallina is also a research associate at New York University, Department of Education. W. P. Mason, B.S. in E.E., University of Kansas, 1921; M.A., Ph.D., Columbia, 1928. Bell Telephone Laboratories, 1921-. Dr. Mason has been engated principally in investigating the properties and applications of piezoelectric crystals, in the study of ultrasonics, and in mechanics. Fellow of the American Physical Society, Acoustical Society of America and Institute of Radio Engineers and member of Sigma Xi and Tau Beta Pi. J. W. McRae, B.S. in E.E., University of British Columbia, 1933; M.S., California Institute of Technology, 1934; Ph.D., California In- stitute of Technology, 1937. Bell Telephone Laboratories, 1937-42, 1945-. U. S. Army 1942-45, where he attained the rank of Colonel and served as Deputy Director of the Engineering Division of the Signal Corps Engineering Laboratories. Returning to Bell Telephone Labo- ratories in 1945, Dr. McRae became Director of Radio Projects and Television Research in 1946; Director of Electronic and Television Research, 1947; Assistant Director of Apparatus Development I, 1949; Director of Apparatus Development I, 1949; Director of Transmission Development, 1949; Vice President, 1951. Legion of Merit, 1945. Presi- dent of the Institute of Radio Engineers, 1953. Member of the A.I.E.E. and Sigma Xi. T. F. OsMER, E.E, Polytechnic Institute of Brooklyn, 1935. Bell Telephone Laboratories, 1920-. As a member of the Physical Research Department until 1952, Mr. Osmer was primarily concerned with trans- ducers, including transmitters, receivers, loudspeakers, and high quality microphones. During World War II he worked on military contracts, and since the war he has been occupied with carbon contact studies, and more recently with studies of the solderless wrapped connection, using photoelastic techniques. R. C. Prim, received a B.S.E.E. degree from the University of Texas in 1941, and M.A. and Ph.D. degrees from Princeton University in 1949'. Following graduation from college, he was employed by General Elec- tric Company in Schenectady until 1944, and then, as an ensign in the CONTRIBUTORS TO THIS ISSUE 777 Reserve, he joined the Naval Ordnance Laboratory at White Oak, Maryland. Here he conducted research on torpedo motion and control theory. He joined Bell Telephone Laboratories in 1949, and has con- ducted mathematical research on non-linear partial differential equa- tions and served as a consultant on military projects. Dr. Prim is a member of the American Mathematical Society, the American Physical Society, Sigma Xi, and Tau Beta Pi. E. D. Reed, B.Sc, University of London, 1941; M.S., Columbia University, 1947. Ardente, Ltd., 1941-43; U. S. Army, 1944-46; Bell Telephone Laboratories, 1947- Mr. Reed is engaged in the development and design of klystrons. Associate member of the Institute of Radio Engineers and member of Sigma Xi. Harry Suhl, B.Sc, University of Wales, 1943; Ph. D., Oriel College, University of Oxford, 1948. Admiralty Signal Establishment, 1943-46; Bell Telelphone Laboratories, 1948-. Dr. Suhl conducted research on the properties of germanium until 1950, when he became concerned with electron dynamics and solid state physics research. Member of the American Institute of Physics and the American Physical Society. R. H. Van Horn, B.S. in E.E., Pennsylvania State College, 1937; M.A., Columbia University, 1947. Bell Telephone Laboratories, 1937-. Mr. Van Horn is a member of the Switching Apparatus Development Department and is in charge of the machine switching apparatus labo- ratory. He has previously been engaged in the development of under- water sound devices and the vibrating reed selector for mobile radio applications. Member of A.I.E.E. and Acoustical Society of America. E. F. Vaage, E.E., Technical University of Darmstadt, 1926; M.E.E., Brooklyn Polytechnic Institute, 1932; Royal Norwegian Air Force, 1918-19; Elektrisk Bureau, 1926-27; A. T. & T. Co., 1927-34; Bell Telephone Laboratories, 1934-. Mr. Vaage is a member of a group en- gaged in systems studies, an outgrowth of his previous work of evaluating transmission systems. Member of American Mathematical Society and A.I.E.E. A. S. WiNDELER, B.S., Rutgers University, 1930; Bell Telephone Laboratories, 1930-. Mr. Windeler has been engaged in the design and development of toll cable, including coaxial, video pair, and microwave, types. He is currently in charge of a group concerned with the develop- ment of expanded polyethylene insulated conductors for multipair cable. HE BELL SYSTEM nicM lournai VOTED TO THE SC I E N TIFIC>w/ A N D ENGINEERING PECTS OF ELECTRICAL COMMUNICATION LUME XXXII JULY 1953 NUMBER 4 THE L3 COAXIAL SYSTipM Foreword ^ '\ ^^^ System Design i, . . "'" ' - C. H. ELMENDORF, R. D. EHBAR, R. H. KLIE AND aS.^IUMSi 781 Equalization and Regulation R. W. KETCHLEDGE AND T. R. FINCH 833 Amplifiers l. h. morris, g. ii. lovell and f. r. Dickinson 879 Television Terminals j. w. rieke and r. s. graham 915 Quality Control Requirements H. F. dodge, B. j. KINSBTJRG AND M. K. KRUGER 943 Application of Quality Control Requirements in the Manufacture of Components R. F. GARRETT, T. L. TUFFNELL AND R. A. WADDELL 969 Abstracts of Bell System Technical Papers Not Published in this Journal 1007 Contributors to this Issue IO15 COPYRIGHT 1953 AMERICAN TELEPHONE AND TELEGRAPH COMPANY THE BELL SYSTEM TECHNICAL JOURNAL ADVISORY BOARD 8. BRACKEN, President, Western Electric Company F. R. K A P P E L, Vice President, American Telephone and Telegraph Company M. J. K E I.L Y, President, Bell Telephone Laboratories EDITORIAL COMMITTEE E. I. GREEN, Chairman A. J. BUSCH F. R. LACK W. H. DOHERTY J. W. McRAE G. D. EDWARDS W. H. NUNN ,. B. FISK H. I. ROMNES [. K. HONAMAN EDITORIAL STAFF H. V. SCHMIDT J. D. T E B O, Editor M. E. 8 T R I E B Y, Managing Editor R. L. SHEPHERD, Production Editor THE BELL SYSTEM TECHNICAL JOURNAL is published six time* a year by the American Telephone and Telegraph Company, 195 Broadway, New York 7, N. Y. Cleo F. Craig, President; S. Whitney Landon, Secretary; Alexander L. Stott, Treasurer. Subscriptions are accepted at $3.00 per year. Single copies are 75 cents each. The foreign postage is 65 cents per year or 11 cents per copy. Printed in U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XXXII JULY 1953 numbbr4 Copyright, 1953, American Telephone and Telegraph Company The L3 Coaxial System Foreword The articles in this issue are devoted to different phases of the develop- ment of a new system for the transmission and utilization of broader fre- quency bands on existing or new coaxial cables. This new system, which is called the L3 carrier system, represents the latest phase of develop- ment activities begun in the late twenties. It permits far more intensive exploitation of the cable medium than its predecessor, affording the op- tion of providing, in each direction on a pair of coaxial tubes, either 1860 telephone channels or 600 telephone channels and a 4.2 megacycle broadcast television channel. These results have been attained through wide extension of previous art. New electron tubes, transformers, inductors, and other circuit elements have been designed for extreme precision in respect to stability and other performance factors. Statistical quality control techniques are being applied to obtain the benefits of closely controlled distribution of the performance of circuit elements and system units. Fundamental to the program has been the devising of techniques for achieving hitherto unobtainable accuracies in the measurement of impedance, loss, phase and other transmission properties. To provide precise attenuation and delay characteristics over the wide frequency band, new techniques of network synthesis have been developed. Refined system analysis and circuit design have derived maximum performance from component capabilities. The highest standards of 779 780 THE BELL SYSTEM TE'JHNICAL JOURNAL, JULY 1953 overall transmission performance and reliability have been adhered to. The new system is now in commercial serivce and large scale applica- tion is planned. The following articles discuss (1) the over-all systems, together with its fundamental design problems, (2) the methods developed for equali- zation and regulation, (3) the broadband amplifying techniques, (4) the circuits for transmitting and receiving television, (5) the requirements established for controlling the performance of component elements, and (6) the application of these requirements in manufacturing. E. I. Green The L3 Coaxial System System Design By C. H. ELMENDORF, R. D. EHRBAR, R.H. KLIE, and A. J. GROSSMAN (Manuscript received March 31, 1953) The L3 coaxial system is a new broadband facility for use with existing and new coaxial cables. It makes possible the transmission of 1,860 tele- phone channels or 600 telephone channels and a television channel in each direction on a pair of coaxial tubes. The principal system design problems and the methods used in their solution are described. The over-all system is described in terms of its components and their location in the system. 1.0 Introduction The L3 coaxial carrier system is a new broadband transmission system capable of transmitting either 1,860 telephone message channels or 600 message channels and a 4.2-megacycle broadcast television channel, in each direction, on a pair of coaxials. The system is designed so that signals transmitted over any of these channels will meet high quality Bell System objectives after 4,000 miles of transmission. The system is composed of auxiliary or line repeaters spaced at ap- proximately four-mile intervals along the cable route and connecting terminal or dropping repeaters where telephone or television signals are translated to or from the L3 frequency band. Equalization equip- ment, power generating and power transmission equipment, and main- tenance equipment are required at 100 to 200-mile intervals. Planning and exploratory development for the system was started late in 1945 with the objective of designing a trunk route system which would provide the maximum channel capacity on the existing coaxial cable consistent with the state of the repeater art. At that time and for the next four years a large amount of new cable employing the 600 channel-three megacycle LI coaxial carrier system was being installed or projected.^ Since a major field of use of the L3 system was to replace the LI 781 782 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 THE L3 SYSTEM DESIGN 783 system on existing routes, the design of the LI system, the cable, and the cable route layouts presented the L3 system with a definite plant framework. The present day network of LI coaxial systems is shown in Fig. L There are about 8,000 route miles of cable installed of which about 70 per cent consists of eight coaxials, the remainder consisting of six and four coaxials. About 70 per cent of this cable uses coaxials with a y^" diameter outer conductor, the present day standard. The remainder uses the older 0.27'' diameter coaxials. All but a few miles of this cable is plowed into the ground or placed in underground conduit. A piece of a typical eight coaxial cable is shown in Fig. 2. Normally, the coaxials are included in a lead sheath with interstitial pairs which are used for Fig. 2 — An 8-coaxial cable. control purposes. In many cases additional quads are included in the cable for other types of transmission systems. The broad objectives of the L3 system planning were: L The existing cable was to be reused. Thus, the cable loss and its variation with temperature, the cable irregularities due to manufacturing and splicing, and the power transmission capabilities of the cable became basic restrictions on the design of the L3 system. 2. The LI telephone terminal equipment was to be reused. This equipment involves channel banks, group and super-group equipment and carrier supplies.^ This hmited the system planning to the use of frequency division multiplex on a single sideband carrier suppressed basis. 3. It w^ould be desirable to reuse existing LI repeater locations and buildings. The LI auxiliary repeaters are spaced at eight mile intervals and housed in 6' x 9' concrete block huts. The LI main repeaters are spaced at 40 to 160-mile intervals largely dictated by geographical and power transmission considerations. 784 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 4. Sufficient bandwidth should be provided so that a black and white television signal of at least four-megacycle quality could be transmitted simultaneously with 600-message signals, the message capacity of the LI system. Alternatively, as many message channels as possible should be transmitted when there was no need for television service. 5. The channels should meet Bell System high quality signal-to-noise and equalization objectives after 4,000 miles of transmission. Section 2 of the paper is devoted to a discussion of the principal system design problems and descriptions of methods used in solving these problems. Section 3 contains descriptions of the components of the system, their locations and their functions. 2.0 Transmission Design With a given cable loss, the line repeaters determine in large measure the bandwidth and quality of transmission and the economics of the system. The basic system plan therefore evolves from a consideration of the signal-to-noise and equalization performance — i.e., the transmission stability — that can be designed into the repeaters. This leads to the development of broad signal-to-noise and equalization analyses which guide and coordinate the system design. 2.1 SIGNAL-TO-NOISE DESIGN Simply stated, the signal-to-noise problem is to adjust the repeater spacing and bandwidth of the system so that channel objectives can be met with the repeater noise, linearity, and gain performance that the electron tube and circuit art permit. In detail this means the following: (1) to translate the broad transmission objectives on message and tele- vision channels into detailed requirements on noise, specific modulation products and compression; (2) analyzing the amount of these inter- ferences that result from various repeater design choices; (3) determining the effect of signal wave form and frequency allocations on both the channel requirements and the repeater performance ; and (4) integrating these studies into a specific system design plan that meets the objectives. 2.11 Telephone Channel Interference Objectives The amount of noise, tone interferences or crosstalk that is con- sidered tolerable in telephone channels is generally determined by judge- ments involving the subjective reactions of representative observers to specific interferences on typical transmitted signals and by the cost of providing a given grade of service. The broad objectives for message THE L3 SYSTEM — DESIGN 785 channels stem from early unpublished work on transmission standards. The interference and load capacity requirements for transmission sys- tems involving large numbers of message channels were developed by Dixon, Holbrook and Bennett. ' In effect, they provide techniques for translating channel objectives into linearity and power handling require- ments on repeaters, taking into account the statistical properties of individual and multi-channel speech. Based on the data and techniques in these papers, the requirements on individual channels shown in Table I can be derived. These requirements in themselves form an im- portant basis for the signal-to-noise design of the system. However, in a highly refined system design it is necessary to extend our notions of requirements somewhat further. In the L3 signal-to-noise design the message channel requirements of Table I were used as the initial basis for study. However, when specific interferences of a complex nature were found to be limiting, the wave forms and the probability of their occurrence w^ere examined in detail. As a result of these studies, two distinctive types of interferences were found to be important when the system is used to carry message and television signals simultaneously. The first of these, due to both second and third order modulation involving multifrequency key pulse signals and components of the television signal, has the characteristics of inter- mittent musical tones. The second, due to the second order difference products generated by the television signal components, produces tones in the message channels which vary in amplitude and frequency as the television signal changes with picture content. Both types of interference were generated in the laboratory and recorded on tapes. From these tapes, records were cut and then used in a series of subjective tests Table I — Summary of Message Circuit Objectives {Allowable Zero Level Interference in 3 kc band) Source of Interference Type of Interference dba (message Weighting) dbm* Unweighted Terminals Line Line Largely spillover between channels, cross-modulation, and crosstalk Noise and multichannel modulation Unintelligible crosstalk and babble Tones All sources +32 +36 +24 +24 +38 -50t -46t -58t Line Total -44t * The translations from dba to dbm are effected by noting that a 3000 cycle band of flat noise with one milliwatt of power equals +82 dba and that one milli- watt of 1000 cycle single frequency is equal to +85 dba. t Interference assumed evenly distributed over 3000 cycle band. I Tones assumed to be at 1000 cycles. 786 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 which were made to determine the maximum permissible magnitudes consistent with other important message circuit objectives. 2.12 Television Channel Interferen ce Objectives The amount of noise and single-frequency interference that can be tolerated in a commercial grade television channel again depends on judgements involving the subjective reactions of observers and the cost of providing a given grade of service. The broad objectives are based on subjective measurements which have been reported on by Messrs. Mertz and Baldwin.^' ^' ^ From this work it has been determined that 95 per cent of the observers consider a signal-to-noise ratio of 40 db (composite signal to rms noise) tolerable, providing the noise has a frequency characteristic that rises about 11 db across the video band. Likewise the tolerable single frequency interference can be set at —70 db (peak sine wave below composite signal) if the interference falls below about one megacycle. The requirement becomes more lenient for interferences falling in the upper part of the band. Again, for a refined system design, more detailed account must again be taken of the requirements on short duration interferences, the prob- ability of interference occurring, and the exact frequency in the tele- vision spectrum that an interference occurs. In the L3 signal-to-noise design the broad television channel objec- tives outlined above were used except when a specific complex inter- ference was found to be limiting. For complex interferences, three ad- ditional types of requirement data were used; (1) tests were made to determine visual thresholds relative to steady tones of short bursts of energy such as occur in the television channel due to switchhook •*bang-up" and multifrequency key pulsing signals in the message chan- nels. Fig. 3 shows the relation between the steady state and transient requirement: (2) advantage was taken from the fact that interferences falling between the 15.75-kc line scan multiples of the television sig- nal would be less interfering than unwanted energy falling directly at the line scan multiples; and (3) a judgement was used that the toler- abihty of an interference depends on its probability of occurrence. The judgement was not made on a quantitative basis but when an inter- ference was found to exceed its requirement by a few db two or three times a day it was ignored in the signal-to-noise design. 2.13 Frequency Allocations The final frequency allocations shown in Fig. 4 are a result of the signal-to-noise design. The principal features were determined on rather THE L3 SYSTEM — DESIGN 787 <0 28 -1 UJ (Q O 24 ?20 o m cr lij 1 = _i tu cc 4 0 \ \ \ \ s s \ s V \ s. \ \ s s \ s ± _L \ J. 0.01 0.02 0.04 0.06 0.1 0.2 0.4 0.6 1.0 2 3 4 5 6 8 10 DURATION OF INTERFERENCE IN SECONDS Fig. 3 — Television bar pattern threshold versus duration. general grounds. When the system is arranged for combined television- message transmission, the television channel is placed above the message channels so that the second harmonic of the television carrier and its immediately adjacent side bands will fall at the top edge of the band where the requirement is more lenient. Likewise, the line repeater noise tends to rise with frequency as does the amount of noise that the tele- vision channel can tolerate. Details of the frequency allocations shown on Fig. 4 will be discussed in later sections. Pilot frequencies, indicated on Fig. 4, are transmitted to control the transmission characteristic of the system as described in a companion paper. ^^ The frequencies, and the power at which the tones are trans- mitted, were selected on two bases; (1) where possible, frequencies used for similar purposes in the LI system were selected for possible economies in pilot supply design and manufacture; these are the 556, 2,064 and 3,096-kc pilots; and (2) the transmission of these pilots should not materially degrade the signal-to-noise or load capacity performance of the system. The latter requirement led to a careful study of cross-modu- lation products involving the pilot frequencies to assure that message and television objectives would be met. 2.14 Repeater Performance The details of the amplifier design and the factors which determine its performance are covered in a companion paper. For purposes of the signal-to-noise design it is sufficient to know the noise power vs fre- quency characteristic, the second and third order modulation coefficients 788 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 o z Q.. O tc o a. lU < 2 CO A Pi ro o Z Q. 00 (U to < in ^ (0 ; § jk C\J '^ < 2 >. o § z !-i V t4-( CO a -) 1 c? Ill tr "^ li. bf) ;^ THE L3 SYSTEM — DESIGN 789 of the repeater as functions of frequency and the overload performance of the repeaters. These factors depend on the repeater spacing and cable loss characteristic, electron tube parameters, achievable feedback, and the bandwidth to be transmitted. Thus, in the design procedures the dependence of these properties on repeater gain and bandwidth are determined and used in adjusting the system parameters for a final compatible design. Figs. 5 and 6 show the noise and linearity properties of the final L3 repeater. The four mile repeater spacing requires a re- peater gain shown on Fig. 7. ■90 5-100 •110 -120 RANDOM NOISE MEASURED IN 3KC BANDS AT REPEATER OUTPUT / f y / / y 1 1 \ 1 0.2 Q.3 0.4 0.6 0.8 1.0 2 3 4 5 6 8 10 FREQUENCY IN MEGACYCLES PER SECOND Fig. 5 — L3 Auxiliary repeater noise characteristic. 2.15 SIGNAL MECHANISMS A signal-to-noise plan which contemplates transmitting the complex wave form of the combined telephone and television channels through 1,000 auxiliary amplifiers and about 200 flat amplifiers with performance factors that are variable with frequency will depend very strongly on the detailed analysis of the interactions between the signals and the repeater system characteristics. In developing this aspect of the signal- to-noise design four related phenomena had to be examined in detail. 2.151 Intermodulation Between Signals in Different Parts of the Band In the classical multichannel modulation theory for a large number of message channels, the modulation noise generated by interaction of the speech signals due to the non linear characteristics of the amplifier is shown to be equivalent in interfering effect to random noise. In ad- dition to this type of interference in message channels, cross modulation between components of the message and television signals result in a host of specific individual modulation products which have been ex- amined by determining their amplitude, duration and probability of 790 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 -28 o -32 -36 -40 -60 •X- FUNDAMENTAL AND HARMONIC VOLTAGES REFERRED TO OUTPUT GRID / / / / / / 20 LOG Ma = 20 LOG ^ y^ / / / y / / / _ X ^^20 LOGM3 = 20 LOG —3 1 ^^^ 1 .1 0.3 0.4 0.6 0.8 1.0 2 3 4 5 FREQUENCY IN MEGACYCLES PER SECOND Fig. 6 — L3 amplifier modulation coefficients. occurrence and then relating them to the requirements previously dis- cussed. Approximately 400 different products or groups of products were studied in the design of the L3 system. All but about thirty of these were found to be of negligible importance for the signal levels and frequency allocations being given serious consideration. On final analysis six of these thirty products were found to be controlling in establishing system levels. Fig. 8 shows the generating signals and the products they form for the six most critical products. The exact way in which the critical pro- 40 -i Ui fi 30 u Z 0.2 0.3 0.4 o.e o.e i.o 234 FREQUENCY IN MEGACYCLES PER SECOND Fig. 7 — ij3 repeater gain characteristic. 5 6 THE L3 SYSTEM — DESIGN 791 ducts entered into the determination of signal levels and frequency allocation will be discussed later. 2.152 Location of the Television Carrier Relative to the Telephone Chan- nel Carriers Among the important modulation product types is one formed by difference frequencies involving components of the telephone and tele- vision signals, see Fig. 8(d). These interferences fall back into the tele- phone band and are of different magnitudes depending, among other things, on which components of the television signal produce them; those (a) (b) (c) (e) (f) MESSAGE SIGNALS A __f TELEVISION SIGNAL (d) I lllllll FREQUENCY ALLOCATION OF SIGNALS I I I I ! I llllll|lllli TELEVISION 2ND HARMONIC SPECTRUM I (2B, 2C,B+C) I mil MESSAGE-TELEVISION SUM PRODUCTS I CA+B,A+C) ! I ' I ' 1 I I I ! lli|miiiiii llu MESSAGE-TELEVISION DIFFERENCE PRODUCTS (B-A,C-A) I I lllllll|lllll ILL Xili TELEVISION DIFFERENCE PRODUCTS I |(C-B)| j 'ill I ' I I I ^ lllliiiiiiiiiiiili I TELEVISION COMPRESSION (3RD ORDER PRODUCTS OF TELEVISION COMPONENTS) (g)U iiiiiiiiii 3RD ORDER CROSS PRODUCTS (A+B-C,A+C-B) ! 2 3 4 5 6 7 FREQUENCY IN MEGACYCLES PER SECOND Fig. 8. — L3 coaxial system. Critical modulation products in combined message- television application. 792 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 produced by the television carrier and adjacent line scan multiples are by far the strongest. The energy in a disturbing telephone channel tends to be concentrated near the 1,000-cycle point in the voice frequency band. By careful choice of the television carrier frequency, the difference products pro- duced by cross modulation between telephone signals and the high magnitude television signal components can be made to fall at fre- quencies such that the high energy portions of these products are greatly attenuated by the cut-off characteristics of channel filters. The message channels are spaced at 4-kc intervals controlled by car- rier frequencies which are multiples of 4 kc. To obtain the maximum advantage from the channel filter cut-off characteristic as described above, it was found desirable to set the television carrier frequency 1 kc below a 4-kc multiple. A direct result of this allocation is a gain of 12 db in television signal-to-noise performance over what could be re- aUzed if the carrier had been set at a 4-kc multiple. Such an allocation would have required a 12 db lower magnitude of television signal in order to meet the message channel objectives. 2.153 Addition of Modulation Products Along the Line It has been established by analysis and experiment, that in a multi- repeater system second order modulation products tend to accumulate on a power basis while certain third order products tend to add on a direct or voltage basis. This direct addition of third order products depends on the slope of the phase curve being the same over small fre- quency intervals from repeater to repeater. In multi-channel telephone systems, the locations of channels in the frequency band are shifted at intervals along the line to avoid this direct addition of third order prod- ucts. In the combined telephone-television application of the L3 system the A+B — C product illustrated in Fig. 8(g) is formed. Since the B and C components are television line scan multiples which cannot be shifted in location, certain components of this type product would add directly in a 4,000-mile system. If this were allowed to take place the requirements would be exceeded by many db. However, by placing the delay distortion equalization only in the television band at approxi- mately 200-mile intervals the phase of these products can be shifted so that rms addition of products accumulated over several 200-mile links of the system may be assumed. 2.154 Wave Form of the Transmitted Television Signal Early studies of L3 led to the conclusion that the most economical method of transmitting the television signal would be by amplitude THE L3 SYSTEM — DESIGN 793 modulation of a carrier with one sideband partially suppressed, i.e., vestigial sideband transmission. There remained, however, three major problems for detailed study; (1) the transmission of dc components of the video signal; (2) the per cent modulation of the carrier which for convenience is defined in terms of ''excess carrier ratio", the ratio of the peak (white) signal to the peak-to-peak composite signal as measured in the carrier frequency envelope; and (3) the sign or sense of modulation, that is, whether increasing or decreasing brightness should correspond to increasing signal voltage on the high frequency Hne. Typical wave forms illustrating the alternatives are shown in Fig. 9. ^___R___^ .IS. (a) VIDEO WAVEFORM IT. 4d Jl _J-L ¥ (b) NO DC COMPONENTS 100%^TOSITIVE" MODULATION FOR WHITE BLANK FIELD JT Jl- ir -i_r Jl u (c) DC COMPONENTS TRANSMITTED 100%''NEGATIVE" MODULATION (d) DC COMPONENTS TRANSMITTED 509^ "POSITIVE" MODULATION (E CR = 2) X---^ (e) DC COMPONENTS TRANSMITTED 100»/o" POSITIVE" MODULATION CECR=0 J^ ^ Ul Jl hi inrir 11 (f) DC COMPONENTS TRANSMITTED GREATER THAN 100% MODULATION (ECR=1/2) Fig. 9 — Typical television signals. Alternative carrier frequency waveforms. 794 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 The solution to each of these problems required an understanding of how the various alternatives would be affected by the system noise and linearity performance and an understanding of representative tele- vision viewing tube performance with respect to susceptibility to differ- ent types of interference. In analyzing the effect of system performance on these problems, it was found that non-linearity (cross modulation) would produce interferences in the television band which, while very complex electrically because of the effect of cross modulation involving line scan components of the signal, would produce the same effect on viewing tubes as single frequency interferences, i.e., bar patterns. Fur- ther simplifications were made in the analysis when it was found that such interferences were most visible in relatively large areas of tele- vision pictures having essentially constant brightness. During the time intervals corresponding to such areas, the video frequency voltage of the television signal is essentially constant and therefore, in the cases of interest, it could be assumed that the magnitude of the television carrier would also be constant during such intervals. Thus, to compute the magnitude of any modulation product which falls into the television band and which has as one of its components the television signal itself, it is found convenient to use in the computation the magnitude of the tele- vision carrier corresponding to either black or white portions of a picture signal. (The reason for intermediate shades of gray being less susceptible than either black or white is discussed below). To evaluate the effect of television viewing tubes on wave-form prob- lems, a number of tests were made to determine blank field threshold values of single frequency interference as a function of frequency for typical viewing tubes. Furthermore, judgements were made as to what might be expected of future viewing tubes with respect to achievable high light brightness, contrast ratio, and operating characteristics. As a result of these tests and judgements, a series of requirements w^ere derived on the basis of long range objectives to be met for these pro- jected characteristics. The results of these tests and judgements are summarized in Table II. Using the parameters and methods of analysis outlined in the pre- ceding paragraphs, the relative system performance achievable with each of the carrier frequency wave forms of Fig. 9 was computed or determined by observation. For example, these wave forms are all drawn to the same peak-to-peak amplitude. If we assume that the coaxial system is limited only by the peak amplitude transmitted we may use Fig. 9 to determine relative signal-to-noise performance directly by measuring the peak-to-peak magnitude of the composite signal voltage (sync tip to white) transmitted. THE L3 SYSTEM — DESIGN 795 Fig. 9 may also be used to obtain relative modulation performance. For this purpose, the following factors must be considered; (1) the magnitude of the signal generating the interference ("black" or 'Svhite" carrier magnitude); (2) whether the interference is proportional di- rectly or to the square of the carrier magnitude; (3) relative interference sensitivity in black or white portions of the picture; and (4) deviations from the Weber-Fechner law as the brightness is varied over its full range. The relationships among these factors were used to establish that for all cases of interest, bar patterns due to cross modulation are always more interfering in either black or white portions of a picture than in an intermediate gray area. Table III shows the relative system performance for the five carrier frequency wave forms of Fig. 9. For comparison purposes, the signal- to-noise and signai-to-bar pattern ratios are all related to Fig. 9(f). Table II — Television Viewing Tube Characteristics Assumed FOR L3 SIGNAL-TO-NOISE ANALYSES 1. Brightness-grid voltage characteristic of viewing tubes follows 5/2 power law: B a el". 2. Maximum high light brightness of viewing tubes will be 150 foot lamberts. 3. Contrast ratio of viewing tubes will be 150:1. 4. Viewing tubes will have interference sensitivities which vary with brightness in accordance with the characteristic of Fig. 10. 5. The visibility of bar patterns will decrease with frequency in accordance with the characteristic of Fig. 11. 6. Deviations from the Weber-Fechner law may be assumed to follow the curve of Fig. 12. This law states that "the minimum change in stimulus necessary to produce a perceptible change in response is proportional to the stimu- lus alread.y existing." It is obvious from Table III that the signal is transmitted most effi- ciently at an excess carrier ratio of one half. The wave form of Fig. 9 (f), which illustrates excess carrier of one half, is the one used in L3. Television terminal circuit problems arising from this choice of carrier frequency wave form are discussed in another paper. 2.16 Signal Levels and Repeater Spacing In a broadband system like L3, the problem of determining the re- peater spacing is made complex by the large number of parameters that must be considered. The approach to this problem that has been used to advantage in the L3 design is to assume several reasonable values of repeater spacing and determine for each the system performance achiev- able with various combinations of important parameter values. This method also permits evaluation of the effects on repeater spacing due to variations in parameters so that it is possible to form judgements as to the most economic design. 796 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Table III — Relative Performance of Alternative Television Waveforms Waveform* Relative Signal-to-Noise Ratio in dbf Relative Signal-to-Modulation Ratio (Bar Patterns) in DBt Group 1 Group 2 9b no dc + 10.2 +6 + 12 +6 0 + 11 +9.5 + 14 0 0 + 11.3 Qc npiE mod + 12.5 9d ECU = 2 + 15.5 9e ECR =1 +6 9f ECR = 1/2 0 * The waveforms are numbered to correspond to those given on Fig. 9. t All values referred to E.C.R. = 3^^; plus values indicate poorer performance. X All values referred to E.C.R. = 3^^; Group 1 products are those whose mag- nitudes are directly proportional to the carrier magnitude. Group 2 products are those whose magnitudes are proportional to the S(5[uare of the carrier magnitude. One of the important factors in setting repeater spacing is the mag- nitudes at which signals are transmitted in the system and the relation between these magnitudes and signal-to-noise and repeater overload performance. In the all telephone system (1,860 channels), the telephone levels (db with respect to the transmitting toll test board) were set to optimize signal-to-noise performance. To avoid penahzing the channels in the upper part of the band where random noise tends to be much higher than at low frequencies, the levels of the three mastergroups are staggered. At the output of any repeater in the high frequency line, the nominal level of mastergroup No. 1 is —21 db, that of mastergroup 16 12 I 2 3 4 5 6 8 10 ^ 20 30 40 60 80100 20( PICTURE TUBE BRIGHTNESs''B"iN FOOT LAMBERTS Fig. 10— Picture tube interference sensitivity assumed for L3. THE L3 SYSTEM DESIGN 797 ct — UJ o5 I UJ -I Q < zz "J Z VALUES ARE IN TERMS OF PEAK INTERFERENCE TO PEAK-TO-PEAK (Erma) SIGNAL FOR BLANK FIELDS OF 1 FOOT LAMBERT BRIGHTNESS AND INTERFERENCE SENSITIVITY OF 24 DECIBELS y /■ y 1 /^ „ 1 ^ -80 >- 0.1 0.2 0.3 0.4 0.6 0.8 1.0 2 3 4 5 VIDEO FREQUENCY IN MEGACYCLES PER SECOND Fig. 11 — Threshold values for bar patterns. No. 2 is —16 db and that of mastergroup No. 3 is —11 db. As a con- sequence of setting levels in this way, the random noise in the message channels is approximately 2 db higher on the average than modulation noise. It can be shown that with both second and third order modulation products contributing, and with third order somewhat predominant, this relation between random noise and modulation noise produces optimum signal-to-noise performance. With these levels, the 1,860 channel tele- phone system has approximately 6 db margin against repeater overload which, for L3 purposes, has been defined as the point at which the re- peater modulation coefficients just depart from their constant small- signal values. The signal-to-noise objective of +29 dba at the — 9 db level is met with about 2 db margin. When the system is used to transmit television and message signals simultaneously, the level of the telephone channels in mastergroup No. 1 at the repeater output is the same as that of mastergroup No. 1 in the all -telephone application, —21 db. The most convenient measure of the television signal is the power of the unmodulated carrier at the output of a repeater. Its value is +6 dbm. Due to the inter-relations o X in in oi J a: UJ xm 4 0 -4 VALUES INDICATE THE ASSUMED DEPARTURE FROM CONSTANT OF THE RATIO OF MINIMUM PERCEPTIBLE BRIGHTNESS CHANGE TO EXISTING BRIGHTNESS ^^ "~~" -8 1? 1 1 1 1 3 4 5 6 8 10 20 30 40 FIELD BRIGHTNESS IN FOOT LAMBERTS 60 80 100 Fig. 12 — Assumed deviation from Weber-Fechner law. 798 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 between the two signals, the limitations on achievable maximum trans- mission levels or magnitudes arise from certain types of second order modulation products rather than from optimizing signal-to-noise per- formance as in the all -telephone application. One of these types consists of sum products of cross modulation between telephone and television signal components. These form bar patterns and, in so far as signal-to- interference ratio is concerned, are independent of the television signal magnitude. Thus, adjusting such products to equal the appropriate requirement has the effect of setting the maximum permissible magnitude or level of the telephone signal. The second type of limiting product is due to difference frequencies formed by cross modulation among the television signal components. These fall into the telephone channels and, after the telephone level has been set as described above, permit calculation of the maximum permissible television signal magnitude. With signals set at —21 db level for telephone and +6 dbm unmodu- lated carrier for television, all of the critical products discussed in sec- tion 2.15 above and illustrated on Fig. 8 have adequate margin. The 40 db signal-to-noise objective for 4,000-mile television transmission is met with about 2 db margin and long haul (4,000 mile) message channels meet the -f 29 dba at the — 9 db level objective with about 5 db margin. A margin of about 5 db is also realized with respect to repeater overload performance. The single frequency pilots are adjusted to have the following values of power at the output of a transmitting amplifier: 7266 kc -16 dbm 8320 kc -26 dbm All others -36 dbm With these values, modulation products produced by cross modulation among the pilots and message and television signals all meet the ap- propriate objectives. 2.17 Frogging of Message Circuits When signals, either message or television, are transmitted over long distances through many amplifiers in tandem, the accumulation of modulation products along the line becomes an important system prob- lem for two reasons: (1) the accumulation of certain types of third order prcxlucts t(»nds to follow a direct or in-phase law and (2) the distribu- tion of modulation products over the band produces more modulation noise in certain parts of the band than in others. Both of these cumula- lion problems are alleviated if, at intervals along the line, the signals THE L3 SYSTEM DESIGN 799 are shifted with respect to one another in the band, a process known as frogging. In the L3 system, signal-to-noise performance is substantially im- proved by frogging the supergroups at intervals of about 800 miles. In the 1,860 channel all-message application, the busy hour signal-to-noise performance of 4,000-mile circuits is alike to within two db with all channels meeting the objective of +29 dba at the — 9 db level. In con- trast, if frogging were not specified, a substantial number of circuits (10 to 20 per cent would fail to meet the objectives while the perfor- mance of other channels would be better than required by six db or more. When the system is used for combined message-television signals, the message circuits are frogged in supergroup blocks at approximate 800-mile intervals except for supergroups Nos. 113 and 114 which must be frogged at 400-mile intervals. This procedure is necessary to prevent second order sum and difference products of message and television signal components from cumulating excessively, expecially those pro- ducts which involve television signal components close to the television carrier. Frogging these supergroups more frequently than others results in a 3 db improvement in television signal-to-noise performance. 2.18 Special Services Transmission During the early design stages, requirements based on the trans- mission of message and television signals were used to set repeater spac- ing, to determine the bandwidth and frequency allocations and to fix important design parameters of the amplifiers. Concurrently, the ob- jectives for the transmission of telegraph, program, and telephotograph signals were studied and before the system design crystallized, analyses were made to assure that these special services objectives would be met. In a few instances it was found that the special services objectives tended to dominate and the system requirements and design were ad- justed accordingly. For the most part, however, channels which meet message circuit objectives are satisfactory for special services trans- mission. In L3, telegraph and telephotograph signals may be transmitted without restriction provided the proportion of these signals does not materially exceed the proportion now installed in the plant. Program signals may be transmitted in the 1,860 all-message arrangement with- out restriction but when television transmission is provided, program circuits are restricted to supergroups Nos. 113 and 114. This restriction is due to the fact that program circuits are usually more than 4 kc wide; interferences of high magnitude which normally fall between 3,300 and 800 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 4,000 cycles or below 300 cycles in message channels of supergroups other than Nos. 113 and 114 would fall close to 4,000 cycles in a program circuit where there is high susceptibility to interference. 2.19 Uncertainties In the early stages of system design, firm decisions have to be made on such matters as repeater spacing, bandwidth, and component char- acteristics. These decisions must be based on a detailed signal-to-noise analysis which in turn involves many judgements of repeater perfor- mance parameters, tolerable system requirements and the effects of signal mechanisms on system performance. It would be easy and safe to engineer the system to provide enough signal-to-noise margin to cover the uncertainties in each of these judgements. Conservative en- gineering of this type could easily have justified a repeater spacing of three miles instead of the four miles actually chosen. Instead, an effort was made to estimate a "mid-range" or most probable value for each performance, requirement or mechanism factor entering into the signal- to-noise design. In addition, a "probable" uncertainty was estimated for each critical parameter. This was usually taken as one third of the maximum foreseeable error in the estimate. Finally, these uncertainties in electron tube modulation, realizable feedback, network impedances, channel requirements, interaction laws between signals and a myriad of other factors were all translated by the signal-to-noise analysis into their effect in db on the television channel signal-to-noise performance. On this basis, the "probable" uncertainties were summed on an rss basis to find the "probable" uncertainty in the overall design. Whereas the direct addition of the probable uncertainties gave a figure of about 20 db uncertainty in the design, the rss addition indicated about six db uncertainty. It was then argued that during the ensuing years of develop- ment the probability of finding all the judgements to be wrong in the same sense was extremely small. On the other hand it was deemed reason- able to provide enough margin so that there would be perhaps a 75 per cent chance of not exceeding the margin. Six db of margin was therefore provided, half by clear margin and half by having available economically feasible changes in system design such as a decrease in the telephone channel frogging interval. Any further error in judgement would then have to be taken up i)y degrading performance below the desired ob- jectives. As the system design proceeded, the early judgements were changed in considerable measure. Likewise, numerous additional system parameters were introduced. However, at no point in the system plan- THE L3 SYSTEM DESIGN 801 ning was the balance of factors such that there was less than three db clear margin. Margin handled in this way becomes a carefully husbanded asset of the whole system. In designing or analyzing a part of the system a major effort must be made to achieve the performance introduced into the initial determination of repeater spacing and bandwidth. The design of each individual part of the system cannot be allowed a margin which can be used up as the individual designer chooses. 2.20 Equalization Design The term "equalization" is used to describe the process of obtaining flat gain and delay characteristics for the system transmission. The system and equipment designs to accomplish this function represent two of the major engineering features of the L3 system. In an overall sense, equalization includes the following: (1) determining deviation objectives for the gain and delay characteristics of the system and its component parts; (2) designing the auxiliary repeater so that the most economical over-all system equalization is obtained; and (3) specifying the location, form and control methods for the mop-up equalizers that are used at intervals along the system. Equalization and its related process, regula- tion, are the subject of a companion paper ;^^ therefore, in this paper only those aspects of equalization will be covered which are necessary for an appreciation of the over-all system design. 2.21 Transmission Objectives The requirements on the gain characteristic of a band used for multi- channel telephony depend on two message channel objectives. One of these is that the gain of a message channel must not vary by more than two db over the 4-kc band. To meet this requirement, broad changes in the transmission characteristic of the message band are held to less than 0.5 db for 150-mile links. The second objective stems primarily from the need to transmit telephotograph signals. Since these signals are rela- tively intolerant of level changes, the transmission characteristics of working lines and protection lines are made alike to within d=0.25 db. The requirements on the gain and delay characteristics of the tele- vision band are based on the subjective determination that an echo de- layed by about two microseconds or more in a representative picture is considered tolerable by 95 per cent of the viewers when the peak-to- peak voltage of the echo signal is 39 db below the peak-to-peak signal voltage." The translation of this echo objective to allowable variations 802 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 in the gain and delay characteristics is straight forward if idealized sinusoidal deviations extending across the whole band are assumed. In practice, the characteristics of the transmission deviations in a long repeater system are very complex and therefore, the idealized objectives are only a tentative guide in system design. Since we do not have a thoroughly satisfactory method of evaluating complex echo pat- terns, the exact nature of the final television mop-up arrangements will be determined after subjective tests on the interfering effects of echoes resulting from the complex transmission deviations of representative links of the system. 2.22 The Mop-Up Plan The deviations from ideally flat gain and delay transmission char- acteristics may be classified in three broad categories; (1) fixed de- viations; (2) slowly varying deviations; and (3) rapidly varying de- viations. The distinction that is made between slow and rapid in the last two categories relates to the frequency of adjustment needed to meet system objectives. Those variations which require adjustment more often than once a week are considered rapid and those requiring adjust- ment at longer intervals are considered slow. Corresponding to each of the three classifications of deviations is a set of equalizers, fixed, manually adjustable, or automatic under con- trol of the pilot or a temperature sensitive element. Networks capable of fulfilling the functions of each are distributed along the line according to carefully prepared rules which enable system objectives to be economi- cally met. The locations of these equalizers, their functions and general characteristics are illustrated in Fig. 13. 2.221 Fixed Equalizers To the extent that the auxiliary repeater is designed so that its nomi- nal gain compensates for the loss of four miles of coaxial, it may be con- sidered as the first step of fixed equalization. In addition to the amplifier, the auxiliary repeaters are equipped with artificial lines, which are used to build out the loss of short sections to the equivalent of four miles of cable, and basic equalizers which provide for differences in the loss characteristics of different types of ('able. The second and final step of fixed gain equalization is known as a design deviation eciualizer. Its function is to correct accumulated de- viations due to the fuilurc of the average auxiliary repeater to exactly THE L3 SYSTEM DESIGN 803 I . I ! IMPEDANCE I {irregularities I I I FIXED LOSS ! SHAPE I LOSS VARIATION 'WITH TEMPERATURE ^-. TERMINATIONS DESIGN I DEVIATIONS! CABLE LOSS COMPENSATION BASIC EQUALIZER ARTIFICIAL LINE t MISALIGNMENT! 'COMPENSATION I RANDOM ■n DEVIATIONS i •(manufacturing)! FIXED GAIN SHAPE RESIDUES INSUFFICIENT SHAPES T- FIXED AND MANUAL DELAY SHAPES "_J REGULATING SHAPE (AUTOMATIC) AUXILIARY REPEATER VACUUM TUBE GAIN VARIATIONS WITH ^^'^^ i j TEMPERATURE ! I MANUAL GAIN SHAPES AUTOMATIC GAIN SHAPES U o'^.^^^^r^SlFr?., I MISALIGNMENT EFFECTS ' I ^Q^^"-'^^" MANUAL GAIN SHAPES ::j AUTOMATIC GAIN SHAPES FINAL MOP-UP I 1 SOURCE OF I J DEVIATION CD note: all SHAPES USED AT MISALIGNMENT EQUALIZER ARE ALSO USED AT FINAL MOP-UP POINT Fig. 13 — L3 coaxial system, equalization plan. compensate for cable loss. These equalizers will be used at every mop-up point, at 40 to 120-mile intervals. When television is transmitted, fixed delay equalizers are used at approximately loO-mile intervals. These equalizers compensate for the delay distortion introduced by the cutoffs of the auxiliary repeater sections. 2.222 Manually Adjustable Equalizers The manually adjustable gain equalizers consist of networks whose loss-frequency characteristics are related to one another by a Fourier series type of representation. The number of terms of the series required to meet system objectives varies with different types of mop-up points and depends on the system apphcation being provided for, all-message or combined message-television service. Equalizers of this type are used in mop-up points at 40 to 120-mile intervals along the line. Manually adjustable delay equahzers are provided at approximately 150-mile intervals when television signals are transmitted. These equal- 804 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 izers supplement the fixed delay equalizers described above and are needed to trim the delay characteristic of the line in finer detail than would be possible with fixed equalizers. 2.223 Automatic EquaHzers The first step of automatic-gain equalization is provided at each auxiliary repeater. The nominal gain characteristic of the repeater is designed to match the loss characteristic of the coaxial at 55° F. The cable loss varies with changes in temperature; the variations, however, have a predictable characteristic, being very closely proportional in db to the square root of frequency. To compensate for these changes, the gain characteristic of the amplifier at auxiliary repeaters is adjustable and follows the loss of the cable under the control of a thermistor. Two types of circuits are used to control the current fed to the thermistor as described in a later section. In the second step of automatic gain equalization, networks are pro- vided to match system gain variations caused by electron tube aging and by changes in repeater hut temperatures. The loss characteristics of these equalizers are controlled by thermistors which in turn are con- trolled by the 308-kc and the 2064-kc pilots. These equalizers are used every 40 to 120 miles. In the final step of automatic gain equalization, networks are pro- vided to compensate for second order effects of the first three rapid variations described above, namely, cable loss variations, and repeater gain variations due to hut temperature changes and electron tube aging. The loss characteristics of these networks are under control of thermis- tors acted on by the 556-kc, 3096-kc and 8320-kc pilots. These equalizers are located at approximately 150-mile intervals. The thermistors which control the loss-frequency characteristics of automatic equalizers are driven by regulators through a simple form of analog computer. The design and operation of this circuit is described in a companion paper.^" There is no automatic control of the delay characteristic in the sys- tem except that provided by the automatic gain equalizers. Every effort is made to have these equalizers match the transmission changes out- side the band so that resulting delay changes in the band are minimized. 2.23 Equalization System Considerations Whether the system is being eciualized for telephone or television it is immediately apparent that the channel requirements described THE L3 SYSTEM — DESIGN 805 earlier applied after 4,000 miles of transmission imply that, with no equalization, stability of the transmission characteristics of the indi- vidual repeaters would have to be of the order of a few ten thousandths of a db. Obviously, stabilities of this magnitude with changes due to temperature, electron tube aging and manufacturing processes cannot be achieved. Therefore, the equaUzation system design must he based on an economical balance between the cost of achieving repeater ac- curacy and stability and the cost of providing and maintaining an elabo- rate system of fixed, manual, and automatic equalizers. The equalization problem involves so many variables that no attempt has been made to evolve a unified theoretical basis for evaluating the factors entering into this economic balance. However, in planning and designing the L3 system a number of principles and points of view have been developed which have guided the equalization planning. 2.231 Misalignment The transmission objectives described above are determined on the basis of delivering satisfactorily equalized signals at terminals. In ad- dition to this function the equalizers must limit the signal excursions along the line so that excessive noise or modulation is not accumulated in the repeater system. The amount of signal misalignment that can be allowed to accumulate before the first mop-up equalizer depends of course on the signal-to-noise allowance that has been made for this purpose. The amount of signal-to-noise performance allotted to mis- alignment must represent a balance between the reduced repeater spac- ing and increased complexity of equalizers that it costs and the increased spacing between mop-up equalizers and increased repeater deviations that it allows. The engineering method for arriving at this balance represents an interesting example of system design by successive approximations. For example, the total gain area available (over an infinite frequency range) in a coupling network is inversely proportional to the capacity across the network and one of the important design choices is the extent to which one tries to utilize this area in the transmitted frequency band. The degree to which the available gain is concentrated in-band is called the resistance integral efficiency. In the very early stages of the ampHfier design it was necessary to choose resistance integral efficiencies and frequency characteristics for the coupling networks. In a definite but complicated way these parameters are related to the sensitivity of the networks to element variations. Efficient networks give improved sig- nal-to-noise performance but also increase the sensitivity to element I 806 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 changes. By examining deviation curves for a number of specific net- work designs, tentative choices were made of 50 per cent resistance integral efficiency for the coupling networks and an allocation of about half the cable slope to the pair of coupling networks and the remainder to the feedback network. With an amplifier employing these networks a detailed study was made of the noise and modulation penalties at two frequencies resulting from misaUgnment in several lengths of line. This study indicated that with certain refinements in the repeater design the misalignment in twenty or more auxiliary repeater sections could be tolerated with a signal-to-noise penalty of about 2 db which was judged to be a reasonable allotment for this purpose. In addition, this study brought out: (1) that randomizing the variations of an element between its normal manufacturing limits resulted in a 4/1 reduction in the re- quired misalignment allowance as compared with accepting large num- bers of elements at one extreme of their limit; and (2) a small amount of gain adjustment at each repeater in the vicinity of the high magnitude television carrier would reduce the required misalignment allowance by about 2/1. Refinements on this plan for handling misalignment had to wait until the signal-to-noise and repeater design were crystallized. However, the study referred to above provided a powerful tool for evaluating proposed element deviations during the design period. When the exact signal levels and the most limiting modulation prod- ucts became known and when the repeater characteristics and final element deviations were determined it became possible to make a re- fined study of misalignment in terms of the noise and modulation im- pairment associated with specific signals and distortion products. At this point performance margins associated with specific interferences could be used to allow more or less misalignment of the particular signal components forming the interference. Likewise, amplifier deviations with specific frequency characteristics could be evaluated exactly in terms of their effect on the number of repeaters between mop-up equal- izers. By studies of this type it was determined that the "A" or mis- alignment equalizers could be spaced at intervals not to exceed thirty- two auxiliary repeater sections. 2.232 Distribution of Element Deviations The methods of statistical (juality control used to monitoi* the process of manufacture provided the necessary techniques for obtaining the desired randomization of deviations. A companion paper^^ presents the techniques that were developed to apply the broad fi(^l(l of knowledge THE L3 SYSTEM — DESIGN 807 on quality control to the specific needs of the L3 system. The most im- portant point to appreciate in this connection is that the control of the process of manufacture (as well as the end electrical requirements) of individual elements is being used as a basic factor in the design of the system. 2.233 Repeater Accuracy In developing the equalization plan it is a logical and straight forward operation to provide shapes and ranges in the equalizers that will com- pensate for the random variations of known elements. Likewise, real but indeterminate parasitic elements can be taken into account by specifying the final characteristics of the line amplifier feedback network and the equalizer fixed shapes (design deviation equalizers) on the basis of measurements on a rigidly controlled group of amplifiers that are deemed to be representative of the final product. However, having once specified the equalization on this basis the design elements and indeter- minate parasitic elements must be held to the values and ranges upon which equalizer location, shapes and ranges are specified. This point of view has led to rigid mechanical control and the omission of component adjustments in the line amplifier which represent a departure from other transmission systems. These features are discussed in detail in the com- panion amplifier paper.^ 2.3 NEW YORK-PHILADELPHIA TRIAL The first installation of L3 has been made between New York and Philadelphia. Since the middle of 1952, this installation has been used to test components, to verify values of important system parameters used in system analyses, and to gather data for the further design and development of equalizers. Random noise measurements have confirmed theoretical values (Fig. 5) to an accuracy of better than 2 db. In general, the measurements have indicated that the theoretical values have been conservative. Measurements of system modulation performance, made with single frequency tones, also confirm the theoretical values used in analyses. Third order modulation measurements are in almost complete agree- ment with theory while second order measurements have been generally two to three db more favorable than the analytic values used. Transmission measurements have confirmed that equalizer networks designed so far are satisfactory for systems to be installed in the near future. Further measurements are required to determine automatic THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 z " S -luj So: Q-z o 9^ i u. ill III a in UL CD . a\n <_I oi- ;f, A THE L3 SYSTEM — DESIGN 809 equalizer shapes to correct for the second order effects of temperature changes and electron tube aging. Also under active study are the prob- lems associated with final mop-up for long television systems. 3.0 System Description 3.1 general In the preceding sections on system design the functions of the aux- iliary repeaters and the need for additional repeaters with varying amounts of equalization have been brought out. Fig. 14 shows the trans- mission layout of a typical L3 system. The auxiliary repeaters contain amplifiers and regulating equipment to compensate for the basic cable loss and its variation with temperature. Since such repeaters are de- pendent on the cable for their primary source of power they are called auxiliary repeaters. At points in the system where additional first order equalization is required to reduce misalignment the complexity of the repeater equip- ment increases and such repeaters receiving power over the cable are called equalizing auxiliary repeaters. The distance which power may be transmitted over the cable to the auxihary repeaters is limited; therefore, repeaters at specified intervals must be capable of supplying power to the cable. These are called main repeaters. They may be equalizing main repeaters where only first- order equalization is required or switching main repeaters where lines are switched or circuits dropped. 3.2 AUXILIARY repeater 3.21 Transmission Circuit The auxiliary repeater is the basic unit of the system and its design determines to a great extent the performance and economics of the system. A block diagram of such a repeater for transmission in two directions on two coaxials is shown in Fig. 15. The power separation filter (PSF) is a six terminal high pass-low pass filter designed to sepa- rate the high frequency transmission signals on the coaxial from the low frequency current transmitted on the center conductor to furnish pri- mary power to the repeater power equipment. At the input to the repeater the low-frequency current is diverted to a power supply while the high- frequency current follows a path through passive networks to the input of the amplifier. At the output, the signal from the amplifier and the low-frequency current from the power supply are recombined in the power separation filter for transmission to the next repeater. 810 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 The power separation filters are basically simple designs, but the reali- zation of the theoretical design was complicated by the following: (1) the components in the low frequency section must pass currents of about 1.5 amperes without change in characteristics, and must with- stand potentials as high as 2,000 volts rms without generating corona noise; and (2) the components in the high frequency section must be such that the loss over the transmission band (300-8,350 kc) is small and stable and of such a shape that it is easily equahzed. To meet these requirements stable inductors and capacitors with a minimum of para- sitic resonances in the band were designed. REGULATOR PSF H ARTIFICIAL I I EQUALIZER LINE I I (PASSIVE) ;i TEMPERATURE SENSITIVE ELEMENT 3 B — Fig. 15 — Auxiliary repeater. The artificial line shown preceding the input to the amplifier is a pas- sive network to build out the loss-frequency characteristic of a short cable section to be equivalent to the loss-frequency characteristic of 4.0 ± 0.2 miles of 0.375" cable or 2.87 ± 0.15 miles of 0.27" cable. These lines are provided in several different sizes, so that, where it is impossible physically to locate the repeater within the specified accuracy of 0.4 mile this accuracy can be obtained electrically. The design of the net- work is such that an accurate and stable characteristic is obtained with a minimum number of components. The equalizer is a means for compensating for small variations in the transmission characteristics of coaxial cables due to variations in the physical construction of the cable. In the case of the most generally used cable, this eciualizer inserts only a small flat loss. The amplijirr is of the feedback type whose gain frequency char- THE L3 SYSTEM — DESIGN 811 acteristic is closely equivalent to 4.0 miles of 0.375" coaxial cable plus the loss of the other passive elements in the repeater. This unit is de- mountable without tools for maintenance and is sealed in a die cast housing as protection against moisture and dust. The detailed electrical and mechanical design are covered in a companion paper.^ The regulator may be one of two types. The first, called the auxiUary regulator, adjusts the gain-frequency shape of the amplifier in accordance with the magnitude of the 7,266-kc pilot transmitted along the line. The second type, the thermometer regulator, adjusts the gain-frequency shape of the amplifier under control of an element representing an average value of cable temperature. This element is a thermistor buried in the ground near the cable. Such a control is, obviously, not as accurate as pilot controlled regulation, but it is adequate for use at one-half of the auxiliary repeaters and its simplicity results in considerable saving in first cost and power requirements. The regulators are demountable units similar to the amplifiers. Their detailed electrical and mechanical design are covered in a companion paper.^^ The pilot alarm unit is provided with auxiliary regulators to indicate pilot deviations beyond a predetermined limit. Its operation will be described a lattle later in connection with the discussion of alarm and control arrangements for the entire system. 3.22 Power Supply Primary ac power for the auxiliary repeater is supplied on a constant current basis from the main repeater over the center conductors of the two associated coaxials. Power generating and control equipment used at the main repeater will be discussed in Section 3.6. At the auxiliary repeater, power supply equipment is required to convert the primary power to suitable voltages for heater, plate and bias use as shown on Fig. 16. Half of the input to the power supply is taken from each center conductor and the output of the power supply is used to power the entire two-way repeater. The heater voltages are obtained by simple transformation which is complicated only by the fact that accurate and low loss transformers are required and the primaries of these transformers must withstand high ac potentials without generating corona noise which might be trans- mitted through the power separation filter to the input of the amplifier. Two separate transformers are used to split the load between the two center conductors even though the secondaries are connected together to feed the repeater. This arrangement eliminates one crosstalk path 812 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 o o o o O O ..^ m? ZQ Z Q -"o 3< o -J ai < I -J 1 1 THE L3 SYSTEM — DESIGN 813 at low frequencies where it is difficult and expensive to design power separation filters to meet the system requirements. The dc plate and bias supply voltages for amplifiers and regulators could be obtained by conventional rectifier circuits except for one com- plication which such arrangements introduce. This complication is the fact that a rectifier terminated in a low-pass filter (conventional ripple filter) reflects a highly distorted current wave into the primary circuit. If the primary current is so distorted the various power supphes in the series circuit will be fed with other than a sine wave of current and will supply different voltages depending on the wave form. Since the heater power depends on the rms value of current while the dc output depends on the peak value of voltage, it is easily seen that the relationship be- tween these two will change with the wave form of the applied current. Furthermore, the line loading to be discussed later must be calculated on the basis of a pure sine wave; appreciable harmonics in the line cur- rent tend to make it impossible to predetermine the loading to any reason- able degree of accuracy. It was found that these problems could be avoided and the power factor of the power supply made very nearly unity if the rectifier (rect i) was terminated in a constant resistance load rather than a low-pass filter. This was provided by paralleling the conventional low-pass filter with a high-pass section terminated in the proper resistance load. To avoid wasting the power in this load a second rectifier was added (rect 2) . The dc output of this circuit is used in series with the main dc supply to provide the higher voltage required for the output stage of the am- plifier. This rectifier must also be terminated resistively although its effect on the main current wave is less than that of the first rectifier, and the power dissipated is smaller. Since there was a further use for a small amount of power for bias in the regulators, rect 3 was added to produce a regulated voltage in conjunction with a conventional gas tube circuit. This rectifier and its load provide the termination for the high pass section of the filter circuit for rect 2. A second gas tube circuit is used to obtain a regulated bias supply for amplifiers and regulators from the 3 15- volt source. The loads on both gas tube circuits are fixed so regula- tion for variation in input voltage only is required. For this reason a low current, highly stable gas tube could be used. 3.23 Power Loading The power transmission circuit of a power loop is essentially a re- sistance-capacity network at the power frequency. The line and the 814 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 power supply resistance are the resistance component and the line and power separation filter capacity to ground make up the shunt capacity. If the circuit were used in this form the primary current at each repeater would be different since it would be the vector sum of the current in the succeeding section and the current in the shunt capacitance. This is undesirable as the objective is to make all power supplies alike. A familiar solution is applied to this problem by inserting an inductive reactance in series with the line. A value of this reactance is chosen for each repeater to compensate for the current through the effective shunt capacitance and thus make the currents through each of the power supplies as nearly aHke as possible. To simplify the loading adjustment in the L3 system a continuously variable loading inductor was developed. This arrangement allows more accurate adjustment of loading without the complications of changing wiring taps in a high voltage circuit. The design of such an inductor presented formidable obstacles as a large range of variation was de- sired (20-120 mh), and relatively high currents and voltages were in- volved. The device used consists of the two inductors which may be rotated with respect to each other, so that the coupling between their magnetic circuits varies ideally between zero and 100 per cent. One in- ductor is inserted in each side of the power circuit and a net result is obtained which is equivalent to varying each inductor. 3.24 Physical Description The type of auxiliary repeater generally used is shown in Fig. 17. It consists of a 6-foot cable duct framework upon which the component panels are mounted. It is completely wired in the factory. The lower third of the bay contains the power supply equipment while the upper part contains two transmission panels. One panel is provided for each direction of transmission and all of the transmission components of the circuit are found on these panels. The demountable units, amplifier, regulator, and pilot alarm unit are interconnected with plugs and jacks, so that they may be removed for maintenance. The other units are interconnected by screw-type terminals and cable as it is expected that they seldom will require maintenance. Other types of repeaters will be available to meet special conditions such as manholes where sealed apparatus cases will be required to pre- vent damage due to water submersion, or telephone offices where stand- ard ll'-6" frameworks are usually desired. THE L3 SYSTEM — DESIGN 815 Fig. 17 — Typical auxiliary repeater in concrete block nut. 3.3 EQUALIZING AND SWITCHING REPEATERS 3.31 Components The equalizing auxiliary and main repeaters use the same general types of transmission equipment as the auxiliary repeater except for the equalizers. They differ principally in the quantity of equalization equip- ment provided and the bay arrangements. Table IV lists a summary of the basic transmission units in each repeater. At these repeaters a line amplifier is used as a receiving amplifier to compensate for the previous section of cable. Flat amplifiers are used as transmitting amplifiers and to compensate for the loss introduced by the equalizers. They have a flat gain-frequency characteristic and no provision for pilot control of their gain. Their design is very similar to that of the line amplifier and is covered in the companion amplifier paper.^ 816 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Table IV — Summary of High-Frequency Line Equipment AT Repeater Stations Line amplifier Flat amplifier Auxiliary or thermometer regulator. . . . Manual equalization (Number of terms) Automatic equalizers Regulators for equalization (kc) Design deviation equalizer Auxiliary Equal- izing Aux. 1 2 10 3 7266 308 2064 Auxil- iary Main 1 2 10 3 7266 308 2064 Switching Switching Main Main (Tele- (Tele- phone phone — only) TV 1 1 5 5 15 25 6 6 7266 7266 308 308 2064 2064 3096 3096 556 556 8320 8320 1 1 The functions of the fixed, manual and pilot controlled equalizers are noted in Section 2.22 of this paper and discussed in detail in the companion equalization paper. 3.32 Equalizing Auxiliary Repeaters This type of repeater will be found after a maximum of thirty-two auxiliary repeaters provided power feed to the cable, dropping, or switching is not required (Refer to Fig. 14). The major components provided are covered in Table IV. In addition to these items, power separation filters, basic equalizers and artificial lines identical with those in auxiliary repeaters are used. A pilot alarm unit is also included to monitor each of the three regulators and transmit an alarm when any one of the controlling pilots has deviated beyond a given limit. Power for these repeaters is obtained from the cable just as in the case of the auxiliary repeater and much of the same type of equipment is used. However, due to the larger amount of power required and the layout of the repeater the auxiliary repeater power units have been re- packaged to provide the optimum arrangements for leads carrying high current or critical bias supply circuits. The design of panels used in this repeater was dictated by the general scheme conceived for the switching main repeater where the maximum amount of equipment is required. This arrangement involves the use of both sides of a duct-type frame. A single panel (again called the trans- mission panel) is used, but an amplifier is mounted on one side and a THE L3 SYSTEM — DESIGN 817 regulator is mounted on the other. This requires access to both sides of the bay, but results in an overall saving in the number of bays and over- all floor space. All of the transmission components and a heater and bias supply unit are mounted in one 7' bay for each coaxial. The plate and primary ac power for two of these bays is mounted in another 7' bay. 3.33 Equalizing Main Repeater This repeater contains exactly the same transmission equipment as the equalizing auxiliary repeater (see Table IV). It differs in the func- tion noted before, that is, it is equipped to feed power to the cable. The equipment to perform this function will be described later in the paper. Since the repeater can feed power to the cable it can also supply the power for its own operation. This power is derived from the primary ac supply used for the line by means of conventional metallic rectifiers for dc circuits and transformers for the ac heater supplies. These power supplies are not a part of the power loop containing auxiliary repeaters, so no special arrangements are required to obtain good waveform or high power factor. The equipment arrangement uses the same units as the equalizing auxiliary repeater, but here conventional ll'-6'' frames are used. 3.34 Switching Main Repeater Usually, this type of repeater is supplied at the point where circuits are dropped or terminated. In order to permit switching from a working line to a spare line in case of trouble, see Section 3.4, more complex equalization is required so that the Knes will be as nearly alike as pos- sible when the switch takes place. Furthermore, the signal delivered to the terminal must meet equalization limits that will result in a satis- factory grade of service. Where the repeater is part of a system required to transmit only message circuits the basic equipment shown in Table IV (telephone only case) is required. In addition to these units facilites are provided which indicate and alarm pilot levels and provide for patching and other maintenance arrangements. Since this repeater always feeds power to the cable it uses the same power arrangements as the equalizing main repeater. When the system is being used for the combined telephone and tele- vision signal this repeater is the same as the "all telephone system" repeater except that it has additional equalization equipment to adjust ^ 818 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the system for the more stringent television requirements. Furthermore, line connecting equipment consisting of branching filters and additional equaUzation is required. Branching filters are used to separate the telephone and television bands so that they can be transmitted to their respective terminal equipment. These are combined high-pass low-pass structures compHcated by strict requirements on stability of the gain- frequency and delay-frequency characteristics. Delay equalization for the Une sections must also be provided in the television branch and a large part of this is combined with the branching filters. Other com- ponents required for long television systems are adjustable gain and delay equalizers and associated amplification. The same type of equipment is used as that described for the other repeaters except that a number of additional transmission panels are required to mount the additional amplifiers and regulators associated with the equalizers. Two 11 '-6'' bays are used to contain the equipment for one through coaxial. One bay contains the receiving equipment which precedes the line switch. The other bay contains the transmitting equipment (transmitting amplifier and hybrid) and any line connecting equipment for combined systems. Fig. 18 shows a typical main repeater installation. 3.4 AUTOMATIC SWITCHING* In order to preserve transmission in the event of the failure of a com- ponent of the system and for transmission maintenance purposes, one coaxial in each direction is operated as a standby. An automatic switch- ing system is provided to permit substitution of the standby fine for any of the working lines. The lines are switched at the input to the trans- mitting ampUfier and at the output of the receiving amplifiers and equalizers. (See Fig. 14). At the receiving end of a switching section, equipment is provided whose function is to recognize failure of a working line and initiate the switching circuits. Information as to the transmission conditions of the system exists in the pilot regulators, the output of which controls a sensitive relay with high and low hmit contacts. The operation of one of these relays provides the switching system with the information that transmission has failed or been seriously impaired. It is necessary to make a switch as rapidly as possible in the case of a total failure in order to reduce the effect upon the transmission circuits. As the relays take appreciable time to operate, the receiving switch equipment is designed * Material written by P. T. Sproul. THE L3 SYSTEM — DESIGN 819 Fig. 18 — Typical main repeater installation. 820 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 to operate directly from the dc output of the regulator associated with the 7,266-kc line pilot. This permits complete switch operation in about 15 milliseconds. Upon receipt of information from the regulators that one or more of the pilots have gone out of limits, the switch initiator signals the trans- mitting switch control equipment at the transmitting end of the switch- ing section as to which line has failed. Signalling is accomplished by the use of tones in the 280 to 296-kc range which are transmitted over all coaxials in parallel in the reverse direction. Use of the coaxials for trans- mission of these signals obtains maximum speed of operation of the switch. All paths are used in parallel to preclude failure of the switch if one channel in the opposite direction would be inoperative. The transmitting switch control equipment causes the transmitting end of the standby line to be switched in parallel with the line in trouble and then signals the switch initiator that this has been done. This verifier tone actuates the line switch at the receiving end to complete the switch. When the trouble clears, the switch is released as the initiator checks every minute to see if service on the working line can be restored. In the event of a prolonged trouble, the switch can be locked manually and the initator will no longer attempt to restore to normal. Release of the switch is accomplished by the transmission of a release tone to the transmitting end while a checking tone returned by the transmitting end indicates completion of release and readies the switch initiator for further switching. For maintenance purposes, manual operation of the switching equip- ment is provided. In effect, manual switches are made by simulating a failure. Alarm features are provided to indicate to the operating personnel failure of the coaxial system or failure of the switching equipment. Care has been taken in the design to insure that failure of the switching equipment in no way affects transmission except by removing the pro- tection afforded by the presence of the switching facility. One ll'-6" bay is required for the switch control equipment for each direction of an 8-coaxial system. The Une switches are mounted in the miscellaneous bay of the main repeater lineup. 3.5 TERMINALS* Television terminal equipment, which is required to modulate the video frequency signals to and from the high-frequency line, is described in a companion paper* and therefore, will not be discussed here. ♦ Material written by C. G. Arnold. THE L3 SYSTEM DESIGN 821 The telephone terminals consist of modulators (and related trans- mission equipment) and carrier and pilot generating equipment. The transmission components of a terminal for an all message system are shown on Fig. 19. The channel, group and supergroup equipment are designs previously used in the LI system. The designs of the submaster- group and mastergroup units employ circuit arrangements similar to those used in the supergroup equipment. The greater bandwidths, higher frequencies and more severe stability requirements required new components and improved circuit and layout techniques. Fig. 20 shows the modulation steps and location in the frequency spectrum of the supergroups, submastergroups and mastergroups when the L3 system is used for telephone and television or all telephone. Mas- tergroup one comprises the first ten sixty-channel supergroups. This mastergroup is placed directly on the Hne in the 564 to 3,084-kc frequency band for both the telephone-television and all-telephone cases. When the system is used entirely for telephone, two additional mastergroups are formed by modulating mastergroup one up into the desired frequency bands. Mastergroup No. 1 is subdivided into two submastergroups. The lower six supergroups, comprising submastergroup one are modulated directly up from the basic supergroup located in the 312 to 552-kc band. The modulation and carrier supply equipment for these supergroups are the same units that are employed in the LI system. The upper four supergroups comprising submastergroup two are obtained by modulating four supergroups located in the same frequency range as the top four supergroups in submastergroup one into the top part of mastergroup one. The supergroup numbering system used for L3 has been adopted for easy identification of supergroups in their high-frequency positions. Each supergroup is given a three digit number. The first digit identifies the mastergroup, the second digit identifies the submastergroup, and the third digit identifies the LI supergroup from which it was originally derived. In the 1860 channel all-telephone allocation, supergroup No. 112, which corresponds to the basic LI supergroup No. 2, may be used for high quahty, long haul message circuits. When the system is used for telephone and television, supergroup No. 112 is restricted to circuits under 200 miles in length because of intolerable second order cross modulation between these signals and the television signal. With these groupings of channels new modulating and carrier supply equipment is required for submastergroup No. 2 and mastergroups Nos. 822 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 zttzi r r--W\^ — [^ ^-R 8 n 1 J A JL m^ (^_ ih PC Pc; I h ri ^ t.ti ^-MiK 2108 KC FROM LI SUPPLY (SG7 CARRIER) 4139KC FOR I TV TEST I FILTER \Ay\^ AAA- 2080 KC 115600 ir\ I KC n^ HARM. GEN 18200 KC K> 14040 KC H> 13000 KC K> TO 4KC HARMONIC GENERATOR t 4KC FREQ SUR-L, FILTERS HARM. GEN ^516 _r\ ~l KC n^ iH>r>41H>43 GENERATORS I— ^AA/ j— yw 41H>|H5S FILTERS 695 KC MOD AA^ AAr ^2080 M KC I r41H> \ I ll8200M\. HARM. I I KC \\^ GEN '-^i> Il3000|r\, 1 KC Tv^ Fig. 21 — Arrangements for i THE L3 SYSTEM DESIGN 827 MULTIPLIERS ^ FILTERS X4 — 2064 KC o COMBINING NETWORKS 2064 KC ^; .iT {> 8320 KC 0-ilH>-^ &ilH>-^ 556 KC 4 306KC 13,000 KC ier and pilot frequencies. 828 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 THE L3 SYSTEM — DESIGN 829 the cable at the main stations is about 2,000 volts rms from center conductor to ground. This potential diminishes about 100 volts per repeater section in going out from the power feed point or as the power section is shortened. (The maximum potential applied to the cable in LI systems is about 800 volts rms between center conductor and ground). Extensive tests on the installed cable showed that corona develops in the cable at potentials varying in a random fashion between 1,200 and 1,600 volts rms. This would allow power feed points to be placed at a maximum spacing of about 100 miles. By replacing the nitrogen, with which the cables are normally filled, with a large molecule gas, sulfur- hexafluoride (SFe), the corona potential of the cable is increased well above the maximum operating potential. Only the cable sections ex- posed to potentials greater than 1,200 volts will be filled with the new gas. Elaborate and thorough tests have demonstrated that no deteriora- tion of the cables will result from the use of this gas. Some additional precautions are required in entering manholes and using high tempera- ture torches for soldering when the SFe gas might be present. 3.7 ALARM EQUIPMENT Since the auxiliary repeaters and certain of the main repeaters are unattended it is necessary that arrangements be provided to indicate at the attended stations when some piece of equipment fails to perform satisfactorily. Auxiliary repeaters using pilot regulators are equipped with micro- ammeter relays which monitor the operation of the regulator contin- uously. These relays provide an indication of the operation of the regu- lator and the power of the 7,266-kc pilot at the output of the repeater. When conditions change from the nominal by a specified amount the relay contacts close and are locked magnetically. This bridges an alarm pair in the cable and operates an alarm at the nearest attended repeater. By means of Wheatstone bridge measurements from this repeater over the same alarm pair, the repeater in trouble can be located and a main- tenance crew dispatched to make the necessary equipment replacements. The relays can also be reset over the same alarm pair to aid in the location process or to clear alarms which were initiated at unaffected repeaters by deviations in the pilot due to troubles at preceding repeaters. At main repeaters, microammeter relays are provided on all six pilots used to control the equalization of the system. Deviations in these pilots operate the automatic switching equipment and initiate the usual office alarms. Alarms are also provided to indicate fuse operation, transfers 830 THE BELL SYSTEM TECHNICA.L JOURNAL, JULY 1953 from regular to standby equipment, and the condition of electron tubes in the terminal equipment amplifiers. Provisions for connection to special alarm systems are made at main repeaters which are not fully attended. These systems extend the alarms to the nearest attended repeater and enable the attendant to determine in considerable detail the condition at the remote repeater. The attendant may also perform certain operations such as switching a working line to a spare line at the remote repeater. 3.8 MAINTENANCE Maintenance of the L3 system requires equipment and methods for routine checking of the system and trouble location. Normally the auxiliary repeaters will be visited at intervals of about three months, when checks will be made of the power voltages and currents, the electron tube bias and change in bias with a fixed change in heater voltage (activity), and the pilot magnitudes. At these times amplifiers and regulators which fail to meet prescribed limits will be replaced, the 7,266-kc pilot will be brought to its normal value by adjusting the regulator gain, and the amplifier gain control in the output beta circuit will be adjusted by observing the 3,096-kc pilot. For these routine tests and adjustments two portable test sets are provided. The power test set plugs into the repeater, amplifier, or regulator and provides for measuring power supply voltage and currents and electron tube cathode-grid voltages to an accuracy of ±1 per cent. The pilot indicator makes it possible to measure the 7,26^ and 3,096-kc pilots to an accuracy of ±0.1 db. For trouble locations at auxiliary repeaters a portable transmission measuring set has been designed. It is capable of measuring the power in a 500-cycle band at any place in the frequency spectrum from 50 to 11,000 kc to an accuracy of ±0.5 to ±0.02 db depending on its specific use and the care used in calibration. At main repeaters, line sections will be checked for noise and modula- tion performance and equalizers will be adjusted at intervals of one week to several months. In addition, loss and gain measurements on sections of the office suspected of being in trouble will be made. For all general tests except equalization line up, point by point measuring equipment is provided, consisting of a 50 to 10,000-kc oscillator, the tuned transmis- sion measuring set referred to above, a milliwatt power meter accurate to ±0.035 db and a complement of attenuators, pads and comparison switches. THE L3 SYSTEM — DESIGN 831 Fig. 23 — An engineer testing pilot transmission in an L3 repeater hut. The adjustment of the manual gain equahzers to an accuracy of db0.02 db in a rapid and direct way is accomplished by special equipment described in the companion equaHzation paper.^^ A visual gain and delay transmission measuring test set, capable of measuring gain to ±0.05 db and delay to ±0.02 microseconds, has been developed for observing the line performance and adjusting delay equalizers when the system is used for television. A maintenance center is provided at about 200-mile intervals along the line to service the equipment removed from repeaters. At these points facilities are provided for the following: (1) electron tube testing; (2) regulator repair and adjustment; (3) transmission measurements on passive components; and (4) ampUfier testing of sufficient scope to permit changing tubes and to determine whether an ampUfier is suitable for further service in the line. 832 the bell system technical journal, july 1953 Acknowledgements A system as complex as the L3 system is the result of a large scale cooperative development effort involving many Departments in Bell Laboratories. More than a hundred engineers and technicians have contributed to the design over a period of seven years. In the system planning aspects of the development covered by this paper particular mention should be made of the large contributions of L. G. Abraham, C. H. Bidwell and S. E. Miller. Bibliography 1. Abraham, L. G., Progress in Coaxial Telephone and Television Systems, Transactions of the A.I.E.E., 67, pp. 1520-1527, 1948. 2. Crane, R. E., Dixon, J. T. and Huber, G. H., Frequency Division Techniques for a Coaxial Cable Network, Transactions of the A.I.E.E,, 66, pp. 1451- 1459, 1947. 3. Holbrook, B. D., and Dixon, J. T., Load Rating Theory for Multichannel Amplifiers, Bell System Tech. J., 18, pp. 624-644, Oct., 1939. 4. Bennett, W. R., Cross-Modulation Requirements on Multichannel Amplifiers Below Overload, Bell System Tech. J., 19, pp. 587-610, Oct., 1940. 5. Mertz, Pierre, Data on Random Noise Requirements for Theater Television, S.M.P.T.E., J., 57, pp. 89-107, Aug., 1951. 6. Baldwin, M. W. Jr., The Subjective Sharpness of Simulated Television Images, Bell System Tech. J., 19, pp. 563-586, Oct., 1940. 7. Baldwin, M. W. Jr., Single Frequency Video Interference Data, Proceedings of the J.T.A.C, Annex 8, 2, Dec, 1948. 8. Morris, L. H., Lovell, G. H., and Dickinson, F. R., The L3 Coaxial System — Amplifiers, see pp. 879-914 of this issue. 9. Rieke, J. W., and Graham, R. S., The L3 Coaxial System — Television Ter- minals, see pp. 915-942 of this issue. 10. Ketchledge, R. W., and Finch, T. R. L3 Coaxial System — Equalization and Regulation, see pp. 833-878 of this issue. 11. Mertz, Pierre, Fowler, A. D., and Christopher, H. N., Quality Rating of Television Images, I.R.E., Proc, 38, pp. 1269-1283, Nov., 1950. 12. Dodge, H. F., Kruger, M. K., and Kinsburg, B. J., The L3 Coaxial System — Quality Control Requirements, see pp. 943-968 of this issue. The L3 Coaxial System Equalization and Regulation By R. W. KETCHLEDGE and T. R. FINCH (Manuscript received April 17, 1953) The equalization and regulation problems of the L3 system are described and a theory of equalization of complex systems is outlined. The location and function of the various equalizers are explained including the roles and design of the various fixed, dynamic and manual equalizer networks. The analog computer used in the regulation system is described together with the cosine-equMizer adjusting technique used with manual equalizers. Finally the circuits and operation of the regulation system and its com- ponents are presented. Introduction Equalization is the process of correcting system gain and delay devia- tions sufficiently to permit the satisfactory transmission of signals. Regulation refers to that part of equalization which, by automatic means, corrects for relatively rapid changes in transmission. In the L3 coaxial system the transmission of television signals through more than 1,000 ampUfiers introduces relatively severe equaUzation and regulation problems. Compared to the LI system, the L3 system has nearly three times the band\\ddth and over three times more stringent transmission objectives. Thus it has been necessary to devote considerable effort towards finding equalization methods that yield practical and economical solutions. In previous systems it has often been the practice to design the bulk of the equalization system after the completion of an initial installation and the determination of the deviation characteristics of the system. In order to expedite the introduction of the L3 system into the field, equalization study and planning were initiated at the very beginning of the development of the system. One of the design methods was to make highly detailed studies using relatively inadequate data in order to find the major problems. As the data improved the studies were likewise 833 834 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 improved. While on the surface it might appear more practical to defer this work until the data are more adequate it has been found that these speculative studies are the only sure guide to ever getting the right data. Faced vsith such a complicated problem it is difficult to select from the vast amounts of things that might be important those relatively few items on which success or failure depends. In the L3 system there have been a number of critical problems which required intensive effort to find acceptable solutions, for example, the design of stable regulators to permit operation of 700 regulators in tandem, the choice of shapes for manual and dynamic equalizers and the selection of equalizer adjust- ment methods. The design of equalization for the L3 system is not yet complete since the dynamic equalizer shapes are still subject to considerable un- certainty and the details of some of the final television mop-up equaliza- tion are yet to be settled. However no major difficulties are anticipated in equalizing the telephone system to be installed from Philadelphia to Chicago during 1953. Although amplifiers having the final gain charac- teristic were operated in the trial line between New York and Philadel- phia for the first time in November, 1952, it was immediately possible to transmit quite satisfactory television pictures over the 200-mile loop using existing equalizers. It therefore appears that equalization will not limit the rate of field installation of the L3 system. THE PROBLEM The 4,000-mile coaxial cable has a gain distortion of nearly 40,000 db between 0.3 and 8.5 mc. Although the amplifiers reduce the distortion to perhaps 200 db, (and 100 microseconds), they leave a residue charac- teristic that is considerably more difficult to equahze. Further it is necessary to deUver service to intermediate offices spaced on the average about 120 miles apart. This requires equalization of high precision at numerous intermediate points. A further problem is the variability of the transmission characteristic due to manufacturing deviations plus time and temperature changes. Also, gain distortion cannot be per- mitted to exceed about 5 db at any point in the line or the signal misalign- ment will result in degraded signal-to-noise ratios.^ The overall transmission objectives^ are of the order of 0.25 db and 0.1 microsecond which, if allocated among the over 1,000 amplifiers, lead to rather unrealistic amplifier requirements. In fact, individual am- plifiers do not always meet the 4,000-mile overall requirements. Thus it is the problem of the equalization designer to provide a mop-up sys- THE L3 SYSTEM — EQUALIZATION AND REGULATION 835 tern that will permit attainment of the transmission objectives at all service points and at all times. EQUALIZATION THEORY One of the steps in the solution of the general problem has been to develop a ''theory" of equaUzation. This theory merely applies informa- tion concepts to the equaUzation problem to determine what information is required, when it is needed and how it may best be used. This theory has stimulated the development of novel equalizer adjustment tech- niques and has been of assistance as a guide to the attack on the general problem. In order to equahze a system the man or machine who is to perform the action must know what corrective steps are required, and for this he must have some kind of information as to the present state of the system and as to the desired state. Second, he must have the necessary tools to convert the system from its present state to the desired state. Consider the first problem, the determination of the corrective steps required. The problem is to determine what we need to know, when we need to know it and especially in what form we are able to utilize the information most efficiently. We can assume we know the desired state of the system; which is usually a constant loss with constant delay over the frequency range of interest. As to the present state of the system we note that sufficient information can never be obtained to equahze a system perfectly because of the finite bandwidth of the system and because the system changes with time. This is not a new fact, nor apparently a very important fact, because the system need not be perfectly equalized for satisfactory trans- mission of signals. It leads, however, to the converse idea, which is important — namely, that out of this infinite amount of information regarding the state of the system one should collect only the minimum amount that is needed. This implies making no more measurements of the state of the system than are absolutely necessary to perform the correction to a degree permitting satisfactory transmittal of the signals. The main purpose of this is, of course, to economize on time and effort required to obtain information, but it should be noted that excess in- formation may be a source of confusion to the equaUzation operator. To put this in the form of a rule, we have : Rule I Collect only that minimum of information as to the state of the system as will permit equalization to the required degree for satisfactory transmission of the signals. 836 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Having established that a minimum of information should be col- lected, we return to the time variation of the system, which in theory can make the information obsolete before it can be used. However, without yet bringing in the practical fact that its most rapid rate is relatively slow, we should introduce the fact that the ways in which the system can vary at its most rapid rate are quite restricted as compared with the manner in which it can vary at slower rates, and we note that even these are relatively limited. (This probably applies to most trans- mission systems, not just to the L3 coaxial.) Thus while the information regarding the more rapid, but simple changes must be collected more frequently it consists of a small amount of information per sample; whereas the information regarding the slow (but more complex) changes need not be collected very often, but it represents a relatively large amount of information per sample. Since the total rate of information collection is proportional to the information per sample times the rate of sampling, the minimum information col- lection principle would say that the rate of sampling should also be held to a minimum. This demonstrates the value of association of particular types of sys- tem change with the rate and amount of their variation, because one may thus eliminate from the more rapid sampling the collection of informa- tion about changes that occur at slow rates. Furthermore, one may es- tablish the sampling rates for the various system effects at the lowest possible value. (There is also a very practical value in knowing the rates and amounts of the various deviations, because system misalignment requirements force the equalization to be suitably distributed along the line.) In the form of a rule, this is: >. Rule II To the greatest practicable extent the overall system behavior should be separated into individual effects each having its own time rate of occurrence and corrections should be made for each effect at the minimum tolerable rate for each. It is of interest to note that, if we couple the logic of Rule II with the fact that the fastest changes (due to changes in the temperature of the repeater huts) take hours to become appreciable, we see that the con- tinuous collection of information from continuous pilots (and the con- tinuous correction by pilot-controlled regulators) is in principle un- necessary and inefficient — except, of course, for its other function of giving alarms under trouble conditions. i THE L3 SYSTEM — EQUALIZATION AND REGULATION 837 The technical problem of determining the equalization states of the system is normally solved by sending some kind of signals over the sys- tem and observing the effect of the system on those signals. The raw data are usually in the form of loss and delay as a function of frequency. On the basis of these data, the equalization operator desires to correct the system by means of some equalizers which have adjustable trans- missions and delays as a function of frequency. Thus the operator has a group of controls to be operated plus some data which has encoded in it the information as to the proper adjustment of each control. From these data, and a knowledge of the effect of each control, the operator must suitably compute the proper adjustments. As this may be too compli- cated a process to attempt on a trial and error basis, (or by numerical methods), it is quite an obvious advantage to the operator to receive the data as to the state of the system, not in its original form, but in the form of the necessary adjustments to his equalizer controls. This new form of the data simply represents a decoding process based on the available controls. An operator with the same data, but different and perhaps more complicated equalizers, would need the data in a form suited to his different equalizers. Consequently : Rule Ilia The information as to the state of the system may best be presented to the equalization operator in the form of the necessary adjustments of the avail- able equalization controls. This rule has a closely related corollary which is based on the fact that the available equalization controls determine the amount of in- formation that is needed. For example, if the independent gain equaliza- tion controls are "n" in number, measurement of the gain of the system at "n" suitably chosen frequencies is sufficient to determine the settings. (If the controls are not independent, fewer than *'n" frequencies need be measured.) This is, of course, a restatement of the fact that ''n" un- knowns may be determined by solution of "n" independent simultaneous equations. The unknowns are the equahzer settings and the simultaneous equations are the relationships of the shapes controlled by each equahzer to the total system error. Thus: Rule 1 1 lb In general, the necessary and sufficient condition for the determination of 'n" independent equalization control settings is the knowledge of the system's 838 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 equalization error at ''n" independent frequencies plus the knowledge of the effect of each of the ''n" controls at each of the "n" frequencies. It should be noted that this statement of the rule assumes analysis by- frequency rather than by transient behavior. This approach is used because, at present, equalizers are usually designed on a frequency char- acteristic basis. The validity of the rule is however more general. From these two rules we can derive another regarding the minimum information principle that is similar to Rule I but is actually quite inde- pendent of it. Rule IV No more information should he gathered from the system than is necessary to provide sufficiently accurate control setting information for the equaliza- tion operator. In this case the superfluous information may actually cause confusion or harm. It will, at the very least, confuse a manual operator to know of an error he is powerless to correct, whereas a mechanized system, on the other hand, would probably go berserk if it obtained too much in- formation. It is evident that the minimum amount of information that would be obtained in accordance with Rule IV would be the same as that obtained in accordance with Rule I only if the design of the equalization were optimum, because then the equalizer shapes (and the number of shapes) would just suffice to permit satisfactory transmission of the signals. Up to this point we have determined in general what information is needed, when it is needed and the optimum form of its presentation. Now let us proceed to examine what to do with the data; which involves the nature of the equalization operator as well as his equalization tools. If we followed Rule II rigorously, we should have several different type of controls; one for each of the effects having different time rates of occurrence. For purposes of illustration, however, we need assume only two rates — one quite rapid and the other very slow. It will be postulated here that very rapid equalization operations are most economically per- formed by machine, such as for example, the automatic regulation for cable temperature variations. Likewise relatively infrequent adjustment will be assumed to be best performed by a suitably informed human operator. The first principle to note is that the only real distinction here is the rate at which data should l)e refreshed and acted upon. In either case THE L3 SYSTEM — EQUALIZATION AND REGULATION 839 the data should be in the same form; a set of numbers (or their equiva- lent) representing equalizer changes. Probably the most useful result of this theory of equalization has been the conclusion that mechanization of the equalization process, particularly in regards to computational techniques, permits substitu- tion of simple logical methods for inefficient trial-and-error adjustment processes. This led to the use of an analog computer in the regulation system and, for manual equalizers, the development of measuring cir- cuits that read directly in terms of equalizer adjustment error. Equalization location and function of equalizers The location of equalizers in the L3 system when only telephone is transmitted is shown in Fig. 1(a). Combined telephone-television equali- zation is shown on Fig. 1(b). When telephone and combined systems use the same spare Hne, that line is equipped for television. The general features of the main-repeater layout have been described in a com- panion paper. ^ The detailed layout of equalization is designed to meet the requirements of both telephone and television service and the need for flexibility in television network arrangements. In addition, switch- ing of telephone or television service to a spare line must not appreciably degrade service. Switching sections longer than 120 miles must be provided with an intermediate step of equalization to prevent excessive signal misalign- '1 0 EHf^^ — 0 — X EHD— Q &^- ^ EK?^' .-125 MILES- J< UP TO 125 J<— OFFICE »|*- -.,1;'^c7SJ??k,c— 4« OFFICE-->l OF LINE ^^ MILES OF LINE I H^ MILES OF LINE I ' OFFICE EQUALIZER ,r.^„^,^ FIXED__.»H , , , FIXED _..,j . , , DELAY \4 U L TELEVISION DELAY fl H ^ H P n rU--- PATH V- ' ' ^ H Be€h^4J L>^ ^IHIh^<] D-"- l~^ I 1 r^^BRANCHING I 1 I O h^ ' FILTER I— |oJ— ' ^TELEPHONE PATH Q S^3&^ " EH5>E}-^ Pig 1 _ Typical equalizer locations in (a) telephone and (b) combined systems. 840 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 ment. This intermediate point is referred to as an equalizing auxiliary repeater. It is provided with the so-called A equalizer consisting in turn of fixed, manual and automatic gain equalizers. It reduces the gain error to less than 0.5 db using the fixed and manual sections and prevents appreciable degradation of this residue during an ensuing three month period by the action of three "office" regulators using pilots at 308, 2,064, and 7,266 kc controlling three regulating networks. One of these networks is the V7 shape of the receiving amplifier.^ The other two are first order corrections for vacuum tube aging and repeater temperature changes. At offices, (switching main repeaters), on telephone systems further equaUzation is required to permit line switching of telephone channels and special service signals such as telegraph. Also it is not practical to provide telephone equalization on a cumulative basis over longer links than a switching Unk because of frogging and dropping.^ No telephone channel rides for more than 800 miles in the same frequency location and dropping breaks up the pattern still further. Thus, for telephone, the switching links, which may number between 30 and 40 in a long system, are independently equahzed. The B equalizer contains manual and automatic sections, the latter being three regulating networks omitted in the A equalizer. Thus an A plus a B forms the final telephone equalization. The residue of five, independently-adjusted AB links must meet telephone requirements without frogging and 30 to 40 links must meet these requirements with 800-mile frogging. This performance must continue to be met in the presence of a normal amount of spare line switching. Also it is very important that the character of the residues be such as not to throw an undue burden on the television equalization. In combined systems the more stringent requirements require the addition of further manual equalization to the switching section. The residue at the output of a C equalized switching section must be suffi- ciently small that a 4,000-mile circuit will continue to meet television transmission requirements in spite of a normal amount of spare line switching. Also the C equalizer in conjunction with the AB must permit several switching links to be connected in tandem without further line equalization. The C equalization also contains an adjustable delay section which in conjunction with a fixed delay equalizer in the office path provides delay equalization for the television part of the band, 3.6 to 8.5 mc. This adjustable section builds out the line to match the fixed unit. The A-B-C pattern is for equalization of individual switching lines THE L3 SYSTEM — EQUALIZATION AND REGULATION 841 and these equalizers are adjusted on a single switching link basis. On the office side of the switches are system components that also require equalization. In the telephone system these include such things as office cabling and hybrid coils. Also there is an office flat loss of nearly 30 db. Thus the lines are, in effect, operated to give a 30 db gain while the offices give a 30 db loss. Aside from the flat losses there are distortion shapes but fortunately these are all well approximated by a Vj shape and a single manually adjustable v? equalizer plus suitable choice of flat loss provides adequate telephone office equalization. This is referred to as the 0 equalizer. It is adjusted to make the particular office have a flat characteristic. In the office circuits of combined systems are branching filters to separate the telephone and television bands. The 0 equalizer is reused in the telephone path. The television path includes the fixed delay equalizer and a manual gain equalizer to correct for office cables, hybrids etc. After the tandem combination of several independently equalized lines and offices further equalization will be required. This will be ac- complished by a multi-control manual D equalizer, having both gain and delay sections, inserted in the television only path at approximately 400-mile intervals. These D equalizers will be used to form 800-mile pilot links which are independently equalized to a degree permitting putting any five such links in tandem to form a 4,000-mile circuit with- out further equalization. FIXED EQUALIZERS The line amplifier can properly be considered as the first step of fixed equalization.^ Also acting at this same level are artificial cable networks used to build out the repeater spacing to 4 =t 0.2 miles, as well as the basic equalizer of the amplifier to take up differences between cable t3rpes. These devices are described in a companion paper. The final step of fixed equalization is the so-called ''design deviation equalizer" associated with all A equalizers. This equafizer comes in two versions, one for use with sections containing 23 to 32 repeaters and one for use in sections of 10 to 22 repeaters. In those few cases of less than 10 repeaters the fixed equalizer is omitted. The function of the design deviation equalizer is, first, to correct for the design error of the average repeater and second, to recenter the manual (cosine) equalizers. Although the average repeater matches its four miles of cable to within ±0.12 db this design error accumulates to over 3 db in 30 repeaters and further the shape is a difficult one to equal- ize. In order to keep the number of designs to a minimum only two sizes 842 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 are used, 19 and 28 repeaters, and the residue is corrected by the manual equaUzers. This residue may be positive or negative depending on whether the fixed ecfuaUzer over or under compensates. Thus there is only a small tendency for these residues to accumulate in long systems. However the inaccuracies of match between the fixed equalizer and the gain of the average repeater section tend to accumulate systematically and must therefore be kept small. This brings out the importance of statistical quality control of amplifier manufacture since any systematic shift in the amplifier gain characteristic will accumulate and may con- sume excessive manual equalizer range or may lead to excessive equaliza- tion errors. For example, a shift of only 0.05 db in the gain of the average amplifier w^ould represent 1.5 db in 30 repeaters, 50.0 db in 1000 am- plifiers and thereby becomes an extremely serious matter. Thus quality control of the amplifier is a vital part of the solving of the equalization problem. The various effects that consume the range of the manual equalizers produce for some shapes an unsymmetrical consumption of range. If uncorrected this would produce larger range requirements in the manual equalizers as well as introduce new shapes to be equalized. The manual equalizer shapes are symmetrical and their errors cancel if equal amounts of positive and negative range occur in the system. Any systematic offset of a particular shape tends to introduce new shapes due to the manual equalizer networks themselves. By appropriate modification of the shape of the fixed equalizer it is possible to recenter the manual equalizers so that on the average the manual shapes are in the center of their range. DYNAMIC EQUALIZERS Any long transmission system suffers from relatively rapid gain changes and in the L3 coaxial system, as in many previous systems, the necessary corrections are performed automatically by pilot con- trolled regulators. Pilot tones are transmitted over the line at a reference level and, at appropriate points, regulators pick the pilots off the line, observe the deviation in pilot levels from the reference values and re- store the pilots to or very nearly to the reference values by the use of regulating networks. There an; fundamentally two causes of fast gain changes, time and temperature. Time produces vacuum tube aging and in spite of their feedback the liru; amplifiers change gain. In one week a 4,000-mile sys- tem is expected to change by as much as 5 db due to the aging of the 6000 tulxjs or so in the transmission path. Because of different thermal THE L3 SYSTEM — EQUALIZATION AND REGULATION 843 characteristics, temperature affects cable and repeaters semi-independ- ently. In a 4,000-mile system one week is expected to engender as much as 8 db gain change by change in repeater temperature. Cable changes can exceed 100 db per week. Thus these effects must be corrected to a very high order of precision to maintain good television or telephone service. These gain changes are not the entire story; when the gain changes in the band it also changes outside the band and usually by an even larger amount. Thus equalization of the in-band gain leaves outband (above 8.5 mc) gain changes which produce in-band delay changes of several microseconds. Because of the difficulty and complication of providing automatic delay equalization it is necessary to equalize these delay changes on a gain basis, by at least partial correction for outband gain changes. Further, since satisfactory pilot transmission is possible only mthin the band, these out-band changes must be predicted from the in-band gain changes. This effect, in itself, indicates the use of equalizers whose individual shapes are those produced by specific sys- tem causes producing a correlated change in many elements. Thus building the regulating networks to match the effects of the individual system causes and matching to 10 or 12 mc rather than just to 8.5 mc permits simultaneous gain and delay equalization. Such a set of shapes is also more accurate because it is matched to the special ways in which the specific system can change rapidly. I In theory the regulating networks could be built to match linear combinations of the cause shapes but there are two difficulties. First, not all of the cause shapes are known accurately or often even roughly at the introduction of a new system into the field. The mixtures of shapes cannot be determined mthout the missing ingredients. Second, a system is not a static design. Experience suggests improvements and thus occasions will arise where one will want to change the correction for a particular cause. If the networks represent mixtures of the causes this necessitates changing aU the networks. If specific networks match specific causes only the appropriate network needs replacement. The Computer The use of cause shapes leads to a problem to which the computer provides the solution. These cause shapes are broad effects covering the entire band and more. Thus no one pilot is a measure of a specific cause. However by a process equivalent to the solution of simultaneous equa- tions the pilots determine the amounts of an equal number of cause shapes that will restore the pilots to normal. Thus the computer trans- 844 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 lates pilot errors into shape errors and drives the appropriate regulating networks to obtain the corrections. Let the equaUzer shapes be given by functions of the form SM) = ^nF.(/), (1) where "n" is the subscript number identifying the particular equalizer. (The capital 'W is reserved for the total number of equalizers.) **^n(/)" is the equaUzer shape (on a "unit basis") as a function of the frequency "f\ "kn' is the amount of shape introduced by adjustment, ^'kn' may be positive or negative. "SnifY^ is the resultant shape put in the system by adjusting Fn{f) by an amount fc„ . The total shape introduced by all "iV" equalizers is obviously; StotM) = E &(/) = ZKFnifl (2) n=l n=l To obtain a match of Stotm to the given equalization error, iS^given , at "M " frequencies from m = 1 to m = M, requires that; 'S'total(/m) = A^givenC/m). (3) at each frequency from /i to fu - Or, in terms of equation (2) /Sgiven(/m) = Z KFnifm) (4) n=l again, at each frequency from /i to fm . All of the important conclusions regarding the action of an equaliza- tion computer are implicit in the "Af " equations indicated by equation (4). Consider a case where there are three shapes. Let the information as to the difference between the system state and its desired state be deter- mined by the deviation of three pilot levels which are observed to be 8i , 62 and 8z at the pilot frequencies /i , /2 and /s . The problem is to find the values of fci , ^2 and ks that will give a match at these frequencies. This means that the following equations must be satisfied: Sl(f^) + .S2(/i) + .S3(/,) = 5, (5) *Sl(/2) + *S2(/2) + S,(fo) = 62 (6) Slifz) + *S:2(/3) -f *S3(/3) = 8, (7) Thus THE L3 SYSTEM — EQUALIZATION AND REGULATION 845 /uFi(/0 + k^Foih) + hF,(f,) = 8i (8) klF,(f2) + hF2{f2) + hF,(f2) = 52 (9) hF,(f^) + hF^if,) + hF^if,) = 53 (10) The solutions for A:i , A: 2 and kz take the form kn = a5i + 652 + c53, (11) where a, h and c depend solely on the shapes F„»(/). Thus the values of the 5's may be decoded into the values of the equiva- lent fc's by the simple process of multiplying each "5" by some fraction that is a function of the equalizer shapes; or, more precisely, by a frac- tion that is a function of the values of the various shapes at the fre- quencies /i , /2 , etc. The circuit of the computer is quite simple; consisting of about A''^ resistors for the control of "iV" networks by the deviations that are measured at M = A^ pilot frequencies. This simplicity is valuable for its own sake, but, as previously noted, there is considerable value in the fact that substitution of new equalizer shapes requires only that changes be made in the values of some of the resistors. Fig. 2 illustrates the principle by showing the computer circuit re- quired for the previous example of three shapes for which information is given at three frequencies, /i , /2 and /a . The three dc voltages repre- senting the deviations 5i , 52 and 53 are decoded by simply cross-connect- ing them to the three pairs of output terminals through fixed resistors chosen to satisfy the relationships of equations (8), (9) and (10). The output voltages will be proportional to the desired equaUzer correction quantities fci , k2 and ^3 . In general the calculation of some of these resistors will give negative values. Thus it will generally be necessary to require that the dc voltages representing the errors 5i , 52 , etc., be available in both polarities. Al- ternately, the errors can be provided in only one polarity and the cir- cuits to which the computer outputs connect can provide the push-pull circuit. This latter course has been used in the L3 regulators as indicated on Fig. 2. The effect of using the computer is as follows. If one pilot changes, all regulating networks correct but in such proportions and polarities as to produce no gain change at any pilot frequency except at the one originally disturbed. If all pilots deviate in proportions corresponding [ 846 TITE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 to one of the shapes, no network corrects except the one corresponding to the original pattern of pilot deviation. If conventional regulator circuits were used with particular pilots assigned to particular shapes the interactions would be intolerable. Thus the computer removes the restrictions on the choice of shapes imposed by regulator interactions. In turn this permits freedom in changing shapes as system data improves. It is important to provide the initial equalizers before adequate data on the system are available. As the system grows in length the equalization must be improved but the data improves also. Thus there is an economic benefit in providing a flexible equalization plan which allows the earliest possible commercial use of the system. Dynamic Equalizer Requirements Because temperature and aging variations can result in both gain and loss variation with respect to the average transmission, the equalizers must be capable of inserting both loss and gain compensation. Also, it is extremely desirable that equal gain and loss settings for the equalizer result in symmetrical transmission characteristics with respect to the average. In a long transmission system, some of the sections mil insert excess gain and others, excess loss. If the equalizer gain and loss char- acteristics are symmetrical, the residue will be related to the system OUTPUT (SHAPE ERROR) ^W^ :5»o+' INPUT (PILOT ERROR) / <> [> TO /U/3 SHAPE ^1 (308 KC) a ^2(2064 KC) ^3 (7266 KC) TO Vf SHAPE Fig. 2 — Schematic of regulation system computer. THE L3 SYSTEM — EQUALIZATION AND REGULATION 847 deviation and some constant part of the equalizer characteristics. If this is not the case the residues can produce new variable deviations shapes and thereby lead to increased complexity in following stages of equalization. As previously described, equalization of both gain and loss is required and this implies active equalizers. Although a number of methods were studied, noise and modulation requirements led the L3 system to the so-called ''block of gain-block of loss" design. The equalizers are passive networks and in order to realize both gain and loss adjustability, the normal setting loss must at least equal the total amount of gain ad- justment. The various equalizers are combined into two to four groups whose loss is compensated by corresponding numbers of flat gain amplifiers.^ The required number of such blocks of loss and gain depends upon the amount of system gain variation to be equalized. The determination of the shapes and magnitudes of system variations is an important system problem. Large amounts of study are necessary in order to evalu- ate the system sensitivity to various changes; the determination of the magnitude of these causes, such as temperature variation, aging rates, etc.; and the determination of maintenance intervals that provide an economic balance between maintenance expense and system cost. These must be studied in detail to provide the equalizer designer with shape and range data for his dynamic equalizer designs. The equalization characteristics and maximum ranges for the two most important dynamic equalizers aside from cable temperature, namely, repeater temperature (T) and vacuum tube aging )u/3 are shown on Fig. 3. In order to main- tain satisfactory transmission during a maintenance interval, the dy- namic equalizers must be able to match any characteristic within the maximum ranges and throughout the transmission band to within 1 to 2 per cent. As noted previously the necessity for simultaneous delay equalization requires the dynamic networks to also make at least an approximate correction in the out-band region. This is shown on Fig. 4. The control element in the equalizers is a thermistor whose available resistance range is 30 to 1050 ohms. (V7 ^^ ^^^^ amplifier^' ^ 125 to 2000 ohms). Thus the regulation ranges shown in Fig. 3 are realized using this one variable resistance element in each equalizer. The following is a summary of the dynamic equalizer requirements. 1. Provide symmetry in regulation characteristic. 2. Minimize flat loss. 3. Match prescribed gain variation to a high degree of accuracy within transmission band. 848 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 w) 2 S 0 z < X u _ -1 to 3-3 / / > / \ / / / / / / / \ / / V V / / U/3/ ■\ X \ \ V^ k \ \ 2 3 4 5 6 7 8 FREQUENCY IN MEGACYCLES PER SECOND Fig. 3 — Regulation shapes and ranges for tube aging (miS) and repeater tem- perature (T). The curves given can occur as either gain or loss and apply to 30 repeaters. 4. Provide satisfactory equalization of in-band delay distortion due to out-band gain change. 5. Each equalizer controlled by a single variable resistance element. 6. The desired performance to be realized in 75-ohm circuits. Dynamic Equalizer Design The design used is the structure commonly known as the Bode Regu- lator, • and shown in block schematic form in Fig. 5. An ideal regulating 4 6 8 10 12 14 FREQUENCY IN MEGACYCLES PER SECOND Fig. 4 — Match to out-band gain change to reduce in-l)M,nUH, /U/LIF Fig. 7 — Circuit of the regulating network for tube aging. THE L3 SYSTEM — EQUALIZATION AND REGULATION 853 3. The controls are easily adjusted using methods to be described. 4. If better equalization is required at any point the existing equahzer is retained, additional harmonic terms are added. 5. The major portions of the equahzers require only one value of inductance and two values of capacitance to form the delay Hues used in the networks. This also assists in the application of distribution re- quirements. 6. The manual equalization can be designed with a minimum of information about the system characteristics. 7. The equahzation is on a least square error basis rather than mini- mum peak error. The networks used to realize these cosine shapes are constant re- sistance Bode regulating networks^ employing second degree all-pass sections. If the phase of the all-pass sections were made proportional to frequency, the transmission performance within the frequency band of interest would be that provided by a Fourier series composed of cosine terms in the variable oj. By appropriate choice of the all-pass sections the frequency-phase relationship can be warped to give greater weighting to a specified portion of the frequency range. The phase of the all-pass section is given by * = -»'-[l(7-s)]- »>> Small h and high fc weights the low frequencies, the Unear phase case h = 1.2, fc = 10.2 mc gives uniform weighting and large h weights the high frequencies. A 6 = 2, /c = 13.75 mc, was selected for the L3 equal- izers because it weights somewhat the higher frequencies where the television signal is transmitted and second, each unbalanced bridge T network section can be constructed with only four elements, two like inductors and two capacitors. For 6's smaller than 2, coupling between the two like coils is required, and for 6's larger than 2, an additional element is required. The all-pass networks are designed on a 75-ohm impedance level and thus the flat, normal setting of each regulating network is 4.18 db maximum. The 75-ohm level makes the series and shunt networks iden- tical and also facihtates manufacturing testing. A special dual variable resistor is used as the control element. It has a resistance range of 15 to 375 ohms and provides a regulation range of ±2.78 db maximum. A schematic of this network is shown on Fig. 8. In the case of the 0 harmonic, flat gain, the phase sections are omitted. For the nth harmonic 854 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 term, n sections are used in both the series and shunt arms. The constant resistance structure permits the complete equaUzer to be formed by a cascade of such networks without interaction effects. This provides a loss characteristic given by Loss = ko -\- ki cos 2^ + ^2 cos Ayp + ^3 cos 61/^ + (22) where ^ is the phase of the individual all-pass section which goes from 0 to 90 degress between 0 and 8.5 mc. The /c's are adjusted by means of the dual adjustable resistors. Note that each term of the series is repre- sented by a corresponding equalizer. Application of this mop-up polynominal to a large number of L3 deviation characteristics indicates that considerably less range than the maximum ±2.78 db will be required for most of the cosine terms. Fig. 9 illustrates this convergence of the series for the eight largest amplifier manufacturing variations. This and other studies show that after the first three harmonics the range may be reduced. It can be shown, ^ for example, that if the system gain deviations are finite within the range of interest (interval of convergence) the coefficients of the approximating polynominal will decrease in magnitude linearly pro- portional to the number of the terms, that is, the nth coefficient will be smaller than some constant divided by n. If the deviation characteristics are continuous and hence have finite first derivatives, then the coeffi- cients of the apprximating polynominal will decrease as the square of 77.15 0.868 OJ I n SECTIONS ^ k 121.3 ■vw- x 308.6 75 75 45 3 0.868 0. 308.6 ^^M^^Y'm-^ " SECTIONS OHMS, /iH»/y>UF Fig. 8 — Circuit of a cosine equalizer without dissipation correction. THE L3 SYSTEM — EQUALIZATION AND REGULATION 855 the number of the term. If all derivatives are finite the coefficients will decrease exponentially. This last case is believed to describe the con- vergence for at least most of the L3 system shapes. Thus for the high order terms that require little range it is possible to reduce the flat loss of the networks. In order to reduce the flat loss without changes in the 75-ohm im- pedance level of the phase sections or in the dual adjustable resistors, pads are inserted between the resistance T and the all-pass sections. In this manner the loss could be reduced to 2.2 db for d=0.5 db range if it were not for the dissipation in the all-pass sections. This dissipation is due to the coils and increases with frequency. It tends to produce a reduced cosine amplitude in the high frequency part of the band. To correct this effect, the pad mentioned above is actually made an equalizer section whose loss change with frequency corrects for the dissipation in the coils thereby yielding cosine amplitudes independent of frequency. The price of this is an increase of the flat loss to 3.4 db for ±0.5 db range. The circuit of the term 10 network is shown on Fig. 10. The range 0.8 - \ / "^ 0.4 0.3 0.2 o 0.1 z V \ \ A '\ s. \ - \ H 0.06 - \ A \ A 2 0.04 \ A \ \ A 0 01 \ D 2 X 5 B 1 0 1 2 1 4 1 6 Fig. 9 — Range required for various cosine terms due to expected manufactur- ing variations of eight critical elements. The magnitudes are based on RSS addi- tion in 25 repeaters. 856 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 of this term is ±0.5 db as is that of all higher terms (to 24). Terms 1, 2, 3 have full range, 4, 5, 6 have 1.5 db and 7, 8, 9 have 1 db. The O term, flat gain, is also 1 db because additional flat shape is obtainable from the flat amplifiers used to make up for the equalizer loss. The shapes provided by the first three terms are shown on Fig. 11. Note that the shapes are cosines of a warped frequency variable and that on this warped scale the shapes are orthogonal. In all, 24 such harmonics plus flat gain are used in the high frequency line for combined systems. For all telephone use only 14 harmonics plus flat gain are required. Fig. 12 shows the construction of one of the cosine networks. These are mounted in groups of five as indicated in Fig. 13. The fixed equaUzer and regulating networks are mounted to the rear of the cosine assembly. Harmonic Adjusting Set To make a mental harmonic analysis of a complicated gain char- acteristic is difficult if not impossible. Therefore a special cosine-equalizer adjusting set has been developed which eliminates trial and error from the adjustment process and which leads to a unique optimum adjust- ment. Broadly the method consists of using sweep frequency methods to convert the gain-frequency characteristic into a repetitive voltage- time function. Gain cosines on the warped frequency scale are converted to voltage cosines of time and the audio harmonic-spectrum components IN Fig. 10 — Cosine equalizer circuit (tenth hMrnioiiic) showinjj; dissipation and range corrections. THE L3 SYSTEM EQUALIZATION AND REGULATION 857 -2 .-'^'' — ^ .-"■ \. y -'-' \ .-" ,,'' COSINE ^ EQUALIZER 1 \ .---^ — -3 ^^^'■" / ^ \ • \ \ / v f V / A COSINE EQUALIZER 2 /\ \ / \ \ y ^-^ y' '> >»^ 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 FREQUENCY IN KILOCYCLES PER SECOND Fig. 11 — Shapes introduced by the first three cosine harmonics. are individual measures of equalizer control-setting errors. By the ad- justment process these audio harmonics are removed thus yielding a gain characteristic describable in terms of only the higher cosine com- ponents not available to the equalization operator. The operation can be explained using the block diagram shown on Fig. 14. The sweep oscillator sends a constant level, variable frequency, over the line and through the cosine equalizer to the detector. The output of the detector on terminals x-x at any instant is a measure of the trans- mission of the line and equalizer at the frequency being sent by the sweep oscillator at that instant. The sweep frequency starts at zero and sweeps to 8.5 mc in a period ^i. As shown on Fig. 15 the frequency- time relationship is warped to correct for the warping of the equalizer Fig. 12 — View of single cosine term network. ^ ^ tf tf tf Fig. 13 — Assembly of five cosine terms. A "C" equalizer requires live of tliese assemblies, a "B" equalizer, three, and an "A", two. 858 THE L3 SYSTEM — EQUALIZATION AND REGULATION 859 phase-frequency relationship. The sweep oscillator therefore scans at a linear rate in cosine degrees vs time. Upon reaching 8.5 mc the sweep reverses and returns to zero. If the cosine shapes were linear cosines of frequency the sweep would be a triangular wave. Assume that the line is perfectly equalized except that the first har- monic term is misadjusted. As the sweep goes from 0 to 8.5 mc the voltage at x-x follows the first harmonic curve of Fig. 16 from 0 to ^i . When the oscillator scans back to 0 the voltage at x-x follows the first harmonic curve from ^i to 2 'i . Then the cycle repeats. The voltage at x-x thus becomes a pure cosine of time oscillation of frequency, }4ti . Although the equalizer shapes are actually cosine on a decibel rather than an amplitude basis this has little practical effect because for small deviations the two are nearly identical. If instead of a first harmonic error the second or third cosine harmonic DIODE DETECTOR ^1 . ""^ SWEEP OSCILLATOR L3 LINE COSINE EQUALIZER < — HARMONIC ANALYZER i\ 1 X V ^ \ — tON POWER 1/ J METER Fig. 14 — Block diagram showing the method used to adjust cosine or other orthogonal equalizers. is misadjusted, the voltage at x-x will follow the appropriate curve of Fig. 16. The dc component measures the zero harmonic but in practice only the ac components are measured and the flat gain is set as a final step to make the pilot levels correct at the line output. If it takes 0.01 seconds for the oscillator to scan up and back, the first harmonic pro- duces 100 cps output from the detector. The second harmonic equaUzer produces 200 cps, the third 300 cps, etc. Therefore at the output of the detector there exist a set of audio frequency harmonics whose amplitudes are a measure of the equalization error of the setting of the cosine con- trols of corresponding periodicity. These harmonics can be separated by convention filtering techniques, for example, by an audio tuned detector or harmonic analyzer as noted on Fig. 14. The analyzer can be tuned to 100 cps and the first harmonic control rotated to remove the 100 cps component. Then tuning to 200 cps the second control is operated, etc. This null method is similar to bridge balancing and may be instrumented to similar high precision. After all of the harmonics corresponding to equalizer controls have 860 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 been removed the process is complete and the equalization residue must be composed solely of those terms not provided by the equalizer. The harmonic analyzer method requires tuning or switching and thus to simplify the method still further the actual field equipment uses a power indicator in place of the analyzer as noted on Fig. 14. Given a spectrum of signals of differing frequencies, removing any one reduces the total power. Therefore the entire spectrum may be applied to a power indicator and the reading reduced by adjusting the various equa- lizer controls. In practice this process is assisted by filtering out the high harmonics which cannot be equalized. This reduces the total power and increases the ease of reading the meter. While the method has been demonstrated using an ordinary 60 cps wattmeter, an electronic watt- meter is used in the field equipment. TIME Fig. 15 — Method of scanning the system gain characteristic to convert cosines of frequency into cosines of time. While the receiver unit can be quite simple, the sweep oscillator is complicated by such things as circuits to control the warping and to hold the sweep limits accurately. In practice the sweep is between 0.3 and 8.5 mc and is at a 37 cps rate. Further there are six pilots on the system and although the dynamic regulators at the adjusting point are paralyzed during the cosine adjustment the pilots to intermediate regu- lators must not be disturbed. Thus the sweep frequency is shifted very rapidly through the pilots. When the sweep frequency gets within about 25 kc of a pilot it is shifted suddenly to the other side of the pilot fre- quency. This materially reduces the interference to the pilot without producing transients in the receiver. While other methods of cosine equalizer adjustment were tested and found to work satisfactorily, the above method was found to be superior. In addition, removal of the filtering permits the power meter to read the rms equalization error and thus the equalization operator can deter- mine the quaUty of the job and observe whether the state of the line is THE L3 SYSTEM — EQUALIZATION AND REGULATION 861 satisfactory. Also the power method is usable with sawtooth as well as triangular scanning and, further, works on any set of orthogonal gain or delay shapes. Thus the basic equipment is readily adaptable to the adjustment of D equalizers if their gain and delay shapes are orthogonal. FIELD PERFORMANCE In the L3 system the function of the dynamic equalizers is solely to prevent excessive deterioration of the transmission characteristic from one manual line-up to the next. During the manual adjustment the dynamic networks are held at that point in their range which minimizes the probability of running out of range in either direction before the /SECOND HARMONIC THIRD HARMONIC ti 2t, 3t, Fig. 16 — Detector output produced by scanning the first three cosine shapes. next manual line-up. This is done to hold the dynamic ranges to a mini- mum. Thus the transmission errors remaining at the completion of a manual line-up are chiefly due to the fixed and manual equalizers. It should be noted however that, when the dynamic regulators are re- stored to operation after the manual adjustment, any residues will be seized by the dynamics and regulated. As yet the field experience with the regulation system and the dynamic shape performance is quite limited. However, the stability of the regu- lation system and the action of the computer have been well established. It would appear that the major remaining regulation system problem will be the determination of the cause shapes to the accuracy required for long television systems. Somewhat more experience has been gained with the cosine equalizers. Using a fixed equalizer design based on ten line amplifiers from initial production, a 100-mile circuit equalized using 15-cosine terms yields the residues plotted on Fig. 17. The ripple at the extreme high end of i 862 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the band is largely due to the failure of the fixed equalizer to match the very sharp cut-off of the line between 8.35 and 8.50 mc. Since telephone channels are not transmitted in this region and since television require- ments are less severe at such high video frequencies (4.2 mc) this effect does not limit the transmission quality. Also note that long television circuits, over 400 miles, will have a further level of equaUzation, D. There is no visible impairment of ordinary television pictures due to insertion of the 200-mile L3 line in their path. With critical types of test patterns there is a slight effect. While there are problems yet to be solved before 4,000-mile transmission can be obtained, the 200-mile performance is most encouraging. In the past the adjustment of manual equalizers has often been a difficult and time consuming task. The cosine adjusting set described o.a - l^ .r- -\> ^ I -0.2 _ ^"v-y A J 0123456789 FREQUENCY IN MEGACYCLES PER SECOND Fig. 17 — Final gain characteristic of a 100 mile L3 line after equalication with 15 cosine shapes. previously appears to have made sizable inroads on this problem. The adjustment of the 25 controls used for television is a three minute job. The fact that the equalization operator is working toward a unique solu- tion and therefore knows when he is done appears to be of material value. Regulation The I^ regulation system is in many respects, similar to the LI sys- tem. However, the design of a stable regulation system for over five hundred regulators in tandem introduces unusual problems. Also the accuracy and stability requirements have led to the use of novel regu- lator circuits including, for example, an analog computer as an element of the system. Six pilots are used, 308, 556, 2,064, 3,096, 7,266 and 8,320 kc. These frecjuencies were selected to l)est measure the anticipated system changes as restricted by where signal allocations would permit their insertion. THE L3 SYSTEM — EQUALIZATION AND REGULATION 863 For example, the wide gap from 3,096 to 7,266 is largely due to the difficulty of inserting and removing pilots in the lower video frequencies of television signals. The problem of finding a satisfactory set of pilot levels and frequencies which will at the same time, be compatible with the desired signals is an important part of the system design problem. ^ The change of four-mile cable loss with temperature is so large (±1.2 db) that regulation is required at each repeater. It takes three months or more for the cable loss to change 2 db but the normal line maintenance interval is of this order. A gain error of this magnitude could not be allowed to accumulate over very many repeaters before the signal to noise performance of the system would collapse. Other effects such as vacuum tube aging and repeater temperature changes can be allowed to accumulate over as many as 30 repeaters before regulation. These facts dictate the location of regulators in the system. At each repeater there is a ''line" regulator controlling a square-root-of -frequency-shape regu- lating network. Then at equalizing points and dropping points "office" regulators correct for the remaining effects. CHAIN ACTION Pilot controlled dynamic regulators derive much of their advantage from the fact that they prevent gain changes from accumulating from repeater to repeater. This advantage is one manifestation of what might be called the ''chain action" of a series of regulators. There is however a corresponding disadvantage, disturbances of the pilot cause the accumu- lation of unwanted gain fluctuations. In previous systems this disad- vantage has been aggravated by positive envelope feedback (l-MiS less than one), at some frequencies, an effect known as "gain enhancement". In the L3 system the "gain enhancement" is nearly negligible but the television requirements still require careful control of certain types of gain fluctuations. The advantage noted above can easily be demonstrated by a simple example. Consider a chain of regulators each having 20 db envelope feedback so that pilot level changes are reduced by 10 to 1. Now con- sider what happens if each cable section changes loss by one db. Table I illustrates the action. The first regulator inserts a gain change of 0.9 db in response to the 1.0 db input change. The 0.1 db error increases the input change to the second regulator to 1.1 db and it therefore inserts a 0.99 db correction. The total resultant error of 0.11 db adds to the change at the third regulator input, etc. Simply stated: The error of the first regulator rides through the system forcing the other regulators to make an accurate 864 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Table I Regulator Number Input Pilot Change Inserted Correction Output Pilot Change 1. 2 3 4 db 1 1.1 1.11 1.111 db 0.9 0.99 0.999 0.9999 db 0.1 0.11 0.111 0.1111 correction. Actually, of course, the above statement is oversimplified but it should be clear that the effective feedback of the regulation system is the (voltage) sum of the feedbacks of the individual regulators. Thus 100 regulators each having 20 db of feedback tend to act like a single regulator having 60 db of feedback. The rigorous treatment of these effects will be developed later. Fig. 18 shows a block diagram of a regulator in a form intended to indicate the feedback structure. The feedback loop includes a pilot pickoff filter, amplifier and rectifier. This converts the output pilot level into a dc voltage. The ''battery", which is the actual input signal for the circuit, represents the equivalent of the desired pilot output level. The signal applied to the dc amplifier is a dc signal representing the error in pilot level. This dc signal is, in effect, converted back to a pilot level by the action of the regulating network and its modulation of the input pilot level. Thus changes in input pilot level are equivalent to gain changes in the m circuit of the feedback structure and are resisted by feedback action just as in any other feedback ''amplifier". It is also valuable to note the respective m and jS roles played by the various com- ponents since the stability requirements, etc. then become clear. For DC AMPLIFIER -=- INPUT \> s PILOT INPUT REGULATING NETWORK DIODE DETECTOR PILOT AMPLIFIER :<] PICK- OFF FILTER OUTPUT Fig. 18 — Block diagram of regulator showing the feedback structure. THE L3 SYSTEM — EQUALIZATION AND REGULATION 865 example, the pilot amplifier is in the beta circuit and therefore must be a highly stable device. On the other hand, the dc amplifier is in the /x circuit and its drifts are reduced by the loop feedback. Having developed the feedback nature of the structure and the roles of the components, the conventional feedback art can be used for the analysis of the individual regulator. One can show that : Change in output pilot level 1 Change in input pilot level 1 — Mi^ (23) System gain change to pilot frequency _ n^ . . Change in input pilot level 1 — Mi3 This result is not very surprising but it can be used to determine the performance of the following regulators in a chain. Many different cases must be considered. Sometimes the pilot levels change because of an effect distributed all along the system. In other cases the change occurs only at the input to the line. Sometimes the pilot level changes are the important effect. In other cases the importance resides in the gain change to the signals. In all cases the results may be complicated by the fact that fjL0 is, in general, a complex number and thus phase as well as amplitude is important. Adopting the notation : APin = fractional change in input pilot at nth regulator, APon = fractional change in output pilot at nth regulator, AGn = gain change to signals (near pilot frequency) of nth regulator, and AGt = total system gain change =^ Gm one can readily show the following : Case I — Disturbance of pilot only at input to system : APqU ~ap7i fe ^ = f^-Y - 1 f APu \l - y?) Case 2 — Equal gain change in each regulating section. \ APc = fractional gain change of section (27) APo„ « = 1 \( ^ X _ l] (28) THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 AGt AP. As an example of the application of these formalas consider the effect of television induced compression. The presence of the television signal reduces the gain of the line amplifier. The effect is small but cumulative. In the absence of regulator action it merely compresses the television signal slightly and makes a negligible change in the contrast rendering of the picture. However the regulators observe a gain change to the pilots and attempt a correction. The very rapid changes are ignored but 60 cps, for example, is partially corrected. This introduces a 60 cps gain change which will lag the picture and therefore must meet 60 cps bar pattern requirements. This problem is solved by keeping the regulator response low at 60 cps. For /xjS of —70 db, 90 degrees, at 60 cps a chain of 700 regulators will insert a total gain change approximately one tenth that of the total compression. If the m/3 were allowed to approach — 50db the total gain change would equal the compression and certain types of pictures would be degraded. The above example brings out one of the important facts: When designing regulators for long systems the )U/S characteristic must be care- fully controlled to losses much higher than is customary in amplifier design. In a conventional feed-back-amplifier loop-cutoff the magnitude and phase of /u/S is no longer of much interest after it drops below — 10 db. In L3 regulators the loop is of vital interest to losses of the order of 70 db. This is, of course, largely due to the fact that the chain action increases the effective system feedback by nearly 60 db. Thus the over- all system is similar to conventional amplifier practice. Loop gain (ai/3) and feedback (l-n^) characteristics for the line and office regulators are shown on Figs. 19, 20, 21 and 22. Note that 1,000 line regulators in tandem give an over-all gain enhancement of only 1.2 db. This would be even less if it were not for a 100 cps roll-off in the dc amplifier to reduce noise. One hundred office regulators give 0.8 db gain enhancement even with their 20 cps roll-off. THE DYNAMIC LINE REGULATOR As indicated in Fig. 23 a crystal filter is used to pick the pilot off the line in the presence of the other signals. The filter impedance goes THE L3 SYSTEM — EQUALIZATION AND REGULATION 867 0 \ -20 \ . -40 -60 \ ^ ^ ^ -80 \ 2 5 0.1 FREQUENCY 2 e. ,0 2 5 IN CYCLES PER SECOND Fig. 19 — Loop gain characteristic for the line regulator. through resonances due to the crystals and introduces a gain character- istic on the hne that cannot, in practice, be equahzed. Thus it is neces- sary to hide the filter from the line with loss. To hold the transmission distortion of signals to 0.15 db with 500 regulators (0.0003 db per regu- lator) requires a voltage loss of 23 db with the filter impedance changing by large factors from its nominal 25,000-ohm level. The power loss bridging on the 37.5-ohm (75 into 75) circuit is, of course, much greater. The 7,266-kc pilot level at the line amplifier output is —16 dbm into 75 ohms. The filter and pad loss totals 24 db (voltage ratio) leaving an .-7^ 6 lU 4 a z I 0 \ y \ \ \ \ \ - GAIN ENHANCEMENT ^ T 5 6 20 30 40 60 80 100 200 FREQUENCY IN CYCLES PER SECOND 400 600 Fig. 20 — Feedback characteristic for the line regulator showing the gain en- hancement effect. THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 40 20 !3 0 UJ ffl o ai o -20 z ^ -40 -80 200 100 ^ D.001 2 5 0.01 2 5 0.1 2 5 i.o 2 5 iq 2 i FREQUENCY IN CYCLES PER SECOND Fig. 21 — Loop gain characteristic for office regulators. 100 available pilot signal of about 0.002 volts in 25,000 ohms. In order to solve drift and stability problems in the dc circuits the pilot is converted to a dc voltage of 60 volts. This requires an amplifier-rectifier of 90 db voltage gain, stable with time and temperature. As indicated on Fig. 23, the amplifier consists of three stages using ai m 2 o xio' -2 \ \ V GAIN ENHANCEMENT f^ \ y ^ X \ / \ V^ _L_ -L -JL. 4 -6 -8 2 3 4 5 6 8 10 20 30 40 60 100 200 400 600 FREQUENCY IN CYCLES PER SECOND Fig. 22 — Feedback characteristic for office regulators showing the gain en- hancement eflfect. THE L3 SYSTEM — EQUALIZATION AND REGULATION 869 --^W^^--T^y^Ar ^^^^> — Vv\ — I -^WV-P 1 f 1- a o c3 a f i u. I 30 870 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 403B Gong life 6AK5) tubes. Feedback is taken shunt-shunt, output plate to input grid, to stabilize the input and output tuned circuits against Q changes with temperature. The regulators are designed to operate from —20° F to +160° F and the amplifier gain does not change by more than 0.3 db over this range. The beta circuit contains an adjustable condenser divider for field gain adjustment. The tuned sections are heavily damped to get tem- perature stability but nevertheless compensate for the cutoffs of the /x circuit. The resultant loop gain and phase are shown in Fig. 24. It will be noted that the phase margin against singing is rather large, about 60 degrees. This permits tube replacement without retuning as well as protection against tuning changes due to time and temperature. Grid-plate capacity places a limit on the permissible interstage im- pedance for stability. Thus the output circuit between the third stage and the rectifier is operated as a reactive transformer giving a voltage step-up of 2. This increases the output voltage obtainable without raising the impedance facing the third tube above 12,500 ohms. About 16 db of local dc feedback is used on each stage to stabilize the 28 24 20 ffi 12 1 1 A! ^^"•^^ '"PHAS \ A \ \ \ \ \ V \ \ \ 1 \gai N / 1 ) \ \ / / 1 1 ^-^. "l V ^y / » ■ 7266 O LU Ij-24 -28 -32 -36 ^ ^ It !66 Kc\ / \ / \ i / \ / \ / \ ^ 1 / \ / \ \, / 6.4 6.6 &0 8.2 6.8 7.0 7.2 7.4 7.6 7.8 FREQUENCY IN MEGACYCLES PER SECOND Fig. 25 — Regulator selectivity without the crystal filter. cathode current. In so far as trans-conductance depends upon cathode current, trans-conductance changes are reduced. This is of material value in further reducing the effects of vacuum tube aging. Fig. 25 shows the external gain of the 7,266-kc amplifier without the crystal filter. The filter response is shown in Fig. 26. The relatively wide 3 db bandwidth of ±1.5 kc is to reduce the contribution of the filter to the gain enhancement problem. The large rejections to fre- quencies further removed from the pilot prevents operation of the regu- lator by signals other than the pilot. Also note that for very strong signals such as the television carrier at 4,139 kc the filter is aided by the am- plifier selectivity. The diode detector is also designed to reduce the effects of inter- ference. The time constant of the detector is made short so that the output will follow envelope fluctuations up to about 40 kc. Thus the output of the detector in the presence of an interfering signal within 40 kc of the pilot becomes the power sum rather than the voltage sum of the two signals. This is readily understood from Fig. 27. Here Ej, is 872 THE BELTj system TECHNICAL JOURNAL, JULY 1953 56 O 40 z 2 3 24 13 ,6 u S B 0 / / / \ / \ / \ / / MIDBAND LOSS = 1 DB \ I \ 1 7230 7235 7240 7245 7250 7255 7260 7265 7270 7275 7280 7285 FREQUENCY IN KILOCYCLES PER SECOND Fig. 26 — Loss characteristic of the 7266 kc crystal filter. the normal rectified pilot. In the presence of the interfering signal Ei a diode detector with a long time constants will deliver a dc output of of Ep + Ei and thus give voltage addition between the pilot and the interference. With a fast time constant the detector output can follow the nearly sinusodial envelope variations and the dc level is changed only sUghtly. The ac component may be suppressed by the cutoff of the dc amplifier, but, even if this is not the case, the thermistor being a thermal device responds to the total power rather than the peak am- plitude. Thus the diode time constant is made long enough to hold over a few cycles of the pilot frequency but short enough to follow the im- portant interference difference-frequencies. The dc amplifier consists of three triode sections essentially in parallel a .^-sr Ep + E-, . i TIME Fig. 27 — Diode detector output in presence of interference Ei. Ep is normal pilot signal. Curve a obtains with long time constant, b with short time constant. THE L3 SYSTEM — EQUALIZATION AND REGULATION 873 but with the biases and feedbacks differing in order to provide an EI characteristic that corrects for sensitivity changes of the thermistor with operating current. This maintains the overall loop feedback rela- tively constant over the 1 to 20-ma current range. The thermistor is directly heated by the plate current of the dc amplifier in order to obtain single time-constant performance of the thermistor. The thermistor transmission, plate current changes as an input and pilot level as an output, is the main frequency characteristic of the regulator loop. LINE THERMOMETER REGULATORS It is possible to dilute the regulation system with less costly, less ac- curate regulators without undue loss of overall performance. This is •190 o MANUAL PILOT LEVEL CONTROL AMBIENT COMPENSATION REGULATING THERMISTOR BURIED THERMOMETER THERMISTOR Fig. 28 — Thermometer regulator schematic. accomplished by the use of thermometer regulators at alternate regu- lating points. These consist of a thermometer thermistor buried in the ground, electrically in parallel with the regulating thermistor. The circuit is quite simple as indicated by Fig. 28. Ground temperature changes vary the resistance of the thermometer thermistor thereby changing the current and resistance of the regulating thermistor. The manual control is used to effect initial alignment of the system. The regulating sensitivity is designed to slightly overcompensate for cable loss changes in order to somewhat ease the burden on the following dynamic regulator. AMBIENT TEMPERATURE COMPENSATION Both types of line regulators require the assistance of ambient tem- perature compensation of the regulating thermistor. Conventional com- pensation circuits would hold the thermistor resistance within about 20 per cent but this would produce an error of one db at a thermometer 874 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 regulator and about 0.2 db at a dynamic unit. Thus an improved com- pensation scheme was required which would connect only to an indirect heater, the bead itself being already controlled by dc heating from the regulators. The compensation circuit adopted is shown on Fig. 29. A second thermistor called the compensating thermistor is mounted in the same glass envelope with the regulating thermistor. The fixed resistor Ri and the compensating thermistor together with transformer T form a bridge which is made a feedback path for tuned amplifier A. The feed- back is positive when the compensating unit is cold so oscillation begins at the tuning frequency (4 kc). These oscillations heat the thermistor and tend to bring the bridge into balance. The bridge stabilizes at a COMPENSATING THERMISTOR Fig. 29 — Ambient temperature compensation oscillator. small unbalance just sufficient to yield a loop gain of unity. The level of oscillation is forced to that value which will maintain this small un- balance. Any changes in balance thereafter produce deviations of loop gain from unity and the oscillation level increases or descreases until the equilibrium is reestablished. Thus the level of oscillation changes with temperature but the resistance of the compensating thermistor is held constant. Because the circuit supplies nearly perfect temperature compensa- tion to the unit in the bridge a suitable fraction of the oscillator power may be fed to the heater of the regulating thermistor to achieve very close compensation of it. Resistors R2 and Rz are adjusted in manu- facture to correct for slight differences between the two thermistors. Note that the compensating thermistor is provided with an unused heater to match the thermal properties of the two units. The amplifici- consists of a single 403B tube with 20 db dc feedback for current (and transconductance) stabilization. This feedback is vital THE L3 SYSTEM — EQUALIZATION AND REGULATION 875 because it also introduces a slight compressive action in the amplifica- tion of the tube and thereby prevents rapid wild changes in oscillation level. Bypassed dc feedback on an amplifier causes dc second order dis- tortion to increase the bias and thereby reduce the transconductance. This effect overcomes the tendency of the third order distortion to create expansion in this particular tube. If the thermistor response were fast compared to the reciprocal of the bandwidth of the amplifier the compression action would be unnecessary. However with an audio frequency amplifier and a 100 second thermistor the compression is essential in preventing motorboating. The field limits on the compensation of the regulating thermistor over the range -20 to +160 degrees F are ±3 per cent in resistance due to all causes including manufacture and aging. Specific units can be ad- justed to yield compensation to a fraction of a per cent. OFFICE REGULATORS The L3 office regulators are similar in design to the line regulator. However the office regulators operate their regulating networks via an analog computer and, of course, a variety of pilot frequencies are em- ployed. Because signals are dropped at offices, higher loop feedbacks are used to insure accurate equalization. However, temperature variations are smaller and conventional thermistor ambient temperature com- pensation is adequate. Also the smaller number of office regulators per- mits less isolating loss for nick effect (except for the 7,266-kc oflSce regulator). The lower levels of the pilots (except 7,266) are compensated by re- duced isolation loss, (12 instead of 23 db), and reduced detector level (40 instead of 60 volts). Thus the gain required is not sustantially in- creased. The pilot amplifier design is therefore different primarily in the tuning frequency and in the simplifications in the lower frequency units permitted by the higher permissible interstage impedances. The diode detector feeds a cathode follower to obtain the dc voltage representing the deviation of the pilot from its assigned value as a low impedance source to feed the computer. The appropriate signals from the computer are fed to the dc amplifier. This amplifier differs from that used in the line regulators in that (1) a push pull input is provided, (2) higher gain is required to produce greater feedback, 30 db, and over- come computer losses, 5 db and (3) the output stage supplies somewhat higher currents, 1 to 30 ma, (except 7,266) because the regulating net- works use a lower impedance thermistor. 870 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 OTHER REGULATOR FUNCTIONS The regulators are also used for alarm and pilot indicator functions. At line dynamic regulators the current flowing through the diode de- tector load resistance is also passed through a relay to obtain an alarm indication whenever the pilot level deviates from normal by more than 3 db. At offices similar arrangements operating on a 2 db error are pro- vided both for alarm and switch initiation purposes. If any pilot deviates by 2 db the service switches to the spare line. In addition fast switch initiation is obtained from the 7,266-kc regulator by direct connection to the detector output. This arrangement avoids the time delay of the relay operation. For pilot level indication the diode load current is read on appropriate meters. This avoids the necessity of providing separate pilot level indicators and is possible because of the reliability and stabihty of the regulator amplifiers. MECHANICAL FEATURES Figs. 30 and 31 are views of a regulator seen from the wiring side and from the top respectively. The chasis consists of two steel end plates riveted to steel angles, with a punched copper plate screwed to this ""j"!!^ ^ *■!{•'; • .i^4>"-;ii' -3fi^ Fig. 30 — Lino dynamic regulator us seen fioin wiring side. THE L3 SYSTEM — EQUALIZATION AND REGULATION 877 Fig. 31 — Top view of line dynamic regulator showing case. structure. The salient feature of this type of construction is that one universal punched plate can be used for all the regulators (including the thermometer regulator), any individual regulator being fabricated by mounting the necessary component cans and vacuum tube sockets on it. Power wiring and all leads that are not critical as to length or placement can be run in the wiring trough around the edge of the chasis shown on Fig. 30. This eliminates the necessity of lacing the wires into a cable, a significant saving in production effort. The component cases are shown on Fig. 31. They are zinc die-castings and contain network elements assembled on stypol forms which fit inside the cans. One universal case accomodates sixty-four different combinations of elements required by the various regulators. The neces- sary wiring of the individual cases may be completed before assembly on the regulator chassis. This feature also saves production effort. The whole chassis is mounted inside a die-cast zinc housing, and all power and test leads are brought into the regulator through airtight connections. The two parts of the housing when assembled together are made airtight by a rubber gasket which fits into the slot around their inner edges. The general construction features and size of the regulator assembly can best be understood by inspection of the illustrations. 878 the bell system technical journal, july 1953 Acknowledgements Space does not permit listing all of the people who contributed to the success of this work. However, we wish to mention S. A. Levin for his work on cause shapes, R. H. Klie for his leadership in system studies, E. T. Harkless for his study of cosine equalizers, and E. Ley for mechani- cal design of equaUzers. Also mention should be made of C. J. Custer and E. G. Morton for their work on regulator circuits, A. R. Rienstra for studies of gain enhancement, C. H. Bidwell for ''chain action" analysis, and F. R. Dickinson for mechanical design of regulators. References 1. C. H. Elmendorf, A. J. Grossman, R. D. Ehrbar, R. H. Klie, The L3 Coaxial System — System Design, see pp. 781 to 832 of this issue. 2. L. H. Morris, G. H. Lovell, F. R. Dickinson, The L3 Coaxial System — Ampli- fiers, see pp. 879 to 914 of this issue. 3. H. W. Bode, Variable Equalizer, Bell Sys. Tech J., April, 1938. 4. R. L. Lundry, Attenuation and Delay Equalizer for Coaxial Lines, A.I.E.E., Trans., 68, 1949. 5. P. H. Richardson, Variable Attenuation Network, patent 2,096,028. 6. H. S. Carslaw, Introduction to the Theory of Fourier Series and Integrals, Third Edition, 1930, MacMillan & Co., Page 148. 1 The L3 Coaxial System Amplifiers By L. H. MORRIS, G. H. LOVELL and F. R. DICKINSON (Manuscript received April 17, 1953) The line ampU£ers for the L3 coaxial system are designed to compensate for the loss of the four miles of cable which separate the repeaters; the flat amplifiers are used to compensate for equalizer loss and as transmitting amplifiers. The two types are basically similar, consisting of two feedback amplifiers in tandem separated by an inter-amplifier network; in the line amplifier this network is variable and is automatically adjusted to com- pensate for variations in cable temperature and for small deviations from the nominal four-mile spax^ing. Coupling networks employing high-precision transformers are used to connect the amplifers to the coaxial cable through the required power sepa- ration filters. The low impedance windings of the transformers are center- tapped and a balancing network provided in order to match the cable im- pedance over the transmitted frequency band. The amplifiers are equipped with plug-in tubes of high figure of merit which were developed for this application. A double-triode output stage is used to obtain improved system signal-to-noise performance. Provision is made for preventive maintenance of vacuum tubes and for a controlled adjustment of gain on an in-service basis. All important components of the amplifier are subject to quality control procedures to assure that the average gain of groups of amplifiers will be held within narrow limits and that individual amplifiers will form a normal distribution around the average. This approach is essential in order to meet system equalization and signal-to-noise objectives. Careful mechanical design and rigid control of the mechanical aspects of manufacture are neces- sary to minimize gain variations which might be caused by variations of parasitic circuit elements and unwanted feedback effects. Special measures were required to keep the temperature rise within the sealed die-cast housing within tolerable values. 879 880 the bell system technical journal, july 1953 Introduction In a transmission system such as the L3 coaxial, the degree to which system objectives are achieved is largely dependent on the quality of the amplifiers which compensate for the cable loss. To a considerable extent the same statement applies to the similar flat-gain amplifiers used to make up for the loss of the equalizers at various points along the route. The development of an amplifier which would meet the exacting re- quirements of the L3 system was in turn dependent on new develop- ments in the fields of vacuum tube design and circuitry, network design techniques, element and network fabrication, and statistical quality control. To these new tools were added the lessons learned in years of manufacture and operation of the preceding LI system. The importance of some of these factors can best be illustrated by examining the implications of the amplifier requirements which follow from the material in the companion papers on system design^ and equali- zation.^ Obviously the figure of merit, modulation coefficients and life of the vacuum tubes will be determining factors in setting the amount of feedback that can be obtained and the signal-to-noise ratio of the system. It is not so inomediately apparent that system requirements could not be met with the present 4-mile repeater spacing if it were not for the use of quality control at every stage of manufacture from elements, and even the raw materials entering into components, to complete amplifiers. As the companion papers show, however, the equalization plan of the sys- tem is predicted on a degree of reproducibility of amplifier gain and delay characteristics obtainable only by quality control applied at every stage of the manufacturing process. The present equalization plan is based on the assumptions that the gain of the average line amplifier will match the loss of the preceding line section to within 0.15 db, and that the average amplifier gain will not vary from one batch of new amplifiers to another by more than about 0.06 db. Under these assumptions a system equalization plan can be worked out which results in reasonable spacing between equalizers, a tolerable signal-to-noise penalty due to misalignment and equalizer loss, and a practicable procedure for adjusting long systems. Any gross de- parture from the basic assumptions as to reproducibility of amplifiers would seriously compromise these objectives, which even now are achieved only by using e(iualization which requires a flat gain amplifier for every four or five line amplifiers. Now it turns out that with the most precise elements that can be made, the gains of individual amplifiers will vary by al)out dbO.O db. We need, therefore, a tool which will permit us to control the gain of the average amplifier to an order of magnitude THE L3 SYSTEM AMPLIFIERS 881 greater precision than we can economically control the individual. This is exactly the effect which modern methods of statistical quality control aim at achieving, and since the entire amplifier is merely the sum of its parts, quality control must start at the roots of the manufacturing proc- ess. In order to apply quality control intelligently, and to be sure that all important causes of gain variation are understood, it has been neces- sary to carry out, side by side with empirical laboratory work, a pro- gram of computing the insertion gain of the amplifier, starting with fundamental element values. These computatuions have also proven of value in obtaining satisfactory stability margins in the design of the low^ and high frequency cut-offs of the feedback loops, and in obtaining preliminary information on amplifier gain deviations for use in equali- zation planning. The severe gain and delay reproducibility objectives also have their effect on the mechanical design of the amplifier. At first sight the unit appears to be a lumped constant structure rather than one in which distributed effects would be of paramount importance. Usually the cir- cuit designer in such a case is interested in the mechanical design only for reasons of neatness and economy, but when we look deeper we find that in the amplifier structure as a whole as well as in the case of certain components, the effects of distributed capacity and inductance could easily defeat our objectives if the mechanical design were not such as to assure precise control of element placement and wiring lengths. For these reasons, an order of mechanical accuracy is specified beyond that w^hich can be justified by the accuracy of transmission measurements on individual amplifiers. This striving for mechanical accuracy is carried all the way through, from element piece parts through subassemblies to the assembly on the final amplifier framework. The logical expectation is that by reproducing the amplifiers as exactly as possible we will con- trol not only the known elements but also the many parasitic effects, and thereby minimize the appearance of shifts in amplifier gain and delay which would be substantial in the system even though diflftcult to detect in the measurement of individual amplifiers. Another design feature of the amplifier based on the system equaliza- tion point of view is the omission of all adjustable elements, or trimmers, to control the transmission. By eliminating gain adjustments, the pos- sibility of adjusting one element to compensate for the short-comings of another, and the possibility of systematic errors of setting due to faulty or inaccurate adjustment techniques, are both eliminated. Either of these possibilities would tend to convert relatively large random effects into smaller but systematic effects, a conversion which would penaUze 882 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the system equalization. Obviously, however, the elimination of gain adjusting elements puts a higher premium than ever on the require- ment that all elements, including tube capacities, show only small random variations about tightly controlled nominal values. In addition to the gain requirements discussed above, the following line ampUfier objectives are to be met: 1. The gain of the amplifier must be continuously variable, under the control of a regulator circuit, to compensate for differences in lengths of cable sections and for variations of cable loss caused by temperature changes. The shape of the gain change should match the square root of frequency shape of the line loss change over the transmitted band to within a few hundredths of a db. This accuracy of shape, which of course is based on equalization considerations, should hold over the entire regulation range, which is about ±6 db at the top transmitted frequency. 2. The input and output impedances of the amplifier should match the cable impedance in order to minimize the effects, on television trans- mission, of echoes caused by line irregularities such as splices. Tolerable values of reflection coefficient at amplifier input and output are about 5 per cent at the television carrier and 10 to 15 per cent at upper band 3. Feedback consistent with system modulation requirements, shaped across the transmitted band to minimize low frequency intermodulation products and to give a smooth, easily equalized shape of gain change as tubes age, must be obtained while maintaining adequate margins against singing. 4. Other requirements include design and selection of elements to assure as small a change of gain versus ambient temperature as possible, mechanical design provisions for keeping the temperature rise within the unit to a minimum both in order to keep the temperature-gain effect small and to obtain long element life, a sealed housing to avoid damage by humidity in exposed locations, and provision of facilities for testing tubes in service. Configuration The circuit configuration of the line ampUfier is shown in Fig. 1. It consists of two independent feedback amplifiers with a regulating net- work between them acting like a two-terminal interstage. Each of the amplifiers is essentially a two-stage circuit — the input amplifier literally so and the output amplifier essentially so since the double- triode output circuit acts like a single stage. Each amplifier is connected to the coaxial through a coupling network which consists of a transformer 883 884 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 60 50 a. < 20 o liJ a. 2 10 i- --, ^ / . ^^^^ 0.2 03 0.4 0.6 08 1.0 2 3 4 5 6 8 10 FREQUENCY IN MEGACYCLES PER SECOND Fig. 2 — Required line amplifier gain. plus gain shaping and impedance adjusting elements. Since there is no feedback around the coupling networks, they directly affect the insertion gain of the amplifier, as do the two beta circuits and the regulating net- work. The required shaping of insertion gain across the transmitted band is obtained and controlled by the design of these five networks. Fig. 2 shows the required amplifier gain for nominal and extreme thermistor settings; these required gains differ from the line loss by the small losses of the associated repeater components. At mid-range thermistor setting about 37 db of gain shaping is needed. The manner in which this shaping is distributed among the five networks has important effects on the feedback which can be obtained and the sensitivity of amplifier gain to element variations. This configuration offers several advantages, one of the most im- portant of which is that the regulating network is between the amplifiers. In most other configurations, the only gain-determining network avail- able for the regulating function is the beta circuit. When the feedback is not infinite, this introduces errors for which it is difficult to compensate, and limits the available feedback by complicating the design. In this con- figuration, the impedances which the amplifiers effectively present as shunts across the regulating network are high and can be allowed for in the design so that the regulation error can be made much smalloi' than the error associated with beta circuit regulation in an amplifier having relatively little feedback. Other major advantages are the superior signal-to-noise performance and the relative simplicity of the feedback loops of this amplifier as compared to alternative designs. the l3 system amplifiers 885 Mechanical Design The mechanical assembly of the amplifier, like the configuration, is divided into three main sections. The input and output amplifiers are mounted on separate chassis which are designed for ready removal from the main base casting. The regulating network is mounted in an enclosed, shielded compartment which also serves to shield the component ampli- fiers from each other. Figs. 3 and 4 show the assembled amplifier. Because of the wide frequency range and close control of parasitic capacity and lead inductance required, the L3 amplifier was designed as an integrated whole, and all networks were designed with, and as part of, the amplifier. In each case circuit elements were placed in space in the"" best possible position for optimum electrical performance, and support- ing structures were then designed to maintain the desired space relation- ships. These supporting structures are made as separate units which mount on the amplifier chassis, so that the networks can be individually tested before final assembly into the amplifiers, and can be removed for repair of replacement if necessary. Heretofore, this method of design has been impracticable because it results in very complicated supporting structures. It was feasible in this case because of the availability of a new type of material, which removes many of the mechanical design constraints. This is a cold casting resin, and parts are produced in a cheap phenolic mold. Since the process is practically equivalent to the sand-casting of metals, complex parts can be economically manufactured even in the relatively small quan- tities required for L3 production. Where necessary to assure accurate ->''i J» Fig. 3 — Line amplifier and housing. THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 location or orientation of parts, the cast resin frameworks are milled or spot-faced. The structure used in the input interstage, shown in Fig. 5, is an example. Circuit elements are wired to pins driven into the casting or, in some cases, to wires imbedded in the material. All wiring can thus be made direct with a minimum of bending and no doubling back, resulting in a reproducible and uniform product. The entire amplifier is housed in a sealed die cast container to protect the components from humidity damage. Dessicant is enclosed in each amplifier. It would have been desirable to make this housing of alumi- num, but the high melting point of this metal makes it impractical for such a large casting to meet the air-tightness requirement. A zinc-base die casting alloy was therefore used. Sealing is accomplished by rubber gaskets at all openings for connections and at the joint between the two parts of the housing. The removable part of the casting which serves as a cover is made as large a part of the total housing as possible, in order to provide maximum accessibility for maintenance of the units mounted on the base. The entire housed unit is arranged to mount on a relay rack mounted panel by means of slides which are self-locking. Signal and power connections are made by means of flexible cords which Fig. 4 — Line amplifier, wiring side. THE L3 SYSTEM — AMPLIFIERS 887 Fig. 5 — Input interstage, illustrating use of case resin frameworks. are available on the panel to plug into the unit after it is in place on the slides. Vacuum Tubes The tubes have been fully described in an earlier paper ;^ their charac- teristics are summarized in Table I. They are plugged into conventional sockets, and single rather than parallel tubes are used in each stage. These are departures from the practice of the LI coaxial system, in which two tubes in parallel were soldered in each stage. The use of sockets increases the parasitic capacities and reduces the obtainable feedback by one or two db, but it was felt that the resulting maintenance economy was worth this sacrifice. With single tubes, the failure of one tube results in a line failure and a switch to the protection line. At first sight, it would appear that using parallel tubes in each stage should greatly decrease the probability of line failure. A study of LI experience, however, showed that most tube failures could either be forestalled by preventive maintenance, or else were of such a nature (for example, shorts within the tube) that the parallel tube would not afford protec- tion against line failure. The reliability advantage of parallel tubes, then, turns out to be small; their use, on the other hand, increases the wiring 888 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Table I — Tube Characteristics, Design Center Values, New Tubes Heater voltage Heater current Plate -cathode voltage . . Screen-cathode voltage. Grid -ground voltage — Cathode bias resistor. . . Plate current Screen current Transconductance Plate resistance Grid-cathode bias. . 435 A 436A Tetraode Tetrode 6.3 6.3 0.3 0.45 180 180 150 150 9.0 9.0 630 315 13.1 23.4 3.2 8.6 16.5 32.0 1.3 1.1 437A Triode 6.3 volts 0.45 amperes 150 volts — volts 9.0 volts 262 ohms 40.2 milliamp — milliamp 47.0ma/volt 970 ohms 1.5 volts Capacitances (Approximate hot values in mmf ) Grid to plate Grid to cathode and screen. . . Grid to heater Plate to cathode and screen . . Heater to cathode and screen. Heater to plate 0.02 0.04 9.2 18.2 0.6 0.8 1.0 1.5 5.4 7.2 1.7 1.9 3.6 16.4 0.1 0.7 5.4 0.2 Modulation* Second order (2F) Third order (3F) Third order "effective' (3F) Equivalent noise resistance, ohms 36 db 67 db 60.5 db * Ratio of product to fundamental at output for a 0.1 volt rms signal from grid to cathode. complexity greatly when sockets are used. The small gain in reliability would not compensate for the degradation in performance caused by the complication of wiring and the resulting capacity penalties. The applied voltages and current drains are shown in Table II. Heat Dissipation III common with many modern designs with high gain vacuum tubes, the problem of heat dissipation was acute in the L3 amplifier. The am- plifier is enclosed in a sealed housing of sufficient size to dissipate readily the heat generated. However, with the usual types of chassis and tulx^ shield construction, vacuum tube envelope temperatures were so high that long tulx* life could not be assured. Consequently a new type of tube shield w.is developed. This shield is of heavy copper tubing equipped THE L3 SYSTEM — AMPLIFIEKS 889 Table II — Power Requijiements of Amplifier Unit 6.3 volts (grounded) 1.5 amperes 6.3 volts (190v off ground) 0.45 amperes + 100 volts regulated 0.5 milliamps + 190 volts 68 milliamps +315 volts 41 milliamps with internal helical springs mounted in such a manner that each turn of the spring makes contact with both the glass tube envelope and with the copper tube as shown in Fig. 6. Thus each turn of the spring provides two metallic heat conducting paths to carry away the heat from the tube envelope. A total of 480 conducting paths are provided for each of the large tubes used in the output amplifier by the use of these springs. Good contact is made between the copper tubes and the chassis, so that heat generated within the vacuum tube is efficiently conducted through metallic paths to the copper chassis of the amplifier. In order not to raise the temperature of the chassis and consequently the temperature of the circuit components, large ribs are provided in the housing, with which the chassis make intimate contact. A continuous metallic con- GLASS VACUUM TUBE ENVELOPE RADIATING SURFACE Fig. 6 — Heat conducting tube shield. 890 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 ducting path is thus provided from vacuum tube envelope to amplifier housing. This design. resulted in a temperature drop in the output am- plifier tubes of about 70° C without any temperature rise in the chassis or circuit elements over that produced when ordinary types of shields were used. The smaller vacuum tubes in the input amplifier do not ordinarily run as hot as those in the output amplifier and a temperature drop of only about 55° C was attained by these methods. Coupling Networks The input and output coupling networks are essentially identical. The low side of each transformer is a balanced center-tapped winding which together with the balancing network acts as a hybrid, to produce a good 75-ohm impedance facing the cable. The use of this type of con- nection gives a signal-to-noise advantage over the use of a brute-force high-side terminating network. The advantage is theoretically 3 db in the case of the output coupling network and would approach the same figure in the case of the input network if tube noise were dominant over the resistance noise of the cable. ^ Aside from the fact that the design of a high-side shunt termination network for an off-ground peaked tran- former is well-nigh impossible, the use of a balancing network in a hybrid connection has the important additional advantage that the adjustment of this network to obtain a good reflection coefficient has negligible effect on the insertion gain of the circuit. A relatively modest share of the total shaping required has been allocated to the coupling networks: 5.5 db each, or 11 db total. One reason for this is that although these networks are outside the feedback loops in the usual sense, nevertheless the impedances which they present to the amplifiers are important factors in the feedback design, and the effects which they produce must not be allowed to become so severe as to limit the feedback to too low a value. It is obvious from inspection of Fig. 1 that only a part of the voltage developed across the input beta circuit by the plate current of the second tube will appear as a grid- cathode voltage to drive the first stage. The proportion of the beta circuit voltage which will be thus effective in producing feedback around the loop will be dependent on the potentiometer division between the impedance of the coupHng network and the grid-cathode impedance of the first tube. The greater the peaking of the input coupling network, the greater its impedance at high frequencies where the grid-cathode capacitive impedance Ls already decreasing, and hence the greater the potentiometer term loss. A similar loss occurs in the output amplifier. The plate current of the output stage divides between the output THE L3 SYSTEM AMPLIFIERS 891 coupling network and the parasitic admittances to ground. The portion of the plate current which returns directly to cathode (ground) through the latter path, without passing through the beta circuit, is not effective in producing loop feedback. A second and even more important limitation is that the sensitivity of the insertion gain to variations in the coupling network elements is increased as the slope is increased. The same considerations lead us to keep not only the slope but also the gain level or efficiency of these net- works relatively low. The maximum possible coupling network gain which can be obtained over the entire frequency spectrum is limited by the capacity across the circuit, as shown by Bode's Resistance In- tegral Theorem. This capacity cannot be reduced without incurring a more severe potentiometer term penalty and thus limiting feedback, so that the total gain area cannot be profitably increased. The in-band gain can be made greater or less as we concentrate more or less of the total gain area in the transmitted band, but it is found that attempting to get high values of in-band gain area leads to networks which are increasingly sensitive to element deviations. A resistance area efficiency of about 50 per cent turns out to be the most acceptable compromise. In spite of these steps, the coupling networks are the most important source of manufacturing deviations, but by improved mechanical design and the use of quality control these effects have been reduced by an order of magnitude as compared to previous designs. For example, the end- capacity of each coupling network is a quartz-disc condenser, and the peaking or splitting coils are tension-wound on ceramic forms to give inductances of 1 per cent tolerance with distribution requirements within this range. The windings of the transformers, shown in Fig. 7, are made by plating the turns on threaded forms machined from optical grade quartz or Vycor glass to tolerances of tenths of a thousandth of an inch. Leakage inductance and parasitic capacity deviations are thus held to a minimum. Split ferrite cores, which must also be held to close tolerances, make it possible to use these methods of fabrication, which previously available tape cores would not have permitted. Since the amplifier configuration gives series feedback on the tubes adjacent to the cable, and cathode feedback on the tubes adjacent to the regulating network, the high impedance winding of each coupfing network is necessarily off-ground. This leads to considerable difficulty in specifying the equivalent circuit. For an on-ground coupfing network, the equivalent circuit of Fig. 8(b) would be adequate for gain and feedback computations — essentially a reactive equafizer plus an ideal trans- former. Even then the capacities and inductances associated with the 892 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 JP ^ > ilg. i'rccision transiormer, windings, core, assembled unit. transformer itself would have to be given as functions of frequency be- cause of the distributed nature of the device. When, however, the high impedance winding is raised above ground potential by the voltage developed across the beta circuit, which is nearly the same magnitude as the voltage across the coupling network, the effects of distributed para- sitic capacities to ground, and the lumped capacity from the junction of transformer and peaking coil, become of prime importance. The coupling network, therefore, cannot be adequately represented by merely lifting the circuit of Fig. 8(b) off-ground. In order correctly to understand and compute the amplifier gain, it was necesary to develop a complex mathematical analysis of the dis- tributed structure of the transformer, in conjunction with an extended program of precise measurements of the transformer constants. Even then, one must be content with an accuracy of a few tenths of a db and a few degrees of phase, as compared with the order of magnitude better accuracy which can be obtained for a two-terminal lumped-constant network. The agreement between measurement and computation of amplifier gain and reflection coefficient is sufficiently good, however, to 75175: 1158n TRANSFORMER (a) — BALANCING NETWORK AAAr 75 il :c; g L4 TO GRID (OR PLATE) ^a TO BETA CIRCUIT C, LOW SIDE PADDING CAPACITANCE C3 HIGH SIDE PADDING CAPACITANCE L4 PEAKING COIL C5 PEAKING CAPACITANCE La L^ >Ro ■^ UUU *"■ V vv < < 8l, < n < ;r, = ^C, := :::C3 ^ C5 1= ^lEoYcn XN E^ EQUIVALENT GENERATOR, CIRCUIT VOLTAGE, Eq. Rq 1158A. X CABLE OPEN Li,Ri MUTUAL INDUCTANCE, DISSIPATION OF TRANSFORMER. Ci LOW SIDE CAPACITY X 1/7^ La C3 LEAKAGE, REFERRED TO HIGH SIDE OF TRANSFORMER. HIGH SIDE CAPACITY OF TRANSFORMER (INCLUDING PADDING CONDENSER). L4,C5 PEAKING ELEMENTS. Ii2-^LEoY,2 TO GRID (OR PLATE) 1 —*- YiCN - Y,2 1 . ■- ~^ TO BETA CIRCUIT 1 p- „ » 1 ^2 ►^l'=oYc, Y2CN ^2 - 1 note: WHERE Yr. IN I? IS ADMITTANCE OF CABLE Fig. 8 — Coupling Network Circuits, (a) Physical elements, (b) On-ground equivalent circuit, adequate for gain and feedback computations in an amplifier configuration employing on-ground coupling networks, (c) Off -ground equivalent Circuit. 893 894 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 assure that all the important effects are sufficiently understood to make intelligent control possible. Using the results of this analysis, the off-ground coupling network can be represented by the equivalent circuit of Fig. 8(c), where the values of the pi of admittances are obtained in terms of the fundamental parameters of the transformer and associated elements. This representa- tion is convenient for insertion gain and feedback computations. The most important difference between the equivalent circuits of Figs. 8(b) and 8(c), from a practical standpoint, is the presence in the latter of the admittance Yicn which is a manifestation of the capacity to ground from the high impedance winding and the junction of transformer and peaking coil. 1 2 and the contributions to Yzcn not directly attributable to the obvious shield-to-shield capacity of the transformer can be set equal to zero with little error, but Yicn affects the insertion gain and feedback by about 2 db. It will be observed that the transformer is shown with two shields, one connected to the bottom of the high impedance winding, which is the top of the beta circuit, the other connected to ground. The first of these, which is physically adjacent to the high impedance winding, acts to collect, and carry to the beta circuit, the distributed capacity of the high ^vinding, thus avoiding intolerably large capacity from this winding to ground. The second shield prevents capacitative coupling of the large beta circuit signal voltage to the low impedance winding, which would lead to a very poor reflection coefficient performance. Typical curves of reflection coefficient are shown in Fig. 9. Since the amplifier is an active device, the reflection coefficient is to some extent a function of the vacuum tube transconductances, and tends to be degraded as tubes age. 512 6 6 4 P 2 1 1 1 «-''~"~!y k ^r y. / !^ / A- ^ ^-^ y^ ^^ 1 1 ^ 1 1 0.4 0.5 0.6 0.8 1.0 FREOUENCy 1.5 2 3 4 5 IN MEGACYCLES PER SECOND Fig. 9 — Reflection Coefficients. 9 f ±\ ® S2 ±\ ® Y 2 Y, 1 -- 1 1 V ' T Y32 \j 1 I Y 42 1 Y4 1 r^ J- Y2 r NODAL DETERMINANT OF CIRCUIT: E, E2 E3 E4 Y,i -Y,2 0 0 I, A = -Y,2 Y22+S2 -Y32-S2 -Y42 I2 s, -Y32 Y33 0 I3 0 -Y42-S2 S2 Y4 +Y42 I4 WHERE Y,i = Yn-Y,2; Y22=Y2+Y,2+Y32+Y42; Y33=Y3 + Y32 FEEDBACKS RETURN RATIO ON FIRST TUBE: T. = ON SECOND TUBE-. T2: Y33Y2' '^Y2' B-hP ' + P3^P4 Y2 i+E Pl2^P4 + P3-^P4 Y33Y2' Y2' B+E WHERE P12 = T7^; P4 Yii Y2 Y4 . p _ j!i. p _ ^ Y4+Y42'^"Y33' ^' Y33 + D TRANSMISSION Yt2Yi . ^ _ Y3 Y32 . ^ _ Y4Y42 Y„ '^"- Y33 ''^"Y4-.Y42 E4 = I.a^l^^i-l2^ USING FOR I|2 AND I2 THE VALUES GIVEN ON FIGURE 8(C) E4=Ec -I i + B + D + E "*"^'^C' si" 1 + B + D+E Fig. 10 (a) — Input amplifier formulas. 895 896 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Amplifier Formulas Using the equivalent circuit of the coupling network shown in Fig. 8(c), the circuit of the input amplifier can be represented as on Fig. 10a, and the formulas shown can be derived from straight forward nodal analysis of the circuit. Similar formulas can be derived for the more complicated output amplifier. From these, using Thevenin's theorem, the gain of the tandem combination can be computed, as well as the feedbacks on the various tubes. Similarly the input ampHfier can be replaced, for convenience in re- flection coefficient calculations, by the pi of admittances and the driving current of Fig. 10(b). The formulas of Fig. 10(b) expressed in terms of the co-factors of the circuit determinant are of general application for the reduction of a multi-node circuit to simpler form. Regulating Network Like the coupling networks, the regulating network between the amplifiers is outside the feedback loops, and the gain of the amplifier is very nearly a direct function of the impedance seen looking into the network. This impedance is controlled by a single variable resistance — the thermistor — which is directly heated by the dc output current of the regulator. The output of the regulator, in turn, is a function of the 20- rg>9 ^>3l X I = X.E, Yg = A22-A21 A12-12 Yg., ^12-12 = Y,. ^^i-^Aiii^r ^y = A12-A21 A12-12 Yz + ^a + OS+PaSaRv '■^' Y33 ^^J WHERE THE CO-FACTOR A, 2-, 2 + Ci cos 40 • • • C2n cos 2n(f> (3) where the relation between w and is given by the band-pass trans- formation 2 ^ fci + sm^ 0 kz — 8ur 4> K = C0clC0c2 03ei - K ^ lower cutoff frequency (.2 mc) «c2 = K ~ upper cutoff frequency (8.35 mc) K2 — 1 and CjA = , — Xj «r cos 2k<^r , ' , - n \- \ fzi ^ r = l,2....n -f 1 (Match points) THE L3 SYSTEM AMPLIFIERS 899 . _ (2r - 1) X . ^'•- (2r+ 1)2- «r = gamatS4 = 0 nA2n + A2n-2 + " ' ' = 0 3. F(Z ) is expressed as a rational fraction containing both natural modes and infinite loss points. where the coefficients of N and D may be found from the continued fraction expansion of F(Z^). The degree of N and D fixed by the allow- able approximation error and the complexity of the network. 4. The roots of N and D (in terms of Z^) are found and then trans- formed to the p-plane using the transformation These fours steps result in a polynomial F{p) satisfying the require- ments on the change in gain. In this specific case F{v) = 1.06(p + 0.0960) (p -f 0.0280) (p + 0.9890) (p + 0.2890) ^ „+y^ (p + 0.1058)(p + 1.803)(p + 0.0300)(p + 0.3492) Solving (1) and (2) obtain e"-'^^ (7) ' n, z.= ^f- ']['! - II (8) kF - I ' ^ {kF - l)(fc - F) where F{p) ^ F Since the design must absorb the interstage capacities, (and also the stray capacities of some of the elements in the physical network) Zi must include a shunt condenser. It can be shown that the desired result is obtained when F — > 1 + gr - as p ^ c» (9) P 900 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Siiice the derived transmission function F, equation (7), does not satisfy the condition expressed by (9), the function must be modified in order to provide a shunt condenser in Z\. To accomplish this we use equation (5) expressed as a bi-Unear form of the nth term of the con- tinued fraction expansion. FiZ') N^{Z') + KnN2{Z') (10) The coefficient Kn is modified so that the new F(p) satisfies equation (9). The modification of K„ does not substantially alter the required in- band transmission and also does not produce non-physical singularities The new F obtained is F{p) = (p + 0.0289)(p + 0.1018)(p + 0.3431)(p + 1-017 =b i 1.628) (H) (p + 0.0310)(p + 0.1132)(p + 0.4426)(p + 0.6158 ± i 1.368) 1.58 AtH 6.73/iljuF 26.tAH -nm^ — I :^2jx/j.F t9l.6MH «66.2>u.H 2e3.9/iH 2608n 2072 a ♦-AAA- 3229a 76.8/a/iF 183/i/ctF 675.8a I— vw- 55.9 fjbfiF 738.8 a 3.7/iH 1035 a "-^MT^AAArrAAAr-f 57.47 a I 44.7/i/xF i-- Fig. 11(b) — Regulating network schematic. THE L3 SYSTEM AMPLIFIERS 901 0.3 0.4 0.6 0.8 1.0 2 3 4 5 6 FREQUENCY IN MEGACYCLES PER SECOND Fig. 12 — Relative gain of regulating network for extreme and mid-range ther- mistor settings. By using this result in equation (8) expressions for Zi and Zi were ob- tained from which the configuration that is shown in Fig. 11(b) was synthesized. Fig. 12 shows the voltage which would be developed across the regulat- ing network versus frequency in response to a constant current driving force, for the mid-range and extreme values of thermistor resistance. The slope across the band for the mid-range value of 7.5 db; this is the regulating network contribution to the total slope of 37 db required of the complete amplifier. To the extent that the differences between the curves of Fig. 12 fail to exactly match the desired square-root of frequency characteristic, the action of the regulating network will introduce an equalization error. This regulation error is shown in Fig. 13. It amounts to a few hundredths of a db for a six db gain change, caused in part by network design imper- fections and in part by the fact that very small second order effects result in the amplifier gain not exactly following the regulating network impedance. Beta Circuits Starting from our basic requirement that a slope of 37 db across the transmitted band must be obtained, and noting that the total of the contributions from the coupling networks and regulating network is 16 db, we are left with about 20 db of shaping to be supplied by the beta circuits. The input beta circuit has been designed to supply most of this remainder, the contributions of the output beta circuit and the 11/3/(1 — ju/3) effects in the amplifiers being only three db. The original design procedure for this network was basically the same as for the 902 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 ao2 0.01 '\ A \ 0 / \ y^ \ -0.01 " ^ ^^ \ \ -0.02 1 1 1 1 0.3 0.4 0.5 0.6 0.8 1.0 2 3 4 5 6 FREQUENCY IN MEGACYCLES PER SECOND Fig. 13 — Regulation error per one db regulation. 8 10 regulating network, but the final values of the elements represent a long process of incorporating secondary corrections as our knowledge of the amplifier grew. Original constraints included the necessity of including as part of the beta circuit not only the relatively large parasitic capacity of the Eimplifier and the coupling network shield to shield capacity, but also the screen resistor and by-pass condenser of the second tube, which as usual is by-passed not to ground but to cathode, for modulation rea- sons. It is also required that the beta circuit have the correct dc resistance to serve as the cathode bias resistor of the second tube, and that it in- corporate provision for metering this bias through suitable decoupling elements which are among the gain-determining elements of the network. Finally, this network was used as the mop-up equalization network of the amplifier, and its element values were readjusted to give the correct amplifier gain after the performance of a representative group of am- plifiers with average coupling networks and tubes had been determined. In doing this tailoring, it is necessary to precorrect for the effects of non- infinite feedback, since the gain of the amplifier is not exactly a direct function of the beta circuit admittance. The configuration of the input beta circuit is shown by Fig. 14; it is a simple two terminal network. Its in-band impedance varies from about 1,000 ohms at 300 kc to 110 ohms at 8.35 mc. The output beta circuit is relatively fiat, which in this case is the optimum condition for signal-to-noise and feedback loop stability con- siderations. Because; of this simplicity, it is possible to incorporate in this network provision for adjusting the gain of the amplifier to reduce the misalignment of the system. THE L3 SYSTEM — AMPLIFIERS 903 Misalignment Adjustment We can distinguish three major effects which will contribute to mis- alignment and consequent degradation of system signal-to-noise ratio. One is design error — the degree to which the design gain of the ampUfier, because of the finite number of elements and other limitations, fails to match the Hne loss. Second is the cumulative effect in the individual amplifier of manufacturing deviations of the elements. Third is the aging of the components of the amplifier, of which the tube aging mil of course be the dominant short term effect. The signal-to-noise performance of the system can be improved by reducing the misahgnment, if this is done without resorting to measures which would introduce systematic instead of random gain deviations. If we study the shapes of gain change introduced by the more im- portant deviation contributors, particularly the element variations, we find as might be expected that in general the effects are small at low frequencies and increase sharply near the upper edge of the band if the thermistor is held constant. In the system, the regulation around the main pilot will automatically act to reduce the deviation at 7 mc to zero, adding a square root of frequency curve that results in a bow shaped deviation as illustrated by Fig. 15. Examining the signal-to-noise effects of the degradation caused by misalignment, we conclude that it is most desirable to reduce as much as possible the misahgnment at 4 mc, the television carrier frequency in the combined system. The output beta circuit has, therefore, been designed to give var3dng amounts of this shape, the total range being dzO.6 db at 4 mc in 0.2 db steps. The suc- cessive steps of this gain adjustment are simple multiples of each other, symmetrical in the two directions of adjustment, so that we put into (CIRCUIT AND TRANSFORMER PARASITICS ) ^--TEST POINT (T FOR MEASURING CATHODE VOLTAGE Fig. 14 — Input beta circuit elements — line amplifier. o +.B 904 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the system a single systematic shape, which will be small in magnitude at the equalizing points because the different settings of this control in successive line repeaters will tend to cancel out. This setting of output beta circuit gain is controlled by a gain adj switch accessible from out- side the amplifier housing. The adjustment mil be made on an in-service basis as part of normal maintenance procedures, using the level at each repeater of one of the system pilot frequencies (3,096 kc) as the index of proper setting. Manufacturing Testing As mentioned above, the mechanical design of the amplifier has been planned to permit the separate testing of the five gain-determining net- 0.6 o z < I o Z -0.4 < ^-0.6 — ^^^.s 0.2 0.3 0.4 0.5 0.6 0.8 1.0 2 3 4 FREQUENCY IN MEGACYCLES PER SECOND 5 6 Fig. 15 — Amplifier gain versus output beta circuit misalignment adjust- ment. works, the tuned interstage of the input amplifier, and the separate input and output amplifiers before their assembly with the regulating network to form complete amplifiers. Variations in environment are minimized by the use of jigs which also make it unnecessary, in general, to solder to the network under test. Visual gain sets which cover the transmitted band and are accurate to two or three hundredths db are used. The component network or component amplifier under test is connected in series with a complementary or equalizing network to obtain a flat transmission characteristic which can be accurately com- pared to the transmission of standard attenuators. The completely assembled ampUfiers are similarly tested. Quality control charts of the resulting measurements on all components are useful in detecting shifts in transmission whi(!h might be caused by loss of control in element manufacture or by shifts of element values caused by subsequent damage in handling. THE L3 SYSTEM — AMPLIFIERS 905 DC Feedback In addition to the loop feedback at signal frequencies, local dc feed- back is used on each stage. The grid of each tube is returned to a +9- volt potential rather than to ground, and about +10 volts is developed across each cathode resistor. The usual stabilizing effects of self-bias are thus obtained in exaggerated degree, each tube having about 20 db of local dc feedback. Care must then be taken to select the cathode by-pass and interstage coupling condensers so that the transition from low-fre- quency local feedback to in-band loop feedback is accomplished smoothly without the instability which might be caused by a balancing out of these two feedbacks in the transition region. For maintenance measurements, the 9-volt bias potential and the dc cathode voltage of each tube are brought out to a multi-pin amplifier test jack through appropriate decoupling filters. Thus the bias on each stage can be measured on an in-service basis. Filament voltage dropping resistors which can be switched in or out of circuit are provided on the repeater panel, so that bias can also be observed for a filament supply voltage 10 per cent below normal. The activity or change in plate current with filament voltage thus measured, or the history of the bias at normal filament voltage, ^vill be used to determine when amplifiers should be taken out of service for tube replacement. The upper triode is, of course, a special case: since it is the plate supply path for the lower triode, its cathode is about 160 volts above ground and its grid is returned to a similarly high potential. About 35 db of dc feed- back is obtained on this stage; the same provisions for measurement of bias and activity are made. To avoid excessive filament to cathode voltage, a separate filament winding which floats at the +190-volt supply potential is used for this tube. Loop Feedback The design of the feedback loops of the input amplifier follows con- ventional practice. The constraints which operate to limit the amount of feedback which can be obtained in the transmitted band are well known.^ Broadly speaking, the figiu*e of merit of the vacuum tubes and the circuit capacities determine the asymototic cutoff, which in any feedback amplifier limits the magnitude of the feedback which can be built up in the band. In multi-loop structures, there are additional limitations. Consider, for example, the formula given on Fig. 10(a) for the feedback on the second tube. If w^e increase without limit the magnitude of the first tube transconductance, we find that the feedback 906 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 on the second tube approaches the value Si Pa Z32 . This is a limit, com- mon to multi-loop structures, over and above the usual limit imposed by the Nyquist stabiUty criterion. A similar S P Z limit can be derived for the feedback on each stage. The same multi-loop mechanisms which operate to limit the feedback also result in making the feedback on any given stage relatively insensitive to variations in other transconductances or impedances. In a single loop circuit a one db change in beta circuit impedance or first stage transconductance causes a one db change in the feedback on every tube. Here the feedback on, say, the second stage would generally change only one-half db at most frequencies. While this is an advantage in the sense that transconductance decay does not decrease feedback as rapidly as it would in a single-loop amplifier, it is a disadvantage in that it militates against improving stability margins by shaping the out-band impedance of the beta circuit. The capacity distributions within the input and output amplifiers are such that while the S P Z limitations on feedback are closely ap- proached, the most stringent limitation on the feedback obtained is the Nyquist criterion. The use of the Nyquist criterion, particularly with respect to defining the margins against singing, is likewise complicated by the multi-loop nature of the circuit. Ignoring some very recent work, the implications of which have not been fully explored at this writing, it can be said of a multi-loop structure that the apparent margins against singing shown by any plot of feedback give no certain information as to how safe from singing the circuit is. The phase, as well as the magnitude, of the feedback on each stage is a function of the magnitude of the other transcon- ductances, and either decay or increase of these transconductances might destroy the phase margin. In these circumstances, it is theoretically necessary to examine for stability every conceivable combination of transconductances. A more practical expedient, of course, is to rely on judgment backed by computation and laboratory experiment on the circuit for a wide but far from infinite number of circuit conditions. Another difficulty arises from the fact that the feedback on each tube is different, so that gain and phase margins obtained for one return ratio do not imply equal gain and phase margins for the return ratio on some other tube of the same amplifier. It does not follow, however, that it is necessary to investigate separately the margins on each stage versus circuit element variations. The point to be stressed here is that we are using the behaviour of the return ratio merely as an index of our real concern: the i)c)sition on the p-plane of zeros of the determinant of the circuit, and the dciterminant is the same for both stages. Rather THE L3 SYSTEM — AMPLIFIERS 907 than asking how many db of gain margin and how many degrees of phase margin any particular return ratio displays, we ought instead to inquire how quickly the apparent margins disappear as we change transconductances and network impedances. The margins on all the return ratios will vanish simultaneously, regardless of the apparent dif- ference in original margin magnitudes. It is therefore satisfactory to examine the behaviour of whichever return ratio is most easily observed. The design choices made in arriving at the coupling network also affect the magnitude and shape of the feedback which can be obtained: as mentioned above, the relative magnitude of the impedance seen looking into the coupling network and the impedance from first tube grid to ground determines how much of the voltage developed across the beta circuit wall reach the first stage as a driving force. Study of the modulation products which will arise in system opera- tion, both for the all telephone case and for the combined telephone-tele- vision signal, led to the conclusion that optimum shaping of the feedback for the L3 system would be to maximize the feedback at low frequencies in order to suppress intermodulation products falling in that part of the spectrum in the combined telephone-television case. Shaped feedback, falling off at the higher frequencies, is also consistent with obtaining a smooth and simple shape of gain-change as tubes age (known as "mu- beta effect"), which is desirable from the equalization standpoint. With these considerations in mind, the interstage of the input amplifier has been peaked well above the transmitted band — at 11 mc — to partially compensate for the input potentiometer term, and to help in achieving this smooth shape of mu-beta effect. If flat feedback over the band were the objective, it would also be necessary to design the grid-cathode ad- mittance of the second tube so that the parasitic grid-cathode capacity would be absorbed in a flat impedance versus frequency, but in this case the desired shaping of the second tube feedback is attained by taking advantage of the way in which the grid-cathode capacity naturally limits the high-frequency feedback on this stage. The loop feedback in the output amplifier is similarly shaped, for the same modulation and equalization reasons. The use of the double triode circuit in this amplifier is an unusual feature. This connection of two triodes, sometimes referred to as the "cascode circuit," has appeared in many contexts in recent years, usually to serve some other purpose than here. It serves as a superior output stage in the L3 amplifier largely be- cause the effective transconductance w^hich can be obtained is about 3 db higher than that of a pentode of the same grid-cathode spacing. The effective transconductance, ignoring for the moment the division of out- I 908 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 put current between the load impedance and the internal plate impedance of the upper triode, is very nearly the transconductance of the lower triode. Since no current is lost to a screen, this is higher than that of a pentode of the same spacing. The output impedance of the upper triode is multiplied by the local feedback consequent on its cathode being off ground by the plate impedance of the lower triode. Since the load pre- sented to the lower triode is a low impedance, approaching the reciprocal of the transconductance of the upper triode, the grid-plate capacity of the lower triode is not enhanced by feedback. The higher value of trans- conductance means less grid drive for the same output current, and more obtainable feedback, both contributing to superior modulation performance. The modulation of the upper triode, and the changes in transcon- ductance of this tube, are suppressed by approximately the sum, in db, of the local and loop feedback. In consequence, the modulation and mu- beta effect contributions of this tube to the complete amplifier are negligibly small. It will be observed that the grid of the upper triode is not connected to ground, but to the top of the beta circuit. In consequence, the grid- plate capacity of the upper triode appears as part of the end-capacity of the coupling network rather than as a parasitic capacity to ground. This results in a gentle potentiometer term in the output amplifier. This connection, however, also has the effect of vitiating to some extent the desirable qualities of the circuit, particularly at the higher frequencies, where the grid-cathode capacity of the upper triode becomes important. Because of this gentle output potentiometer term, it is not necessary to tune the interstage of the output amplifier, which consists simply of the circuit capacity plus a network which has the characteristics of a 10-mmf capacity in the transmitted band. Above the band this network shapes the gain and phase characteristics of the feedback loop to obtain the desired stability margins. The incorporation of this network reduces the in-band feedback by about 3 db, a sacrifice which unfortunately is necesary to assure stabihty when the thermistor in the regulating net- work is at its minimum value. For this value of thermistor, the phase and magnitude relations of the regulating network impedance and the grid to cathode capacity of VT3 produce a potentiometer term in the feedback loop which appreciably reduces the margins around 30 mc. The stability margins for the mid-range value of thermistor would be satis- factory without this sacrifice of in-band feedback. When the thermistor is at maximum resistance, some degradation of phase margin at 10 mc occurs, again because of the potentiometer term effect mentioned above, THE L3 SYSTEM AMPLIFIERS 909 but in this case the remaining margin is sufficient, since the circuit ele- ments are still under good control at this frequency. Because the plate- cathode impedance of VT2 is very high, the similar potentiometer term at the output of the input amplifier causes only negligible changes in input amplifier stability margins as the thermistor changes. In the 70-mc region there are two almost equally important feedback loops in the output amplifier — one through the transconductance of the lower triode, the other through the grid-plate capacity of this tube. Balances between these feedback paths are observed in the 70 to 100-mc region in the course of measuring the feedback, sometimes accompanied by 180° shifts in the phase of the loop transmission at frequencies above the balance point, an effect which theoretically depends on just how the two vectors go through the balance point. The occurrence of these balances is accompanied by a few degrees loss of phase margin in the 30-mc cut-off region, which must also be allowed for in setting the 30-mc stability margins, since sufficient control of parasitics to prevent these 70-mc effects is out of the question. Parasitic resonances between the lead inductances and the capacities of the circuit, which tend to cause instabilities in the very high-fre- quency region about 200 mc, are damped by small resistors in the leads, and the lead inductances are kept small by careful mechanical design. In this frequency region, neither measurement nor computation of stability margins can be trusted as anything but a rough guide. On the other hand, adding damping resistors in grid leads and other critical points to prevent 200-mc sings causes a phase margin penalty in the 30-mc region, so a nice judgment of how much damping to add is called for. Final values of damping were chosen so that typical amplifiers could not be made to sing by increasing critical lead lengths or by substantial increases in parasitic capacity, thus assuring that manufacturing varia- tions of elements and wiring will not cause high-frequency sings. The return ratio of VT4: is shown on Fig. 16, the mu-beta effects of both amplifiers on Fig. 17. Signal Levels, Modulation, and Noise Fig. 18 shows the signal levels within the amplifier in db relative to one volt from grid to cathode of VT4:, which is a convenient point to use as a reference for system signal-to-noise studies. It mil be noted that as a result of using the input beta circuit to give so much of the shaping of amplifier gain, the input amplifier has little gain at low frequencies. In consequence the input tube of the output amplifier and the regulating network are important thermal noise sources at low frequencies. The 910 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 50 30 o 10 z z -20 -30 / X VIN / /- '^^, '^-v, '^PHASE \ \ / "-, \ ■\ r--, -v^ ^^y ■\ r- /^\ V \ / ^ / J 300 200 O aOI 0.02 0.05 0.1 0.2 0.5 1.0 2 5 10 20 50 100 200 FREQUENCY IN MEGACYCLES PER SECOND Fig. 16 — T4, return ratio of lower triode. 0.35 0.30 0.25 S0.20 UJ Q 2 0.15 «0 ^ 0.10 tc o z - 0.05 z < \ _4 / J / ^' ""^ 1 1 -0.05 0.2 0.3 0.4 0.6 0.8 1.0 2 3 4 5 6 8 10 FREQUENCY IN MEGACYCLES PER SECOND Fig. 17 — Mu beta efFect, amplifier gain change for a one db decrease in each transconductance. THE L3 SYSTEM — AMPLIFIERS 911 5 20 ^ 10 (r 0 tn 1 -10 s Z -20 Equt \^ E4 ^^ E5 ^ \ -^ ^Ss,^ Jl?._ "*, Eo"" ^ ^ «n "^^-> ^E,^_ ■::»-»r r^ > -40 -—'' -'" E46 •"--.^.^ 01 23456789 FREQUENCY IN MEGACYCLES PER SECOND Fig. 18 — Relative levels of signal voltages in line amplifier referred to voltage at grid of lower triode. Thermistor at mid-range value. relative magnitudes of the noise sources are shown on Fig. 19, which gives the noise at ampHfier output as a function of frequency. Comparison of the grid to cathode voltages of VT2 and VT4: shows that the former will be an important modulation contributor, since the driving force on these tubes is nearly equal, particularly at low fre- quencies. Typical amplifier modulation values are given in Table III. Computations using the measured feedback and the performance of UJ < Q5 ...ID -90 -100 -110 Rc CABLE, TRANSFORMER LOSS, INPUT BETA CIRCUIT Rt1 SHOT NOISE, FIRST TUBE R4 RESISTIVE COMPONENT OF REGULATING NETWORK i Rt3 shot noise, third TUBE Z TOTAL COMPUTED ASSUMING AVERAGE AGE TUBES y^ 'y 7 AN J A 611 C hm THERMI5IOR ^ r ^' A ^ ^^y^-^' ;-' ^:^^' ^ ^ ^- Rt: J .— _ CLIh "T^ -120 -130 R ^^ rrrr: K 'Rti' r^ .-' 1 1 1 1. 0.2 0.3 0.4 0.6 0.8 1.0 2 3451 FREQUENCY IN MEGACYCLES PER SECOND Fig. 19 — Thermal noise at line amplifier output. 912 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 single tubes without feedback check the measured values of amplifier modulation to within a couple of db if the third order coefficient of the tube is corrected to take account of the fact that some third order modu- lation is generated by the interaction of the fundamentals and the fed- back second order products. In general, the effective third order coeffi- cient of the tubes is approximately equal to the voltage sum of the tube's uncorrected third order coefficient and a coefficient 6 db worse than the square of the tube's second order coefficient. If this interaction correction is not taken into account, the correlation of tube modulation, feedback and amplifier modulation is unsatisfactory. The analysis leading Table III — Modulation Products, in db Below One Milliwatt AT Amplifier Output, for Fundamentals 5 db Above One Milliwatt at Amplifier Output Type Fundamentals Mc Product Mc Product — dbm 2F 0.5 1.0 72 1.0 2.0 65 3.0 6.0 55 4.0 8.0 51.5 3F 0.25 0.75 95 0.667 2.0 97 2.0 6.0 87.5 2.66 8.0 81.5 to this result, which is due to F. B. Llewellyn, S. E. Miller and R. W. Ketchledge, is too long to give here. Load Carrying Capacity The load carrying capacity of an amplifier is difficult to define with exactness. One possible definition is the load at which the modulation coefficients of the amplifier have departed appreciably from the small signal power series values because of loss of feedback as the transcon- ductance is cut off during part of the cycle. The signal carried without serious overload, in terms of a single frequency, is practically constant in the transmitted band as a consequence of the fact that the output voltage and the lower triode grid voltage have nearly the same shape versus frequency, as shown on Fig. 18. The output coupling network shaping approximately compensates for the potentiometer term division of current between the load impedance and parasitic paths to ground. Departure from the small signal power series behaviour just begins to THE L3 SYSTEM — AMPLIFIERS 913 be appreciable at +14 dbm of any single frequency. At +18 dbm the dc effects of overload show up as slight changes in the transmission of the pilot frequencies. At +26 dbm, the second order modulation is 3 db, the third order modulation is 6 db, higher per line amplifier than would be predicted from small signal behaviour. Flat Amplifier The fiat gain amplifier, which is used as a transmitting amplifier and to make up for equalizer loss at various points in the system, is basically the same as the line amplifier, with only the obviously necessary modifica- tions. The input beta circuit is nearly flat, the regulating network has been replaced by a fixed gain network which contains a single variable element whose adjustment at the factory compensates to some extent for variations in the coupling networks, and the input coupling network has been modified so that the peaking used in the line amplifier is re- placed by a drop in the high-frequency gain of this network. The output beta circuit has been modified to give flat gain control of ±1.0 db in 0.2 db steps. The interstage designs are changed to readjust the feedback so that the modulation suppression and the change in gain as tubes age will be nearly the same as in the line amplifier. Somewhat more feedback is obtained in the output amplifier since no network is needed in the output amplifier interstage to adjust the 30-mc phase for an unfavorable regulating network setting. The nominal gain of the flat gain amplifier has been set at 34 db and is flat to within ±0.2 db over the transmitted band. The amplifier circuit capacities are low enough so that the inter-amplifier network could be built to give considerably more gain than this; the limit has been set so that flat gain amplifier noise contributions to the complete system noise will not exceed about 1.0 db at the television carrier. Acknowledgements As in any corporate development, many members of the Laboratories have made important contributions to the L3 amplifier design. Particular mention should be made of the work of S. E. Miller on fundamental ampfifier design, S. Darlington and T. R. Finch on network design, C. W. Thulin on the precision transformer, B. J. Kinsburg on quaUty control problems, E. Ley on mechanical design, and E. F. O'Neill on the flat gain ampfifier and on high-frequency stability problems in both amplifiers. 914 the bell system technical journal, july 1953 References 1. C. H. Elmendorf, A. J. Grossman, R. D. Ehrbar, R. H. Klie, The L3 Coaxial System — System Design, see pp. 781 to 832 of this issue. 2. R. W. Ketchledge, T. R. Finch, The L3 Coaxial System — Equalization and Regulation, see pp. 833 to 878 of this issue. 3. G. T. F'ord and E. J. Walsh, Electron Tubes for a Coaxial System, Bell Sys. Tech. J., Oct. 1951, Part II. 4. H. W. Bode, Coupling Networks, U. S. Patent 2337965, Dec. 28, 1943. 5. H. W. Bode, Variable Equalizer, Bell Sys. Tech. J., April, 1938. 6. Sidney Darlington, The Potential Analogue Method of Network Synthesis, Beli Sys. Tech. J., April, 1951. 7. Sidney Darlington, Network Synthesis Using Tchebycheff Polynomial Series; Bell Sys. Tech. J., July, 1952. 8. Hendrik W. Bode, Network Analysis and Feedback Amplifier Design, D. Van Nostrand, 1945. The L3 Coaxial System Television Terminals By JOHN W. RIEKE and R. S. GRAHAM (Manuscript received April 17, 1953) Television terminals are required at circuit ends of the L3 coaxial system; at the transmitting end to condition video signals for carrier transmission and at the receiving end to detect the transmitted signals. Special signal characteristics, e.g., a degree of modulation which exceeds the value commonly referred to as 100 per cent modulation, require departures from standard modulating and detecting processes. The high degree of modulation requires both careful control of transmitted wave form and at the receiver product demodulation with phase synchronous carrier (homodyne detection). Carrier regeneration requirements result in the choice of one step fre- quency translation from the video frequency spectrum to the allocated vestigial sideband carrier spectrum. The one step process using a single modulator results in unusual balance requirements for the modulator itself and an unusual circuit configuration. Transmission quality objectives for the terminals are such as to permit six pairs of television terminals to operate in tandem in a transcontinental circuit. This permits a degree of interconnecting flexibility in operation with other systems, e.g., LI coaxial or microwave systems. These objectives place severe requirements upon the transmission stability of various filters and other circuits within the terminals. New network techniques both in design and fabrication are brought to bear in the effort to achieve required performance. The transmitting and receiving terminals are described, illustrating the functional operation and mechanical and electrical arrangements of the equipment. Introduction The main features of the L3 coaxial cable transmission system have been described in a companion paper/ This paper describes the television transmitting and receiving terminal equipment of the L3 system. The 915 916 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 transmitting terminal conditions a television signal for carrier transmis- sion over the system simultaneously with a group of 600 telephone mes- sages. The receiving terminal reconverts the carrier signal at each re- ceiving point along the cable route. Primarily, the transmitting terminal IS a modulator which translates the composite video picture spectrum of frequencies up to the carrier band of frequencies and the receiving terminal is a detector which retranslates the carrier spectrum back to its original band of frequencies. Particular characteristics of the transmitted television signal, which are intended to aid in achieving optimum transmission quality, have necessitated the departures from past techniques in modulation and demodulation processes that are described in the following. Described also are the methods employed to achieve transmitted picture quality adequate for tandem operation of as many as six pairs of transmitting and receiving terminal equipments in a 4,000-mile television transmission circuit. Operation with several pairs of terminals in tandem occurs when L3 coaxial systems are interconnected with LI coaxial systems or micro- wave radio systems. L3 television terminal development has been in progress since early in 1948. Two transmitting and two receiving terminals have been built on a preproduction basis and currently are being tested under field con- ditions as part of the L3 system field trial. Development effort is con- tinuing on the terminals with emphasis on equipment reliability, includ- ing means for maintaining and improving transmission quality. Frequency Allocations The L3 coaxial system was designed to have as broad a transmission band as the economics of repeater spacing together with presently rea- lizable feedback ampUfier performance permit.^ The band extends from 300 kc to 8.5 mc. In comparison with this band a broadcast television signal occupies the frequency spectrum from zero frequency up to 4.5 mc. From the foregoing it is evident that the television spectrum will not occupy fully the available system transmission band. It is feasible and attractive to allocate part of the transmission band for television trans- mission and the remainder for transmission of message channels. De- tailed allocations then result from a compromise among transmission performance, cost and the number of message channels made available. From these considerations vestigial sideband transmission of the tele- vision signal rather than double sideband transmission is called for. The smaller the vestigial band of transmitted frequencies is made the I THE L3 SYSTEM — TELEVISION TERMINALS 917 smaller will be the total television band required. However, both the cost of band shaping filters and the difficulty of maintaining satisfactorily- low values of vestigial sideband quadrature distortion increase as the vestigial band width is reduced. The compromise of these factors re- sulted in the choice of a 500-kc vestigial sideband. Another choice made was to transmit the television signal in the upper part of the L3 band and the message channels in the lower part. This allocation was determined by considering the noise distribution in the transmission band together with the modulation distortion, (har- monic distortion), produced by the repeaters. By transmitting television in the upper part of the band a minimum modulation distortion is achieved since the harmonics of the television signal largely fall outside the transmitted band or at high frequencies where their effects in the picture are relatively less visible than low frequency distortion. This factor outweighs the higher noise level in the upper part of the band. With respect to the television carrier location, it is placed at the bot- tom of the television band at 4.139 mc, with the vestigial sideband extending down to 3.64 mc and the main sideband extending upward to 8.50 mc. Alternatively the carrier could have been located at the top of the L3 band with a main lower sideband and vestigial upper sideband, but this choice would be disadvantageous because of higher noise levels and poorer repeater gain stability near the top edge of the transmitted band. The final allocation of frequencies is shown in Fig. 1. Just below the television band from about 3 mc to 3.5 mc is a dead space. This fre- quency space is needed for the filters which are employed to separate the telephone signals from the television signals at television and telephone dropping points. A very high value of loss is required of these filters in their attenuating bands and if this is built up over too small a frequency « < n 2 (TV CARRIER) O CD to ^t Q telephone band (600 channels) 8 TELEVISION BAND TRANSMISSION ALLOCATIONS 2 2000 3000 4000 5000 6000 7000 FREQUENCY IN KILOCYCLES PER SECOND PILOTS Fig. sion. 1 — Frequency allocations for L3 combined television-telephone transmis- 918 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 band the resulting delay distortion introduced into the television band becomes very difficult and expensive to equalize. Below this ''guard band" is the "master group" of 600 telephone channels which is trans- mitted simultaneously with the television signal. The detailed allocation of specific frequencies, e.g., TV carrier and the pilot frequencies, results from consideration of effects produced by these frequencies in telephone channels as a consequence of modulation dis- tortion in repeaters. These considerations are described in detail in the system design paper.^ Modulation Process In the LI coaxial cable system^ frequency translation by the television terminals is accomplished in two stages. The two step process employs a first modulator supplied with a very high frequency carrier to translate all video frequencies to a band far outside the final transmitted band of II VIDEO ll INPUT LOW-PASS FILTER MODULATOR VESTIGIAL FILTER 1 I L3LINE SIGNAL I I OEMOOULATOR LOW-PASS FILTER 1 r 4.139 MC X 4.139 MC 1. 1. r8.278 MC Jl ll OUTPUT FREQUENCY IN MEGACYCLES PER SECONDS Fif?. 2 — Television terminal modulation processes. 12 THE L3 SYSTEM — TELEVISION TERMINALS 919 frequencies where the upper side band is suppressed. A second modula- tion then translates the vestigial sideband signal back down in frequency to the final band. In contrast the L3 terminals employ only a single step of modulation to convert the signal directly to the assigned band as shown in Fig, 2. In general this can be accomplished if the carrier fre- quency is at least half the sum of input and vestigial band\vidths. Then the lower modulation sideband does not fold over the zero frequency axis to produce frequencies which fall back into the vestigial or upper sidebands. Some foldover is evident in the L3 case shown in Fig. 2 at low frequencies of the modulator output. Single step modulation is advantageous in that the very high frequencies encountered in the multi-step process are avoided. The disadvantage of the one step process is that many extraneous products of modulation, which in the two step process can be suppressed with filters, must be reduced to tolerable levels by balances in the modulator. The modulator, Fig. 3, is a combination of two double balanced modu- lators of a form often employed for modulation of telephone signals.'' The effect of a double balanced varistor modulator of the type repre- sented by either of the two in Fig. 3 is to multiply the input signal by a square wave function having the period of the particular carrier fre- quency. Since the square wave contains all odd harmonic multiples of VIDEO INPUT HARMONIC INPUT CARRIER MODULATOR ►1 :^-^Vv AAAr ]__L HYBRID NETWORK n HYBRID TRANS- FORMER AAV AAAr DOUBLE SIDEBAND OUTPUT THIRD HARMONIC MODULATOR Fig. 3 — Modulator for L3 terminals with third harmonic carrier product bal- ance feature. 920 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the carrier frequency its multiplication by the input signal generates a series of double sideband output spectra, each centered about one of the harmonics of the carrier. It happens in this case that the lower sideband of the third harmonic spectrum of the carrier modulator contains fre- quencies low enough to overlap the high frequencies of the carrier spec- trum upper sideband and this overlap results in quite visible picture distortion. It is possible, by employing a second modulator paralleling the first but driven with a frequency three times the carrier frequency, to generate a signal spectrum centered at carrier third harmonic which will cancel the corresponding output of the carrier modulator. Successful translation of the video spectrum to the L3 carrier band in a single modulation step depends upon the maintenance of this and other modulator balances to unusually stringent requirements. The carrier supply oscillators for the modulators and demodulators are of the Meacham bridge type^ with quartz crystal frequency control and thermistor amplitude control. Frequency stability of two parts per million is required for successful carrier regeneration at the receiver. A constant temperature oven for the quartz plus the inherent stabihty of the bridge type circuit is expected to provide the required frequency stability between monthly maintenance periods. A feature of the signal transmitted over the L3 system is a degree of modulation which exceeds the value commonly referred to as 100 per cent modulation. The resulting waveform contains a maximum ratio of information to peak carrier, important from the standpoint of optimum signal to noise performance. Fig. 4 shows progressively the reduction in peak carrier amplitude which may be effected by subtraction of carrier component from a modulated signal. Figs. 4(b), (c) and (d) each contain the same amplitude of video modulation. Fig. 4(b) represents a video modulated carrier signal with maximum carrier occurring at tips of synchronizing pulses and a minimum carrier, equal to 20 per cent of maximum carrier, corresponding to picture white. Fig. 4(c) represents the same signal as Fig. 4(b) except that the 20 per cent excess carrier has been subtracted. This is the 100 per cent modulation case. Fig. 4(d) shows the effect of further carrier subtraction, (addition of negative carrier), to reduce to a minimum the peak amplitude of the modulated signal. The waveform of Fig. 4(d) employed for L3 transmission requires lyi db less maximum carrier power than that of Fig. 4(b) for the same transmitted information. The term "excess carrier ratio" has been de- vised to describe degrees of modulation which exceed 100 per cent. It is the ratio of peak carrier amplitude to the peak-to-peak modulation THE L3 SYSTEM — TELEVISION TERMINALS 921 (a) VIDEO MODULATING SIGNAL (b) EXCESS CARRIER RATIO EQUAL TO 1.2 (C) EXCESS CARRIER RATIO {0} EXCESS CARRIER RATIO EQUAL TO 1.0 EQUAL TO 0.5 Fig, 4 — Carrier waves variously modulated by a composite video signal. amplitude. Excess carrier ratio, (ECR), for the waveforms of Figs. 4(b), (c) and (d), respectively, are 1.2, 1.0 and 0.5. Modulated signals of the forms of Fig. 4(b) or 4(c) may be detected by rectification, i.e., envelope detection. However, rectification of the waveform, Fig. 4(d), produces a spurious envelope wherein video signals which exceed a particular value are inverted. It is necessary to employ homodyne detection, that is, a demodulator driven by a locally generated carrier which is synchronous in phase angle and frequency with the carrier component of the signal wave. As described later, homodyne detection also makes possible the necessary suppression of the quadrature distortion associated with vestigial sideband transmission. Quadrature distortion associated with envelope detection is tolerated in the LI coaxial system but with tandem operation of terminals required in the L3 system, would accumulate to intolerable values. Vestigial Sideband Considerations A vestigial sideband signal is produced by a band shaping filter follow- ing the modulator. In this filter the lower sideband is suppressed com- 922 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 pletely except for those frequencies which are within 500 kc of television carrier. Lower sideband frequencies within 500 kc of carrier are sup- pressed only pa^rtly as also are upper sideband frequencies mthin 500 kc of carrier to achieve a symmetrical response function in the vestigial sideband region. . It is convenient in a discussion of vestigial sideband transmission to consider the transmission as made up of two components, each symmet- rical about carrier frequency, a real or in-phase component and a quadrature component which is a distortion term. The process is illus- trated in Fig. 5. Here the response function shown in Fig. 5(a) represents idealized conditions for vestigial sideband transmission. The main side- band is shown extending from carrier frequency Fc to the upper cut-off Fu. A vestige of the lower sideband extends from carrier frequency to the lower cut-off Fv. Constant envelope delay is required in the entire band from Fv to Fu. In the frequency region Fc =b Fv the response characteristic is so shaped that the sum of responses at corresponding frequencies above and below carrier add to a constant value. The sum- ming of signal components in the vestigal bands above and below carrier is accomplished by the receiver demodulator. The response function of Fig. 5(a) may be considered to be the sum of the two response functions 5(b) and 5(c) which have even and odd symmetry respectively about the carrier Fc. Both 5(b) and 5(c) are double sideband functions. The component in Fig. 5(b) represents the VESTIGIAL SIDEBAND Fv (aj VESTIGIAL SIDEBAND RESPONSE FUNCTION §olL (b) REAL COMPONENT 1 (C) QUADRATURE COMPONENT Fig. 6 — Response function of a vestigial sideband currier system. THE L3 SYSTEM — TELEVISION TERMINALS 923 normal double sideband output of the modulator supplied with video and carrier signals. The other component, Fig. 5(c), represents the out- put of a modulator supplied with carrier and signal voltages each shifted in phase by 90° and with the amplitude attenuated as shown in the vestigal region Fc d= Fv. The modulation transmitted by a circuit with the response function of Fig. 5(c) is called the quadrature component and is related to the normal modulation, depending upon the shape and extent of the vestigial sideband. Fig. 6 illustrates the real and quadrature components of an idealized rectangular wave form demodu- lated after transmission over circuits having the response functions of Fig. 5, respectively. The 90° shift of carrier frequency in the quadrature component of the vestigial sideband signal makes possible the suppresssion of this component. The transmitted vestigial sideband signal may be written V{t) = P{t) Cos ct + Q{t) Sin ct, (1) where c = 27r times carrier frequency and P{t) and Q{t) are "real" and quadrature modulating functions^ typically as represented on Fig. 6. The demodulator may be regarded as an ideal multiplier of signal and carrier supply. Let the carrier supply be denoted: C{t) = Cos {ct - 0), (2) where 0 is the phase of the receiver carrier supply relative to the carrier factor of the ''real" component of the signal. The demodulator output is the product V{t) X C{t). A low-pass filter rejects the output com- ponents in the band of frequencies about twice carrier frequency so that the demodulated video signal output is the lower frequency component. Thus, neglecting a factor of 3^, Vo(0 = P(0 Cos 4> -f Q{t) Sin 0. (3) It is seen that real and quadrature video components in the demodulator ji _^y ..(p: IT— ^- \ ^'quadrature, (Q) Fig. 6 — Real and quadrature components of a rectangular pulse after vestigial sideband transmission. 924 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 output exist ill the same proportion as the components of the demodula- tor carrier supply in phase and in quadrature respectively with the real carrier component of the vestigial signal. By providing carrier exactly in phase with the real component of the signal the quadrature component in the output may be suppressed completely. It has been determined that to suppress the quadrature component resulting from the L3 ves- tigial band shape to barely perceptible (threshold) values the phase angle of the carrier regenerated at the receiver must be maintained to an ac- curacy of plus or minus 2.5 degrees. A requirement for one demodulator, when six pairs of terminals contribute to produce quadrature distortion at threshold value, becomes 2.5 degrees divided by the square root of six, or about one degree. The regeneration of carrier at the receiver is one of the principal L3 terminal features. Here a 4.139-mc carrier must be provided to de- modulate the "over-modulated" L3 signal. The required carrier must be reconstituted from information carried in the signal itself. It would be possible, of course, to transmit separately a signal from which carrier frequency could be derived but carrier frequency is really the smallest part of the required information. It is the phase angle of the carrier of the received signal which must be duplicated closely at the demodulator and separate transmission of carrier phase angle does not seem feasible. A phase controlled oscillator is employed for the carrier supply at the receiver, with phase control obtained from information residing in the signal itself and frequency synchronization an additional burden upon the phase control system. The basis for synchronizing the receiver oscillator to the carrier of che received signal lies in the phase angle of the carrier frequency com- ponent of the vestigial sideband signal averaged over a period of time of the order of one frame scanning period. Referring to Fig. 5(b) and 5(c) again, it may be noticed that the quadrature response function is zero at carrier frequency. This means that the quadrature component of the transmitted signal contains no carrier frequency component and will not affect the determination of the real carrier component phase angle based upon averaging over a sufficient period of time. Another signal characteristic presents more serious problems. The degree of modulation employed in L3, shown in Fig. 4(d), makes the, average carrier polarity indeterminate. That is, the carrier polarity for a video amplitude cor- responding to picture white is opposite to that corresponding to picture black or sync pulses. The polarity reverses as the composite signal changes through its half peak-to-peak value. The average polarity determined from a predominantly white picture is thus opposite to that THE L3 SYSTEM — TELEVISION TERMINALS 925 determined from a predominantly black picture. A carrier oscillator, phase synchronized to the average carrier phase of the signal would execute 180° phase reversals as picture content changed from average white to average black, producing sudden video polarity reversals at the demodulator output. This signal carrier polarity ambiguity which is momentary in charac- ter can be exchanged for one which is not time variable by a multiplica- tion operation. The modulated signal is squared, i.e., multipHed by itself on an instantaneous basis, in a square law circuit. Such an operation squares carrier amplitude and doubles carrier frequency and phase angle, the latter effect converting 180 degree phase reversals into 360 degree changes which are indeterminable in the average phase detector. Under these conditions the phase synchronized demodulator carrier supply, stably locked to the average phase of the squared signal, experiences no phase reversals with change in picture content. The ambiguity now is in the determination of incoming signal polarity. The squaring operation eliminates any basis for determining polarity so that the demodulator carrier may with equal likelihood lock to either polarity relative to the signal and thereby at the demodulator output produce video signal wave- forms of either polarity. The method used to secure phase synchronization of the local receiver oscillator to the received signal is described next with reference to Fig. 7. Signal from the line together with the output from the carrier oscillator are brought to the demodulator where the desired video output signal is obtained as the lower sideband of the modulation product. This process has already been described, equations 1 to 3. In the carrier regeneration process signal and carrier phase shifted by 45 degrees (equations 4 & 5) are each squared in square law circuits. V(t) = P Cos ct + Q Sin ct, (4) C(t) Z45° = ± Cos (ct - - t/4:). (5) Band pass filters select from the squaring circuit output signal frequencies in the neighborhood of twice carrier frequency. From the signal squarer, i(P' - Q') Cos 2ct + PQ Sin 2ct (6) and from the carrier squarer, i Sin {2ct - 2<^). (7) The two squared signals at twice carrier frequency are multiplied to- gether in a product modulator. This product contains signals in two 926 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 bands of frequencies, one band in the region of four times carrier fre- quency and the other in the video frequency band starting at zero frequency. The lower frequency component of this product is selected by a low pass filter following the product modulator yielding, if - Qt) Sin 2-i, + PQ Cos 2. (8) The dc component of the low pass filter output is a suitable control voltage for synchronization and is obtained in the limit as the cut-off frequency of the low pass filter is lowered. Average values as produced by the low pass filters are applied as a frequency control voltage to the carrier oscillator. The first term in equation (8) {P' - Q') Sin 20 when averaged is, for small errors in carrier phase angle, proportional to the error angle . The factor of proportionality is recognized as the difference in mean squared values of the ^'real" and quadrature modulat- ing functions, P and Q, illustrated typically in Fig. 5. This difference is always positive when the modulating signal contains energy components VESTIGIAL SIDEBAND SIGNAL INPUT / <[' P COS Ct + Q SIN Ct\ l± [P COS 0 + Q SIN 0]\ VIDEO i 'y^ CARRIER OSCILLATOR |cos(ct-^)|- DEMODULATOR UOW-PASS FILTER POLARIZER SIGNAL OUTPUT note: BRACKETED EXPRESSIONS INDICATE VOLTAGE FUNCTIONS EXCEPT FOR CONSTANT MULTIPLIERS C = 2 77" X CARRIER FREQUENCY ▼ {^ (P^- Q2) cos 2 Ct + PQ SIN 2 Ct]> SQUARER BAND -PASS FILTER MODULATOR 45» PHASE SHIFT SQUARER BAND- PASS FILTER 2fr. LOW -PASS NETWORK ,- {2 (P^~ Q^) SIN 2 0+ PQ COS 24>\ H SIN (2Ct -2 0)} Fig. 7 — Functional diagram of the L3 homodyne demodulation process. THE L3 SYSTEM — TELEVISION TERMINALS 927 within the bounds of the vestigial sideband since the quadrature re- sponse function, Fig. 5(c), is attenuated relative to the ''real" response function in this band. In the present case, with a 500 kilocycle vestigial bandwidth and a composite video waveform for a modulating function, the amplitude of Q^ for control purposes is negUgible compared with P^. The second term of equation (8), PQ Cos 20, contains no dc component since Q itself contains no dc component and all other frequency com- ponents of Q are shifted 90° in phase relative to corresponding compo- nents in P, The dc control voltage therefore is not modified by the existence of the second term of equation (8). However, the function PQ does contain sum and difference frequencies due to the cross products of the spectra of P and Q. These frequencies in the control voltage tend to be large compared with corresponding frequencies due to the products P^ and Q^ since the trigonometric multiplier Cos 20, equation (8), is large when the phase angle error is small. The effect of the term PQ Cos 20 is that of phase modulation of the receiver carrier supply and its suppression determines the characteristics required of the low pass filter which averages the control signal. At the penalty of sluggish synchronization and restricted oscillator pull-in range the phase modula- tion can be reduced to arbitrarily small values. For our purposes a pull-in range of d=20 cps can be achieved with phase modulation less than ±0.1 degree with adequate margins. The control voltage is applied to a tuning element in the receiver os- cillator, in this case a small saturable reactor made with ferrite as a core material. This reactor is part of the series resonant quartz crystal cir- cuit which determines the oscillator frequency and is capable of shifting the frequency in response to the control voltage by db 20 cps, a figure chosen as safely less than the first sideband components of the trans- mitted signal which are ± 30 cycles from carrier frequency. This pre- caution avoids possible synchronization of the local oscillator to a signal sideband frequency rather than to the carrier. Sufficient gain is provided in the carrier frequency control loop just described so that the maximum frequency difference encountered be- tween transmitting and receiving oscillators is corrected by the phase control voltage due to a steady state phase angle error, 0, less than J^ degree. The control characteristic of the saturable reactor may be ex- pressed, A/= A(P' - Q') Sin 20, (9) where A/ is the frequency shift introduced by the reactor and A is the factor proportional to required loop gain. One other factor to be considered is the stabiUty criterion of the 928 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 frequency control circuit as a feedback loop/ In this case two factors contribute to loop phase shift. First, the phase angle variation of the oscillator output in response to the control voltage is an integration process. The control voltage changes the oscillator frequency and the resulting phase change can be expressed as the integral mth respect to time of the frequency shift, phase, 0, = 27r J ^fdt. (10) The integration with respect to time introduces a 90 degree "low-pass" phase shift into the control loop at all frequencies. Second, the averaging low-pass filter introduces phase shift in the same direction so that care must be exercised to avoid instability. In this case a phase stability margin of 45 degrees is provided over a wide range of frequency by de- signing the low-pass filter as a series of resistance capacitance steps of loss. These are staggered in frequency to produce a cut-off rate of 3 db per octave with a phase shift of 45 degrees over a wide frequency band. The polarity ambiguity resulting from the squaring process has been demonstrated in the derivation of the phase control voltage. In equation (5) the plus or minus designation indicates that either polarity of carrier signal might be assumed without affecting subsequent expressions. How- ever, in the derivation of the output voltage from the main signal demodulation, equation (3), the output signal polarity reverses if the carrier, equation (2), is assumed with reversed polarity. The carrier polarity estabUshed at any given time depends largely upon initial phase conditions when signal is applied. Correct video polarity at the receiver output is established by a new device called a polarizer which follows the demodulator. This circuit recognized video polarity on the basis of standard features in the com- posite television waveform. The particular features used in this case are the vertical blanking discontinuities expected once each sixtieth of a second and the duty factor of sync pulses. These two characteristics taken together form a sufficient condition for the determination of polarity of any composite video waveform independent of picture con- tent. The polarity, once recognized to be inverted, is corrected. II .i;\i(>Nic Distortion A significant consideration in transmission problems is the generation of distortion by the non-linear amplitude characteristic of the transmis- sion apparatus. In the c^jise of video transmission, non-Unearity results THE L3 SYSTEM — TELEVISION TERMINALS 929 in the distortion of brightness values of the transmitted picture and presumably will distort chromaticity values of color television signals. With carrier transmission apparatus, in addition to these effects non- linearity produces extraneous interference patterns at harmonics of the carrier frequency. In L3 with a carrier modulated spectrum concen- trated near 4.139 mc, the second harmonic distortion of line repeating amplifiers produces a new spectrum concentrated near 8.278 mc, which is demodulated by the receiving terminal to the region near 4.139 mc' This form of distortion is considerably more disturbing in the final picture than a comparable distortion of brightness values and constitutes a limit to transmission signal-to-noise performance. Advantage is taken of the spectral distribution of energy of television signals to ameliorate somewhat the effects of second harmonic distortion. A pre-emphasis network is employed in the transmitting terminal to accentuate the amplitude of high frequency components of the signal before transmission. At the receiver a restorer network introduces a complementary frequency characteristic to make the overall transmission characteristic constant Avith frequency, (see Fig. 8). The restorer net- work, de-emphasizing the high frequency components, likewise sup- presses the second harmonic distortion signals. A limit to the amount of predistortion permitted is set by the maximum amphtudes expected of the high frequency picture components particularly in anticipation of a high frequency color sub-carrier in color television systems. Tentatively, the characteristic of Fig. 8 is chosen as a compromise of these factors. TOTAL ~~ ——J — .^ PRE-E MPHASJ V ^v / / / / X \ \ / / / r V / ' \ \ REST ORER^^ ,.-' \ ^y 12 3 4 5 6 7a FREQUENCY IN MEGACYCLES PER SECOND Fig. 8 — High frequency pre-emphasis characteristic. 930 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Filters and Equalizers A low-pass filter is required before the video signal is modulated, to limit the bandwith of the signal, eliminating possible sources of disturbing cross products, both in the modulator and in the repeaters of the L3 system. This filter has a cut-off at 4.3 mc and provides over 40 db dis- crimination to all frequencies greater than 4.8 mc. It consists of three m-derived sections and introduces about 1.3 microseconds of envelope delay distortion near its cut-off frequency. This is equalized after modu- lation by the delay equalizer to be described later. Following the modulator is the vestigial sideband filter which passes the upper sideband, and provides 60 db discrimination against all fre- quencies in the lower sideband less than 3.7 mc. The large discrimination is required to avoid interference with the telephone channels during transmission over the coaxial Une. This filter provides a controlled loss characteristic to frequencies in the band 3.64 to 4.64 mc which satisfies the requirement for vestigial sideband transmission. The response func- tion for this band is shown on Fig. 2. A flat transmission characteristic including the effective pass band loss of the video LP filter is maintained over the entire upper sideband from 4.6 to 8.44 mc. A second low pass filter after the modulator provides at least 50 db discrimination against third and higher harmonics of the TV carrier (4.139 mc). A four section high-pass filter designed by the insertion loss method^ is used to supply 50 db of the discrimination at frequencies less than 3456 789 10 11 FREQUENCY IN MEGACYCLES PER SECOND 12 Fig. 9 — Insertion loss of the transmitting terminal filters. THE L3 SYSTEM TELEVISION TERMINALS 931 to a «/) o O 2.0 z '-^ o |..o «0 Q J 0.5 S 0 "X \ \ i \ / <» \ \ ^ y 1 3.5 4.5 5.0 5.5 6.0 6.5 7.0 7.5 8.0 FREQUENCY IN MEGACYCLES PER SECOND 8.5 Fig. 10 — Envelope delay distortion of the transmitting terminal filters. 3.5 mc. This filter has a low-pass shunt network at the input to maintain its stopband impedance at 75 ohms. This is required because to produce a uniform frequency characteristic the impedance facing the modulator has to provide a reflection coefficient not exceeding 3 per cent. The re- maining discrimination at frequencies less than 3.7 mc and the major part of the vestigial sideband shaping are provided by a group of six constant resistance equalizer sections, as shown on Fig. 9. The low-pass filter to suppress carrier harmonics consists of 2 m-derived filter sections. The delay distortion of the complete set of filters and loss equalizer sections is shown on Fig. 10. This includes the equivalent delay distortion of the video low-pass filter, translated in frequency for equalization after the modulator. The distortion must be equaUzed to a constant delay over the television band. A delay equalizer to do this is incorporated in the filter. It was designed by a potential analogue method and consists of 24 all-pass sections, each having the schematic as shown in Fig. 11. A redundant capacitor is used to avoid excessively small capacity values. I _i_ I I — u. ■^15J^ ^h-^^^ Fig. sitic). 11 — Schematic of a delay equalizer section, (dotted capacitors are para- 932 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 The capacitors shown dotted are parasitic elements which must be com- pensated for by modifying the design values of the other elements. This group of sections has an insertion loss varying from 2 to 12 db across the pass band of the filter, caused by the dissipation of the elements. This loss, of course, must be taken into account in the design of the loss equalizer sections. The objective is to obtain equivalent video transmission through the terminal flat to limits varying from d=0.02 db to dzO.lO db, depending upon the frequency characteristic of the deviation. To achieve this, close control of the dissipation loss is essential. The inductor losses which cause the major part of the delay equalizer loss characteristic vary up to ±15 per cent from nominal values. As a means for controlling inductor Q to ±0.5 per cent or better, a "Q adjusting screw" is employed. The inductors are solenoids wound on molded tubes with a threaded hole through the center. A threaded magnetic dust core is used for induc- tance adjustment. Its travel can be limited to the distance from the center to one end of the form without losing adjustment range. By introducing an additional core made of solid magnetic iron into the field of the solenoid, using the opposite end of the form, an adjustment is provided which reduces the Q in a continuous manner as the second screw is advanced into the form. The reduction in Q is caused by the losses in the iron, and normally these would cause a reduction in inductance also. However, the permeability of iron causes an increased concentration of field which tends to increase the inductance. A balance between these tendencies to decrease and to increase the inductance is obtained by controlling the geometry of the Q adjusting core. As a result, a reduction of up to 50 per cent in Q can be obtained, accompanied by a change of less than one per cent in the inductance. Models of the inductor and the adjusting screws are shown on Fig. 12. This adjusting screw in con- junction with the magnetic dust core provides an accurate and economical means for adjusting simultaneously both inductance and dissipation in each inductor of the delay equalizer. The flat transmission level for the upper sideband and the shaped cut-off" for the vestigial sideband were obtained by including loss equal- izer sections, assuming Q factors for the all-pass sections of about 20 per cent less than the nominal Q of the inductors. As a final step in the design, the Q factors were modified to absorb in the loss of the all-pass sections the residual loss distortion uncompensated by the loss equalizer sections. This in eff'ect provided the use of 24 additional parameters for shaping the loss in the pass band and resulted in an improved loss chara(;teristic. THE L3 SYSTEM — TELEVISION TERMINALS 933 I *Q" ADJUSTING CAA COIL Wm INDUCTANCE SCREW ADJUSTING SCREW Fig. 12 — Inductor with adjusting screws. The close limits on delay distortion can be met only by close control of the adjustments on the individual all-pass sections. In order to obtain reproducible results to the order of ±0.1° for the phase shift and ±0.01 db for the insertion loss of the individual sections, a special fixture is employed to make the connection between the section and the measuring circuit. This is shown in Fig. 13. The fixture can be clamped on the net- work terminals quickly without soldering and provides coaxial patch cords with plugs for connection to the measuring circuit. Each section is mounted in an individual container with shielding between the inductors to reduce couphng. Each inductor is resonated with its associated capaci- tors and the dissipation is adjusted by adjusting the two cores. The construction of a typical section is illustrated by Fig. 14. An important consideration in obtaining smooth loss and delay char- acteristics for the delay equalizer is the reflection coefficient of each section. Poor reflection coefficients cause reflections and interactions between all-pass sections. Due to the large phase slope of the equalizer these tend to produce frequency characteristics with large numbers of loss and delay ripples across the frequency band for which transmission requirements are most severe. Reflection coefficients of 2 per cent or less at all frequencies in the TV band have been obtained for all delay sections by taking the following precautions: 1. Mutual coupling is limited between the two inductors in each sec- tion by use of a shield in the section container. As little as 0.1 per cent I 934 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Fig. 13 — Fixture for delay equalizer section adjustment. Fig. 14 — Model of a typical delay equalizer section. THE L3 SYSTEM — TELEVISION TERMINALS 935 coupling coefficient can cause a reflection coefficient of 1 per cent in certain sections at the frequency of 180° phase shift. 2. The Q factors of both inductors are adjusted to be equal in each section. 3. The values of the capacitors are modified to compensate for the presence of the parasitic capacitances associated with the shunt arm, Fig. 11. The measured insertion loss characteristic and phase shift deviation from linear phase slope for one model of the transmitting filter and delay- equalizer is shown on Fig. 15. This has been reduced to video frequency to show the detailed residual distortion which results from the addition of the vestigial lower and upper sideband. The delay equalization is maintained for about 200 kc above the loss cut-off to provide for at least 30-db insertion loss at frequencies where the delay distortion becomes large. Without this precaution the transient response is charac- terized by a severe "cut-off ring" distortion which is a slowly damped oscillation at cut-off frequency generated by high frequency signal components. Z »/) ^ III 10 >- (a) w \ \ \ \ \ \ \ \ CRITERION I — n- \ \ — H \ \ \ y \ \ \ \ \ \ \\ ^.i^^^**— ^ \S. \ (b) \ V \ \ V \ \ \ \ \ \\ N' k V' LESS C \ CRITERION I-- A \ \ \ n- \ \ \ i \ \ \ \ \ \ \ \ \ \ V \ \ \\ \ S<,- 0.70 0.60 =:^ 0.30 ) O.IA 0.2A 0.3A 0.4A 0.5A 0 0.1A 0.2A 0.3A 0.4A 0.5A DEVIATION OF x" FROM NOMINAL Fig. 7 — Operating characteristic curves for control chart method. 964 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 ago 0.70 0.60 0.40 a3o 0.20 A ] I \ l\- CRITERION n -\A 1 \ \ \ \ \ \ \ \ \ \ \ \ \ \ (a) > ""§-iAl VALUES 1 \1 0.3 5 A W ■0.40 A N \ \ \ CRITERION in \ \ ^ 1 ^ II 'A \ la ijA * I iV\ (b) \ k-. 0.1A 0.2A 0.5A 0.3A 0.4A 0.5A 0 0.1A 0.2A 0.3A 0.- DEVIATION OF x' FROM NOMINAL Fig. 8 — Operating characteristic curves for batch method. standard deviation to meet its limit, each lot is judged solely on the data obtained from the sample from the lot. The OC curve for the batch method (Criterion III) is shown in Fig. 8(a) for the case where a' = 0.3A. For comparison purposes the corresponding curve for Criterion II of the control chart method is also shown. It is noted that the batch method gives a somewhat sharper discrimination between good and bad distri- butions than Criterion II of the control chart method, due primarily to the use of a relatively larger sample. Fig. 8(b) shows how the OC curve is modified for other values of process standard deviation, a'. It is seen that the batch method is rela- tively less sensitive than the control chart method to changes in a pro- vided the deviation from a standard value of d" = 0.3 A is not too great. 4.3 CHARACTERISTICS OF THE THREE-CELL METHOD The operating characteristics of the three-cell method cannot be evalu- ated probability-wise in the manner given for the other two methods. THE L3 SYSTEM — QUALITY CONTROL REQUIREMENTS 965 However, the manner in which the three-cell method serves as a con- tinuous corrective influence over the distribution of delivered product can be indicated by a few diagrams, for all of which a Normal distribu- tion with a' = 0.3 A is assumed. The running average of small segments of product delivered in pack- ages of 5 is held closer to the nominal by the three-cell method than by the control chart method or the batch method, even w^hen the process is statistically controlled at the nominal. A comparison with the control chart method is illustrated in Fig. 9. In the upper chart are shown av- erages of random samples of 5 units each, plotted on a control chart Avith A5 and PA limits. These are samples obtained experimentally from a Normal distribution whose average, X', was at the nominal for the first 20 samples (Series A). For the next 20 samples (Series B) the average was 0.15A above the nominal and for the last 20 samples (Series C) the average was 0.1 5A below the nominal. The same units were then classified and packaged by the three-cell method. In this experiment, as the units of each sample were classified, as many units were packaged in 1-3-1 or 0-5-0 distributions as possible. Of the first 100 units (20 groups of 5), 95 were packaged. After 200 units (40 groups of 5) were sorted into cells, tu < I N UJ K— -SERIES A --HK- — SERIES B-- — ■*{ \*- SERIES C "H N+0.5A A5 LIMIT J - / \ CONTROL CHART METHOD " L ^ . / \hhh /^ PA LIMIT N \ /\ ; \A r/V 0 V U 8 \ NOMINAL 1 Y «-^ ^V - f ^^VAaAV N-0.5A - A5 LIMIT " " " " 1 3- CELL METHOD £^^^5^%z^;:?s;^ |<- GROUP t >\ \* GROUP 2 *\ \*- GROUP 3 ■*{ Fig. 9 — Comparison of control chart and three-cell method, averages of packages of 5. 966 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 175 were packaged. Of the 25 units not packaged, 24 were in the upper cell and 1 unit did not meet the A limits. At the end of the experiment 295 of the 300 units had been packaged. Of the 5 units not packaged, 3 remained in the upper cell, and 2 did not meet the A limits. The averages of the packages are shown in the lower chart of Fig. 9. For comparison, the PA limits are also shown on this chart. It is apparent from these two charts that the three-cell method yields packages whose averages are held closer to the nominal than are averages for packages from the con- trol chart method. The corrective effect of the three-cell method is further illustrated by Fig. 10, which shows the average of product packaged by the three-cell method as a function of the process average. This curve is for "long term" conditions, that is, it represents the expectancy for any given level of process average. This corrective effect is purchased at the expense of not packaging a portion of the product while the process average is not at the nominal value. However, as already noted, the unpackaged por- tion may be packaged with subsequent product if the process average subsequently deviates from the nominal in the opposite direction. The percentage that can be packaged is also shown in Fig. 10 as a function of the deviation of the process average from the nominal. It should be noted that this curve also represents the expectancy for any given level of process average. Of course, continued production at a fixed level other than nominal would result in a steadily growing ac- cumulation of unpackaged units, a situation that would call for corrective O.IA 0.2A 0.3A 0.4A PROCESS AVERAGE, X', (DEVIATION FROM NOMINAL) 0.5A Fig. 10 — Expectancy curves for three-cell method. THE L3 SYSTEM — QUALITY CONTROL REQUIREMENTS 967 action on the process. Close to 100 per cent packaging can be assured by introducing a negative bias in the process to compensate for the effect of a prior positive bias, and vice versa. 5.0 Conclusions The L3 system's need for holding transmission performance close to the design center, both Avithin short segments and over the full span of the transcontinental line, has called for a high degree of statistical uni- formity of critical characteristics of component elements. The statistical quahty control methods are imposed from the point of view of the user in the interests of the over-all economy of system design. The control procedures are designed to provide at all times a parade of suitably dis- tributed batches of production units, and at the same time to furnish incentives for controlHng manufacturing processes at the design center. Any enterprise of this kind, involves the closest of interplay and ad- justment between design and production interests. Many cases of in- compatibility of design desires and production capabilities had to be cleared in the early stages of the work. Intensive process quality con- trol work and the development of a number of ingenious processing techniques on the part of the Western Electric Company have con- tributed greatly to what has been achieved. Experience will undoubtedly indicate the need for some refinements or adjustments in the plan. Acknowledgments The authors wish to express their appreciation to members of the Western Electric Company's engineering organization for cooperation in the development of the general plan, to Miss M. N. Torrey and Dr. R. B. Murphy for development work on statistical features of alternate and final plans, and to the Misses E. F. Lockey and J. Zagrodnick for conducting sampling experiments and making computations. The L3 Coaxial System Application of Quality Control Requirements in the Manufacture of Components By R. F. GARRETT, T. L. TUFFNELL and R. A. WADDELL (Manuscript received April 27, 1953) The application of quality control procedures^ in addition to conventional maximum and minimum limits, is an important factor in the manufacture of components for the L3 carrier repeaters. In this application, control chart techniques are used for providing assurance that the average of each charac- teristic subject to control is held close to a desired value and that, collectively, individual units have a desired distribution about this average value. The three-cell method is frequently used under certain conditions encountered in the manufacture of these components when sampling procedures cannot be applied. This method consists of measuring each unit of product, classifying conforming units into one of three cells and the selection of groups of five units each to provide the desired distribution. Case histories of a number of factory applications of these methods are presented. 1.0 Introduction 1.1 GENERAL Statistical quality control methods are well kno\vn and useful indus- trial tools for economically controlling quality during manufacture. These methods have as one of their goals the shipment of product meet- ing the end requirements for a particular quality characteristic. The application of such methods also makes possible the delivery of a product whose quality is statistically uniform as, for example, having a dis- tribution whose average is maintained consistently close to the design center. Bell Telephone Laboratories engineers have made use of these principles in the development of the new L3 long distance carrier system which will employ hundreds of repeaters in tandem. An important factor in the application of statistical quality control methods is the specification of distribution requirements in addition to 970 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 corn-eritioual maximum and minimum limits for the characteristics of many of the components manufactured for the new repeaters. The in- troduction of distribution criteria where maximum and minimum are customarily specified requires a number of operations which are supple- mentary to normal procedures. The relatively simple act of identifying good product becomes complicated by the need of more extensive meas- urements, the recording of data, computations, plotting of charts and the active participation of technical personnel in the administration of the procedures. The first step in the program was the development of practical sta- tistical quality control techniques which would be applied under the special circumstances attending the design and manufacture of compon- ents of L3 carrier repeaters. This required careful study by Bell Tele- phone Laboratories and the Western Electric Company and resulted in the development of a general specification which provides procedures and criteria for maintaining the average value of a quality characteristic close to a nominal value and for obtaining as nearly as possible a random dis- tribution of individual values around the nominal. The purpose of this paper is to discuss the procedures thus developed with emphasis on their relationship to manufacturing processes and to describe the problems en- countered and solved in the course of application to the manufacture of components of L3 carrier repeaters. Detailed mathematical derivations and terms will be generally omitted since the theories underlying the principles involved are covered by another article. 1.2 PARTICIPATION IN DEVELOPMENT Normal practice in the creation of new product designs is for the de- velopment and construction of the first models to be handled by the design engineers. This work usually includes discussions with the manu- facturing organization in order to minimize the costs and to utilize exist- ing or most effective manufacturing facilities and various preferred or stocked materials. In the case of the critical components of the L3 carrier amplifiers, a design change in one component resulting from the transi- tion from development to production requires especially close study and may require an adjustment in other components in order to compensate for the one being (jhanged. Knowledge of the behavior of regular manu- facturing facilities and methods used in the fabrication of preproduction units provides considerable assistance in establishing specification limits which are compatible with the produc^t design and manufacturing process capabilities. THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 971 1.3 SEQUENCE OF MANUFACTURING OPERATIONS The application of distribution requirements places added emphasis on the proper sequence of the various operations required for the fabrica- tion of the product. Normally, any assembly or finishing operation follow- ing a process adjustment of a particular characteristic is designed to keep that characteristic within maximum and minimum limits in the final state. Such procedures often fail to satisfy the desired distribution and it is necessary to rearrange the sequence of operations. Once the proper sequence is established it must be rigidly maintained. 1.4 TESTING As a result of refinements in design and in production methods em- ployed for the critical components of the L3 amplifiers, the design en- gineer has in many instances been able to specify limits closer than ever before attained. The specification of such close limits may tax the pre- cision of factory testing equipment and in many cases it has been neces- sary to develop and construct new electrical and mechanical inspection facilities. Measurement reproducibility as well as accuracy in terms of absolute values is important since the measuring instrument ordinarily indicates variations in repetitive readings, even though the product be- ing measured remains constant. Once the characteristics of the measur- ing instrument are determined and used as a basis for the specification of limits, the measuring facility becomes an important part of the dis- tribution control system and must be controlled the same as all other elements of the system. This means careful watch over the maintenance of factory inspection facilities so that these characteristics are controlled. Obviously, an adjustment made on the measuring instrument in the course of regular maintenance which introduces a significant bias or shift, even though well within accuracy limits, may have to be taken into consideration in the use of the instrument. One method of minimizing this problem is to employ stable fixed standards whose characteristics are numerically equal to or near the nominal value of the products being tested. Such standards can be used for either calibrating the instrument or as comparison standards in the actual measurement of the product. These auxiliary standards must still be periodically checked and extreme care taken to prevent any shift in their characteristics. 2.0 Quality Control Methods 2.1 conventional statistical quality control methods The concept of control as used here includes the use of data resulting from measurements made on product produced under the same essential 972 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 conditions. The curve which can be used to represent the observed frequency distribution of data obtained under controlled conditions may, for most practical purposes, be illustrated by the shape shown in Fig. 1. The characteristics of this curve are mathematically represented by the average X (arithmetic mean) and the standard deviation a (the root- mean-square deviation of individual values from their average X). The control chart techniques as originally developed by Walter A. Shewhart^ of Bell Telephone Laboratories provide economical methods for measuring and evaluating the characteristics of such distributions. In practice they are useful in obtaining an estimate of the capabilities of manufacturing processes, sometimes referred to as the "natural toler- ances" of the process and in maintaining control at that quality level. In general, a process having a controlled distribution with an average X Fig. 1 — The ideal frequency distribution for observations obtained under con- trolled conditions. and a standard deviation a will result in practically all of the individual units of product falling within the band X zt Sa. Conventional quality control techniques are used by both the design and manufacturing engineers. The design engineer, in order to specify tolerances compatible with the design needs and the capabilities of ec- onomical commercial manufacture, applies the techniques to a reason- able number of preproduction models or possibly to a limited quantity of initial regular production. The manufacturing engineer in turn uses the techniques for determining the capabilities of existing or new manu- facturing facilities in order to select the most effective facilities and methods. The techniques are also effective tools for locating and elim- inating assignable causes of manufacturing variations, while their con- tinued use as a regular part of the manufacturing process provides an excellent contribution to effective quality control. In these applications there usually exists a substantial margin between the =b3o- variation around the nominal value and the specification limits. This is illustrated THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 973 as specification limits, 1-1 in Fig. 2. The margin is attained either by- improvement of the manufacturing processes or by widening the specifi- cation limits. This means that occasional out-of -control conditions can be indicated by the control chart without the factory being faced with a shut do\vn in production, provided that corrective measures are taken before the magnitude of the deviations results in any significant quantity of the product failing to meet the specification limits. The contrasting situation occurs where the spread between the maxi- mum and minimum specification limits becomes equal to or less than the 6-sigma spread of the manufacturing process, as shown in Fig. 2. For specification limits 2-2, production must be stopped every time an out- of -control condition is indicated or 100 per cent inspection introduced until the cause of the condition is located and eliminated. In the case of specification limits 3-3 in Fig. 2, 100 per cent inspection must be continu- ously applied in order to eliminate non-conforming product as shown in the shaded portion under the distribution curve. 2.2 SPECIAL STATISTICAL QUALITY CONTROL METHODS The special statistical quality control techniques developed for L3 carrier components present a special problem. Here they become an ac- tual part of the product specification, rather than an aid in meeting it, although conventional methods may still be useful for process control. It was recognized that any attempt to express specification limits in terms of a it3(r variation around the nominal would have to include LIMITS SPECIFIED FOR INDIVIDUAL PRODUCTS MINIMUM 1 2 MAXIMUM 2 1 g N Fig. 2 — The relations between a frequency distribution of individual units of product and various specified maximum and minimum limits. 974 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 allowances for the process average to vary a reasonable amount around this nominal. The methods developed and included in the general speci- fication allow the process average to vary within a band of dbj^cr around the nominal which results in limits for individual units of product, desig- nated as "A^' limits in the specification, equal to nominal (A^) =b 33^ a. The value chosen represents a balance between the needs of the operat- ing requirements of the product and the difficulties of maintaining closer controls in the factory. In Fig. 3 the permissible variation of individual units of product is represented by a Normal distribution curve displaced zbj^o- from the nominal N . Since the specification limits are represented by ±A, the allowable variation in the process average is ±0.1A. This is a severe requirement, for in spite of the care employed in col- lecting data during the preproduction period for computing the "natural tolerance" of a manufacturing process, there is always the danger that all the variations which are an unavoidable part of regular production do not occur during the time data are collected. There could be periods after manufacture has started when the factory could not determine whether the out-of -control condition was one which should be promptly ehminated or whether some important characteristic of the process which had not occurred earlier had made its appearance. At this stage it is important to have available some method for sorting product al- ready manufactured which will meet the desired distribution pattern LIMITS SPECIFIED FOR INDIVIDUAL PRODUCTS MINIMUM MAXIMUM Fig. 3 — The relations between a frequency distribution of individual units of product and specified maximum and minimum limits when the average or nominal is allowed to vary ±^^o. THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 975 SO that large inventories will not accumulate in the factory and delivery commitments can be met. Such a method requires the measurement and classification of the product into groups or cells, followed by the selec- tion of units from such cells in accordance with a required distribution. Although this method requires 100 per cent inspection instead of sam- pling inspection permitted by the use of control charts it was estimated that this method would be used extensively until new manufacturing processes were thoroughly proven and sufficient data collected to fully establish a process capability. 2.3 THE GENERAL SPECIFICATION Anticipated application required consideration and development of methods for the wide variety of manufacturing conditions likely to be encountered in the production of such items as coils, condensers, resistors and vacuum tubes. These products may be manufactured at widely sep- arated factories for assembly in equipment at still another location. In order to assist in the description of the factory applications, to be presented later, the methods developed which are referred to as "dis- tribution requirements" in the general specifications are listed and briefly described below. With the exception of the Records method, distribution requirements are applied to only one characteristic of a product. If ap- plication to more than one characteristic is desirable a method suitable for use as a basis for shipment of the product is selected for the most important characteristic and the Records method is specified for the remaining characteristics. 1 . Continuous production method. 2. Batch production method. 3. Three-cell method. 4. Records method. 2.31 Continuous Production Method This method is for application where production comprises a reason- ably steady succession of individual units or small groups of units of product from a common source so that individual units as produced may be kept in the order of their production. The criteria of this method apply to a series of units or groups of units of product arranged in the order of their production and are based on the use of control charts for averages and ranges for samples of 5 units each. Examples of typical control charts are shown in Figs. 4 and 5. The inclusion of an allowance for the long time variation in the level 976 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 of the process average results in the use of two sets of control limit lines for averages in place of the one used for conventional methods. These consist of : (1) A relatively narrow band equal to N zk O.IA {N :h }i(r), desig- nated in the general specification as "Z>" limits. (2) The normal 3-sigma limits for averages of samples of 5 units o o o o J»0-H (1-51) CONTROL SHEET Q-ZJ-SH, IS07F INDUCTOR Qbi, iNOucraNce. CHARACTERISTIC UmT OF MEASUREMEHT M/rHfiH£NHIP< fJHaUf ^J2 ■ Q I WSP. METHOD SPECIflCATIOH iiuiTS;-MIN. 2.1.7 MAX. 2.S. I INSPECTOR _ .MANUFACTURED BY .MACHINE HO._jr GROUP HO^ OPERATOR NO. -re^T ssT |«^ i. ^^i^l •If^vi w»v,» « « 9 S V >; •*• »| H ss S«4 .<>;•^>;^ *i( y «C •>; OC Nl "^ «^ «> M *5 « M 85 «^>N( ^ > »1 35 iis OM«4 Vr «^'^'^VflU^ "J^o>^ lill O^V^i M 5».>< v( »J t\j *> -;^^N< ^V>ooo\ ^•4 N m 1^0 N • • o •" <^n ^$: Fig. 4 — A typical control chart (1507E inductor). THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 977 placed outside the first band equal to N ± 0.5 A, designated in the gen- eral specification as *'C" limits. The control limit for ranges is the customary upper 3-sigma value computed by regular methods,^ and equals lASA. This is designated in the general specification as the "^" limit. The product specification re- quirements are expressed in the form N zL A so that the calculation of O o o o If SICM (3-S2) CONTROL CHART- ALLENTOWN PLANT PRODUCT j.i otk/sa i^oo ^33AA3Z3jOlt9/S:00 3i2i0 3Z5ZO\32Z6a\30 7S0,309^0\3Z/66 ±^300\ s/a/1 [jsooo ,sooo\-9/oo ,/4ao [■¥9ao:^'^3ao'3aoo l_3fae. tZMa.^aoo ^lOOO.3O/OO\30Q00 30/00.31200.31 too. 3300032900.3^000. 3/300 3A00&3Z7O0 3'9^/QXl .iA000:3JOS0jaS00 3Z0QO.33IQ0.3Z0Q0.3f8QQ 3/Zi!a3ZOOO.3ZZ0Q 3ZZO0.3Z000^3a0QQ. ^1300 ^ 7600.3Z/ 00.30300 . 32000 33000 33S0Q. 33300 33800. 32500^^3300. 33 600. 3Zi,0CL Al^Q0^30900.3/900 30300 3/000^8900 3Z/Q0^3J000.3//00.32^00.3/30a3290(i3Z80£i 30300 33/00 31000 3Z/0O 3Z30033000 33200 32/003/300 3ZO00\33 /0O\309/)O\3/OaO ZSS/OL/S^JtCC wz/) 3/ Ate #* /la. lSJfTCQjSZ?<}aj5S9Q0/SS1M>0/ia;tao/SSOOOy63600./6//00 /6Z800./6/O0O./<¥/0&^/4Z/0O./6ZSOC 3JSVa .30S¥ff.3//80 .3/0SO.32Oy0.3/600 .3Z7^aSZZZ/l3£JiO:3£Z0dS2SZO32V£432SO^ •""• ""- .^^^OOiL., ZJOa^H^OJl ^/200 , 23D0^^7S0 . 00.3Q1/^0, 32^00x3/200 32/0O3.=i/0O 3ZS00 334}Oa 32B0O 33900. 33SOO 333^0 3Z800 3000e\32900\3220Q 3Z/00 3¥aaa^ 3/ao/L 3¥saa /* t. /eo. /6oaooyt 7toay6?3ikJ /63*oo jtS3oa. /60900 33ZZOi3;i/€a.33SZa33-9^0.32660.330i0^32JAa j'soo . 3aao. //oo \ /200 . 390a /sao .. ^ooo /j:s 20a i/ea ¥00^62 vaa 3/3ifOl3£0£Ci32¥SO. \^2SaOl330Q\_Z^ai/ -^ J»i_ \vi .3^[aa£L^32300,33J00.30300.33900.3/900. 33/00. 33300, 32¥M.33a00i3Z/00 3U30a,3/3eO.3a2Oaj2fO0.3//O0 .33000,32000.33300x3/200 Z8300 [3/300 AZ&Qit 32aaA3z/oji^ 33//io.J2a/iaiJ/oaai3/saa^33aao\3/iMO\33ooo[3a30Q /ssraovsstoo /64Sda 3/y*0 3/7£a vhv.>i,^ui}},.i^yvi\!kyhhhik}iJ.}.M^y9hhJk}^^^^^^ ^mmiimm^mm^mmmmmjmummmm:xf/in 3/3O0 30000 Jf/Q/L 3330/L 3S^aa^ 53 3/o^a. 30300. 33iOa^ 3z&ra\ 2/00 I znaa 370C 1 isaa I £900 1 Fig. 5 — A typical control chart (436A vacuum tube). 978 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 the various control limits for factory use is a relatively simple operation. A typical example is the 1507 A Inductor where the inductance require- ment is expressed as 283.9 =b 20 microhenries. Then the "C" limits = 283.9 =t 0.5 X 20, = 293.9 max., 273.9 min. microhenries, and the "E" limit = 1.48 X 20 = 29.6 microhenries. Since the distribution requirements are a part of the specification, the product must meet rather exact criteria in terms of the limiting condi- tions previously described, even though they may appear complicated in terms of the usual manufacturing procedures. In the case of the Con- tinuous production method the criteria are applied to the samples of 5 units each, in two steps, (1) to establish eligibility, either at the start of production or when eligibility is lost and (2) to maintain eligibility when once established. Prior to the establishment of eligibility or when eligibility is lost the three-cell method is applied. Criterion 1 Eligibility is established as soon as 7 consecutive samples of 5 units satisfy the following: (a) The averages, X, all fall within the C limits. (b) The ranges, R, all fall below the E limit. (c) Seven consecutive averages, X, are not all outside the same D limit (not all above the upper D limit or all below the lower D limit.) Criterion 2 Eligibility is maintained as long as each current sample of 5 units satisfies the following: (a) The average, X, either 1. falls within the C limits or 2. falls outside the C limits but at the same time all of the 6 preceding consecutive averages fall within the C limits. (b) The range R either 1. falls within the E limit or 2. falls outside the E limit but at the same time all of the 6 pre- ceding consecutive ranges fall within the E limit. (c) Seven consecutive averages X (the current sample and the six preceding samples) do not fall outside the same D limit. 2.32 Batch Production Method This method is for application where production consists of intermit- tent batches of 50 or more units, all of which have been made under the THE L3 SYSTEM QUALITY CONTROL IN MANUFACTURE 979 same essential conditions with respect to materials, parts, workmanship and processing. The criteria of this method apply to a series of batches or lots of product arranged in the order of their production and are based on the use of control charts for averages and standard deviations for samples of 50 imits each. The control limits for averages are for 50 units of product and also include an allowance for the long time variation in the level of the process average. They are represented by a band N ± 0.23 A which is equal to the customary^ 3-sigma control limits for averages of 50 units placed outside a band corresponding to iV dr 0.1 A (N zb Ko-) and are desig- nated in the general specification as "C" limits. In this method the ''E" limit in the specification is the control limit for the standard deviation (a) of a sample of 50 units and equals 0.41 A. For this method a batch, represented by the sample, is considered conforming if the sample meets the following criterion. (a) The average X, falls within the C limits. (b) The standard deviation, LtF ± 1/2% °/o WITHIN CELL LIM1TS--40 % ELIGIBLE TO SHIP 23 % MERCHANDISE LOSS__60 RESULTS after: 1 M N< DM N^ L j\ 1 \ MOLDING A /^ k. / / ^ y ^ x/ N s>^ -— -t: ^ ^ ^— ^^ ^ ^- . ■ D3 3C )4 3( D5 3( D6 3< 37 3( )8 3( )9 3 0 3 11 3 2 3 3 314 315 316 317 31 UNIT CAPACITANCE Fig. 10 — A summarized frequency distribution of a 310.5-mmf mica condenser at various stages of manufacture. 3.32 Quartz Disc Capacitors The quartz disc capacitors are made by the addition of a silver coating to fused quartz discs which have been lapped to an exact thickness. The final adjustment for capacitance is obtained by the mechanical removal of small areas of the silver before the application of a protective finish. The nature of the manufacturing process and the quantities produced requires the three-cell method. A typical requirement is 9.8 ±0.1 mmf which requires the measurement and classification of the product into cell widths of 0.067 mmf each. Capacitance measurements, including the use of reference standards, combined with careful adjusting procedures, have resulted in shipping approximately 99 per cent of the capacitors produced. 3.4 BORO-CARBON RESISTORS 3.41 Introduction The stringent requirements for L3 carrier line amplifiers and imped- ance matching and balancing networks prevented the use of commer- THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 991 cially available composition type resistors due to their high temperature coefficient and lack of stability. Fortunately a new unit, known as a boro-carbon resistor was under development and could be utilized. The boro-carbon films employed in this unit have a maximum temperature coefficient of resistance of 0.01 per cent per °C compared to ordinary deposited carbon films which may have an average temperature coeffi- cient of 0.03 per cent per °C. Two types of boro-carbon resistors covering approximately 45 re- sistance values are being produced for the L3 system. These types are the 200A resistors, held to a tolerance of ±3 per cent and the 200B re- sistor held to a tolerance of ±1 per cent. Both types are available in resistance values from 5 ohms to 9999 ohms inclusive. Of the 45 values used in L3 equipment, approximately 40 are of the 200A type and 5 are of the more precise 200B type. 3.42 Manufacturing Procedure The core of the 200-type resistor consists of a high grade alkaline earth porcelain of special composition. The resistance film is produced by plac- ing the cores in a heated furnace containing a reducing atmosphere of hydrogen and injecting a combination of a hydro-carbon gas and boron trichloride into the furnace. Subsequently, cracking occurs depositing a thin film of boro-carbon over the surfaces of the cores. The resistance value of the film thus formed is dependent on the relative gas concentra- tions and the duration of exposure in the furnace. By suitable control, blanks having a resistance range from 5 to 250 ohms are produced. After removal from the furnace a band of silver paste, consisting of silver flake in a suitable binder, is applied to each end of the coated ceramic and baked. These bands form the contact surfaces for the terminals. The resistors are next sorted into resistance groups preparatory to adjustment to the desired resistance value. Resistors having values be- tween 35 ohms and 50 ohms are used to produce the higher resistance values between 250 ohms and 9999 ohms. The resistors falling outside the 35 to 50 ohm range are held for simple adjustment by rubbing. The method used to obtain higher resistance values is to cut a helix through the carbon film using a diamond cutting wheel. Resistance values may be raised by as much as 1000 times by proper selection of the pitch of the spiral groove. After helixing, these resistors together with those initially outside the 35 ohm to 50 ohm range are fitted with a lead assembly at each end. Final adjustment of the resistance value is obtained by abrading the entire 992 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 coated surface uniformly to reduce the coating thickness. An adjustment of 15 per cent increase may be obtained by this means. After adjustment, resistors are given a protective coating, code marked and subjected to ten accelerated aging cycles over a temperature range of —50° C to 85** C. 3.43 Use of Control Charts Since the manufacturing processes are essentially common within certain ranges of resistance it has been found possible to maintain con- trol charts associated with the three-cell method for combinations of different values of resistance. In order to handle such combinations the characteristic under control is expressed as a percent deviation from the specified nominal. The control chart for averages then consists of averages of these percent deviations and the range chart is expressed in similar terms. This procedure permits a single chart to represent a large number of resistance values. Control charts are used at two points in the manufacture of the 200- type resistors. The first chart is applied to the resistance value after adjustment either by rubbing, in the case of lower values, or by the combination of helixing and rubbing used for the higher resistance values. Estimates based on preliminary studies predict that at half life this type of resistor will age upward by an average of 0.5 per cent. Man- ufacturing requirements for the nominal resistance values are set 0.5 per cent low in order to compensate for this condition. The second chart is kept on the finished product after the accelerated aging cycles. This chart is also kept in terms of the percentage deviation from nominal. A comparison of the charts for the product after cycling, which represents the units which are candidates for stock, with those charts at the earlier resistance adjusting operation will show up changes in value due to processing. Appropriate changes in processing or adjust- ment bogies are made to obtain a more nearly centered value. 3.44 Shipment of Product Although resistor production consists of a series of batch operations with each batch readily segregated and identified, the small quantities involved have resulted in the decision to package all of the product in accordance with the three-cell method. The use of control charts after adjusting and after cycling has resulted in a more uniform product for the 200A resistors than would have been THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 993 possible by conventional manufacturing techniques. Initial production experience indicated that the "natural tolerance" of the 200A resistor is better than the ±4 per cent "A" limit originally specified, and this limit was reduced to ±3 per cent. 3.5 2504-TYPE TRANSFORMER The 2504-type transormer, due to its critical function in the operation of the L3 amplifier circuit, has necessitated the use of completely new materials and methods not usually associated with transformer pro- duction. For instance the windings of this transformer are formed by diamond grinding the threads in fused quartz forms upon which a silver coating is bonded by a firing process. The grooves are then filled with electroplated copper, thus producing a winding intimately bonded to the quartz forms. Since the whole manufacturing process is unique and must be controlled in each small detail, application of statistical quality control procedures is essential. Control of the leakage flux of this transformer, capacitance across the high impedance winding, and the capacitance to ground from the high potential terminal of its high impedance winding is of particular im- portance. To achieve the desired control of these characteristics, care must be taken to maintain a number of mechanical dimensions to toler- ances far more precise than those on any transformer previously made by the Western Electric Company. For example, the inner diameter of the quartz form, bearing the outer winding, is held to 0.7280 =b 0.0005 inches, while the thickness of the form between the inner diameter and the root diameter of the threads is held to 0.0310 ± 0.0005 inches. Dimensions of other component parts of the transformer must be held to comparable tolerances. To do this, it has been found necessary to keep all facilities — chemical, electrical and mechanical — under careful control at all times. The facilities provided for the machining of the component parts are such that all critical dimensions which affect the final overall electrical characteristics of the transformer are held to approximately one-third of the design tolerances. By this means, it is expected the resulting transformers produced will meet the desired distribution. In the initial stages of production, records are being kept on 14 mechanical dimensions on each unit, checking them at each critical state of manufacture. The use of extensive mechanical tolerance controls on the components of this transformer is necessitated by the non-adjustable nature of its design. If controls were not applied, it would be impossible to obtain the close 994 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 electrical uniformity required by the end product. By means of special circuits and testing techniques developed for electrically measuring the characteristics of this transformer it has been found possible to indicate microscopic variations in secondary mechanical dimensions not directly under control. The problem of bringing this transformer under control is not com- pletely solved. Progress to date in the analysis of the data obtained indi- cates that the mechanical variations of the parts have complex inter-rela- tions to the electrical requirements; however, manufacturing variations in the parasitic impedances mentioned above are being controlled to an order of magnitude better than on any transformer previously produced. 3.6 VACUUM TUBES 3.61 General Three new vacuum tubes, the 435A, 436A and 437A have been de- veloped for use in the L3 amplifiers. These tubes were described in detail in an article appearing in The Bell System Technical Journal of October, 1951.^ Two of them, the 435A and 436A are high transconductance tetrodes and the third tube, the 437A is a high transconductance triode. By applying the latest advances in design and in manufacturing tech- niques to tubes of conventional basic type, substantially higher levels of broadband amplifier performance have been realized. The key to improvement in broadband amplification lies in an increase in figure of merit or transconductance to capacitance ratio. Figure of merit is a direct measure of the bandwidth over which the required level of amplification can be obtained. In general a given increase in figure of merit can be directly reflected in a wider transmission band which will provide more communication channels. The higher transconductances and higher figures of merit obtained with the 435A, 436A and 437A over earlier broadband amplifying tubes are a direct result of advances in the art of manufacturing fine pitch grids of sufficient accuracy and rigidity to permit extremely close grid to cathode spacing. The basic objective is to provide a grid which can be placed very close to the cathode to act as a uniform potential plane con- trolling the cathode current without offering any physical obstruction to the passage of the current. This objective is approached by winding the grid with many turns of very small diameter wire. The conventional method of grid manufacture consists of winding the grid lateral wire in a spiral around two side rods, usually of nickel. A groove is cut in the side rods at each point where the lateral crosses it THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 995 and the lateral is placed in this groove. The groove is closed by swaging to fix the lateral in place. In the L3 tubes, the grid lateral wire is ap- proximately 0.0003" in diameter and the grid to cathode spacing is approximately 0.002". A ten per cent change in grid cathode spacing would result in a change in transconductance of fourteen per cent. It is important, therefore, that the control grid be maintained very accurately to insure proper grid to cathode spacing. A conventional grid made with such small diameter lateral wire would not be self-supporting in the length of span required and could not be made with the accurate control of diameter needed. Since the size of lateral wire used and the close spacing of adjacent turns preclude notching and swaging, some other method of holding the laterals in place must be employed. Accordingly, a new type of grid construction is used. This consists of making a supporting frame from two large side rods joined together by cross straps located at the ends of the grid proper. On this rigid frame the fine lateral wires are wound. This frame type of grid was described in previous articles.^' ^ In the L3 grids, the lateral wires have been bonded to the support rods by a glass suspension sprayed along the edge of the support rods. This glass glaze is sintered at a temperature of approximately 700°C to hold the laterals firmly in place. The earlier design used a gold brazing opera- tion to secure the laterals. This brazing was done at a temperature of approximately 1070° C. The newer method at the lower temperature produces less stretching of the laterals and as a result higher residual tension is obtained. This is a distinct advantage in reducing noise and the possibility of grid to cathode shorts. With the new method of holding the laterals in place, the grids are gold plated after the glazing operation is completed. Gold plating is necessary in order to minimize thermionic emission from the grid wires due to the closeness of the grid to the hot cathode and the possibility that active cathode coating material may be deposited on the grid during processing and operation. 3.62 Distribution Requirements Distribution requirements haved been place on transconductance as well as the most critical inter-electrode capacitances, the input capaci- tance, Cg.^kg, , for the 435A and 436A tubes and the grid to plate capaci- tance, Cg-p , for the 437A. In the case of the 435A and 437A tubes, control of modulation is also required. Since transconductance is of primary importance, this characteristic has been selected as the one which governs shipment of product. Inter-electrode capacitance and modula- 996 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 Table IV — Characteristics of Vacuum Tubes Studied 435A and 436A Cathode current (h) Grid-plate capacitance {Cg^^p) Plate-cathode, screen grid cap. (Cp-kg^) Grid-heater cap. (Cg^^h) Heater-cathode, screen grid cap. (Ch-kg^) Heater-plate cap. (Ch-p) 437A Plate current (lb) Plate resistance (Rp) Heater-cathode cap. (Ch-k) Heater-grid cap. {Ch_g) Heater-plate cap. {Ch-p) Grid-cathode cap. {Cg-k) Plate-cathode cap. {Cp-k) tion are subject to the requirements of the Records method previously described. In addition to the characteristics mentioned above, control charts were maintained on a number of other electrical characteristics at final test in order to determine process capabilities and to aid in setting final test specification limits. Among the characteristics that were studied were those given in Table IV. It was recognized at the beginning of production of these vacuum tubes that control of the test characteristics could not be achieved without similar control of the critical piece parts and processes going into the assembly of the tubes. Accordingly, control charts were started on the parts and processes, given in Table V, and have been maintained throughout the manufacture of the product. Production of vacuum tubes represents a complex interlinking of many separate processes, each of which is essentially a batch type of manu- facture. These batches vary considerably in size depending on the process or part involved. No part or process can usually be singled out as the controlling effect which would isolate one group of tubes from another. For example, a given lot of 100 tubes would probably be made from cathode blanks taken from a supplier's production run of 5 to 10 thou- sand parts. The cathode coating would be applied in batches of 50 to 200 parts. The control grid side rods used in making the grid frames would have come from a production run of perhaps 5000 parts and the Table V — Parts and Processes on Which Controls Charts Were Used Cathode blank Outside diameter Coated cathode Outside diameter Control grid support rod Outside diameter Control grid wire diameter Unplated Gold thickness Control grid wire Control grid Minimum lateral resonance frequency Screen grid Minor axis, outside diameter Plate Inside diameter THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 997 fine grid wire used in winding the grid might be from a lot of 300 to 1000 meters etched to size at one time. The mounts are generally sealed into their glass envelopes in lots of 50 to 200 and pumped in lots of approxi- mately 50 tubes. Since the individual batches of parts and processing lots are of such different sizes, the final product does not result in groups of tubes that can be readily segregated into distinctive lots. As the tubes are assembled on a regular running basis the production is treated as being of a con- tinuous nature and the Continuous Production method is applied. Mounts made from each week's production are identified by a serial number and, at test, samples from each week's production are used in determining conformance to the requirements of the distribution specifi- cation. In the application of control charts to the L3 carrier vacuum tubes, the charts kept on the characteristics required by the test specification have been plotted against the C, D and E limits derived from the specifi- cations. The charts kept on the parts and on other test characteristics have control limits derived from the process capabilities. In some cases these natural limits have been compatible with the original drawing limits and in others they have been at variance. Wherever incompati- bility was established, an attempt was made to improve the process or where correlation studies justified the action, agreement was reached with the design engineers to increase the drawing tolerances. 3.63 Application of Control Charts 3.631 Cathodes The cathodes used in the 435A, 436A and 437A are purchased from an outside supplier. A maximum variation from nominal of d=0.0002" was specified for the minor axis outside diameter. These limits are tighter than for any previous similar cathode used, the narrowest limits hereto- fore being d=0.0005". A control chart established on the first lot of cathodes produced for the 436A and 437A tubes indicated that a stand- ard deviation (a) of 0.00021'' was obtainable. The 3-sigma limits thus derived of ±0.00063'' in addition to the displacement in the average (X) for the lot of +0.00026" represented a completely unsatisfactory con- dition. By selection of cathode blanks the nominal was reduced to 0.03121", approximately equivalent to the maximum permitted by the drawling. A much narrower range resulted from this selection. The stand- ard deviation for selected cathodes amounted to 0.000066" corresponding to 3-sigma limits of ±0.00020". However, this distorted distribution 998 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 was still unsatisfactory and attempts were made to obtain cathodes more nearly conforming to the design requirements. In the meantime a sizing operation was introduced in the shop to adjust the existing cathodes to size. By the additional sizing operation, the nominal value was reduced to 0.03102" almost exactly the desired value. However, the standard deviation of 0.00014" and 3-sigma spread of =b0.00042" were still twice the design intent. A second run of cathodes was obtained in which the average measured 0.03099" and the 3-sigma limits were ±0.00021". This substantially fulfilled the design requirements but made it imperative that the average be held precisely at nominal. To avoid differences between measuring instruments an agreement was reached to use a common type of elec- tronic micrometer at both the suppliers' plant and the Western Electric Company. Control charts are kept by the supplier using a modification of the Continuous Production method outlined in the distribution speci- fication. The application of this procedure has resulted in maintaining close control of the averages. Correlation studies on tubes manufactured with these cathodes, have shown that the limits for individual cathodes could be expanded to ±0.00033". 3.632 Coated Cathodes Three characteristics of oxide coated cathodes are particularly im- portant in producing adequate emissive qualities, satisfactory life and the close spacing required in the L3 carrier tubes. These characteristics are weight of coating, density and coated dimensions particularly length and minor axis. In order to eliminate cathode blank variations from the measurements, dummy cathodes of known weight and precise diameter are included with each spray rack load of 41 cathodes. Measurements of gain in weight and of the increase in diameter due to coating are made on these dummy cathodes. The density can be calculated from these measurements. In addition to the measurements made on the dummy cathodes, control charts are kept on coated length and coated diameter on sample cathodes from each coating lot. 3.633 Control Grid Close control of the minor axis for the frame type of control grid used in all three tube types is obtained by precision manufacture of the molyb- denum side rcxis used in making the grid frame. The design specifications require a tolerance of ±0.0001" on the diameter of these rods. Adjust- ment of this diameter is obtained by tumbling the parts, which rounds THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 999 the ends of the rods at the same time to permit easy insertion of the grid in the support micas. Initial production was measured with a barrel micrometer read to the nearest 0.0001". The resulting data indicated a standard deviation of 0.00013". The 3-sigma limits of d=0.00039'' were four times the desired spread. Some portion of this spread was undoubt- edly due to the measuring instrument which could not be read with sufficient precision. Subsequent production was measured with a Brown and Sharpe dial indicating barrel micrometer which was calibrated with standard gage blocks. Production measured under these conditions indi- cated the standard deviation to be 0.00007" for process limits of itO.0002". Fortunately, correlation studies indicated this tolerance to be acceptable. Control charts are kept on the product after the tumbling operation. The desired average is maintained by sampling the side rods for diameter several times during the tumbling operation. The lateral wire used in winding the control grid is tungsten, etched to a diameter of 0.00029" ± 0.00001". The diameter of the wire is measured with a Bausch and Lomb metallurgical microscope equipped mth a Filar eyepiece. The magnification used is approximately 450 X. During initial production, it was not certain that wire of this size could be obtained consistently to a given nominal value and a chart was pre- pared adjusting the specified winding turns per inch to accommodate wire sizes from 0.00027" to 0.00032". Control charts kept on the wire diameter have shown that it is possible to make mre consistently to a diameter of 0.00029 =t 0.00001". Correla- tion studies have been made comparing grids of various wire sizes and corresponding turns per inch against the related tube characteristics. From these studies it was determined that with wire controlled to 0.00029" d= 0.00001", a single winding pitch for each tube could be substituted for the chart of turns per inch originally used. Subsequently, studies were made which enabled a common grid to be used in the 436A and 437 A tubes. After winding, the grids are gold plated. The desired increase in lateral wire diameter as a result of plating is 20 micro inches with limits of =blO micro inches. This measurement is also made with the microscope equipped with a Filar eyepiece. Considerable difficulty has been encountered with measurement of wire diameter. Variations between operators have been encountered as well as differences between successive readings made at the same point on the wire. This difficulty has been minunized by careful training of the operators and by taking several readings at each point and averaging them. 1000 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 The difficulties with instrumentation encountered with measurements of gold plating thickness make it more difficult to establish the standard deviation of the process. By determining the standard deviation of the measuring method used, it appears that a value of 4 to 5 micro inches may be realized. This corresponds to 3-sigma limits of 12 to 15 micro inches. Fortunately, correlation studies have established that this control has resulted in tube characteristics which have more than met the test requirement. 3.634 Screen Grids The screen grids used in the 435A and 436A tubes are of conventional design. The tolerance for minor axis diameter is ±0.001''. The grids are first wound in strips on a mandrel which is shaped to the approximate dimensions of the desired finished grid. The strips are next degreased and given a preliminary heat treatment at 700° C for 5 minutes and then stretched longitudinally to straighten the side rods. The strips are then cut into individual grids and the loose grid lateral turns are trimmed from the ends. A heat treatment at 925° C for 15 minutes follows and finally the grids are sized, which consists of stretching the major axis on an expanding set of sizing blades to obtain the desired shape and size. It was felt that it would be difficult to control the minor axis around a desired nominal since so many operations took place between the winding of the grid and the final sizing operation. Corrections for deviation of the center of the distribution usually consisted of slight changes in the amount of stretch imparted at the sizing operation and major corrections were usually attempted by a change in winding location on the tapered winding mandrel. Process capability studies revealed a parent distribution of approxi- mately ±0.001" which is within the specified tolerance but provides no allowance for shift in average from one lot to another. Successive lots of grids processed in the same manner showed a considerable shift in center of distribution, amounting to as much as ±0.0007". Studies were then made to determine the important variables which needed to be controlled in order to minimize the shift in average value. It was found that for a given spool of grid wire, considerable control of the process average could be obtained by careful attention to two points in the processing. The grid minor axis size varied directly as the tension of the lateral wire was increased at the winding operation. By variation in tension, as much as 0.0005" shift in process average could be obtained. Secondly, when a more precisely controlled heat treating oven was used THE L3 SYSTEM — ^ QUALITY CONTROL IN MANUFACTURE 1001 which employed a thermocouple in each heat treating boat, it was established that the final minor axis dimension varied inversely with the heat treating temperature. Temperatures between 850° C and 1000° C could be used satisfactorily with a corresponding shift in process average of 0.0006''. In Fig. 11(a) is shown the variation of grid minor axis with temperature and in Fig. 11(b) the variation of size with tension is indicated. A procedure was then set up which resulted in a more uniform grid production. Each time a new spool of mre is used, a group of 25 grids is wound using a standard value of tension. These grids are processed through final sizing using 925° C for heat treatment and the standard sizing blades. The distribution of these 25 grids is then plotted. The position of the average is noted and correction is made for displacement of the average by a change in winding tension, a change in heat treating V -2 o o §-3 (a) ^ ^ ^^ \ \, \ ^ 850 875 900 925 950 975 FINAL HEAT TREAT TEMPERATURE IN DEGREES C (l5 MINUTES) 1000 20 40 60 80 100 120 140 LATERAL WIRE TENSION IN GRAMS Fig. 11 — Control of grid minor axis in terms of temperature and tension. 1002 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 temperature or a combination of both. Homogeneous lots of grids, are next run from this spool of wire and are processed up to the sizing opera- tion using the same standard conditions. This test lot of 25 grids is run whenever a new spool of wire is introduced. The first few grids from each lot are sized on the standard sizing blades and checked against a reset-run chart, ^ a form of sequential analysis. The set-up is satisfactory if the distribution of the minor axis dimension is found to be within 1-sigma (O.OOOS'O of nominal. If grids pass the reset-run chart the balance of the lot is sized on the standard blades. If the sample does not conform to the reset-run chart, a final correction is made by substituting a larger or smaller set of sizing blades as required. The distribution of the minor axis after sizing is plotted on control charts. A sample of 5 grids is taken from each tray of 36 grids and X and R plotted. If in control the lot is passed. If out of control, the lot is 100% inspected and packaged into three cells. The grids are then shipped to the mount assembly line in the usual 1-3-1 or 0-5-0 distribu- tion. By application of the above controls of processing, variation in the process average from lot to lot has been reduced to ±0.0003''. In order to keep an accurate record of the grid lots, relating the pro- cesses used to the resulting grid distribution, a special lot ticket has been developed. This ticket contains information such as specified tension and temperature along with the reset-run chart and XR chart. A ticket of this type accompanies each lot of grids through processing. 3.635 4S7A Plate Since the 435A and 436A are tetrodes, the dimensions of the plate in these tubes are not nearly so critical as for the 437A tube, a triode. In the 437 A, the plate assembly is made up of two sections welded together on a mandrel. An air press sizing operation is employed to control the inside diameter of the assembly. The drawing requirement for the plate assembly minor axis inside dimension was set at 0.0750 db 0.0005". Initial production was measured using a tool maker's microscope sighted on the edge of the plate assembly. Misalignment of the part under the microscope and burrs on the edge of the material contributed to variation in the measured values. In addition, the measurement was made at the edge of the assembly rather than the center which was desired. Results of measurements made with the tool maker's microscope are shown on the first portion of Fig. 12. The plate assemblies appeared to be oversize, although this was at variance with electrical test results on the tubes made from these parts. The standard deviation of 0.00101" and 3-sigma limits of 0.00303" were entirely too wide to meet the design intent. THE L3 SYSTEM — QUALITY CONTROL IN MANUFACTURE 1003 o o o o CONTROL CHART- ALl FNTQWN PLANT <^763 m^ \?76/ ^g£^ C7^/ 0777 076 (, 0763 076i n76a ,o7<;7 C76€> \ 0766 07f7,076f,(?7(;f, 07^0\Q76?\0746 O?^'/ r>76¥ 0767 , 07S0 o/yp3 07S^ 07?y 0777^S^ 077d 07S7 C7770 ^77/ Q2Zi^^QJ^S^^^17 073V\/)779 .^77/ 078fi /7£/£\n77/ 077^ ^7aa \077/ 077S ./)7S.H /P7^a f?772 a?^^ ^73fi £22.2±. £>/:>£ 0^/7t?sy ai7C>6_^/f/?/S 1^ .£?7'79,£>7J3\ C>7i'^ C>73^. 7V£\ ^735 . £>7'i'7 i 7yp,£2JA:i77i'lX^2y^,/2Zi^^76 1^ 07y^\C>735 07$7 /P7jr7 ^^9^7^ i773^\^7^7^ ^^-ZTi /y7^2 /:7y6 C733^779 ^7^^ /17^3: ^7ff .7f3 \ 7J$ \ <273&^ji77J^3l^?^^ ' i?7m^a2IIL\jl71A V^^/g I C^Z2\^/77¥ ; CCJZ/ /77?^9l^ffILX i7^^4^2i2C/' /7^^6^e/£ i27f^i>7f7^7¥7"7^^Jl2S3lje7¥6 \ 07VfV77'7S\7P7yiy??^. \ ^7jf7 ^7!ri' ^7f£ ^'^Lra /PJy? /77'M 7)7¥M a7¥7 ^>7J?C . 07^£ ^¥r 7)7S^ . ^7y7 £>7¥e 77^¥6 /975-y C7^3 /?7y7/>7j^3 £>7¥S\ /77^<\ /J7fS-\/)7yAr,/)7¥S\ ^7SS\ ^7¥i /?7S3\/J7yf r>7yi. /77¥9\ a 7 $7) /^7jf7 \ /J7S0 07¥s\ a7¥S \ /J7¥S gYSO >^^i^jKj?pg t7/7/t? \ 777>/a i ^AfA I 2g7/i>^ n/MJ 7 I a/)a9 \ M7a7 /}/M 9 i y^Xl^S ! i9>?/7>» I /7^7j^ ! /?^^^ I /?/7/!> 9 <77¥¥ ^739 ^737 JP732 '77^/} Cryf C7^3 ^.ZIf_717£AL /22^^7222£. i?7V6 (77SC^^7£¥ t279'7_ i22££L Zj^^ i/?7S2.^7: a2it£. ^.S£^/2I62_ ^?V7 ff76^ 3f C^7¥7 777^-3 ^7^1 C>753 72776 is introduced and en- closed between two very close analytic limits in section 2. In sections 3, 4, 5 precise definitions are given and information is collected for the three main types of functions which are to be connected. Section 6 contains the connection formulas. Section 7 reviews the status of knowledge achieved. Section 8 is an appendix on integral equations which are more general than those developed earher in the text, but which appear to be of no use for the main purpose of this paper. Walker, J. G., see R. M. Bozorth. Wick, R. F., see W. P. Mason. 1 Bell Telephone Laboratories, Inc. Contributors to this Issue Frank R. Dickinson, B.S., Union College, 1927. Bell Telephone Laboratories, 193 1-. Mr. Dickinson has been primarily concerned with the development of carrier telephony and is currently working on the L3 carrier system. During World War II he was engaged in the development of airborne radar bombsights. Member of Eta Kappa Nu. H. F. Dodge, S.B., Massachusetts Institute of Technology, 1916; Instructor, Electrical Engineering, 1916-17; A.M., Columbia Univer- sity, 1922; Engineering Department, Western Electric Company, 1917- 25; Bell Telephone Laboratories, 1925-. Mr. Dodge was associated with the development of submarine detection apparatus, telephone trans- mitters and electro-acoustic devices until 1924. He then joined the newly organized Quality Assurance Department and as Quality Results En- gineer has been concerned with the development and application of sampling inspection methods, quality control techniques, and quality rating plans based on statistical methods. His group is also responsible for the preparation of Laboratories Inspection Practices. Consultant to Secretary of War, 1942-44. Shewhart Medal, American Society for Quality Control, 1949. Award of Merit, American Society for Testing Materials, 1950. Member of American Society for Testing Materials; Fellow of American Society for Quality Control, American Statistical Association, and Institute of Mathematical Statistics. Chairman of the American Standards Association Committee Zl on Quality Control, the American Society for Testing Materials Committee E-11 on Quality Control of Materials, and the Standards Committee of the American Society for Quality Control. Co-author with H. G. Romig of Sampling Inspection Tables (John Wiley and Sons, 1944). Member of editorial board of Industrial Quality Control. R. D. Ehrbar, B.E., Johns Hopkins University, 1937. Bell Telephone Laboratories, 1937-. Mr. Ehrbar is in charge of equipment design and field operations related to the development of the L3 coaxial system. During World War II he worked on radar development for the Signal Corps. Member of the Institute of Radio Engineers and Tau Beta Pi. 1015 k 1016 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 C. H. Elmendorf, III, B.S., California Institute of Technology, 1935; M.S., California Institute of Technology, 1936. Bell Telephone Laboratories, 1936-. Transmission Systems Development Engineer, 1952. Since joining the Laboratories, Mr. Elmendorf has been associated with the development of the coaxial repeater system and is currently in charge of the group responsible for the development of the L3 coaxial system. During World War II he participated in the development of microwave components and airborne radar systems. Member of I.R.E. Tudor R. Finch, B.S., University of Colorado, 1938; M.S., Uni- versity of Colorado, 1939. After joining the Laboratories, Mr. Finch spent two years in the study of relay contacts. From 1940-46 he de- veloped networks and circuits for radar applications and more recently networks for the wide-band L3 coaxial transmission system. He is cur- rently engaged in transistor network development for both military and telephone applications. Member of the Institute of Radio Engineers. Robert F. Garrett, Graduate Electrical Engineer, Johns Hopkins University, B.S.E.E. 1926. Western Electric Company, 1926-. Worked since graduation as an engineer and as an engineering supervisor on various assignments with the Western Electric Company in the En- gineer of Manufacture organization. These assignments include the design of factory testing equipment, the supervision of various depart- ments engaged in the engineering planning for field maintenance test sets, spiral four carrier and microwave equpiment. Member of American Society for Quality Control. R. Shiels Graham, B.S. in E.E., University of Pennsylvania, 1937. Mr. Graham has been principally concerned with the design of equalizers, electrical wave filters, and similar apparatus for use on long distance coaxial cable circuits for both telephone and television transmission. During World War II he designed circuits for electronic fire control computers for military use, and later developed methods for computing network and similar problems on a digital relay computer. Member of the A.I.E.E., Tau Beta Pi, and Pi Mu Epsilon. E. I. Green, A.B. Westminster College (Fulton, Missouri) 1915, graduate student University of Chicago 1915-16, B.S. in E.E., summa cum laude, Harvard University 1921. Professor of Greek at West- minster College 1916-17; Captain Infantry Overseas Service 1917-19. American Telephone and Telegraph Company, Department of Develop- ment and Research, 1921-34; Bell Telephone Laboratories 1934-. From 1921 to 1940, and again from 1946 to 1947, Mr. Green was engaged in development work on toll transmission systems, principally in multiplex wire transmission. During the war, 1941 to 1945, he was responsible for CONTRIBUTORS TO THIS ISSUE 1017 development of microwave test equipment for radar systems, radio moni- toring and jamming equipment. In 1948 he was made Director of Trans- mission Apparatus Development, and in 1953 was appointed Director of Military Communication Systems. He is a Fellow of the A.I.E.E. and a Senior Member of the I.R.E. Alexander J. Grossman, E.E., Rensselaer Polytechnic Institute, 1925. Bell Telephone Laboratories, 1925-. Transistor Network Engineer, 1952. Mr. Grossman has been engaged in the development of trans- mission networks since joining the Laboratories. Author of Electric Wave Filters in Electrical Engineers' Handbook (Pender and Mcllwain, 4th ed.). Member of the Institute of Radio Engineers. R. W. Ketchledge, B.S., Massachusetts Institute of Technology, 1942; M.S., Massachusetts Institute of Technology, 1942; Bell Tele- phone Laboratories, 1942-. During World War II Mr. Ketchledge as- sisted in research related to infra-red detecting devices and in the de- velopment of sonar devices. After the war he spent two years working on the development of the Key West-Havana submarine cable system and from 1949-53 he was in charge of systems design for the L3 coaxial system. He was recently appointed Electronic Apparatus Development Engineer and is responsible for gas tube and storage tube development. Member of Sigma Xi. Boris J. Kinsburg, B.S., University of Southern California, 1926; M.A., University of Southern California, 1928. Southern California Edison Company, 1928-30; Bell Telephone Laboratories, 1930-. Since joining the Laboratories, Mr. Kinsburg has worked on research and development of broad band carrier systems using coaxial cable as the transmission medium. This includes amplifier development, study of cross-talk in coaxial conductors, requirement studies for coaxial equip- ment, equalization studies and television echo requirements and, cur- rently, quality control studies of the L3 system components and re- liability studies of the long-range submarine cable development. Member of the Institute of Radio Engineers, American Association for the Ad- vance of Science, and Society for Social Responsibility in Science. Robert H. Klie, B.E.E., Polytechnic Institute of Brooklyn, 1945. New York Telephone Company, 1930-42; Bell Telephone Laboratories, 1942-. After spending two years in the Commercial Relations Depart- ment, Mr. Klie entered a group engaged in the development of radar systems. Since 1946 he has worked on coaxial systems development. Member of Tau Beta Pi and Eta Kappa Nu. M. K. Kruger, B.S., St. Lawrence University, 1920. Engineering Department, Western Electric Company, 1920-25; Bell Telephone Lab- 1018 THE BELL SYSTEM TECHNICAL JOURNAL, JULY 1953 oratories, 1925-37; Western Electric Company, 1937-49; Bell Tele- phone Laboratories, 1949-. Mr. Kruger spent a few years as an instructor in the student assistant course and then became engaged in the design of filters, networks, and transmission testing equipment. He devoted twelve years at Kearny to the design of shop testing equipment for transmission apparatus. Since 1949 he has been concerned w^ith the application of quality control methods for the L3 carrier system and, more recently, with general quality assurance work. Member of the American Society for Quality Control and Phi Beta Kappa. G. H. LovELL, B.S. in E.E., Texas A. & M., 1927; M.S. in E.E., Polytechnic Institute of Brooklyn, 1943. N. Y. Edison Company, 1927- 28. Bell Telephone Laboratories, 1929-. From 1929 to 1948 Mr. Lovell was concerned with the development of crystal filters for carrier systems. Since then he has worked on the development of broadband amplifier networks. Lester H. Morris, B.S., College of the City of New York, 1935. Bell Telephone Laboratories, 1928-. Mr. Morris' first assignments were in the calibration of standard telephone instruments and, later, the de- velopment of acoustic impedance bridges. From 1930 to 1935 he con- ducted research in loudspeakers and since 1935 he has developed repeat- ers for coaxial systems. Member of Phi Beta Kappa. John W. Rieke, M.S. in E.E., Purdue University, 1940. Bell Tele- phone Laboratories, 1940-. A member of the Transmission Systems Development Department, Mr. Rieke worked on television circuits before and after World War II. During the war he assisted in the develop- ment of radar indicators, and currently he is engaged in the development of broad band television transmission systems. Member of American Institute of Electrical Engineers, Tau Beta Pi, and Eta Kappa Nu. T. L. TuFFNELL, Bell Telephone Laboratories, 1927-1941, Western Electric Company, 1941- With the Laboratories, Mr. Tuffnell worked on terminal equipment for transatlantic telegraph and telephone service, and the design of vacuum tubes for carrier telephone systems. His work in the Western Electric Company has been chiefly concerned with engineering problems related to the manufacture of vacuum tubes. R. A. Waddell, B.S. in E.E., Rose Polytechnic Institute, 1936. M.S.E.E., Ohio State University, 1938; Westinghouse Electric and Manu- facturing Company , 1939-1941; Western Electric Company, 1941-. Since January, 1941, has been in various phases of product and test planning on test sets, varistors and coils used in the telephone system. Engineer Resistances and Spool Wound Coils, 1951-. Member Eta Kappa Nu. HE BELL S Y S T E IV /' meat louma^ t^l^ECTRICAL COMMUNICATION B. E. ALLEY, JB. 1155 )isplay of W. R. BENNETT 1173 ^'^ TmSr ^"''^ °^ ^"*''*''" Telephone TheCard"ianslatorforNatio„wideDialing ""=^-^ ^"^^ n^,. ] ^' ^' ^^^^TON AND J. B. NEWSOM 10S7 P« I -Th«.,y „, ,h, l,i,i,ao„ „t U,, Short A„ Pol,.aytael„..hWT*ph,„eC.b,. . """"" "" Contributors to this Issue ^^^^ ;>£■: 1267 COPVH.CHX X.a —..K«PHO.. ..„,,,,„,,,, ,„^~ NT THE BELL SYSTEM TECHNICAL JOURNAL ADVISORY BOARD S. BRACKEN, President, Western Electric Company F. R. KAPPEL, Vice President, American Telephone and Telegraph Company M. J. KELLY, President, Bell Telephone Laboratories EDITORIAL COMMITTEE E. I. GREEN, Chairman A. J. B U S C H F. R. L A C K W. H. DOHERTY J. W. McRAE G.D.EDWARDS W. H. N U N N J. B. FISK H. I. ROM NES R. K. HONAMAN H.V.SCHMIDT EDITORIALSTAFF J. D. T E B O, Editor M. E. STRIEBY, Managing Editor R. L. SHEPHERD, Production Editor THE BELL SYSTEM TECHNICAL JOURNAL is published six times a year by the American Telephone and Telegraph Company, 195 Broadway, New York 7, N. Y. Clco F. Craig, President; S. Whitney Landon, Secretary; Alexander L. Stott, Treasurer. Subscriptions are accepted at $3.00 per year. Single copies are 75 cents each. The foreign postage is 65 cents per year or 11 cents per copy. Printed in U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XXXII SEPTEMBER 1953 number 5 Copyright, 1963, American Telephone and Telegraph Company Transmission Design of Intertoll Telephone Trunks By H. R. HUNTLEY (Manuscript received June 12, 1953) At the 1952 Minneapolis summer meeting of the A.I.E.E. a symposium* on the nationwide toll switching plan went into such features as the funda- mental plant layout, numbering plan, toll switching and automatic account- ing equipments. The present paper is intended to round out this coverage of the plan with a further discussion of the transmission features. THE PROBLEM In the new nationwide toll switching plan using switching machines the layouts of toll circuits and the routings of traffic will be quite different from that of the earlier plans which were based on manual switching. Individual calls can be switched so fast and cheaply that switching is no longer a limiting factor and circuits can be laid out and used in such a way as to obtain maximum economy with few, if any, limitations from the switching standpoint. An example of these changes is given in Fig. 1 which shows in (a) the circuit groups which would be used to handle a given (assumed) flow of traffic on a manual basis and in (b) the groups which would be used to handle the same traffic on a dial basis. In (a) there are 44 different * Trans. A.I.E.E., 71, Part I, Sept., 1952. 1019 1020 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 (B4\ (a) RINGDOWN OPERATION q regional center ^Iprimary outlet • toll center A5 (b) NATIONWIDE TOLL DIALING FINAL GROUPS □ ''Iffg^ ^^^ O PR'^ARY OUTLET MU ronnp. A SECTIONAL ^ ^^^^^^ ^^^'"^^ HU CROUPS ZA CENTER « JOLL CENTER Fig. 1 — Typical intertoll trunk networks. TRANSMISSION DESIGN OF INTERTOLL TELEPHONE TRUNKS 1021 circuit groups and in (b) there are 26 circuit groups. More specific ideas regarding the effects of these differences can be obtained by considering how calls between specific centers (for example, Al to Bl) would be routed in the two plans. From the transmission standpoint the principal impact of the new plan is that the situation will be changed from one in which as much of the traflftc as practicable was handled over direct circuits with a minimum of switched traffic (circuits in tandem) to one in which two or more (up to a maximum of eight) circuits will be used in tandem on many calls and in which different numbers and make-ups of circuits may be en- countered on successive calls between the same two telephones, as a result of the alternate routings employed with machine switching. This means that the losses of circuits must be low in order to provide adequate transmission on all calls and to avoid large differences in transmission on successive calls between the same two places. The ideal method in such a situation would be to operate all circuits at zero loss since this would make the results independent of the number of circuits in tandem. However, the distances involved in the Bell System are so great that the propagation times, which affect echo, and the cross- talk between circuits require that even carrier circuits be operated at finite losses. Also, the plan must accommodate many voice frequency circuits on which the noise and singing conditions, as well as echo and crosstalk, may be more severe than on carrier circuits. The practical plan, therefore, is to: 1. Operate every circuit at the lowest loss practicable considering its length and the type of facilities used. 2. Assign circuits with different transmission capabilities in accordance with the parts they have to play in the operation of the over-all network. The principal problem is to determine how low circuit losses can be made without getting into trouble due to one or more of the limitations mentioned above. This problem is complicated by the fact that the effects of these limitations are not directly proportional to circuit length or to the number of circuits in tandem. For example, if circuit (a) can be oper- ated by itself at a loss of X db and circuit (b) can be operated by itself at a loss of Y db, the loss permissible when circuits (a) and (b) are .switched together is less than X + F. Ideally, therefore, each circuit should have a different loss in each different connection in which it is used. However, this is not practic-able and a compromise must be adopted. This compromise provides that in some connections a particular circuit will operate at its lowest practical loss while in other connections higher losses will be employed to give over-all figures that will be ade- 1022 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 quate from the standpoint of echo, crosstalk, etc. The general procedure is as follows : 1. When a toll circuit is switched to another toll circuit at both ends work it at a loss which is called 'Via net loss" (VNL). 2. When the circuit is switched to another at one end only (the other end being at the point of origin or destination of the call) work it at a loss higher than VNL by an amount which we shall call ">S" C'>S" being a generic term derived from the fact that it may be associated with switching pads — usually called "S'' pads). 3. When the circuit is used by itself (i.e., the origin and destination of the calls are at the ends of the circuit) increase its loss by "S'' again — that is, work it at VNL plus 2S. This is known as "terminal net loss" (TNL). Via net loss is, of course, to be the "lowest loss practicable" referred to above, and the next step is to establish methods of deriving VNL and of selecting the best value for "S'\ Since it would be a very complicated process to work simultaneously with all four of the limiting factors mentioned above, (echo, crosstalk, singing, and noise), the practical approach has been to select one of them as the basis of design and then check the results against the other three, modifying the final solution as necessary so that all four are kept under control. Since long experience indicates that echo is likely to be the most difficult and complex factor to control, it has been used as the starting point in the solution of the problem. As will be evident later, there are a large number of solutions possible from the echo standpoint and the one which has been selected has been affected to a considerable extent by the other factors. "The next part of the material in this paper is, therefore, devoted to an analysis of circuit design from the echo standpoint. DETERMINING LOWEST PRACTICABLE CIRCUIT LOSS FROM ECHO STAND- POINT The over-all objective is to have practically no cases in which objec- tionable echo will be observed by customers. If circuits could be precisely adjusted to the requirements in each different connection the probability of echo would be the same on all connections and the computations would have been carried out on the basis of a very small probability — say, 1 in 10,000. However, losses can be changed only in discrete steps (S) so that in a very large propor- tion of cases the losses will be higher than are theoretically necessary. TRANSMISSION DESIGN OF INTERTOLL TELEPHONE TRUNKS 1023 Hence it seems sensible to compute the theoretical losses on the more liberal basis of 1 in 100, relying on the excess loss in most connections to reduce the over-all probability to the very small value desired. The echo problem with which we are concerned is illustrated in Fig. 2. As shown there, part of the speech power which is being transmitted to the listener "leaks" across the hybrid (or four- wire terminating set) at the listener's end and returns to the talker. This is known as "talker echo." Actually, of course, some of the echo which returns to the talker can leak across the hybrid there and go back to the listener. This is known as "listener echo." However, with modern plant listener echo will not be important if talker echo is adequately controlled. J r- ■i> 4-WIRE TERMINATING SET -o .HX, BALANCING NETWORK _. AAAr s TERMINATING (a) PAD IN INTERTOLL TRUNKS ■—t>^ -— TRUNK LOSS UP TO 2DB TERMINATING (b) PAD IN TOLL CONNECTING TRUNK — THROUGH ^S on a judgment basis. PROVISION OF s The loss S can be provided in any of the following ways as appropriate. (See Fig. 4.) 1. As a switchable loss pad in the intertoU trunks. 2. As a fixed loss pad in the toll connecting trunk. 3. As part of the conductor loss of toll connecting trunks. This can be done only if the structural return loss of the connecting trunks against the balancing network is reasonably good. 4. If there is to be no switching to other intertoll trunks or to con- necting trunks with more than 2 db loss, S may be provided simply by increasing the circuit loss by 2 db. VIA NET LOSS FACTORS Table IV lists typical VNLF's of Bell System intertoll trunk facilities for the condition S == 2. T3rpical losses at which circuits would be worked with S = 2 and with the via net loss factors tabulated in Table IV 1032 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Table IV- - Typical Via Net Loss Factors FacUity VNLF (db pel mile) 2-Wire Circuits 4-Wire Circuits iqH-88-50 0.03 0.02 0.01 0.014 19H -44-25 0.010 O W Voice 0 W Carrier 0.0017 K or N Carrier 0.0019 L Carrier 0.0015 Radio . . 0.0014 are given in Table V. The advantages of high velocity, four- wire cir- cuits (carrier and radio) are obvious from these tables. ECHO SUPPRESSORS Even if the intertoU trunk plant of the Bell System were all carrier' the length of some connections would be so great that some method of controlling echo other than simply increasing circuit loss is desirable. Lower losses can be obtained on such connections through the use of an **echo suppressor," an electronic device which under control of the talker's speech currents places a high loss in the return path at the right time to intercept the return echo currents. Echo suppressors perform very well so long as not many circuits equipped with them are connected in tandem and there is not too much time delay between them. With manual operation the switching is so limited that the chances of connecting circuits with echo suppressors in tandem are small and it has been practicable to apply echo suppressors on the basis of round-trip delay of the individual circuits. However, with dial operation it will be possible to establish connections which are long enough to require an echo suppressor but which are composed of cir- cuits each too short to require an echo suppressor based on its round-trip delay. For example, an echo suppressor would not normally be used on a 500-mile carrier circuit, but if eight such circuits were connected in tan- Table V Type and Length of Trunk VNL(db) TNL (db) 60-inile Nl Carrier 0.5 1.9 2.4 1.4 4.5 50-mile 2-W H-88 5 9 200-mile 4-W H-44 6 4 500-mile K Carrier 5 4 TRANSMISSION DESIGN OF INTERTOLL TELEPHONE TRUNKS 1033 dem giving a total length of 4,000 miles, an echo suppressor would be imperative, if over-all loss is not to be excessive. It is not practicable to take care of this problem merely by reducing the delay time at which an echo suppressor is applied, since if this were done it is conceivable that eight circuits each with an echo suppressor might be connected in tandem. It has been necessary, therefore, to establish more or less arbitrary rules to insure at least one echo suppressor on long connections and to make it very improbable that more than two will be encountered. In general, these rules specify that echo suppressors will be placed on: a. All RC-NC circuits. b. All RC-RC circuits. c. On high-usage group circuits when the desired losses can not be met without them. Our ideas as to when suppressors of Item c will be required may change with the trend from voice-frequency towards high-velocity carrier cir- cuits. Experience will be a valuable guide, for it is not likely that an intolerable situation will build up overnight and without casting some shadow of coming echo ; and the echo suppressor, being a discrete equip- ment unit, can be installed after it is found to be needed without appre- ciable lost motion or additional cost. ALLOCATION OF FACILITIES If the intertoll plant were homogeneous the over-all problem would be solved at this point — each circuit would be designed in accordance with the preceding and that would be that. But the plant is not homogeneous — it consists of everything from loaded voice frequency circuits to circuits on microwave radio with VNLF's ranging from 0.03 to 0.0014. It is, therefore, necessary to allo- cate these facilities among different circuit groups in such a way that as far as practicable the higher performance facilities are used in the more demanding parts of the network. As an aid to allocating facilities, charts like Fig. 5 are used. This chart shows ranges of losses within which circuits in different parts of the network are expected to fall. The losses shown there are exclusive of S which must be added, as indicated before, at both ends of each connection. It should be emphasized that these losses are not ''limits" in the usual sense, neither are they attempts to divide up over-all' losses among circuits. They simply help in allocating facilities in the non- homogeneous plant among different circuit groups. As the use of carrier 1034 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 is extended, the plant will become more homogeneous and the need for such charts will gradually disappear. Fortunately, from the transmission standpoint, while the machines will set up a wide variety of connections, the routing patterns will be rigidly controlled. Thus, it is practicable to know for each circuit group the maximum number of other circuits with which it can be used in tan- dem. The lower velocity circuits, two-wire circuits, narrow band circuits, etc. can (within practical limits) be allocated to groups which have relatively easy requirements. TWO-WIRE SWITCHING — OTHER CONSIDERATIONS Present views are that even the ultimate plan will involve two-wire switching at many points, mainly at the smaller switching points where 2-W NC= NATIONAL CENTER RC = REGIONAL CENTER SC= SECTIONAL CENTER P0= PRIMARY OUTLET TO= TANDEM OUTLET (SAME AS PO BUT WITH SWITCHING equipment) TC= TOLL CENTER [is] = ECHO SUPPRESSOR (REQUIRED)* [ks] = ECHO SUPPRESSOR (IF NECESSARY)* = FINAL GROUP = HIGH USAGE GROUP IF ECHO SUPPRESSOR IS USED, THE ASSIGNED LOSS IS 0.5 DB UNLESS A HIGHER VALUE IS REQUIRED FOR CROSSTALK. TC"A" TC"B' Fig. 5 — Intertoll routing pattern between two regions showing typical circuit groups. TRANSMISSION DESIGN OF INTERTOLL TELEPHONE TRUNKS 1035 the amount of traffic will not support the cost of complete automatic alternate routing features. This will cause some additional complication, for each such point introduces another source of echo due to the fact that the capacitance and resistance of the office cabling reduces the balance obtainable when two intertoll trunks are switched together. The effect of such switching on VNL's can be cared for by adding appropriate loss increments which will be small if a careful job of impedance matching is done and the distances from the toll terminal equipment to the switches is held within bounds. No increment is added if the return loss for about 84 per cent of the cir- cuits in the group is 24 db or more. These increments increase to about 0.2 db for a return loss of 20 db, 0.4 db for 18 db and so on. They are added to VNL of circuits between a two-wire switching point and a four-wire switching point and of circuits between two two-wire switching points. Impedance matching is usually accomplished by adding capacitance across the compromise network and in some cases across the shorter cable runs in an office. All circuit losses referred to in this paper are 1000-cycle values, i.e., no allowance is made for the effects of noise and frequency distortion. Careful design, layout, and coordination of individual transmission systems are depended on to keep noise within proper bounds; and all new carrier systems going into the plant have transmitted bands wide enough to require no assignment of distortion transmission impairment (DTI). Circuits having excessive noise and those circuits with large DTI's are earmarked for improvement by any means that may come along. But beyond this, frequency distortion does not enter into VNI calculations since it can not be offset by reducing circuit losses without encountering trouble from the echo or other standpoint. While we have considered only circuit design in this paper, it is evi- dent that the success of the whole plan also depends on how closely circuit losses are maintained. This is important from two aspects. 1. The expected variations determine the allowance which must be made in the assigned loss. As indicated previously, it is expected that an allowance of 0.4 db per link will be adequate for the near future and it is hoped that as time goes on this figure can be reduced. 2. A more important factor is that unless circuit losses are maintained fairly precisely, large positive or negative excess losses can be accumu- lated on multi-switched connections. Avoidance of such large excesses is particularly important with dial operation since detection and avoid- ance of unsatisfactory transmission conditions by operators will be much less effective. I 1036 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 While the maintenance problem is at least as complex and difficult as the design problem, it is beyond the scope of this paper. SUMMARY To summarize the preceding discussion : For the particular conditions in the Bell System, a formula has been set up to give adequate approxi- mations of the lowest practicable loss for practically all intertoll trunks as follows: VNL = VNLF X L + A + B, where: VNL = "Via Net Loss" (db) of the trunk. VNLF = "Via Net Loss Factor;" i.e., a factor which depends on and is appropriate to the type of facilities used in the trunk. L = Length in miles. A = Design allowance for expected variations of circuit loss in service (0.4 db). B = Amount to be added if two-wire switching is used; the mag- nitude depends on the passive return loss obtainable on such connections at the two-wire switching office. At each end of the connection a loss of 2 db (/S = 2) is added by ap- propriate means as discussed earlier. CONCLUSION Let it be emphasized that we have been talking largely of planning for the future in all that has preceded, for the switching plan as outlined is a growing thing and it will be a couple of years before much complex automatic alternate routing is done. And we would be very much sur- prised to escape growing pains and change of ideas as the plan develops. We are confident, however, that the plan is sound economically and transmission-wise; and flexible enough to adapt itself to further de- velopments and experience. ACKNOWLEDGMENT As in most papers like this it would be prolix to mention all persons who took an active part in the preparation or in the development of the background data. But the author would be remiss if he did not call by name L. L. Bouton, who just prior to his recent retirement from Bell Telephone Laboratories, did much of the basic work on the mathematical concepts involved, on the simplification of these concepts for practical application, and on the re-evaluation of data that was required in these applications. The Card Translator for Nationwide Dialing By L. N. Hampton and J. B. Newsom (Manuscript received August 24, 1953) Nationwide operator and customer dialing requires the existence of a num- ber of switching centers equipped with automatic systems having a much higher order of mechanical ^^ Intelligence^' than previous systems. One of the most important components of this new switching system is the Card Trans- lator. Its function is to take the telephone address of a call and determine how to advance this call toward that address. This translator has to meet unique requirements in that it must accommodate a very large number of addresses; must provide a great amount of information for routing the calls; and must enable quick and convenient changes to be made in its stored in- formation. It must also meet, of course, the normal basic requirements of reliability, economy, long life, etc. The fundamental principle of this trans- lator is that of a card file, containing individual coded cards for each des- tination, with routing information recorded on each card. Whenever the routing information for a specific code is needed, the system selects the ap- propriate card, and reads the information by means of electronic circuits employing phototransistors and transistor amplifiers. INTRODUCTION The "Card Translator" was developed for the 4A toll crossbar system used at Control Switching Points (CSP) in the nationwide dialing net- work of offices. Although the many problems and conditions presented in the development and implementation of a nationwide dialing plan have been discussed in papers* by A. B. Clark, J. J. Pilliod, H. S. Os- borne, W. H. Nunn, F. F. Shipley and others, it is necessary to restate some of these because of their effect on the translation problem and the features the card translator had to have in order to meet the nationwide dialing requirements. Translation is the process of converting the called destination code into information that is needed for the proper routing of the call. * B.S.T.J., 31, pp. 823-882, Sept., 1952. 1037 1038 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 NUMBERING PLAN For nationwide dialing it is necessary that each customer have a distinctive universal number. This numbering system is accom- plished by dividing the country and Canada into about ninety num- bering areas. Each of these areas is assigned a distinctive three digit code which, in order not to conflict with local office three digit codes, has either the digit "1" or "0" as the second digit. Within these num- bering areas each local office will have a distinctive non-confficting, name and number code. Since each customer in an office has a distinctive number, a corresponding distinctive nationwide universal number is thus provided. To reach a customer outside the local numbering area will require the dialing of three digits for the area code, three digits for the office code and four or five digits for the line number. Thus by dialing ten or eleven digits a connection can be made to any customer an3rwhere within the country and Canada. Fig. 1 shows the present numbering area code assignments for the United States and Canada. TOLL LINE SWITCHING NETWORK A second requirement for nationwide dialing is the provision of about 70 strategically placed automatic switching toll offices called control s^vitch points (CSP) throughout the United States and Canada. The switching system used at each of these 70 or so CSP offices have sev- eral new features as follows: 1. Six digit translation. 2. Ease in changing and adding routings. 3. Automatic alternate routing. 4. Code conversion. 5. Storing and sending forward digits as needed. BASIC SWITCHING ARRANGEMENT In the CSP offices the transmission paths are established through crossbar switches mounted on the incoming and outgoing link frames as shown in Fig. 2. The setting up of the connection through these switches and the linkages is controlled by equipment common to the office which is held in use only long enough to set up each connection. The major items of common control equipment are the senders, de- coders, markers and card translators. The sender's function is to receive and register the digits of the called destination, to transmit the area and, if required, the office codes CARD TRANSLATOR FOR NATIONWIDE DIALING 1039 to the decoder, and then, as directed subsequently by the marker, to send digits ahead as may be required. The decoder's function is to receive the code digits, either 3, 4, 5 or 6, from the sender and to submit them to the translator for translation and to make selections of alternate routes as required to route the call to the destination. The decoder also gives instructions to the sender and marker to enable them to carry out their functions. The marker gets access to an outgoing trunk group through the trunk block connector and selects an idle outgoing trunk in this group, then chooses an idle linkage between the incoming and outgoing trunks, operates the crossbar switches to close the transmission path, and gives the sender information for pulsing ahead as may be required. The operation of these common control circuits is briefly as follows. On the arrival of a call the incoming trunk is connected to a sender through the sender link frame. When the code has been registered the sender makes connection to the decoder through the decoder connector cu-cuit. The decoder passes the code to the translator for translation. There is a translator associated with each decoder which contains the three-digit cards for the local offices and area codes, also perhaps some six-digit code cards. However, most of the six-digit cards (there may be several thousands of them) will be in a group of foreign area translators used in common by all decoders. The decoders obtain information from the area code card for selecting the particular one of these translators. Connection to them is made through an appropriate translator connector. The translator gives information for the selection ot an outgoing trunk and passes other information for routing the call both to the decoder as well as to the marker. The marker proceeds with trunk test and the operation of the switches to establish the connection. When the proper information has been given to the marker, then the decoder and trans- lator release from that call. When the marker has given information to the sender for routing the call and has established the talking connec- tion, the marker releases. The sender releases as soon as it has finished pulsing ahead. In this way these common control units handle many calls in rapid succession. SIX DIGIT TRANSLATION This feature is needed in nationwide dialing because from a particular CSP to points in another numbering area there may be several routes or trunk groups. To reach a point in such an area it is necessary that the office code as well as the area code be translated to select the route to k 1040 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Ift . 604 06CEMKR 19S2 OKLAHOMA AND ONTARIO ARE TO BE DIVIDED INTO TWO AND THREE NUMBERING PLAN AREAS RESPECTIVELY. THE NEW AREA CODE NUMBERS HAVE NOT BEEN ASSIGNED Fig. 1 — Nationwide toll dialing CARD TRANSLATOR FOR NATIONWIDE DIALING 1041 CONCOR0>{A. BOSTON MASON CI LANSING' \ A \ • ,'' DETROIT / SYRACUSE f- \ • ». ALBANY* I ^^,3-— r-J^ ^-V 716 j ^^ KINGSTON- .-r^TT \ >«0^0l 0 ijy ..xrv.^,= L \ ^^-^^/^ PROVIDENCE •jl^MONTlCELi^O \ ^^^^^^^ « ^, i| 1 >' _^ y / y^ 1 1 SCRANTON , . ...ILWAUKEE V % \ 'iSZjL^' 4^ •KANE\ -,<-. -O NEWARK _, "'^'""°°^'^''i7nrprRK'/° •'^^t!^t!^s2^^^'ir'\^ sk ^ ^"//'"T'^NEwroRK CANTON 1^'2 , • , ;;^^^,,t^£^201 FORT DODGE (WATERLOoN^ "" ' 1 1 I ROCKFORO* I ~ 2 I 515 1 319 ; CHICAGO*^^^-^ \ TOLEDO T 216 \aip\*^''°<^' 312 1 SOUTH BEND I ^jg |_ CANTOnH'^ 1 * 219 \ MANSFleLD^ ,• ''Wpittsburgh — ' • *-'' " ' — ' BALTIMORE 0* CF 815 'v' I DES MOINES OTTUMWA«^> f< w / PEORIA ^ r- • 1 INDIANAPOLIS •. --> iw y - "u/A.iuikirTf CHAMPAIGN 1 . 3,^ J ^.^ ^ 614 ^LAptsBURG^r^^ MANSFIELD^ 1 """TCOLUMBUS \ DAY TON \« 301. -302 816 \»^OBERLY. SPRINGFIELD jTERRE HAUTE— < ^^^.e. ^CINCINNATI r* 304 ^HUNTINGTON RICHMOND 703 KANSAS CITY \ Vr' , / .^^^''^J^O^^^'^^^ \ch'"^"^°^' ■— 1 \ ST. LOUIS/ cENTRALlA/.e^'" r^»^ • X ^^ • NORFOLK^ . • — -~^^ \ ,, , \. ^.„ f^-N./^ LEXINGTON /V/"^ ROANOKE -\ -o 314 X^eisj' ^Q2 ^ \ CAPE»\ ^1 ' \ GIRARDEAU •joPLiN ^V J « ' ^ 704 ^=^ Nashville knoxville y^ charlotte 501 little rock 504 MEMPHIS 601 JACKSON °^' CHATTANOOGA, \ ATLANTA BIRMINGHAM \ AUGUSTA*^ • COLUMBIA .0. \ \ 404 V MONTGOMERY J JACKSONVILLE 713 305 le United States and Canada. 1042 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 the desired point in the area. To other areas there may be only one route and in this case the area code will suffice. The provision of facilities for the translation of six digits greatly affected the design of the switching system for the nationwide dialing CSP offices. It led to the development of the basically new card trans- lator. In previous toll common control systems translation is done by means of relays. The code digits, never more than three of them, re- sulted in the operation of groups of relays in certain combinations and led to the eventual operation of a route relay for the particular com- bination of code digits. This route relay with cross-connecting facilities from its contacts is used for the identification and selection of trunk groups and other information as might be required for the particular code routings. To change a routing with this system of translation re- quired the removal and reconnection of many cross-connections. With the nationwide dialing plan in operation, routing changes or opening of new offices in one part of the country will necessitate trans- lator changes in many offices, some of them far removed from the scene of the event that forced them to be made. The changes in any one CSP may, therefore, be frequent under certain conditions, and to make them by running cross-connections would be cumbersome and expensive. The new translator uses punched cards instead of relays making it possible to effect changes by the simple process of removing old cards and insert- INCOMING LINK FRAMES OUTGOING LINK FRAMES INCOMING TOLL LINES < OR TRUNKS TANDEM TRUNK u OUTWARD TOLL BOARD CALLING SUBSCRIBER SENDER LINK FRAME SENDER ^i;- DECODER CONNECTOR MARKER CONNECTOR o TO DISTANT ' TOLL OFFICES TO LOCAL OFFICES TRUNK BLOCK CONNECTOR FOREIGN AREA TRANSLATOR DECODER TRANSLATOR -FOREIGN AREA TRANSLATOR CONNECTOR Fig. 2 — Schematic diagram of crossbar switching system for CSP's. CARD TRANSLATOR FOR NATIONWIDE DIALING 1043 ing new ones in the machine. This can be done in a very short time and requires less out-of-service time for the equipment. CARD FILE A card is provided for each area code and also one for each office code that must be translated in a particular CSP, the cards representing destinations. The cards are lined up in a box as in a filing drawer with tabs along the bottom of the cards resting on select bars which run the length of the box. It is by operating the select bars in combinations, depending upon the code, that the particular card for the destination is selected. Each card, as shown in Fig. 3, has tabs, one for each select bar along the bottom edge. The information presented to the card translator for the selection of a card is in the form of code digits on a two-out-of-five basis. Each card is coded by removing all of the tabs ex- cept those that represent the particular combination of select bars for the particular code. When a code is presented to the translator, a combination of select bars corresponding to the code is lowered and the card having all tabs removed except those that were resting on the lowered selects bars will be selected while all other cards will remain in their normal positions. The groups of tabs labeled. A, B, C, D, E and F (Fig. 4) are for the six code digits. For each digit, two tabs remain, since the digits are ENLARGED HOLE FOR USE WITH (AS REQUIRED BY CODING) CARD LIFT BAIL FOR USE WITH CARD LIFT BAIL AND TO UNENLARGED HOLE "^OR USE WITH CLEAR BIN BAR (118 BEFORE CODING) BULK HANDLING TOOL WIDE TAB FOR CARD \ ALL TABS REMOVED EXCEPTING ^ '^ ?,'?,>' ^.^k.^I,^, u SUPPORT BAR THOSE ASSOCIATED WITH OPERATED CODING AND BULK FOR USE WITH ^^^^ SELECT AND CARD SUPPORT BARS HANDLING TOOLS CODING TOOL Fig. 3 — Typical coded card as seen from the phototransistor side. 1044 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 registered in the sender on a two-out-of-five basis and the leads from the sender will cause the select bars to be operated. If the card represents an ordinary three digit code, all the tabs will be cut off except two each for the A, B, and C groups of tabs, and, for reasons that will be discussed later, two of the four CG tabs, card group, will also be removed. In addition, either the VO or NVO tab may be removed. The VO and NVO tabs are used when the group of toll lines over which the call will be routed is divided into one sub-group of a transmission grade suitable only for terminal traffic (NVO, meaning ''not via only") and another subgroup for either terminal or switched-thru traffic (VO mean- ing "via only"). If the card represents a six-digit code, two tabs will be left in each of the six code digit positions and a different pair of card group tabs will be used. The cards, then, are in different groups and are selected by combinations of code select bars together with the card group select bars. As noted there are four of these card group bars, CGO, CGI, CG2 and CG4. They are used in the six possible combinations, two at a time as follows : For the regular three-digit code cards CGI and CG4, for the regular six-digit code cards CG2 and CG4; the other four combinations CGO and CGI, CGO and CG2, CGI and CG2, and CGO and CG4 are used for selecting the four groups of alternate route cards, which may be of the three-digit or six-digit variety. CODE CAPACITY The card translator by means of the code select bars and card tabs provides facilities for a great number of different codes and routings. There are 40 select bars provided, 36 of these are used in the combina- tions as has been described, two are reserved for possible future use and the remaining two are used for aligning the cards as will be discussed later. The total possible card code combinations is sufficient for growth in nation-wide diaUng for the foreseeable future. TRANSLATOR CARD CAPACITY In some CSP offices there will be many thousands of cards for the destinations to be reached. It was not mechanically practicable to design a single translator capable of accommodating all these cards and more- over it would not have been economical, particularly for many of the smaller CSP offices. Also for service hazard reasons and to provide for the simultaneous translation of several calls, needed to handle traffic during heavy load periods, more than one translator is required in a CARD TRANSLATOR FOR NATIONWIDE DIALING 1045 CSP. Considering these factors, it was found desirable to design the card translator to accommodate over a thousand cards. TRANSLATORS Since several translators are needed in a CSP, for further economy and consistent with service hazards, the translators are segregated ac- cording to the groups or kinds of cards they contain. These are as follows : Decoder Translators These translators contain the local three-digit office code, the three- digit area code, the alternate route code and, where space permits, some of the high usage six-digit foreign area code cards. Several of these translators are furnished, one for each decoder, depending upon the volume of traffic handled by the particular CSP. Foreign Area Translators These are furnished, one or two per office (maximum 19), for each 1000 or so foreign area code cards. These translators are in a common pool and the particular one is selected when needed for the six-digit code to be translated. Decoder Foreign Translators If there are sufficient high calling rate, six-digit, foreign area cards, to justify it, these translators may be provided, one per decoder. Emergency Translator Provisions are made so that the emergency translator can be sub- stituted for any other translator by changing the cards of the trans- lator in trouble to the emergency translator. TRANSLATOR OUTPUT The translator output information required for a CSP office for na- tionwide dialing must be very extensive to accommodate the many varieties of routes over which calls must be completed. The need for automatic alternate routing, code conversion and the storing and sending forward of digits also affects the need for increased translation output. The output is provided by means of the 1 18 holes in the face of each card. 1046 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 CARD TRANSLATOR FOR NATIOKWIDE DIALING 1047 By enlarging these holes in combinations the output information for the particular route is obtained. As seen in Fig. 4 the top holes beginning at the left of the card are used for ''pretranslation" purposes. The senders are not provided with facilities which enable them to predetermine when to present for trans- lation the first three digits received or when to wait for more than this number as when six-digit translation is required. Therefore the sender always requests translation when the first three digits are received. So for six-digit calls, the sender must be informed to disconnect from the translator after three digits have been received and wait for six digits. The CA6, (come again six) hole in the card is used for this pur- pose. The CA4 and CA5 holes are used similarly for calls to certain four- or five-digit operator codes, informing the sender to apply again for translation T\4th four or five digits. The NCA (No Come Again) hole is used for three-digit calls. The "OGT" holes are used to inform the common control equipment on which train of s^\dtches the outgoing trunk appears to enable the associated circuits to select the proper switching train. The remain- ing holes on the top line are for controUing operation of traffic meters. On the second line, the translator box number holes are used on area code cards to indicate which translator contains the particular cards for the called area when six-digit translation is required. The INDI hole on the second line and the IND2 hole on the fourth line which commonly are referred to as index holes are never enlarged. They serve as an indication that a card has dropped and that all is ready for trans- lation output detection. These index holes also aid in trouble detection in case of light failure, for routing of certain calls where cards are de- liberately omitted and for calls where a blank code was dialed in error. The class holes are used for indicating the type of outpulsing and the kind of signalling channels used on trunk groups out of the office. The AREA CODE CONTROL holes on the third line are used for deter- mining the number of digits to be transmitted forward to the next office and for supplying undialed code digits needed primarily in connection with automatic alternate routing. The alternate route pattern NUMBER holes are used for the selection of the series of alternate routes to be used. The holes on the fourth line are for making proper disposition of calls when all trunks are busy and to inform the associated circuits how many digits should be received for the particular code. The CODE CONVERSION holes on the fifth line are used to supply the sender with information as to the outpulsing of certain arbitrary digits 1048 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 as may be required through step-by-step toll trains. Facilities are pro- vided for the outpulsing of one, two or three digits as may be required. The VARIABLE SPILL CONTROL holes On the sixth line inform the sender when to pulse forward all digits as received, or to omit sending the first three or six code digits. The remaining holes on the card define the location on the switching frames having the desired outgoing trunk appearances. The notches around the outer edges of the card are for proper positioning of the card in the stack and for card removal purposes as will be discussed later. CARD OUTPUT DETECTION The enlargement of the holes in the face of the card to obtain the translation output as previously stated is recognized by means of modu- lated light beams falling on phototransistor detectors. With all of the Fiij. 6 — Normal (above) and dropped card views as seen from the photo- transistor side. CARD TRANSLATOR FOR NATIONWIDE DIALING 1049 CARDS LIGHT SOURCE LIGHT PHOTO AC TRANSISTOR AMPLIFIER AC GAS TUBE DC ■ RELAY TRANSISTOR -^ Fig. 6 — Block diagram of channel circuit. cards in their normal positions they are aligned and the holes in the cards form unobstructed, horizontal tunnels called channels through the entire stack. When a particular card has been selected and dropped a distance slightly greater than the height of an unenlarged hole, these light beams through all of the channels are blocked by the dropped card except for those holes which have been enlarged. This results in a pattern of clear channels which represents the translator output information. The change in the silhouette of the stack from the condition of all cards normal to that of one card dropped is shown in Fig. 5. THE CHANNEL CIRCUIT Fig. 6 is a block diagram of the circuit used to determine whether a particular channel is interrupted by a dropped card or not. Each block, with the exception of the light source, represents a piece of equipment provided individually for each channel. The light source is common to all channels. If the hole in the dropped card for a particular channel has been enlarged, the light will pass completely through the stack of cards and fall on the phototransistor. The phototransistor converts the light into an electrical signal which, after being increased by the transistor amplifier, is used to trigger a cold cathode gas tube. The gas tube in turn operates the channel relay. This relay is located in the associated equip- ment which uses the information supplied by the translator to process the call. In making a detailed examination of the channel circuit, it is con- venient to consider it in two parts: the optical section and the electrical. The optical section includes everything up to the point where the light falls upon the germanium of the phototransistor. This part of the chan- nel is shown functionally as Fig. 7. The light source is a standard pro- jection type lamp normally rated at 500 watts. To obtain long life it is operated in the translator at about half of its rated voltage at which level its input is approximately 170 watts. This type of lamp was chosen because of its high concentration of light in a small plane area. The light from the lamp passes through a motor driven perforated disc which modulates it with an approximate square wave at a 400-cycle rate. Modulated light is used because it is more economical to use ac 1050 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 MODULATING DISK PHOTO- TRANSISTOR LAMP Fig. 7 — Optical section of channel circuit. than dc amplifiers and also since the difference between the light and dark currents of the phototransistor is the important factor rather than the absolute value of either. The modulated light is collimated by a lens to minimize the loss as the beam passes through the holes in the card. Unless interrupted by a dropped card, this light beam will bass through all of the cards and fall on the lens which focuses the light on the sen- sitive area of the phototransistor. The light intensity at the lens of the phototransistor is 34-foot candles minimum. This is equivalent to about 12 millilumens at the phototransistor, a figure relatively small when compared to the light intensity that is required by conventional photo- electric cells. The electrical part of the channel circuits starts with the photo- transistor and is shown in Fig. 8. The light acts as the emitter of the phototransistor. The collector is of the conventional type for point contact transistors. As is normal in grounded base transistor circuits, the collector of the phototransistor is biased in the high impedance di- rection. A variation in the light intensity causes a variation in the col- lector impedance of the phototransistor. The type used has an impedance of about 10,000 ohms when dark which is reduced to approximately 3,000 ohms when illuminated. The output of an illuminated photo- transistor when coupled to the amplifier ranges from 1.3 to 12 volts positive peak at 400 cycles depending upon the age and condition of the transistor. Since the discrimination by the channel circuit between a clear or blocked light path depends upon the presence or absence of an ac output from the phototransistor, noise of sufficient magnitude, if present when the channel is dark, would cause a false indication. To guard against such false indications each phototransistor is checked during manu- facture for dark noise. During a five minute interval the dark voltage must not exceed 75 millivolts. CARD TRANSLATOR FOR NATIONWIDE DIALING 1051 The phototransistor is coupled to the amplifying transistor by trans- former Tl. This permits convenient matching of impedances and separa- tion of the dc bias voltage. A voltage limiting varistor, V, is connected across the input of the transformer to limit surges which might other- wise damage the amplifying transistor. The circuit of the transistor amplifier is a conventional arrangement. Voltage gain of the amplifier, including the input transformer to the gas tube, varies from 40 to 100. However, when operating in the trans- lator, the phototransistors normally will drive the amplifier to saturation TRANSISTOR AMPLIFIER MODULATED] T LIGHT sion. PHOTO TRANSISTOR I I + 130V 6 R T -24V ^ J^||^ ASSOCIATED EQUIPMENT *1 + 130V Fig. 8 — Electrical section of channel circuit. which limits the output to 160 volts positive peak. For the purpose of guaranteeing operation, a minimum output voltage of 38.5 has been set as a rejection point for a phototransistor-amplifier combination. The output of the transistor amplifier is normally sufficient to break down the control gap of the cold cathode gas tube. Sufficient current flow^s in this control gap to insure reliable transfer to the main gap when the output control relay O in the associated equipment operates. To aid deionization of the gas tubes, the bias of —24 volts is removed from the control anode just before channel operation is required. The relay R, which replaces the bias voltage with ground, is operated by a circuit which checks that the card being dropped is completely down. This "down" check circuit utilizes the two index holes in the card which are never enlarged and employs two phototransistors to detect the presence or absence of light through these holes. Fig. 9 shows the card down check circuit. It differs from the routing information channel circuits in that the operation of a relay is required 1052 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 when the light is blocked. Therefore the transistor amplifier-gas tube circuit is not appHcable. For the down check channels, the output of the phototransistors is amplified by the first section of a conventional double triode therminonic emission tube VI. The ac signal at the plate of VI is rectified by a conventional full wave rectifier consisting of transformer T3 and tube V2. The rectifier voltage is negative with respect to ground and when it is impressed on the grid of the second section of tube VI, that section is driven beyond cutoff. Therefore, as long as light falls on the phototransistor, no current flows through the relay. When the ii PHOTO TRANSISTOR 4000 n tt isoon. i65n ' WW— vw 165A Fig. 9 — Card down check circuit. dropped card blocks the light, the negative voltage on the grid disap- pears and the tube conducts, thus operating the associated relay. While the prime purpose of this down check is to signal the associated equip- ment that the card is in position for recording its output code, advantage is taken of this circuit's ability to distinguish between a light and dark phototransistor at any time to provide alarms in the event of a lamp or modulating disc failure. After the card has been checked down and the associated equipment is ready to accept the output of the card, that equipment connects a positive 130-volt battery through its channel relays to the main anodes of the gas tubes. All gas tubes associated with illuminated channels will have their control gaps fired and will transfer to the main gaps. This operates the corresponding channel relays in the associated equip- CARD TRANSLATOR FOR NATIONWIDE DIALING 1053 merit. The relays in operating lock to ground and thereby extinguish the main gap discharges thus increasing the hfe of the gas tubes. Those channels which have been blanked out will not have the control gaps of the gas tubes broken down. Therefore when the 130 volts is applied, the relays associated with the darkened channels will not operate. The operation or non-operation of the relays in the associated equipment completes the function of the channel circuits. The capacitor and resistor network at the main anode of the tube is to prevent transients due to the operation of other channels from falsely firing the main gap of a dark channel. CHANNEL PACKAGES The phototransistor is mounted in a metal tube along with a lens that focuses the colliminated light on the transistor. Fig. 10 is a cutaway LENS COLLECTOR' ^GERMANIUM BLOCK Fig. 10 — Cut away view of phototransistor. view of the 3A phototransistor showing the relationship of the lens to the transistor. To mount in the translator, the tube is slipped into an accurately positioned hole and clamped in place using the slotted ear. This mechanical fastening is also the ground connection. The output lead from the collector is attached using a slip-on connector. The amplifying transistor, transformer Tl, varistor V, the resistor and two capacitors of the amplifier are packaged in a convenient "plug- in" unit. Fig. 11 is a photograph of these amplifiers. As shown, the transistor is mounted under a removable cap on the package so that it may be conveniently replaced if necessary. The gas tube, transformer T2, and the associated resistor and capaci- tor are also assembled as a packaged unit. TRANSLATOR CIRCUITRY The card translator is mounted on an associated translator table as shown in Fig. 12. The translator table contains the transistor amplifiers 1054 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 and the cold cathode tubes one for each of the channel circuits as pre- viously discussed, and, in addition, the miscellaneous relays through which the operation of the translator is controlled. As already stated, the connection between a translator and the associated decoder is through suitable connector relays which are either a part of the decoder or on a separate connector frame. The phototransistors may be adversely affected by temperatures greater than 130°F. Therefore, to provide satisfactory operation during sustained heavy traffic load periods and with high ambient room tem- peratures, an air circulating and filtering unit, as shown in Fig. 13, may be provided. This unit mounts on the translator in place of the regular end cover and otherwise requires no further apparatus change. For convenience in ordering, the "IB" translator is specified when the air cooling unit is desired. Otherwise the "lA" translator will be furnished. The translator table contains relays for controlling the dropping, checking, and restoring of the cards in the translator. First are the relays that operate the card "pull-up" and "pull-down" magnets. Then there are select code bar control relays, one for each bar, operated from the sender or decoder, which in turn operate the associated select code bar solenoids. These relays are necessary since the lead resistance from the sender to the translator would adversely affect the operating time of the code bar solenoids. Finally, there are relays that check the code bars for proper operation on a "two-out-of-five" basis. Two, and only two code PLUG -IN BASE TRANSFORMER, VARISTOR AND RECTIFIER PLUS TWO CAPACITORS MOUNTED IN THIS PORTION Fig. 11 Transistor amplifier. CARD TRANSLATOR FOR NATIONWIDE DIALING 1055 r liii i ilii 8^ Fig. 12 — Card translator and table. bars for each digit must be operated, before an attempt is made to drop a card. The operating cycle of the relays and the translator in selecting a card and restoring it is then as follows: When the translator is first seized the pull-up magnets are energized and when the cards are suspended the latches are operated. The decoder then closes the leads over which the code bar relays operate and they in turn operate the code bar solenoids. When the proper number of these are operated a check for this is made which releases the latch. 1056 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 With the latch restored to normal the pull-up magnet is de-energized. All the cards then drop until they meet the code bars, about 0.016". The card for the particular code, however, continues 0.180" further, because the bars for all the tabs have been lowered until it rests on the pull-down magnetic pole face. The index channel relays then operate causing the decoder to read the output of the card. When the decoder has checked that it has received proper information from the card, and on certain calls when the marker has selected an WARM AIR OUTLET i AIR INLET FILTER r-VENTILATiNG LOUVERS / ///- / Hum Mfi'tif HOtifu ««»*%'■ mm feffi: mmit '"'""" mini'' Fig. 13 — Card translator equipped with cooling unit. CARD TRANSLATOR FOR NATIONWIDE DIALING 1057 outgoing trunk, the decoder releases the translator. Just prior to this, the decoder operates an automatic restoral relay in the translator, which causes the card to be restored to its normal position in the stack. This takes place by again energizing the pull-up magnet, and, when the cards are again suspended, the latches are operated and the code bars are released, restoring to their normal positions. The card that was down is restored by the code bars to its position in the stack. All the relays in the translator table then release except for a slow releasing relay which remains operated holding the pull-up and latch magnets energized so that a subsequent call does not have to wait for these magnets to operate and suspend the cards. This feature saves time under heavy traffic. TRANSLATION TIME The call carrying capacity of a translator and the number of trans- lators for a particular CSP is directly related to the time required for translation, that is, the selection, reading and restoral of the translator cards. For this reason, particular attention to translation time was used in the design, not only of the translator, but throughout all the associated circuits of the switching system. Since the translators in service are always controlled by the decoder circuits and since the number of decoder translators and decoder foreign area (DFA) translators are directly proportional to the number of de- coders, the times for the various types of translation and alternate routing will be given in terms of the holding time of the decoder. PRETRANSLATION This requires approximately 235 milliseconds. On this type of call translation of three digits, when either four, five or six digits are needed for the routing, informs the sender to release the decoder and wait for the rest of the code. THREE DIGIT TRANSLATION The time for this type of call is approximately 330 milliseconds. This is when three digits are sufficient for a routing. SIX DIGIT TRANSLATION The time for this is approximately 550 milliseconds, assuming that the six-digit card is in the decoder translator. This does not include the 1058 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 pretranslation time. Pretranslation does not occur if the office code as well as the area code is registered in the sender before the decoder is connected. This often occurs during periods of heavy traffic. On six-digit translation, two cards must be translated, the three-digit area code followed by the six-digit destination code card. SIX DIGIT TRANSLATION IN FOREIGN AREA TRANSLATOR (fAT) This type of call requires approximately 560 milliseconds. The area code card in the decoder translator gives the information for selecting the particular FAT in which the six-digit card is located. The trans- lation time will be extended if there is a delay, due to another decoder using the FAT. Pretranslation time, as stated above is not included. SIX-DIGIT TRANSLATION-PRINCIPAL CITY AND VACANT CODE ROUTING (fAT) In this case translation requires approximately 615 milliseconds. This is for routings to areas where there is a principal city (PC), usually an- other CSP, through which all calls to that area can be completed, although in the area there are other destinations reached over direct high usage trunks. In this case, to save cards and perhaps translators, the six-digit cards for all destinations reached directly through the PC are deliber- ately omitted. The time given is for a call to such a destination. The three-digit card for the area has on it information for the proper routing of the call to the principal city. For those destinations where the six- digit cards are omitted, as well as for vacant codes in such areas, the call is routed to the principal city. Pretranslation and foreign area trans- lator delay times, if any, are not included. THREE-DIGIT CARD-TO-CARD TRANSLATION This type of card-to-card operation is used where there are several sub-groups of trunks or routes to the destination and the decoders do not have facilities for determining which group of trunsk has an idle trunk. The routing information for each trunk group must be presented successively to the marker in selecting an idle trunk. The translation time, considering that an idle trunk is selected from information on the first card, is approximately 330 milliseconds, assum- ing no marker delays. When the trunks for the first card are found busy and routing is made from the second card, the total time is about 550 milliseconds. This increases to about 770 milliseconds for routings CARD TRANSLATOR FOR NATIONWIDE DIALING 1059 from the third card and finally to 1060 milHseconds for the fourth card routing. SIX-DIGIT CARD-TO-RELAY TRANSLATION This type of operation is used where the first group of trunks to the destination is of the type that requires test by the marker for selecting an idle trunk. There are alternate routes, however, in which the decoder can determine which group has an idle trunk. In this case the area code card provides information for the selection of the first six-digit card. This card has information for routing the call over the first group of trunks. There are alternate route cards for each subgroup of alternate routes. There may be as many as five alternate routes, each of which may have as many as four sub-groups of 40 trunks each. The translation time, assuming the routing is from the first card, is about 550 milliseconds. For routings from any one of the alternate route cards, the time increases to approximately 800 milliseconds. Pretransla- tion and FAT delay time, if any, are not included. THREE-DIGIT CARD-TO-RELAY TRANSLATION The time required is about 350 milliseconds for routings from the first card and about 580 milliseconds for routing from an alternate route card. SIX-DIGIT RELAY-TO-RELAY TRANSLATION This type of operation is used where all of the trunks including the alternate routes are tested by the decoder in determining in which group there is an idle trunk. The translation time, assuming the routing is from the first six-digit card, is approximately 575 milliseconds. For routings to succeeding alternate routes the time is approximately 800 milliseconds. MAINTENANCE FACILITIES Although the card translator and all of its components are designed for relatively long and trouble-free life, adequate testing facilities and maintenance procedures are essential because of the importance of the individual CSP's in nationwide dialing. Adequate guards and methods of procedure have been made available in case of almost any catastrophe that would incapacitate any CSP office. Moreover the complete break- down of a translator or even several of them in a CSP office would not completely stop calls through that office although the call carr3dng 1060 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 Fig, 14 — A trouble record card is being removed from the perforator and in- formation from a punched card is being entered in the office records. capacity of the CSP office or the switching system would be reduced. However, to insure satisfactory operation of the card translators auto- matic trouble recording, test circuits and maintenance methods are provided. TROUBLE RECORDER In CSP offices, where the card translators are used, the card punching type of trouble recorder will be provided as shown in Fig. 14. In the event of a failure of any translator, the associated decoder circuit will block and time out. This causes the connection of a multiplicity of leads between the trouble recorder and the decoder as well as to the translator. The trouble recorder will then automatically punch and drop a properly designated paper card (Fig. 15) that shows which translator and decoder are in trouble. The state of the various elements of the translator and decoder will also be recorded. In addition, a record is made showing which code bars, pull up and pull down magnets, latches and important relays are operated. Also the important key relays which are operated in the decoder will be shown. The associated sender will be identified as will also the marker if one is connected. Then too, the state of the marker will be recorded. From this trouble record the cause of the failure can, in most cases, be determined. When the trouble recorder card has been punched, the associated CARD TRANSLATOR FOR NATIONWIDE DIALING 1061 equipment is directed to make another trial. This usually will be with an alternate decoder and translator so that the call is usually completed with a delay of a little more than one second required for punching the trouble recorder card. The trouble recorder, once it has completed punching a card, is immediately available for recording another failure. The trouble recorder is also available for recording failures of other CSP equipment such as the controllers, senders and markers in a manner similar to that described for the decoder and translators. TRANSLATOR TEST There are facilities provided on the trouble recorder frame, through keys and connecting relays, for adding and removing cards from the translators. On adding a card, a check can be made to verify the card output to make sure that it is in agreement with the template from which the card was coded. Test calls using all decoders and markers can be made to verify the selection of a trunk in accordance with the routing information on the new card as well as on any card in any translator. These tests can be caused to recycle automatically and to drop a trouble recorder card in case of failure. This feature is useful in isolating in- frequent failures in any of the associated translators, decoder or markers. Facilities are also provided for removing the translator selector unit for periodic inspection and adjustment as may be necessary. Also timing tests can be made from the trouble recorder frame to check the time required for the translator to drop and restore a card. To determine if any of the output channel elements of the translator, which includes the photo-transistors, transistor amplifiers and cold cathode tubes, for all the channels, have satisfactory operating margin, means are provided for making a mass test of the channels. This test is made under a controlled voltage (36.5 volts) which is considerably below the worst service condition. In case of a failure of an element under this test, a trouble record card will be dropped which will show, by a punched hole, each element that is operating satisfactorily. This mass test is provided to detect any channel element that is approaching end of life and thereby assures that there is at all times ample operating margins of these important channel elements. Portable Test Set In addition to the test facilities provided on the trouble recorder frame, there is also a portable test set for the translators as shown in Fig. 16. This test set is connected to the translators by means of multicontact 1062 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 0-5 10 15 20 25 29 it S3 S2 51 SO R8 R7 R6 R5 04 R3 R2 Rl RO 0 M C OT MT TV CT |TRI TR2 OECOOER TRI TR2| 0 1 2 3 4 5 6 7 8 9 10 II 12 13 14 15 16 17 FIF MFT CFROSIIMSTl RTRF TST SOT 06T 0 1 2 3 4 5 6 7 8 9 1 0 1 2 3 4 5 6 7 8 9 "T7»l Of ».tco«o 1 so«. r« rtNS RO PRO 10 12 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 1 0 1 2 1 0 1 2 H EM TO Tl UO 1 2 3 4 5 6 7 8 U9 TO Tl T2 UO 1 2 3 4 5 6 7 3 U9 AC 1 2 4 A7 BO 1 2 4 B7 CO 1 2 4 C7 00 1 2 4 07 £0 1 2 4 E7 fO 1 2 4 F7 AO 1 2 4 A7 BO 1 2 4 B7 CO 1 2 4 C7 00 1 2 4 07 tO 1 2 4 E7 FO 1 2 4 F7 30 60 6DA VO NVO NROCKICFMPF TSA TSB TSC SBOl LI L2 L3 L4| | VO NVO COO 1 2 C&4 IcSI C52 RA RAI RA2RA3 GSO GSI GS2 6S3 GS4 GS5 GO Gl G2 G3 GB RLS MB RO ROIT ROTC 1 RSI R53 OF MBR ROR NCA CA4 CA5CA6I IT TC ITcl 0 1 2 |TPc| 0 t 2 | H TO Tl UO 1 2 4 7 |T0 Tl UO 1 2 a U7 1 1 ALTtBNATt ROUTE 1 1 BOUT INO 1 N S T. 1 COLcInaC AC AHA AFaI TO I 2 4 T7 UO I 2 4 U7 | I 0 I 2 4 7 I 0 I 2 4 7 InSK SK3 SKBI HN TN UN HO 1 2 4 H7 TO 1 2 4 T7 UO 1 2 4 U7 DImIrO TO 1 T2 UO I 2 4 U7|0 1 2 4 7 1 TO Tl UO 1 2 4 U7 TO Tl UO 1 2 4 U7 OtCODER ROUTING .NSTBUCT.ONS 1 MARKER REGliT RAT,O~-CO0E CONVCRS ION CC CR RR FOFFMBFROFST PCR NPCR I HN TN UN HO I 2 4 H7 TO 1 2 4 T7 UO 1 2 4 U7 MARKER RECSTRATION 1 MKR- iOR TRANSMITTED CODE CONVERS ION MB RO PRO FOF FMB FRO FST HLD MLCT CLCT OOG 40G 50C NDcl HO 1 2 4 H7 TO 1 2 4 T7 UO 1 2 4 U7 M OC MF SXD LPO XOOXSG 0Lc"sxR2OC 00°C 4DG 5DC NSK SK3 SK6 | IC OC Fa"7b K^Fo'Ve "fF FO Fh| CKG HTK SMI SMC CK3CCK CBK NCT NC VC R COPI C0P2 ARST CAK HBASMCOTIO TBY OBS RHC R6D TCK 60K IT TC RCRR ME RCD RCA HBI DCBDCB2 ATB GPL ARS TCO TKSRORL ORL RLT |chk| CKG RCK TCK TBK CCK TKS SCT ATB TB SG OCK ICK OFK IFK SK AK BK CK CHS A C MM& OCT 7R TiRl CONI MT B OSCMRL RL |cLA CLB CLC CDA SKA HA HB TA TB UA UB TSA 0&a|r6DT Fig. 15 — Trouble recorder card showing the multip plugs, cords and jacks. This set can be used for adding and removing cards and verifying that the new card will drop. Bins are provided for storing cards in the process of adding, removing, and transferring cards. Timing tests can be made not only of the over-all time of the translator, as from the trouble recorder frame, but also of the individual component parts of the translator. The test set can also be used for removing the translator selector unit and for bench testing this selector. In this con- nection, a re-cycling test can be made of the code bars, card support bars and latches to check that they operate smoothly and evenly. Cur- rent flow adjustments can be made of the code bar solenoids, latches, and pull up and pull down magnets. DESIGN OBJECTIVES When, in the course of the development, the controlling require- ments that had to be met became apparent, the design objectives were considered. The reasoning applied is set forth in retrospect in the fol- lowing paragraphs. It will be apparent from the foregoing that administrative problems would be involved should it be necessary to arrange and maintain the individual cards in any particular order in the card stack and, therefore, indiscriminate loading became a design objective. Since the cards will be changed from day-to-day and sometimes on CARD TRANSLATOR FOR NATIONWIDE DIALING 1063 IS IPS TB SI MABKta cpossix-) JP ILS OLS JS SM SMI SMO TL RCK TKS TM? TM3 TR TRL STRMRL TIf TQf U9 I 20 ?l ZZ 23 Z4 25 26 27 28 29 30 31 32 3> 34 3S 56 37 38 It 32 33 34 3S 36 37 FR T B CONN OUT CON). 0 I E 0 E 0 8 19 20 21 ii 23 24 2S 26 27 28 29 IMC CO"N CONN »BCr CONTROL Fj C 0 ICNO CNE TC8 OCB ICBlCHB RT» RT8 BTC WTO 13 14 li, 16 17 i/MCTOB CON iO__L 8 l-q i i * S NCF CHKMSK XPS SMG SMBI LLCR CONNCCrOB I LR FB-TCMS LK.FR -UNITS T 2 3 4 bio l?^ 7|o I 24 7| : 3 4 I 1S0H.TY« DP MfP 0! FR GR CLO CLI OT PC OTC OTC PA PB SA SB SC HO IPX PX SX HX ICLX orricE OAT£ 107 ms that may be recorded by punched holes. short notice, the provision of cards capable of being readily coded in the field became another objective. It was estimated that from five to ten per cent of the cards used in nationwide dialing would have to be replaced annually because of routing changes, that on the average a card would be selected for routing pur- poses about one million times before its replacement due to routing changes would be necessary and that the average routing change interval would be from five to ten years. It was recognized that in the course of this period, cards might have to be loaded and unloaded several times. These facts established ruggedness consistent with high-speed opera- tion as still another objective. Ruggedness indicated the use of metallic cards. The comparatively large number of input tabs and output holes required made for sub- stantial size. These considerations together with the comparatively large number of cards that are to be stacked together made it obvious that considerable weight would have to be contended with. Yet it was known that highspeed operation is mandatory. These conditions sug- gested that the cards be made from magnetic material so that their manipulation might be assisted by suitably applied magnetic forces such as may be developed by the pull-up and pull-down magnets that already have been referred to and illustrated. Thus provision for the use of cards made from magnetic material became an added objective. 1064 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 Portable test set. It was estimated that a card translator, on the average, would be called upon to function from twenty to thirty million times a year. Accordingly, provision for direct-acting, high-speed, long-life and readily replaceable components became another objective. A further objective was the provision of means for reading the routing information without mechanically contacting the cards, as for instance by utilizing photo-detection circuirts, such as have been referred to, as it was rea-soned that in this way reliability over an extended period could best be assured. ELECTRO-MECHANICAL DESIGN Phototransistors are utilized as the photo-sensitive elements, the light source is a standard projection lamp, the light beams are modulated and then directed through the card formed tunnels by dual collimating lenses and then upon emission from the card stack are focused on the photo- transistors. (See Figs. 17 and 18.) To a large extent the card, which is CARD TRANSLATOR FOR NATIONWIDE DIALING 1065 1066 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 CARD TRANSLATOR FOR NATIONWIDE DIALING 1067 I ••*'*• Fig. 19 — Coded template (below) and associated card. manufactured as the 200A blank, was controlling and, therefore, it will be duscussed in some detail. CARD (200a blank) The general form of the card is illustrated by Fig. 4. The working card, however, is unmarked as may be observed by reference to Fig. 3. Mark- ing is not required because the cards are coded in accordance with information furnished on a paper template, such as is shown in Fig. 19 and this template is retained as part of the office records for ready refer- ence. A card may readily be identified for reassociation with its template by reading its tab code. The template provides considerable administra- tive data. The dark half of the holes and the triangular representation of the tabs are printed in red as is a strip along a portion of the left- hand edge. After coding, the card is placed over the template. If the red edge portion is visible the card must be turned end-for-end. If, after 1068 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 having done so, any red can be seen, more holes have been enlarged and/or more tabs have been removed than the template calls for. If, on the other hand, openings in the template are visible through holes in the card that have not been enlarged, the hole coding is not complete. If tab notches appear in the template other than in registration with the portion of the card from which tabs have been removed, the tab coding is not complete. Accordingly, the template also serves as a convenient means for checking the coded card for accuracy. As a convenience, the holes of the template were made round and its tabs triangular. The holes of the card before being enlarged are of a form and size that simulates the filament face area of the light source. To accommodate the 118 holes, the optimum dimensions of the holes were determined to be %" wide, 0.140'' high and the optimum vertical and horizontal spac- ing proved to be 0.535'' and 3^", respectively. The tabs are of a form and size determined largely by the mounting space required for the solenoids that are used to operate the code select and card support bars, the vertical displacement required for shuttering the holes and ruggedness considerations. Thus, the nominal size of the tabs associated with the code select bars became 3^" wide X 0.205" long with a spacing of J^e''- The tabs associated with the card support bars — one at either end — being used for each translation instead of COLLIMATING LENSES ^ n a^-X PULL-UP MAGNET UPPER CARD " GUIDE BAR / c=)i=!(=ic=3i=]l_Ji=]^i, D / c=ic=ii=ic=ii=ic=ir=] \ / -^^^ / \ /c=Diizic=ic3c=ic=icniz=]cz3\^ ^[Z!t=iic3c::3Ci]cnCIICIIII^ ?' c=ic=3i=DnncDnna \ CIldZlciDcrjcncDcncr] / pi CD D 1=1 1=3 n CD CqJI^ t=II=II=lI=)l=ICZ]l=ll=IC= i=ii=]cni=ii=]t=ii=ii=ic=i I=ll=lI=ll=lt=li=lt=ICZIC=l CARD SUPPORT BAR f/PATH OF FLUX\\; \-Q^ c=i t=] n t= (=1 u c , BAR DOWN STOP UNOPERATED MAGNETIC CODE SELECT POLES BARS CARD BIN 7x BAR BAR UP STOP ^ I BAR DOWN STOP / ■ tHI I\ oc, nr-r,zr. T " ^, SELECTED CARD DOWN STOP CARD SUPPORT LIFT MAGNET PULL-DOWN "LOWER CARD GUIDE MAGNET BAR (FIXED) OPERATED CODE SELECT BARS Fig. 20 — Card and directly associated components. CARD TRANSLATOR FOR NATIONWIDE DIALING 1069 Fig. 21 — Card coding tool. only selectively as in the case of the tabs associated with the code select bars were made 3 2 -* Fig. 4 — Combination of inhibited gates into and not gate. ties are of major interest, the load has been assumed to have the same conductance as the signal generator. A constant current bias /& is im- pressed' at the midpoint of the gate. The control input voltage is repre- sented by Eh. The internal control generator conductance, which should be large, is assumed to be included in the corresponding diode conduct- ance, for computing purposes. The diodes are assumed to have a large conductance G, or a small conductance g depending on whether they are forward biased or reverse biased. Fig. 5 shows the gate enabled, with the series diodes in the conducting state and the control diode in the reverse bias, or non-conducting condition. If the G and g are inter- changed the gate is in the disabled condition. The relative magnitudes of Fo, Fi, V2 evidently determine whether the diodes are biased forward or backward. Their magnitudes are im- mediately obtainable. The equations for the case shown in Fig. 5 are: * In ca49e the internal conductance of an actual bias source is not sufficiently small to be neglected, its main effect will be on transmission loss. In computing the loss, the bias conductance may be added to the conductance of the control connection. SEMICONDUCTOR DIODE GATES 1143 (Go + G)V - GVi = 7o (1) - GVo +ig-\- 2G)V, - GV2 = A + gEt (2) - GV, + (Go + G)V2 = 0 (3) Their solutions are : h(g + 0+ pr^) + ih + gE,)G ° g6o + gG + 2GoG ^^' _ hG + {h + gg>)((?o + GO ,. ^' gG, + gG + 2G,0 ^°' The corresponding equations for the disabled gate are obtained by interchanging g and G in the above. ENABLED GATE Considering the enabled gate first, it should be evident from the figure that there are four requirements if the gate is to be properly enabled: Fi > Vo (7) Fi > V, (8) Vi 0 (10) Vi is a positive voltage, so equation (6) insures that (8) will always be satisfied. Putting the values of Vi and Vo in (7) gives (/. + .^.)>(^ + ^J/o (11) In the case that g is much smaller than Go (which is normally true) gEb is effectively the current from a constant current control generator and the above inequality has a simple interpretation : In determining whether (7) is satisfied, and the input diode held con- ducting, the above inequality compares the total current which the bias and control generators put into midpoint (Vi) of the gate with the 1144 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 maximum current which the signal generator could put into the same point when that point is grounded. If the control and bias current sum is larger, the current in the input diode cannot reverse. The inequality (10) is to insure that the output diode remains con- ducting. Substituting the value of Vi in it gives: h + GE, > -G Go + G (12) The only way that the output diode could be cut off (with positive bias and control) is by a large negative signal current. The above inequality requires : To hold the output diode conducting, the sum of bias and control generator currents must be greater in magnitude than the maximum negative signal current that the signal generator could put into the midpoint when the midpoint is grounded. A zero potential on the midpoint is the boundary condition between the diodes being conducting or non-conducting. The two inequalities together compare the currents that the generators can put into the grounded midpoint. They require: The sum of bias and control generator currents should exceed in magnitude the maximum current of either polarity, that the signal generator can put into the grounded midpoint. There remains the inequality (9) which is necessary if the control diode is to remain non-conducting. This gives: Gh Go + G + h <2 GoG Go+G E. (13) This compares the same bias and signal generator currents with the current which would flow in the input and the output circuit if Vi were replaced by Eb. If the inequality is satisfied Vi can never get as large as Et and the control input diode remains cut off. lb V, 9 I Fig. 6 — Transmission type diode gate. SEMICONDUCTOR DIODE GATES 1145 DISABLED GATE The condition under which the gate is held disabled is much simpler than the enabling conditions. All that is necessary is that Vi be held more negative than the most negative signal generator voltage : Fi 5.05 10"' -1.1 76 + 10-' Eb > 5.0 10"' No negative signal voltages were used, so inequality (12) is not involved. The above inequalities limit h and Eb as shown in Fig. G. h and Eb must be chosen from the shaded area. The values chosen were h = 5ma and Eb = 15 volts. Comparing experimental results with analytic, gives: Voltage loss 1.0 db (computed 1.6 db) Pedestal 5.05 volts (computed 5.45 volts) Further experimental results, which are all in reasonable agreement with expectations are given on Figs. 7, 8 and 9. Fig. 7 shows how the pedestal voltage varies with gate voltage (Eb). This is also a measure of the load capacity, since the signal output can swing from zero to twice the pedes- tal. The useful range is above the point where the pedestal ceases to increase with gate voltage. In this range the control diode is cut off. Figure 8 shows the pedestal against bias current /&. The limiting is not as sharp here because the forward conductance of the input and output 1.2 5 10 2 j 2 4 6 8 10 12 14 16 18 20 22 24 Eb IN VOLTS Fig. 6 — Bias restrictions on transmission gate. 1148 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 diodes are involved and they do not change as rapidly with current in milliamperes as the reverse conductance does with volts. Fig. 9 shows the actual input output relation for the signal. It is linear over 80 to 90 per cent of the 5-volt range and then limits as expected. The discrimination was not measured. It computes to better than 60-db voltage loss and discrimination of that order of magnitude has been measured in gates of this type. SWITCHING GATE C A form of gate which is useful for pulse systems, since it lacks the pedestal, is shown on Fig. 10. This is, of course, basically the same con- figuration as that shown on Fig. 5, but it is operated quite differently, with pulses or dc potentials applied to the two control inputs and an output obtained by switching the bias current from flowing in a control path to flowing in the load. More specifically, if both Ei and E2 are suf- ficiently negative, diodes Di and D2 are both conducting, Vb is negative, and Dz is non-conducting. Thus practically all the bias current h flows in Di and Z>2, and V2 is zero or slightly negative. If one of the control voltages is increased until its diode cuts off, the bias current can still flow in the other control path and the change in the output voltage is extremely small. If both the control voltages are increased until the two control diodes are cut off, then Vb becomes positive, Dz conducts and practically the entire bias current flows in the load, producing an output voltage The above operation gives a two control and gate with no pedestal 2.4 2.0 1.6 0.4 jT ° y / A ) / / 1^ = 5 MA / / 0 : > i ( 5 i GATE } 1 VOLTS 0 1 2 1 4 16 Fig. 7 — TranHmisHion gate output (pedestal) potential versus gating control potential . SEMICONDUCTOR DIODE GATES 1149 3.0 2.5 ■ ^^ — ° ' y< r / LIMITING / / Eb = 15V / / / / / f 5 1.0 o 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 BIAS MILLIAMPERES Fig. 8 — Transmission gate output (pedestal) potential versus bias current magnitude. difficulty. One of the controls could instead be used as an inhibitor. For example, if E2 were normally biased sufficiently positive it would have no effect on the output, which would be controlled by Ei alone. How- ever, a negative, or inhibiting pulse, sufficient to make D2 conducting, would permit the bias current to flow in that path and prevent an out- put, whatever the state of A- This circuit could be analyzed in exactly the same manner as was done with the transmission gate. However, after a value of Rl has been chosen, a simple first approximation to a design may be carried out by assuming that the diodes are ideal, switching between zero and infinite resistance. In choosing Rl there are three major considerations: 1. Rl must be small compared with the reverse resistance of the diode, D2, or there may be an appreciable negative output when the gate is disabled. 2. Rl must be large compared with the forward resistance of D3 for efficient operation of the enabled gate, — preventing an appreciable voltage loss due to the voltage drop in D3. 3. The peak amplitude of the output pulse is IiRl- The value of Rl, which is chosen from the wide range of possibilities, is a matter of practical compromise, depending on the impedance levels in the system and the constant current generators which are available. Having chosen Rl and lb there remain only the control voltages, El and E2. The voltages which are necessary to hold the gate enabled can be obtained by noting that (in the ideal diode case) V.= V, = RlI Lift (19) To hold the control diodes non-conducting the voltages Ei and E^. must be greater than Vb. This gives: 1150 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 JO 4 / / / ,^^^- LIMITING 1 DB LOSS^ / / / Ib= 5MA Eb=15V A > / r 4 5 6 INPUT VOLTS Fig. 9 — Transmission gate signal output potential versus input signal potential. To enable the gate, voltages must be impressed such that E2 > RJb (20) In disabling the gate, either one of the control paths may have to carry the full bias current. If, for example, the bias current were flowing in D2 it would bias that control path positive by an amount R2lb- To over- come this, and keep Vb negative, the requirement is: To disable the gate, voltages must be impressed such that El < -Rih E2 < -Rih (21) These sets of conditions give the magnitudes of the biases which are necesssary to hold the gate either enabled or disabled, and the dif- ference between them is the minimum magnitude of the necessary switch- ing pulse. EXPERIMENTAL GATE An example may be given, using 400B diodes. The bias was chosen as 5 ma and the load resistance, 2,000 ohms. The control resistances were made small, as is desirable for reasons which will be discussed later. For this gate the output voltage is 10 volts. From (20) and (21) the gate could be enabled by 10 volt positive control pulses and disabled by very small negative control values. A larger than necessary control voltage was put on control 2, E2 = 15 volts SEMICONDUCTOR DIODE GATES 1151 and the output measured as a function of Ei. The results are shown on Fig. 11. Since the diodes are not ideal, there is a transition region, but, as predicted the output is very small at small negative control voltage and the output is the full 10 volts when Ei is 10 volts. The curve also shows what happens if one of the enabling biases are too small. A case is shown in which E2 was only 5 volts. There is no significant difference until the output gets up to 5 volts. Above that voltage the diode, D2, becomes conducting and the output is clamped at that voltage. GATE CHARACTERISTICS The main virtues of this type of gate is that there is no pedestal and a constant amplitude pulse is produced. It is also simple and has good discrimination. There are limitations: 1. Unless a very low control path resistance is used, there is a large loss — that is the output pulse is much smaller than the control pulse. For example, if the control resistance is equal to the output resistance there is a two to one loss. 2. A rather large load is put on the control generator, partly because it must produce the enabling voltage across a small resistance and also because, in some cases the total bias current flows in the control generator output. 3. A phenomenon called "hole storage", which is present to some extent in all semiconductor diodes can make trouble. When a diode has been resting in the conducting state with a current flowing in it and the voltage is reversed, the diode does not immediately change to high im- pedance. A reverse current flow for a short time — up to a few micro- seconds. This can result in very inconvenient, spurious, output pulses being produced by a gate which is supposed to be disabled. D3 E, E2 Fig. 10 — Switching type diode gate with two controls. 1152 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 10 .^Jk Inl 1 y ^ E2 = 15 VOLTS 6 lb = 5MA Rl = 2000 OHMS y .. 1 1 ., 1 TS A E->='^ vni 4 / 2 / f I I tj) 0.4 [ Z ^ 0.1 ^ ^ 0.06 ^ 0.04 O X *^ s ^ 0.01 N 0.006 0.004 0.002 0.00 1 \ ®\ V \ J -16 -12 20 B -4 0 4 8 12 l( INPUT, El, IN VOLTS Fig. 11 — Switching type diode gate output potential versus control potential with a second control in the "enable" state. SPECIAL CIRCUIT It is not intended, in this paper, to attempt to list the numerous modi- fications which have been made of diode gates to overcome limitations and satisfy particular requirements, but it seems desirable to note an example of means to minimize the limitations. One simple way of minimizing the difficulty, is to use point contact diodes in places where a spurious pulse could make trouble and use junction diodes elsewhere, since junction diodes have better impedance ratios but worse hole storage. For example, the output diodes in the switching gate could be a point contact unit, while the better discrimina- tion of the junction unit made use of for control diodes. A second means of avoiding hole storage effects is to avoid leaving diodes with large currents flowing in them, when they must be switched rapidly to the non-conducting state. SEMICONDUCTOR DIODE GATES 1153 Fig. 12 illustrates both these design ideas. One of the control inputs has the control pulse impressed by means of a transformer and so has a very low dc impedance without making excessive demands on the control pulse generator. The biases are so adjusted that, in the disabled state, all the control diodes are just on the edge of conducting except the one in the transformer path. Because of the low DC impedance, practi- cally all the bias current flows in this path. Thus there is no possibility of hole storage except in this one diode. If a positive control pulse is impressed while the potentials on the other controls are at their more negative value, the bias current just switches into those control paths; the output diode remains non-conducting and any spurious hole storage D4 Fig. 12 — Switching type gate which minimizes "hole storage" effects. current from the diode in the transformer control also goes into the other control paths. Since there is no storage in the other control paths, posi- tive pulses may be simultaneously impressed on all the controls, the diodes in all but the transformer control path will immediately become high impedance and any hole storage current from the one diode will harmlessly add to the bias current flowing into the output. Junction diodes are used in all the control inputs except the one with a transformer. A point contact unit is used here to minimize the hole storage. A point contact unit is also used in the output position. This is a critical location where good diode action is more important than a very high discrimination. There is an additional advantage, in this configuration, that the rela- tively large bias current flows in the control generators only very briefly — while the disabled gate is being pulsed by the transformer control. 1154 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 ACKNOAVLEDGMENTS It is impossible to give due credit to all the people who contributed to the development of the gates discussed here. Credit, however, should be given to W. D. Lewis, L. A. Meacham, and A. J. Rack who developed and explored the basic transmission gate analyzed here, and to J. II. Harris who made the modifications in the gate to avoid hole storage difficulties. Acknowledgment is also due that part of the work reported on hero was done in connection with a Joint Services contract. A Review of New Magnetic Phenomena By R. E. ALLEY, Jr. Manuscript received May 25, 1953 As a result of new developments, the classical concepts of magnetic materials, characterized by hysteresis loss and eddy currents, are no longer adequate. Study cf the ferrites has revealed new and important magnetic phenomena. These materials, because of their high resistivities and cor- respondingly low eddy currents, exhibit useful magnetic properties at fre- quencies well above those at which magnetic alloys are applicable. This paper reviews the new phenomena — domain wall motion and dimensional effects in the low megacycle region, and ferromagnetic resonance and the Faraday effect in the microwave region — and relates them to modem theory. Some possible microwave applications are discussed briefly. I. INTRODUCTION Up until a few years ago, the classical concept of magnetic materials, characterized by hysteresis and eddy current effects, was adequate for the communications engineer. Recent new and important developments, however, have made it necessary for him to have a broader knowledge of magnetic phenomena than is given by the old picture. Because of the low resistivities of existing magnetic materials, their properties at high frequencies have been dominated by eddy currents which in many cases have completely masked other magnetic effects. In contrast, the newly developed ferrites have resistivities from 10 to 10 " times greater, and eddy currents are usually negligible. The ferrites are, therefore, useful at far higher frequencies than previously available materials. Further- more they have revealed new and important magnetic phenomena. It is the purpose of this paper to present the modern picture of mag- netism from the standpoint of the engineer. It describes the new phe- nomena and relates the experimentally observed behavior of magnetic materials at high frequency to the present physical theory of magnetism. The new phenomena include dimensional and domain wall motion effects in the low megacycle region and ferromagnetic resonance and the recently observed Faraday effect in the microwave region. 1155 1156 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Although we are concerned more with magnetic phenomena than with the properties of particular materials, we will relate our discussion to ferrites, since these are the materials in which high frequency phenomena have been explored. We begin, therefore, with a brief description of these materials. II. DESCRIPTION OF FERRITES The term "ferrite" as used here refers to a class of ferromagnetic oxides that are structurally the same as magnetite (the naturally occur- ring magnetic mineral commonly known as lodestone) and as the min- eral spinel from which the structure derives its name.^' ' These com- pounds form extensive solid solutions of both the substitional and the subtractional type. Nickel zinc and manganese zinc ferrites are important examples of the substitutional type. In these, the zinc and nickel or manganese are thought to be in solid solution in magnetite (Fe304) where they have directly replaced equivalent amounts of iron in the lattice. An example of the subtractional type of solution is T-FeoOs- Here, oxygen is considered to be in solution in magnetite, not, however, having replaced iron, but having eliminated it, thus leaving vacant sites in the lattice. Magnetically, the ferrites are thought of as consisting of two interpenetrating lattices of metal ions whose magnetic moments point in opposite directions. Since, however, these moments are in general not equal, the material has a net magnetic moment. Ferrites are manufactured by carefully mixing oxides of the constituent materials. The resulting powder is then pressed into desired shapes. These formed parts are fired at a temperature of 1000°C or more to produce the finished materials. The finished product is technically classed as a ceramic, and among its properties is extraordinarily high resistivity compared with magnetic alloys. Mechanically, ferrites have some of the characteristics of ceramics. They are extremely hard and brittle and cannot be machined by ordinary methods. They may be ground and lapped by use of abrasive cutting tools. Experience has sho^vn that the properties of the finished product depend upon composition (both what elements are present and in what proportions) and upon heat treating conditions (atmosphere, maximum firing temperature, and time of firing). It is apparent that, since there are so many possible variables in manufacturing procedure, one may expect a wide variety of electrical and magnetic characteristics. Ferrites have been commercially available for several years. NiZn ferrite is being used extensively in deflection coils and flyback trans- formers in television sets. MnZn ferrite has found specialized but im- A REVIEW OF NEW MAGNETIC PHENOMENA 1157 portant use in inductors for networks and in transformers in telephone circuits. Magnetic recording tape makes use of 7-Fe203. An increasing amount of information of interest to the design engineer is becoming available in manufacturers' catalogs. III. DC CHARACTERISTICS For convenience in the following discussion, the frequency range under consideration has been divided somewhat arbitrarily according to the magnetic phenomena which have been observed. We will begin by discussing dc behavior of magnetic materials. As far dc magnetic properties are concerned, the modern picture is the same as the classical one. The study of ferrites has revealed no essentially new phenomena, at least so far. Hysteresis loops for typical ferrites are shown in Fig. 1, together with loops for iron and for permalloy. It will be observed that some ferrites compare favorably with permalloy with regard to hysteresis loss (pro- portional to the area of the hysteresis loop). Their saturation flux density is considerably lower, the maximum so far obtained being be tween 4,000 and 5,000. By applying pressure to a sample and thereby introducing strain, some investigators have found it possible to produce ferrite cores having prac- tically rectangular hysteresis loops,^ just as similar effects have pre- viously been obtained with permalloy and other magnetic material. 12 - CD -4 •16 - { ^ /^ IRON 1 ^ ) ...,j 1 4-79 Mo PERMALLOY r - TYPICAL FERRITES ___^JJ^ y 6-2 0 2-6 H IN OERSTEDS Fig. 1 — Hysteresis loops of iron, permalloy and typical ferrites. The thin ferrite loop is for a MnZn ferrite, while the larger loop represents a NiZn ferrite. 1158 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Such cores find an important application in memory circuits employed in connection with digital computers. Suitably cut ferrite single crystals also exhibit rectangular loops when properly annealed.^ Initial permeability of ferrites has a wide range of values depending upon the material under consideration. It may be as low as 4 for magne- tite and as high as 3,000 for manganese-zinc ferrites. Curie temperature (the temperature above which the material no longer exhibits ferro- magnetic properties) is around 80°C for the high permeability MnZn materials and is several hundred degrees C for the low permeability nickel ferrite. Resistivity also depends upon the composition of the material. Typical data gives values of 100 ohm-cm for a MnZn ferrite and 10® ohm-cm for a NiZn ferrite. The commonly used metallic magnetic materials have resistivities of the order of 10""^ ohm-cm. As mentioned above, this much higher resistivity is the feature of ferrites which makes possible their application at frequencies where ordinary metallic ma- terials are generally not usable. At dc the dielectric constant of ferrites is high. Determination of dielectric constant is rather difficult, but the best measurements to date indicate values of from 10 to 30. IV. LOW FREQUENCY PHENOMENA (0 tO 1 mc) A convenient and commonly used method of determining low fre- quency characteristics of magnetic materials consists in making bridge measurements of inductance (L) and effective series resistance (R) of a uniform winding placed on a toroidal core of the material. Subtraction of the dc winding resistance gives a value of resistance (Rm) which represents the core loss in the material. Permeability may be calculated from the inductance measurements. The method of analysis described by Legg** may be applied to powdered or laminated alloys. The method consists essentially in determining the coefficients in the equation Rm = cixfL + auBmJL -f c/x/L, (1) where Bm is maximum flux density, ju is permeability, / is frequency, and c, a and c are constants. For alloys in laminated or powder form these constants are associated respectively with eddy current, hysteresis, and residual losses. From measurements of L and Rm at two or more fre- quencies with a fixed flux density, jB„, and two or more values of flux density at a fixed frequency, /, the coefficients e, a and c can be deter- mined by solving the simultaneous equations obtained from equation (1). Equation (1) is equally applicable to the ferrites at frequencies below that at which domain wall resonance and dimensional effects (discussed in Sections VI and VII) begin to appear. This frequency, which is some- A REVIEW OF NEW MAGNETIC PHENOMENA 1159 what below that indicated by /i in Fig. 3, depends upon the particular ferrite. The range of applicability of equation (1), therefore, varies. Regardless of frequency, the eddy current and hysteresis terms in equa- tion (1) are valid. However, additional terms are required to cover other losses which become quite high and in comparison with which the "residual" term in equation 1 may be negligible. Frequently, the separation of losses indicated in equation (1) is not called for. The practice then is to lump all losses together and express them in terms of the material Q, which may be a function of frequency and fiux density. Here Q is the ratio of the reactance of the winding on a toroidal ring to the core loss expressed as a series resistance, The product juQ is convenient in describing magnetic characteristics. Sometimes air gaps are introduced in ferrite magnetic cores in order to provide higher coil Q's or greater stability of ac permeability with super- posed dc magnetizing force. It can be shown that the product juQ re- mains constant even though the core is divided by one or more air gaps. Fig. 2 shows curves of ^lQ versus frequency for various ferrites and also for certain other materials. An extensive discussion of methods of meas- urement and of results of measurements at low frequencies is given in a recent paper by Owens.^ It should be pointed out that eddy currents in ferrites are so small that solid shapes are usually used whenever ferrites are applicable. This is a considerable advantage, eliminating the necessity for thin tapes or insulated fine particles which are required in many applications of me- tallic cores. However, so far, no ferrite has been developed which has permeability nearly as great as that of some of the permalloys. V. CHARACTERISTICS ABOVE 1 MC So far, our discussion has covered the frequency range in which mag- netic materials have traditionally found many important uses and in which the ferrites have properties generally like other magnetic ma- terials, differing from them only in degree. We now come to the fre- quency range in which new magnetic phenomena have recently been observed — a range above that in which magnetic materials have here- tofore been generally applicable. In discussing the higher frequency characteristics, it is desirable to introduce a somewhat different set of parameters by which the properties 1160 THE BELL SYSTEM TECHNICAL JOURNAL, SEPLEMBER, 1953 10* 10 /iQ MnZn FERRitE '/ZQ NL Zn FERRITE ^-^ //Q CARBONYL IRON POWDER (// = 13) \ V---//Q Mo PERMALLOY POWDER (// = 14) JJ.Q 0.001" 4-79 Mo PERMALLOY // FINE METALLIC POWDER 10-1 10 102 10^ FREQUENCY IN KILOCYCLES PER SECOND 10^ 105 Fig. 2 — Comparison of the frequency variation of fx and nQ for ferrites and other magnetic materials. may be described. This comes about because the ferrites, in which the new effects are observed, have both dielectric and magnetic properties. The material is most conveniently described in terms of two complex quantities: the permeability, fi = fi' — jfx" , and the dielectric constant, e = e' — je" . n' corresponds to the usual low frequency permeability and Q = ii! I\i" . Similarly e' corresponds to the usual low frequency dielectric constant, while e'Ve' = tan 5, the loss tangent of the material. Thus the quantities y." and e" are measures of the magnetic and dielectric losses per cycle respectively, in the material.^ The fact that m" and e" represent loss per cycle means that much higher values of these quantities can be tolerated at low frequency than at microwave frequencies. At frequencies above a few megacycles, it becomes very difficult to make meaningful observations on wound toroidal cores. Such difficulty may be overcome by making measurements on a toroidal sample placed in a coaxial line. Details of the experimental procedure may be found in references 9 through 12. The same procedure may be applied at micro- wave frequencies with waveguide used instead of coaxial line. In this case. A REVIEW OF NEW MAGNETIC PHENOMENA 1161 the sample is in the form of a slab, cut to fit snugly against the walls of the waveguide. Fig. 3 shows in a qualitative way the behavior of a typical ferrite. In this figure we have plotted the real and imaginary parts of permea- bility as functions of frequency. Examination of the results obtained by various investigators for a number of different ferrites leads to the con- clusion that they all behave in a fashion similar to that shown in Fig. 3. Frequency /i usually lies in the region between 1 and 100 mc and ji frequently lies in the neighborhood of 3000-4000 mc. Available data are not sufficient to provide similar curves for dielectric constant, but what there are indicate a decrease from the very high apparent value at low frequencies to a constant value of the order of 10. This decrease generally occurs somewhere around 10 mc, although it depends upon the material and also probably upon sample dimensions as will be discussed below. The consensus is that the high experimental values of dielectric con- stant observed at low frequencies result from the peculiar structure of ferrites. They are considered to consist of grains of conduction material (of moderately high conductivity) separated by thin layers of dielectric material having a dielectric constant of the order of 20. Measurements on such a structure would give very high apparent dielectric constant and low Q. Such behavior was observed several years ago in samples of powdered permalloy, and has also been found in samples of plastic in which finely divided particles of copper have been dispersed. ^ A >U B >|< — C-- 1 1 1 1 A 1^" '•^,__ ,-'' \ /=^ FREQUENCY u f. ^2 Fig. 3 — Frequency characteristics of a typical ferrite. The components of complex permeability (/x = m' — i^") as functions of frequency. The behavior in the neighborhood of /i is attributed to domain wall motion or to dimensional effects. /2 represents the frequency at which ferromagnetic resonance occurs. 1162 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 For convenience in the following discussion, Fig. 3 has been divided into three regions as indicated. It will be observed that there are two peaks in the curve of ix" versus frequency, one in Region A and the other in Region B. The frequencies at which these occur are those at which relatively large amounts of energy are absorbed by the material. These two absorption peaks are due to entirely different mechanisms within the ferrite and it is, therefore, of interest to consider them separately. We begin with Region A. The behavior indicated in Region A is typical of polycrystalline samples of ferrite (as distinguished from single crystals which will not be considered here), yl rises somewhat above its constant low frequency value and then decreases rather suddenly. If /i denotes the frequency at which the peak in \i" occurs, we find that in general a high value of ii' at low frequency is associated with a low value of /i, and vice versa. At the present time, we are not sure which of two experimentally ob- served effects in the ferrite is responsible for the behavior shown in Region A. It is quite likely that what one observes in a given sample is actually a combination of the two effects, domain wall motion and dimensional resonance, each of which will now be described. VI. DOMAIN WALL MOTION The basic unit of magnetism is the spinning electron. In an atom of ferromagnetic material, there is an excess of electrons with spins in one particular direction. As a result, the atom has a net magnetic moment. Any ordinary sample of ferromagnetic material consists of many small, irregular volumes called domains, each of which may contain many atoms. Each domain is completely magnetized along some direction. Both the size of the domains and the directions of their magnetizations vary from point to point throughout a sample. In an unmagnetized material, the random orientation of individual domain magnetizations results in a mutual cancellation of their effects. However, if a magnetic field is applied, certain domains will be in a preferred orientation, having their magnetic moments more nearly in the direction of the applied field than others. These will grow at the expense of less favorably oriented domains by a process of motion of the walls separating adjacent do- mains. When an alternating field is applied, the walls will be subject to an alternating force which will tend to move them first in one direction and then in the opposite direction. Now it has been shown^^ that, under these conditions, the permeability of the material is proportional to the ease of displacement of the domain walls. Therefore, if we can predict how a wall will move as the frequency of the applied field changes, we can predict how the permeability will change with frequency. A REVIEW OF NEW MAGNETIC PHENOMENA 1163 The walls have been found to exhibit properties of mass and stiffness and to be subject to damping. Thus it is possible to write an equation which describes the motion of a wall under the influence of an alternating field. ' If x is the displacement of the wall from its equilibrium posi- tion, then m^,+Pj^ + ax=^ M.He'"', (2) where m = mass/unit length of wall, jS = damping coefficient, a — stiffness parameter, Mg = saturation moment of sample, He^"^ = applied field, This equation should look familiar to electrical engineers. It has the same form as that for an RLC circuit subject to a sinusoidal applied voltage.^^ If the damping constant is not too great, we would expect the wall motion to have a resonance at a frequency, /o, for which Fig. 4 shows the expected variations of jjl' and m'' in the vicinity of reso- nance under these conditions. If, however, the mass of the wall is negligible, the situation is some- what different. In this case, /z" has a maximum at a frequency for which 0) = a/^, while m' decreased monotonically from its low frequency value, reaching 3-^ this value at a; = a//?. It is apparent that this behavior is analogous to that of an RC circuit. It is commonly known by the term ''relaxation". We see, therefore, that, depending upon wall mass, there are two possible ways in which wall motion may contribute to the be- havior shown in Region A of Figure 3; namely, through domain wall resonance and through domain wall relaxation. VII. DIMENSIONAL RESONANCE The second phenomenon which may aid in accounting for the behavior indicated in Region A is dimensional resonance. In 1950, Brockman, Dowling, and Steneck^^ reported the results of some experiments on a manganese-zinc ferrite known commercially as Ferroxcube III. Using blocks of this material, they built a closed rectangular core. Measure- ments on a winding placed on this core showed behavior like that of region A with /i lying between 1 and 2 mc. When they decreased the 1164 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Fig. 4 — Typical behavior of the components of complex permeability (jx — ill") in a ferrite in the neighborhood of domain wall resonance. cross section area, /i moved to a higher value. Brockman, Dowling and Steneck attributed this to a cavity type resonance effect resulting from a high dielectric constant which gave wavelengths in the material of the same order of magnitude as the dimensions of the sample under test. Another way" of looking at the phenomenon is to recognize that in the ferrites there are displacement eddy currents which correspond to the conduction eddy currents in magnetic alloys. Analysis based on this approach gives resonance frequencies which are the same as those cal- culated by Brockman, Dowling and Steneck. In any sample it is possible that both mechanisms — domain wall motion and dimensional resonance — contribute to the behavior of the material in Region A. Futhermore, it is evident that considerable cau- tion is required in interpreting results of experiments designed to meas- ure M and e as functions of frequency. One must always bear in mind that what one obtains is the effective ji and e of the sample under test, and the actual fx and e for the material may be different from the ob- served values, depending upon the effect of sample dimensions. It is clear that in a design problem the communications engineer must take into account the dimensions of the ferrite part as related to the permeability and dielectric constant of the material and to the frequency at which it is being used. This may impose a practical limitation on the size of a part for a particular application. VIII. FERROMAGNETIC RESONANCE The behavior indicated in Region B of Fig. 3 is attributed to ferro- magnetic resonance. This phenomenon was first observed in magnetic metals by Griffiths^® and has been studied intensively in ferrites.^^ A REVIEW OF NEW MAGNETIC PHENOMENA 1165 Consider a single electron. Because it is a spinning charge, it has a magnetic moment which lies along its axis of spin. Because it has mass, it has mechanical angular momentum. The ratio of these two quantities is the magnetomechanical ratio, 7. If a steady magnetic field, Hq, is applied, there will be a torque on the electron as a result of the interaction be- tvv'een Ho and the magnetic moment of the electron. The electron will, therefore, precess about the direction of Hq with a frequency which has been shown to be given by coo = yHo. This phenomenon is the well-known Larmor precession. ^^ We may say then, that the electron has a reso- nance frequency coq. Now suppose a sample of ferrite to be placed in a magnetic field ^0. By virtue of the contributions of its many spinning electrons, the sample has a magnetic moment, M. If Hq is a strong field, M will be parallel to i^o. However, there will be, as in the case of the single electron, a fre- quency at which M will precess about the direction of Hq. This precession frequency is proportional to an effective field. He, and to M/J, where / is the vector sum of the individual angular momenta of the electrons. He is a function of Hq and also of the demagnetizing fields within the ferrite. If an alternating field of this frequency is supplied perpendicular to ^0, then absorption will occur. The amplitude of the precessional motion will become such that the energy supplied by the alternating field is equal to the energy transformed into heat in the sample. At the resonant frequency, fi is equal to 1 and /i" reaches a maximum. Although the above discussion postulates a strong external field, a more detailed analysis leads to the conclusion that resonance may be expected even with zero external field. This is attributed to the presence of internal fields which result from such things as crystal anisotropy, magnetostrictive strain, and internal demagnetizing fields in the ma- terial. These demagnetizing fields are generally the most important factor in determining the resonant frequency of a demagnetized ferrite. A number of investigators have studied ferromagnetic resonance in ferrites. Some experimental methods and results are described in Refer- ences 19 through 23. Much of the investigation has been carried out on single crystals of ferrite. In these cases, the experimental results depend upon the orientation of the crystal axis with respect to the external field. In the case of polycrystalline samples, the resonance is still present, but the resonance is not as sharp. Most of the experiments in which ferromagnetic resonance has been studied have been with fixed frequency and varying magnetic field. Fig. 5 shows some typical results of such an experiment. From a practical ex- perimental standpoint, this is preferable to varying frequency with a fixed magnetic field. However, it is apparent that if one holds the magnetic 1166 THE BELL SYSTEM TECHNICAL JOUKNAL, SEPTEMBER, 1953 / \ / \ '' '^-\ v " Fig. 5 — Behavior of a magnetic material in the neighborhood of ferromagnetic resonance. This represents the usual experimental situation, where frequency is kept constant and the external magnetic field is variable. field constant at some value iJo and varies the frequency, one will obtain curves similar to those of Fig. 5. This is indicated in Region B of Fig. 3. Furthermore, for a given sample, as i7o increases, the frequency at which resonance occurs increases. IX. MICROWAVE FARADAY EFFECT If a linearly polarized wave of microwave frequency travels through a ferrite which is magnetized in the direction of propagation of the wave, the plane of polarization will be rotated. The sense of the rotation depends only upon the direction of magnetization of the ferrite and is independent of the direction of propagation of the wave. Thus the effect is anti-reciprocal. This phenomenon, which derives its name from the analogous optical effect was demonstrated experimentally by Roberts^"* and has been ex- tensively investigated by Hogan." It occurs at frequencies above the ferromagnetic resonance frequency, that is, in Region C of Fig. 3. In principle, the effect might be expected in any ferromagnetic material but, so far, only the ferrites are sufficiently transparent to microwaves to allow the effect to be detected. The effect is illustrated in Fig. 6. The linearly polarized microwave in waveguide A passes through a transition section into the circular guide B. A tapered cylinder of ferrite is inserted in B. A solenoid, external to B, supplies a steady field parallel to the axis A REVIEW OF NEW MAGNETIC PHENOMENA 1167 of propagation. Upon emerging from the sample, the wave passes into C, a circular to rectangular waveguide transition which may be rotated for maximum transmission of energy down section C. The angle of dis- placement of C with respect to A is a measure of the rotation of the plane of polarization of the wave in its passage through the ferrite. The rota- tion per centimeter of material depends upon the longitudinal field in the sample, increasing with this field and reaching a constant value when the ferrite is saturated. Hogan has given a discussion of the theory of this ferromagnetic effect. The incident linearly polarized wave may be described as a com- bination of two oppositely rotating circularly polarized waves. The real part, ^t , of the permeability of the ferrite varies with magnetic field in a different way for the two circular polarizations, as shown in Fig. 7. The velocity of propagation of the two polarizations is therefore different, and in passing through the ferrite they will fall out of phase by an amount proportional to sample length. Upon emerging they will com- bine to form a linearly polarized wave whose plane is rotated with re- spect to the incident wave. Reference to Fig. 7 shows that the most useful region for obtaining this effect lies below the field required to pro- duce ferromagnetic resonance (i.e., at frequencies above the ferromag- netic resonance frequency) in the region where the two curves are prac- tically parallel. In this region the device is relatively insensitive to small field changes and is somewhat ''broadband" with respect to frequency. For the practical application of the Faraday rotation, it is desirable that the attenuation per degree of rotation be small. Attenuation varies widely for different kinds of ferrites and is a function of frequency. For SECTION C MAY BE ROTATED FOR MAXIMUM TRANSMISSION SOLENOID FOR PRODUCING LONGITUDINAL FIELD Fig. 6 — Apparatus for demonstrating the microwave Faraday effect. Energy is supplied to section A. Rotation of plane of polarization occurs in section B and is controlled by controlling the longitudinal field. Section C is rotated for maxi- mum transmission. The angular displacement of C with respect to A is a measure of the rotation. 1168 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 NEGATIVE CIRCULARLY POLARIZED COMPONENT~->, V V POSITIVE CIRCULARLY ^-"^N. POLARIZED COMPONENT X MAGNETIC FIELD Fig. 7 — Real part (/*') of permeability of a ferrite versus applied magnetic field for the two circularly polarized components into which an incident plane wave may be resolved. The useful region for the Faraday effect is that for low H, where the two curves are approximately parallel. one sample at 9,000 mc, Hogan found a rotation of 60 degrees per cm of path and an attenuation of about 0.5 db per cm. X. EFFECT OF CROSS FIELD When a steady magnetic field is applied to a ferrite in a direction per- pendicular to the path of transmission of electromagnetic waves through the material, the effective ac permeability of the material varies with the applied magnetic field. For a given frequency, the permeability starts out positive. As the magnetic field increases the permeability goes through zero and approaches a large negative value as ferromagnetic resonance is reached. Above resonance the permeability is positive and gradually decreases with increase in magnetic field. Since the characteristic impedance of the ferrite relative to an empty waveguide is -/'. it is apparent that Z may be varied by changing the applied field. When /x is zero, the ferrite appears to be a perfect reflector, while when m = «, it provides a perfect match to the empty guide. Since it is possible to A BEVIEW OF NEW MAGNETIC PHENOMENA 1169 vary the characteristic impedance over a wide range, this effect has possibilities of application in attenuators, modulators, and phase shifting devices. XI. NEW MAGNETIC APPLICATIONS The engineer is naturally interested in some of the uses to which the new magnetic phenomena may be put. Several applications will be described briefly. 1. The gyrator. This is a four pole element for which there is a 180° phase difference between the two directions of propagation. In other words, the transfer impedances in the two directions are equal in mag- nitude but opposite in sign. Thus the device violates the reciprocity theorem. Hogan built the first microwave gyrator using the arrangement shown in Fig. 8. In this device, a wave traveling from left to right has its polarization rotated 90° counter-clockwise in the twisted section and another 90° in the same direction by the ferrite — a total rotation of 180°. For a Avave traveling from right to left the two rotations, that in the ferrite and that in the twisted section, cancel each other. Thus, if A and B represent points of the same phase for a left-to-right wave, they represent points of 180° phase difference for a right-to-left wave. 2. One way transmission system. If the input and ouput waveguides in Fig. 6 are oriented with their planes at 45° to each other and if the solenoid current is adjusted for 45° rotation in the ferrite, the result is a one-way transmission system.^^ Such a device is broadband. An arrange- ment of this sort may be employed in a microwave system to isolate the transmitter or receiver from the waveguide. It has the advantage that loss in the forward direction can be made quite small by proper choice of material. SOLENOIDAL COIL Fig. 8 — Schematic diagram of a microwave gyrator. From left to right, the plane of polarization is rotated 180°; corresponding to a phase shift of tt. From right to left, the plane of polarization is not rotated and the phase shift is there- fore zero. 1170 THE BELL SYSTEM TECHNICAL JOUENAL, SEPTEMBER, 1953 3. The polarization circulator. This is a modification of the one way transmission system in which there are two connections, with polariza- tions at 90° to each other, on either side of the ferrite rotating element. This is shown schematically in Fig. 9, along with a symbol which has been suggested for this element.^^ Energy sent into the device with polari- zation A emerges with polarization B, polarization B is rotated into C polarization C is rotated into D, and polarization D emerges as minus A. One practical application of this device is as a TR box in a radar system. Another, recently suggested by A. G. Fox, is as a device for separating the various channels in a multichannel communication system. Referring to Fig. 10, the signal comes in at A. Branch B is terminated in a filter which accepts one channel but reflects the remainder of the signal which is passed on to C. Here another filter accepts the second channel but passes the remainder on to D. D in turn feeds a second circulator. This process can go on until all channels are taken care of. 4. Measurement of magnetic field strength. The phenomenon of ferro- magnetic resonance suggests a means of making measurement of mag- netic field strength by observing the resonance frequency for a ferrite when subjected to the unknown field. ^^ Allen^^ has recently described a magnetometer in which an unknown field is measured by observing the Faraday rotation which it produces in a standard sample. 5. Other applications. There are several ways in which the interaction of the steady field with the microwave field may be utilized in designing switches, attenuators, and modulators. For example, one might set the two rectangular guides in Fig. 8 with their transmission planes at 90°. Then by varying the current in the solenoid and thereby varying the magnetic field applied to the ferrite, one may vary the amount of energy accepted by the second waveguide. This is then an electrically controlled attenuator. This same device offers the possibility of providing modula- tion of the microwave signal by modulating the current in the solenoid. FERRITE A "45° ROTATION c IB POLARIZATION CIRCULATOR CIRCUIT SYMBOL Fig. 9 — Schematic representation of polarization circulator. The ferrite is adjusted to 45° rotation by an external field, not shown. The circuit symbol for the circulator is shown at the right. A REVIEW OF NEW MAGNETIC PHENOMENA 1171 INCOMING — SIGNAL FILTER TO PASS CHANNEL t FILTER TO PASS CHANNEL 2 FILTER TO PASS CHANNEL 4 FILTER TO PASS CHANNEL 3 Fig. 10 — Schematic diagram showing the proposed use of circulators to sepa- rate the various channels in a multichannel communication system. Reggia and Beatty^^ have recently described a coaxial line variable attenuator in which the transmission loss is controlled by variation in an external cross field. XII. CONCLUSION From the discussion which has gone before, it should be apparent to the communications engineer that a whole new field of applications of magnetic materials has opened up. It is therefore essential that the engi- neer be acquainted with the modern picture of magnetism including the phenomena which have been described here — low frequency resonance, ferromagnetic resonance, and microwave Faraday effect. Some applica- tions have already been made of the high frequency characteristics, particularly of the Faraday rotation. Knowledge of the general high frequency characteristics of magnetic materials will enable the engineer to interpret new experimental information as it becomes available and intelligently to utilize the new materials in a variety of engineering ap- plications. I wish to express my appreciation to J. K. Gait, A. G. Ganz, C. L. Hogan, and V. E. Legg for several discussions which aided in the clari- fication of certain points described herein and to Messrs. Ganz and Legg for their careful criticism of the overall presentation. REFERENCES 1. J. L. Snoek, Non-Metallic Magnetic Materials for High Frequencies, Philips Technical Review, 8, p. 353, 1946. 2. A. Fairweather, F. F. Roberts and A. J. Welch, Ferrites, Reports on Progress in Physics, 15, p. 142, London, The Physical Society, 1952, 3. R. L. Harvev, I. J. Hegvi, H. W. Levernz, Ferromagnetic Spinels for Radio Frequencies, R.C.A. Review, 11, p. 321, 1950. 4. H. J. Williams, R. C. Sherwood, M. Goertz and F. J. Schnettler, Stressed 1172 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Pf Ferrites with Rectangular Hysteresis Loops, Scheduled for publication in ' A.I.E.E. Transactions. 5 J K Gait, Motion of a Ferromagnetic Domain Wall in Fe304, Physical Re- 'view, 85, p. 664, 1952. 6 V E Legg, Magnetic Measurements at Low Flux Densities using the A. C. Bridge, Bell System Tech. J., 15, p. 39, 1936. 7 C D Owens, Analysis of Measurements on Magnetic Ferrites, Proc. I.R.E., 41, p. 359, 1953. 8. C. Kittel, Theory of the Dispersion of Magnetic Permeability in Ferromag- netic Materials at Microwave Frequencies. Phys. Rev., 70, p. 281, 1946. 9. J. B. Birks, Measurement of the Permeability of Low Conductivity Ferro- magnetic Materials at Centimeter Wavelengths, Proc. Phys. Soc. (London), 60, p. 282, 1948. 10. G. T. Rado, R. W. Wright and W. H. Emerson, Ferromagnetism at Very High Frequencies. III. Two Mechanisms of Dispersion in a Ferrite, Phys. Rev., 80, p. 273, 1950. 11. Techniques of Microwave Measurements, (M.I.T. Radiation Laboratory Series, 11), McGraw-Hill Book Co., N. Y., 1947. 12. W. H. Surber, Jr. and G. E. Crouch, Jr., Dielectric Measurement Method for Solids at Microwave Frequencies, J. Appl. Phys., 10, p. 1130, 1948. 13. J. K. Gait, Initial Permeability and Related Losses in Ferrites, Ceramic Age, p. 29, Aug., 1952. 14. R. Becker and W. Doering, Ferromagnetismus, Julius Springer, Berlin, 1939. 15. M.I.T. Staff, Electric Circuits, John Wiley & Sons, N. Y., p. 323, 1943. 16. F. G. Brockman, P. H. Dowling and W. G. Steneck, Dimensional Effects Resulting from a High Dielectric Constant Found in a Ferromagnetic Ferrits, Phys. Rev., 77, p. 85, 1950. 17. P. M. Prache and H. Billottet, "Magnetodynamique des Semi-Conducteurs", Cables et Transmission, 6A, p. 317, 1952. 18. Griffiths, Nature, 158, p. 670, 1946. 19. C. Kittel, Ferromagnetic Resonance, Le Journal de Physique et le Radium, 12, p. 291, 1951. 20. H. G. Beljers and J. L. Snoek, Gyromagnetic Phenomena Occuring with Ferrites, Philips Tech. Rev., 11, p. 313, 1950. 21. W. A. Yager, J. K. Gait, F. R. Merritt and E. A. Wood, Ferromagnetic Reso- nance in Nickel Ferrite, Phys. Rev., 80, p. 744, 1950. 22. R. M. Bozorth, Ferromagnetism, New York, D. Van Nostrans Co., p. 803, 1951. 23. H. G. Beljers, Measurements on Gyromagnetic Resonance of a Ferrite Using Cavity Resonator, Physica, 14, p. 629, 1949. 24. F. F. Roberts, A Note on the Ferromagnetic Faraday Effect at Centimeter Wavelengths, Le Journal de Physique et le Radium, 12, p. 156, 1951. 25. C. L. Hogan, The Ferromagnetic Faraday Effect at Microwave Frequencies and its Applications, Bell System Tech. J., 31, p. 1, 1952. 26. Theodore Kahan, Ultra-High Frequency Magnetometer, U.S. Patent 2597149. 27. P. J. Allen, A Microwave Magnetometer, Proc. I.R.E. , 41, p. 100, 1953. 28. F. Reggia and R. W. Beatty, Characteristics of Magnetic Attenuators at UHF, Proc. I.R.E., 41, p. 93, 1953. The Correlatograph A Machine for Continuous Display of Short Term Correlation By W. R. BENNETT (Manuscript received April 24, 1953) An analog device has been constructed which displays short term cor- relation as a three-dimensional plot in which the rectangular coordinates are running time and lag time and the intensity of the pattern represents the correlation function. Preliminary tests on the properties of such a device are reported. The place of the electrical spectrograph^ as a signal analyzer has be- come well established in laboratory technology. It has occurred to many investigators however that the spectrum is not the only property of a signal which may be worthy of study, and in recent years there has been a considerable interest in other features, notably the correlation functions. On the basis of the accepted mathematical definitions the auto correla- tion function is the Fourier cosine transform of the power spectrum and in this sense would contain equivalent information presented in a differ- ent form. However, the mathematical definitions apply to a very long time interval and in practice we often deal with short segments of non- stationary processes. The spectrograph does not evaluate the true spec- trum in such cases but gives instead a spectrum-like function of fre- quency which changes with the observation time. We may regard the resolving filter as performing a weighted analysis in which the most recent parts of the signal contribute most heavily to the instantaneous response. The resulting "short term" spectrum depends on the charac- teristics of the resolving filter as well as the signal, but over a useful range of filtering selectivity the individual peculiarities of the signal are distinguishable even though the structural background may be charac- teristic of the filter. "Short term" correlation functions, in which the averaging interval is finite, have also been investigated and show phenomena analogous to 1173 1174 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 the short term spectrum. Comparison of the two short term functions is a much more involved matter than Avhen the time interval is large in both cases. The subject has been extensively treated in the literature,^ but the conclusions are still somewhat obscure. In an investigation which was started by the present author and the late Liss C. Peterson in 1950, we noted that a complete correlation analogue of the audio spectrograph had apparently never been constructed even though suggestions had appeared concerning the possibility.^ It seemed that some of the ques- tions concerning the merits of the correlation method of analysis could not properly be answered without actually building and testing such a machine and it also appeared possible that we would thereby add another useful tool for the problems associated with speech and other signals of interest. We accordingly undertook the design and construction of a "correlatograph" following lines closely parallel to spectrographic ex- perience in order to economize in new shop designs. In a spectrograph, such as used in visible speech for example, a three-dimensional display is obtained in which time and frequency are two rectangular coordinates and the short term power spectrum is represented by the intensity of light or density of marking. The func- tional mechanism is illustrated in Fig. 1. The incoming signal is delivered to a bank of band pass filters with midband frequencies uniformly spaced throughout the frequency range of interest. The envelopes of the filter output waves are picked up in turn by a rotating switch arm to control the voltage impressed on a marking stylus. The stylus moves across the paper in synchronism with the collector arm and the paper advances after each stroke. The analogous diagram for a correlatograph is shown in Fig. 2. We recall that the correlation of two functions is defined as the average of BAND FILTERS SIGNAL Fig. 1 — Mechanism of spectrograph. THE CORRELATOGRAPH 1175 their product with fixed time lag; i.e., when T is large, the expression ^i2(r) = I jy,{t)f2{l - t) dt gives the cross-correlation of /i(0 and/2(0, and the expression Mr) -\[ m)m-T) dt gives the autocorrelation of fi{t). In short term correlation T is finite. In Fig. 2, the different lag times are obtained by a tapped delay line and each tap is followed by its own multiplier and integrator. The integrated values are picked up by a rotating switch arm to control the marking stylus voltage as in Fig. 1. The rectangular coordinates are now t and r instead of t and /. The marking intensity represents the correlation function. In actual spectrographs it is usually found expedient to replace the bank of filters by a single filter and use a swept frequency oscillator and modulator to heterodyne the signal across the filter band. A rotating condenser plate tuning the oscillator thus takes the place of the rotating switch. The sweeping frequency must not change too rapidly for the analyzing filter to respond adequately, and also must not change so slowly that short signal bursts are imperfectly registered. A preliminary recording of the signal wave with a subsequent reproduction at a different MULTIPLIERS INTEGRATORS MARKING STYLUS Fig. 2 — Mechanism of correlatograph. 1176 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 speed forms a practical technique for securing the desired resolution in both frequency and time. Likewise in the correlatograph we can dispense with individual multipliers and integrators if we are willing to give up some of the otherwise available integration time for each r value. One multiplier and one integrator are then put in series with the switch arm and the contacts go directly to the delay line taps as shown in Fig. 3. A complete analogy with the simplified spectrograph would require a line of variable delay with a single output tap. This could be done with magnetic recording and moving heads, thereby eliminating the rotating switch. We felt however, that the fixed delay line would furnish the more stable and accurate component and chose the arrangement of Fig. 3. An auxiliary recording and reproducing process made it possible to accommodate a wide variety of signals with the one delay line. DETAILS OF APPARATUS Fig. 4 shows the arrangement of apparatus chosen. Our program was aimed at signals such as might occur in a nominal speech band extending from 200 to 4,000 cps recorded on magnetic tape at 15 inches per second. The reproducing element was a spinning double-ended pickup coil which successively scanned a one-inch loop of tape with one end beginning its scan just as the other end left the tape. The coil made 60 revolutions per second and hence the reproducing speed was 120 inches per second or eight times the recording rate. Our speech band was thus made to occupy the range from 1,600 to 32,000 cps, and this was therefore the range chosen for the delay line. A recording speed other than 15 inches TAPPED DELAY LINE CO To-Ar '.. V Ar a ^'MECHANICAL DRIVE-, / MULTIPLIER INTEGRATOR \ - X T ^ MARKING STYLUS tt SIGNALS AUTO I CROSS CORRELATION CORRELATION CO Fig. 3 — Mechanism of correlatograph with common multiplier and integrator. THE CORRELATOGRAPH 1177 1178 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 per second brings a correspondingly different signal band into the range of the delay line. The magnetic tape advances relatively slowly during the scanning process and for any one scan of a one-inch section may be regarded as stationary. The revolving pick up thus delivers segments of signal 1/120 second (8.33 ms) long for the correlation analysis. Our delay line consists of 1.5 ms total delay with 200 taps spaced 7.5 iis apart. During the first 1.5 ms of each scan, the delay line contains parts of the response from two successive scans and hence is not suitable for correlation measurement for lag times corresponding to all taps. We make provision therefore for excluding this interval from the analysis and use only the last 6.83 ms of each scan. The rotating switch advances one tap on the delay line for each one inch scan, so that the value of short term correlation corresponding to one value of lag time is computed every 8.33 ms. 200 values are computed in 1.67 sec after which a time of 0.33 sec is allowed for the return of the marking stylus to its initial position. The rotating smtch thus makes one revolution in two seconds but the last sixty degrees of the revolution are not used for display of correlation. The recording paper advances 0.01 inch after each stroke of the stylus. The rate of advance of the signal tape is adjustable by means of a gear train. In terms of the original signal wave, one second of recorded time is represented by the distance the paper moves in one second (0.005 inch) multiplied by the ratio of recording speed to speed of tape advance past the scanning head. The tape scanning mechanism with the synchronized motion of stylus and paper is due entirely to I. E. Cole, who designed this part of the system and supervised the necessary shop work. Mr. Cole also cooperated in the choice of a design plan for the rotating switch, which was manu- factured to meet our special requirements by Applied Science Corpora- tion of Princeton, N. J. The switch output is followed by an electronic gating circuit which removes the effect of time jitter in the beginnings and ei?ds of the contact intervals and trims off the previously mentioned 1.5 ms interval during which the tail end of one scan of the tape loop remains in the delay line. The gating wave is generated from the common 60-cycle power supply which drives all the mechanical apparatus. The output of the electronic gating circuit consists of segments of signal 6.83 ms long with 1.5 ms separation and with the delay increasing in steps of 7.5 microseconds between one segment and the next. This con- stitutes one input to the multiplier; the other input is the undelayed signal in the case of autocorrelation or an independent signal for cross- correlation. The multiplier consists of a bridge of germanium varistors with the two THE CORRELATOGRAPH 1179 inputs applied across the two diagonals and the output taken off one pair of input terminals through a low pass filter. Each varistor is operated in a substantially square law range. If A and B represent the two inputs, we try in effect to produce an output proportional to {A + Bf - (A - Bf = ^AB. The necessary conditions are conveniently expressed in terms of sine wave inputs in order that analyzer measurements may be used as a check on accuracy. Let P cos pt represent a typical component of the signal impressed as one input to the multiplier and Q cos qi a typical component simultaneously applied to the other input terminal. The product is PQ pQ {P cos pi)iQ cos qt) = -^ cos ip -\- q)t -]- -^ cos {p — q)t. Since our purpose is to integrate the output of the multiplier over a time interval relatively long compared to the periods of components within the signal band, we have no interest in the product component of fre- quency p •\- q and in fact filter out such components immediately along with the original signal components to prevent loading the product amplifier with unessential waves. A significant test on the accuracy of the multiplier is therefore the fidelity with which the amplitude of the difference frequency term cos {p — q)t follows the product of the ampli- tudes of cos pi and cos qt. This is not sufficient in itself however because it does not give a check on the balancing out of the squares of the in- dividual inputs. To test the latter we superimpose the two components P cos pi and Q cos qt on one input circuit with no signal applied to the other input and measure the output component of frequency p — q. We also repeat the measurement with the two sine waves impressed on the second input and nothing on the first input. Typical results are shown in Fig. 5. The varistors were selected by R. R. Blair from persistent screen cathode ray tube displays of the characteristics. The square law region is enlarged and the output increased by applying a direct current bias to the bridge through a series resistance. This form of multiplier copies a design worked out by R. R. Riesz for a different purpose. The product output of the multiplier is weak at best because a rela- tively small range of varistor inputs fit the necessary law. A fairly high gain product amplifier is therefore provided. Fortunately we do not have to amplify a band which extends all the way down to zero frequency. The significant component when calculating the autocorrelation function of P cos p^ for lag time r is (PV2) cos pr, which is constant only when p = 0 or r is constant. Our lowest value of p corresponds to 1600 cps. The value 1180 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 100 8 6 4 2 10 a 6 4 1- § a z '"^ 1 X10-3 / / / /: / /^ P IN INPUT NO. 1 Q IN INPUT NO. 2 ^ / / J f / / / / / / P,Q IN INPUT NO. 2 «^ 4 / ^ / X^ / / 2 ^ 2 J / / D ^ -^ F^ Q IN INPUT NO. 1 \ 1 1 - 0.01 1 10^ 10-2 10" PRODUCT OF RMS INPUT VOLTAGES (VOLTS X VOLTS) Fig. 5 — Performance curves of multiplier. of r increases in small steps from 0 to 1.5 ms in 5/3 sec. and therefore can be approximated by r = hi, where h = .0015/(5/3) = 0.0009. Hence the lowest product frequency is roughly 1600 X 0.0009 = 1.44 cps. The actual low frequency cutoff of the amplifier is made about 0.1 cps to allow for changes in the signal components and to preserve good trans- mission within the nominal band. The upper cutoff frequency is likewise made somewhat higher than the nominal value of 32,000 X 0.0009 = 28.8 cps. The integration is performed by a series condenser and shunt resistance at the amplifier output. A good approximation to integration is accom- plished by a large time constant such that the indicial admittance THE CORRELATOGRAPH 1181 remains linear during the integrating interval. Such a long time constant would carry over too much charge from previous intervals if continuous integration were permitted so a shorting relay is provided to discharge the condenser quickly to ground after the integrated value is sampled. The sampling is done by a two-way clamp circuit with the timing pulse generated from the trailing edge of the pulses which gate the switch outputs. The shorting relay operates directly from the 60-cycle supply and is of a type specifically designed to give a brief closure of about one ms. every half period of the driving wave. The sampled outputs of the integrator are applied to a balanced modulator to which is also applied a 6-kc carrier. The resulting double sideband suppressed carrier wave is amplified to form the marking voltage applied to the stylus. Instantan- eous compression is incorporated in the marking amplifier to extend the range of input magnitudes which are encompassed by the relatively narrow recording range of the paper. The stylus responds equally to positive and negative voltages and does not resolve the individual high frequency oscillations. The result is like full wave rectification of the correlation functions. Grateful acknowledgement is given to A. J. Rack, A. E. Johanson and P. A. Reiling for suggested physical configurations and design informa- tion suitable for the circuits which generate the various control pulses, and which sample and hold the integrated outputs. G. W. Blake tested and adjusted these circuits after they had been constructed by the wiring shop. Performance runs were made by F. H. Tendick and N. K. Poole. Also at various stages of the project assistance was given by W. A. Klute, F. W. Kammerer and R. L. Carbrey. We have also received helpful advice from many other associates too numerous for explicit mention. The 200-tap delay line was constructed by the transmission networks department. It consisted of one low^ pass filter section per tap with mutual inductance between sections to maximize the linearity of the phase curve. The taps were taken off high impedance shunts wdth the output fraction tapered down the line to give constant loss along the taps. C. E. Jakielski planned, assembled, tested, and adjusted the delay line. The intricate task of connecting the 200 output taps to the cor- responding 200 contacts of the switch was performed by M. Biazzo. The delay line and switch appear in the photograph. Fig. 6. Additional electronic apparatus not shown in this photograph was placed on inde- pendent panels for experimental convenience but can be arranged in a chassis in the same cabinet with the other apparatus, now that the components have been determined. 1182 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Fig. 6 — Rear view of correlatograph showing rotating switch and delay line. PRELIMINARY RESULTS Tests on the complete correlatograph have only been carried far enough to date to verify that the operation is as planned. Fig. 7 shows correlatograms obtained with purely sinusoidal inputs of frequencies 200, 400, GOO, 800, 2,000, 3,000, and 4,000 cps recorded on the tape at 15 inches per second. In terms of the original signal the values of r extend from zero to 12 ms. The profiles represented by the light and dark bands should be rectified cosine waves starting with a peak at r = 0 and repeating at intervals of one half the period of the wave. In the total range of r = 0.012 sec, a frequency of 200 cps should show 4.8 periods, 4,000 cps would show 96 periods, and in general / cps would show 0.024 f periods. The sudden changes in density parallel to the stripes were caused by manual adjustments of the marking amplifier gain. Fig. 8 fchows correlatograms of a 1,000-cps sine wave embedded in various amounts of flat thermal noise extending throughout the entire input band. The sequence from bottom to top is noise alone, signal and noise power equal, signal power down on noise power by 5 db, 10 db, 15 db, and 20 db. A long time correlation analysis would show zero correlations for the noise alone except for small values of r. Short THE CORRELATOGRAPH 1183 term correlation gives a mottled background level. The signal pattern shows through this background quite plainly when the two powers are equal. As the signal is reduced relative to the noise the signal pattern is obscured and seems entirely missing at 20 db below the noise level. This is a limitation based mainly on the integration time of this particular apparatus. It is possible to improve the resolution by using longer in- tegration periods with corresponding sacrifice of ability to detect fast changes in the applied signal. As pointed out by C. B. Feldman, the autocorrelation type of analysis suffers the same sort of limitations in the low signal-to-noise ratio case as filtering after detection imposes in spectral analysis. That is, finite integration time in the autocorrelation case and finite filter band width in the postdetection filter both allow errors from interaction of noise with noise to swamp the relatively small desired effect of signal.* Cross-correlation on the other hand is like filter- ing ahead of the detector in that the interaction of noise with noise is suppressed leaving as the dominant error the relatively small interac- tion of noise with signal. We cannot obtain this advantage of cross- correlation if we do not have available a noisefree signal to cross-correlate with, and analogously an accurate knowledge of the frequency to be selected is helpful in securing the best possible results from predetection filtering. It Avill be noted that alternate stripes of the signal pattern fade at first as the signal is decreased relative to the noise. This effect appears to be associated with incomplete suppression by the multiplier of the components proportional to the squares of the individual noise inputs. If the noise inputs were steady, the squares would produce mainly direct current which is not transmitted by the product amplifier. Variation in Fig. 7 — Correlatograiii of single-frequency wave. * The autocorrelation method however has an advantage if the signal can be recognized at all in that it is capable of measuring the frequency of the original components, which the postdetection filter cannot do at best. 1184 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 Fig. 8 — Correlatogram of single-frequency wave and random noise. one noise input because of the idle interval during which no signal is supplied by the delay line is a source of change in the direct current which is partially transmitted through the product amplifier to give a biased integrated value and hence an unequal treatment of positive and negative correlation. Effects of this sort would be particularly noticeable when the noise is large relative to the sinusoidal component. Fig. 9 shows a sample correlatogram of the sentence ''He beats his head against the posts." spoken by G. E. Peterson. The locations of the sounds were marked on the tape by observation during an audio playback and from these marks the corresponding positions on the correlatogram were found. The lower legend gives the ordinary English letters and the upper the symbols of the international phonetic alphabet. The characteristic frequency indicated by the vowel sounds is of the order of 600 cps, which coincides very well with the first formant frequency of the speaker's voice. It is a characteristic of this method that the pattern is dominated by the largest component present. To show the higher formants, which have weaker amplitudes, it would be necessary 0.012 - RUNNING TIME — ^ Fig. 9 — Correlatogram of speech sentence. THE CORRELATOGRAPH 1185 to filter out the strong low frequency components from the input. The ''s" sounds show closely spaced stripes indicating a concentration of energy at the top of the band. Some of the consonants do not show much under the conditions of this test. It is of interest to compare correlatograms with spectrograms of the same signal. For the single frequency input, the stripes on the correlato- gram would be replaced by a single line on the spectrogram. It would be possible to construct signals such that these patterns are interchanged. For example, a rounded band of noise transmitted over two paths having different times of transmission would have a correlatogram with two stripes corresponding to lag times of zero and the delay difference while the spectrogram would show a periodic array of stripes corresponding to the interference pattern of the two paths. It appears that there may be complementary fields of usefulness for the two kinds of analysis and further study is planned. Besides the many individuals previously named as contributing the various phases of the project, I would like to acknowledge the inspiration and guidance of R. K. Potter in initiating and carrying through the program. L. C. Peterson shared equal responsibility with the author in the project and but for his untimely death would have been a co-author of the present paper. November 6, 1952 REFERENCES 1. W. Koenig, H. K. Dunn andL. Y. Lacy, The Sound Spectrograph, J. Acoust* Soc. Am., 17, pp. 19-29, 1946. 2. Y. W. Lee, T. P. Cheatham, Jr. and J. B. Wiesner, Applications of Correlation Analysis to the Detection of Periodic Signals in Noise, Proc. I.R.E., 38, pp. 1165-1171, 1950. 3. K. N. Stevens, Autocorrelation of Speech Sounds, J. Acoust. Soc. Am., 22, pp. 769-771, 1950. A Statistical Study of Selective Fading of Super-High Frequency Radio Signals By R. L. KAYLOR (Manuscript received May 7, 1953) The results of two months of comprehensive frequency-sweep measure- ments of selective fading in the hand between 3750 and 4190 mc ever a radio relay path in Iowa are reported. An abridgement of the data, general conclusions derived from the data and an example of the use of the data in connection with frequency diversity measures for radio relay systems are given. INTRODUCTION It is well known that, in the high-frequency range during fading con- ditions, radio signals on different frequencies may exhibit at any instant radically different behavior. This may be true even though they are in the same frequency band and exhibit the same statistical behavior, when observed over a longer period of time. It is also kno^vn that h-f fading may be frequency selective enough within the narrow limits of a single radio channel to cause severe distortion of modulated signals. It has been established that the cause of these phenomena is multipath trans- mission. This knowledge, which is of long standing in the high-frequency range, raised questions concerning the prevalence of similar phenomena in the super-high-frequency range about which relatively little has been known until recently. During recent years studies of super-high-frequency propagation and fading have been made which have been previously reported. In these tests a frequency-sweep method was used to determine how the loss of a particular radio path varied with frequency at a given instant; and short-pulse methods were used to determine the path length differences which were involved when multipath transmission occurred. These are ^ A. B. Crawford and W. C. Jakes, Jr., Selective Fading of Microwaves, and 0. E. DeLange, Propagation Studies at Microwave Frequencies by Means of Very Short Pulses, Bell System Tech J., 31, Jan., 1952. 1187 1188 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 §SiM 9§m tig. 1 —Typical records obtained during the Lowden-Princeton, Iowa, fre- quency-sweep transmission measurements. The horizontal scale is proportional to frequencies between 3,750 and 4,150 mc (the single spike on the right is at 4,190 mc); and the vertical scale is proportional to the amplitude of the received signal. SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1189 standard methods for studying multipath transmission (or selective fading) and the results obtained were in good agreement. Some frequency-sweep transmission measurements were made during the summer of 1950 over a typical radio relay path in the mid-continent region of the United States for the purpose of obtaining additional data cf a statistical nature which could be used in system engineering. The objective of the tests was to determine what per cent of the fading was frequency selective and the degree of its frequency selectivity. Such data were needed to evaluate the advantages of using frequency diversity as a means of minimizing the effects of fading, and to provide quantitative data for designing frequency-diversity measures. Recordings were made throughout the months of July and August which serve as the basis for a statistical picture of the fading occurring during those months in the frequency band between 3,750 and 4,190 mc. Path Over Which Measurements Were Made The measurements were made over a 30.8-mile path between the Lowden and Princeton (Iowa) towers of the Chicago-Omaha Radio Relay System. This path was chosen as a typical radio relay path as to length, clearance above terrain, and climatic conditions. Height-loss runs made by the American Telephone and Telegraph Company (when originally selecting the path as part of their radio relay route) indicated that ground reflections on this path were unimportant. The reflection coefficient was less than 0.1 Method of Tests The frequency -sweep equipment used in these tests w^as similar to that used in the previous studies.^ With this equipment about 50,000 record photographs were obtained of a cathode-ray tube presentation of the path-loss vs frequency characteristic of the path between 3,750 and 4,150 mc. Fig. 1 shows three illustrations of these record photo- graphs, which are more fully discussed below^ The taking of these records was distributed throughout the months of July and August in such a manner as to give complete coverage of all the fading during that pe- riod. In addition the single-frequency path loss at 4,190 mc was con- tinuously recorded throughout the two months. The data from these records were analyzed on a statistical basis; and families of curves were obtained depicting several aspects of the nature of the selective fading encountered during the tests. These are described more fully below. 1190 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 GENERAL DISCUSSION OF FADING PHENOMENA The variations in the strength of a received radio signal known as "fading^' are caused by variable or temporary conditions in the trans- mission path. These conditions fall into two broad classes: those causing partial obstruction of the path, and those causing multipath transmis- sion. The latter class is believed to be the principal source of frequency- selective fading. Multipath transmission involves the reception of more than one signal ray, each of which travels over a different path between the transmitter and receiver. Generally each such path has a different length between the transmitter and receiver. Multipath transmission involves either or both reflection or refraction of at least one (and in some cases all) of the re- ceived rays. Since the conditions of the atmosphere are continuously varying, the paths of the received rays are variable with time. The re- ceived signal is the resultant of all the rays accepted by the receiving antenna. The relative phase of the different received rays depends on (1) the differences in the lengths of the paths over which they have travelled and (2) the signal frequency. If the rays arrive at the receiving antenna in nearly the same phase, they add and enhance the received signal; if they arrive in phase opposition they partially cancel each other and fading results. This fading is not only variable with time but also with signal frequency, and is called "(frequency) selective fading". Since the conditions which cause this kind of fading are substantially random, the variation of fading with time on a statistical basis might be expected to approach the Rayleigh distribution. There has been experimental con- firmation of this. A number of other factors are involved which will not be treated here, since the mechanism and effects of multipath transmission have been discussed quite thoroughly in the previously mentioned reports by Crawford and Jakes, DeLange and in an earlier paper.^ Results of Tests Fig. 1 shows three illustrations of the type of records obtained. On each record the horizontal deflection of the cathode-ray- tube trace is proportional to the radio signal frequency. The vertical deflection is linearly proportional to the amplitude of the received signal, which because of constant transmitter power is inversely proportional to the « H. T. Friis, Microwave Repeater Research, Bell System Tech. J., 27, Apr. 1948. SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1191 path loss.^ In each record one second was required for the trace to travel through the entire frequency range covered. The three records shown are consecutive, being taken at 3:05:21 AM, 3:05:25 AM and 3:05:27 AM. The first record (frame number 9524) was taken with 10 db more attenuation in the input to the measuring equipment than when the later two records were taken (frame numbers 9525 and 9526). There is obvious overloading in the left hand portions of the records on frames 9525 and 9526; but the accuracy of the right hand portions of these records (showing the deeper part of the fade) is unimpaired. The noticeable difference between the shapes of the curves near the deeper parts of the fade on frames 9525 and 9526 is typical. These changes occurred within two seconds. Generally, the deeper parts of the fades show more rapid changes than the less deep parts. From the 50,000 record photographs all those pertaining to fading of 30 db or more were segregated and analyzed. The remaining records were analyzed on a sampling basis, except that every record showing unique effects was analyzed. About 1,800 path-loss versus frequency curves such as those illustrated in Figs. 2 and 3 were obtained. These curves were separately studied, and were also treated statistically. Because the levels during the deeper part of the fade are too low to show on frame 9524, and because portions of frame 9525 showing the less severe part of the fade are affected by overloading, it was necessary to combine two records (shown on Fig. 1) to obtain the single path-loss versus frequency curve shown as Fig. 2(a). Theoretically, it should be possible to synthesize by means of an addi- tion-of- vectors method each of the path-loss versus frequency curves obtained from these tests. Each vector term of the equation would correspond to a component of the received signal and would be of the form R cos coT, where: R is the magnitude of a particular component (normalized to the magnitude of the direct signal component), T is the delay (in seconds) between the time of arrival of the direct signal com- ponent and the particular interfering component, and oj is 27r times the frequency. In practice, however, it has been found in the case of deep fades that components of relatively small magnitude and relatively long delay are of importance in determining the shapes of the curves near maxima of path loss. Also it has been found that in most such cases quite a few components are involved. These factors make accurate analysis of many 2 A logarithmic amplitude characteristic (db scale) would have been preferable; but time did not permit modifications of the test equipment before the start of the 1950 fading season. 1192 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 curves impractical without a computer capable of handling a relatively- large number of components several of which may be relatively small. However, experimentation with a trial and error method of solution on a few selected curves has shed considerable light on the nature of received signals which could produce the types of curves included in the data under consideration. Curve (a) on Fig. 4 is one of those selected for analysis and was syn- thesized by using a direct signal component and six interfering com- ponents as shown in Table I. If components 5 and 6 had been omitted from the synthesis, curve (b) would have resulted; and, if components, 3, 4, 5 and 6 had been omitted curve (c) would have been obtained. The difference between curves (a) and (b) illustrate the radical effect which can be produced on the shape of a curve near maxima of path loss 10 20 30 -J < i 40 O 9 0 g 10 o $30 z ^40 8 5 50 a. 0 (a) ^^ ^ \ / ■N^ -A \ / \ v (b) -^ ^__^ / ^ '^ r 1 1 30 (c) ^^ ^^ "^ 3800 3900 4000 4100 FREQUENCY IN MEGACYCLES PER SECOND Fig. 2 — Typical path-loss versus frequency curves observed on the Lowden- Princeton, Iowa, path during July and August, 1960. SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1193 10 20 _j 40 < a. o 50 < o o 5 10 o a. it 20 if) 5 30 o (a) , ^ X \, / 1 \ II II (b) V / ^ \ \ /^ ' ^^ / V i \ / J b 10 20 30 50 (c) 1 N, y ^ 0 s \j ^ / \ 1 1 ! 3800 3900 4000 4100 4200 FREQUENCY IN MEGACYCLES PER SECOND Fig. 3 — Typical path-loss versus frequencj^ curves observed on the Princeton- Lowden, Iowa, path during July and August, 1950. by relatively small components such as components 5 and 6. In the case of deeper fades much smaller components may be of importance in determining the shapes of the curves near maxima of path loss; hence the curves for deep fades are quite difficult to synthesize. The shape near the maximum of path-loss on the curve in Fig. 2(a) is quite typical of the shapes to be found in the data under discussion. Based on the experience gained in attempting to synthesize some of the cur\^es, it is possible to recognize the significance of certain features of other curves, and to generalize the nature of all the curves. The prin- cipal conclusions are: 1. No deep fades were found which did not show definite frequency 1194 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 _i 0 < a. o z r a q: -I ^ 15 O UJ o z {^520 O 25 3750 a 42.5 23 11 1/ \ 1 0.5 0.2 0 1 99.9 1 1 1 1 c ) \ ( ) 1< I 1 6 2 0 2 4 2 8 3 2 DIFFERENCE IN PATH LOSS IN DECIBELS Fig. 6 -;- Statistical distribution of differences between maximum and minimum path loss in bands 20 mc wide. (Center frequency of 20-mc band random with respect to frequencies corresponding to maxima and minima of path loss). SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1197 400-mc observed band was 8 db less than mid-day normal; and in one case it was more than 10 db less than mid-day normal. Another family of curves is shown by Fig. 6. These curves were ob- tained by dividing the 400-mc observed band into a number of 20-mc bands, chosen at random with regard to the shapes of the path-loss versus frequency curves. These data show the difference between the maximum and minimum losses within a single 20-mc broad-band channel which might accompany a fade of a given depth. Such data are of use in estimating possible distortions of a modulated signal occupying a band width of 20 mc. Frequency Diversity The fact that the instantaneous fading may be different on different frequencies within the same frequency range offers a means for mitigating transmission impairments caused by fading. During periods when there is fading in excess of a specified value on the regular carrier frequency, the carrier can be shifted to an alternate frequency in the hope that the fading on the alternate frequency may be less severe. The merit of using this type of frequency diversity can be gauged from the statistical distribution of fading on the alternate frequency during periods when there is fading in excess of the specified value on the regular frequency. The data obtained in these tests indicates that there was no correlation between the fading on frequencies separated by: (1) 40 mc or more during periods when there was fading of 10 db on one of the frequencies and (2) 160 mc or more during periods when there was fading of at least 20 db on one of the frequencies. However, during periods of severe fading, the data indicate considerable correlation between the fading on regular and alternate frequencies separated by 80 mc or less. Fig. 7 shows the distributions of fading on alternate frequencies at specified frequency separations from the regular frequency, when there is fading of a specified depth on the regular frequency. Table II indicates which of the curves on Fig. 7 applies to a particular set of conditions. Curves G and H on Fig. 8 show the distributions of depths of fade at 4190 mc during the entire months of July and August, respectively. Comparison of the 4,190-mc data with data from tests made over other paths indicates that the fading occurring during the Iowa tests was nearly as severe as during the "worst month" observed on any path to date. Curve B on Fig. 7 shows the statistical distribution of fading on any frequency (within the range under consideration) during periods when important fading is prevalent. The fading shown by this curve is con- 1198 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 4 8 12 16 20 24 28 32 36 40 44 PATH LOSS IN DECIBELS (RELATIVE TO MIDDAY NORMAL) Fig. 7 — statistical distributions of fading measured on Lowden-Princeton, fk^^'^irJ'^ during July and August, 1950. These curves are useful in predicting the efficacy of specific frequency-diversity systems as described in the text. siderably more severe than is shown by the curves on Fig. 8. This is because the latter curves include all the time within the months under consideration. Therefore, they include much time when the fading mechanism was not present in the path, and the path loss remained steady near mid-day normal. There is evident similarity between the shapes of Curves A through F (on Fig. 7) with the shape of the curve which is based upon the Rayleigh distribution. This is consistent with the theory that deep fading asso- ciated with multipath transmission is caused by random phasing of a large number of vectorial components. SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1199 These data have been applied practically to the design of frequency- diversity measures for minimizing circuit outages caused by fading in TD-2 radio relay systems. Fig. 9 shows an illustration of the practical use of the data. Curve H (which is the same as Curve H on Fig. 8) shows the distribution with time of the fading on a typical radio relay path without frequency diversity measures. Curve J shows the distributions of fading if certain specific frequency diversity measures are used (based solely on fading considerations). The difference between Curves H and J shows the improvement gained by using that specific kind of frequency diversity. For example, Curve H shows that without frequency diversity fading in excess of 30 db will occur 0.075 per cent of the time and fading in ex- cess of 40 db will occur 0.02 per cent of the time. But, if the kind of fre- Table II Depth of Fade on Regular Separation Between Regular and Alternate Curve Frequency Frequencies 10 db 40 mc or more A 20 40 mc C 20 80 mc or more B 30 40 mc E 30 80 mc C 30 160 mc or more B 40 40 mc F 40 80 D 40 160 mc or more B quency diversity to which Curve / corresponds is used, fades deeper than 30 db will occur only 0.0012 per cent of the time, and fades deeper than 40 db will occui only 0.00008 per cent of the time. Thus the improve- ment resulting from this type of frequency diversity is a reduction of fades deeper than 30 db from 0.075 to 0.0012 per cent of the time, and fades deeper than 40 db from 0.02 to 0.00008 per cent of the time. To explain how Curve J was derived, let us assume that during any time when there is a fade of 30 db or more on the operating frequency of a given channel the signal Avill be switched to another channel fre- quency. Let us further assume that the alternate frequency will be sepa- rated by 160, 240, 320, or 400 mc from the assumed operating frequency and that the choice of alternate frequency is random. If we also assume that the conditions of August, 1950, on the Iowa path prevail, we will find from Curve H on Fig. 8 that the assumed operating frequency will have a fade of 30 db or more 0.075 per cent of the time. Then a new dis- tribution curve can be prepared, based on (1) Curve B (on Fig. 7) for 1200 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 0.075 per cent of the time, and (2) that portion of Curve H (on Fig. 8) which corresponds to fades of less than 30 db for the remaining 99.925 per cent of the time. Curve / on Fig. 9 is such a curve. If frequencies separated from the assumed operating frequency by 80 mc were used instead of those assumed above. Curve C instead of Curve B on Fig. 7 would have been used. This would have shown a somewhat smaller improvement because of some correlation between the fading on the assumed operating frequency and an alternate fre- quency separated from it by 80 mc. However, the improvement gained from the use of this system would be ample to prove-in its use. The question is sometimes raised concerning the reason that the curves on Fig. 9 show an apparent improvement for fading of less than 30 db, since the circuit is switched only when fading deeper than 30 db occurs. This is because the curves are cumulative distribution curves; -* 0 4 8 12 16 20 24 28 32 36 40 44 48 PATH LOSS IN DECIBELS (RELATIVE TO MIDDAY NORMAL) Fig. 8 - Statistical (nsfribution of fading loss Princeton-Lowden, Iowa, path July and August, 1950. SELECTIVE FADING OF SUPER-HIGH FREQUENCY SIGNALS 1201 ^ e O _l X ^10-3 V STATISTICAL DISTRIBUTION OF PATH LOSS \ \, WITHOUT FREQUENCY DIVERSITY WITH FREQUENCY DIVERSITY \ \ \ X X. % ^ ^H \ \ ^^ \ \ \ \ \ \ \ \ 'x ''n. X^ '^^ V 4 8 12 16 20 24 28 32 36 40 44 PATH LOSS IN DECIBELS (RELATIVE TO MIDDAY NORMAL) 48 Fig. 9 — Effect on transmission of use of frequency diversity in 3,750-4,150 mc range. and reduction of the per cent of time when there is fading deeper than 30 db also reduces the per cent of time when there is fading deeper than 25 db, etc. CONCLUSIONS Quantitative information on fading phenomena is essential in the engineering of super-high-frequency systems and in evaluation of fre- quency diversity arrangements. The data obtained from the field tests and the statistical analysis reported herein, while limited to a single path and particular season, fill a gap in previous knowledge. They have found practical application in the design of diversity systems for improving the reliability of radio relay systems. The principal conclusions that can be drawn from the available data are: (1) all of the deep fading is definitely frequency selective, and is caused by a complex multi-path transmission; (2) deep selective fading 1202 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 is ordinarily accompanied by a 6 to 10 db signal depression over a band of at least several hundred megacycles; (3) fading on frequencies sepa- rated by 160 or more megacycles shows little correlation, but fading on frequencies more closely spaced shows an increasing correlation as the frequency difference is diminished; and (4) frequency diversity offers a practical means of mitigating circuit impairment due to fading, if the transmission can be shifted to a frequency far enough away to minimize correlation with fading on the original frequency. ACKNO WLEDGEM ENTS The author wishes to acknowledge the important contribution of J. Mallett, who conducted the field tests during which the data were obtained. The field tests and data analysis were under the supervision of R. P. Booth. Acceleration Effects on Electron Tubes By F. W. STUBNER (Manuscript received June 19, 1953) This paper discusses methods of measuring shock and vibratory accel- erations to which electron tubes may be subjected in various equipments^ and the influence of these disturbances on the performance of the tubes. An outline is given of the design problems connected with the elimination of tube damage or faulty operation of tubes under adverse shock or vibra- tion conditions, and of the methods used for simulating these conditions by means of production test machines and test methods. INTRODUCTION The rapid expansion of the use of electronic equipment by industry and the armed services has created increasingly new demands on both the electrical and mechanical characteristics of electron tubes. Since these tubes are electronic devices, it is only natural that their structural designs are dictated to a large extent by electrical requirements. How- ever, the experience gained with conventional tubes in some of the new equipment applications has revealed certain mechanical shortcomings which reflect on the proper functioning or life of the tubes. The realiza- tion of the increasing importance of mechanical design has resulted in an increased effort for structural improvements to assure more reliable tube performance. The term ''reliable", as used here, denotes that a tube has a high degree of dependability when subjected to specific conditions, either electrical or mechanical. Thus, the requisites for reliability may differ for various applications. Although, in designing a tube, many re- quirements must be taken into consideration, only some of the problems connected with dependable tube performance under mechanical distur- bances, i.e., shocks and vibrations, will be discussed here. Since the performance of a tube depends on the geometry of its com- ponent parts, minute changes in element spacing may produce variations in its characteristics. Because of the necessarily delicate stmcture of some tube elements, permanent or transient dimensional changes may be produced by mechanical forces acting on the tube unless the tube is 1203 1204 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 specifically designed to minimize the effects of such distrubances. For a rational design, it is, therefore, necessary to have some knowledge of the nature of the disturbances likely to be encountered by tubes during their service life. If one considers the numerous conditions under which elec- tronic equipment is required to function, it becomes clear that the me- chanical requirements are manifold indeed. Equipment applications can be divided into three general groups, each one imposing in many cases special requirements on tubes. These groups are: (a) stationary equip- ment, such as central office telephone installations, home radio and tele- vision receivers, etc., (b) mobile equipment used in land vehicles, ships or airplanes, and (c) portable equipment. In many instances, military equipments straddle above subdivisions and superimpose additional re- quirements. It must be stressed, too, that pre-service conditions encount- ered through handling and shipping must be taken into consideration, since a tube is of no use to the customer until it is installed and operating. A knowledge of expected service conditions is not only useful in the initial design stage but also aids the manufacturer in devising suitable test gear to check the quality of the product at the factory. Although a wealth of data has been collected on shocks and vibrations to which elec- tronic equipments are subjected under actual service conditions, little is known how these disturbances are altered by the mechanical structures of the equipments before they reach the tubes. In general, the nature and magnitude of mechanical disturbances can be expressed in terms of acceleration, velocity, or displacement. Since electron tubes may respond to a wide frequency range of vibrations, the most sensitive measure is acceleration, which varies as the square of the frequency of the element displacement. Velocity or displacement instruments are usually not suffi- ciently sensitive to give a true record of disturbances in the higher modes of vibration due to the small velocity and displacement values involved. The investigation of the nature of accelerations at tube sockets offers special problems. The accelerometers must approximate the weights of the tubes used in the respective sockets so that the disturbances are not modified by the substitution of acceleration pick-ups for the tubes. For the same reason, the method of fastening the accelerometers in the 80MASS (a) (b) Fig. 4 — (a) Radial miniature tube compression type accelerometer and (b) radial miniature tube shear type accelerometer. ACCELERATION EFFECTS ON ELECTRON TUBES 1211 tension and compression stresses respectively, thereby producing the desired charges in these elements. The characteristics, i.e., the resonant frequency and sensitivity, of these elements are largely determined by the dimensions of the metal member. Relatively high internal capacities can be obtained by this construction. A recent variation of such a struc- ture is shown in Fig. 6(b). In this design the inner metal component of the cantilever described above has been eliminated. The entire structure is made of the active material, in this case in cylindrical shape for ease of 320P 300° 290" 260° 250' 240P 23Cf COMPRESSION TYPE 220° 190° 180° 170° 160° Fig. 5 — Relative angular response of compression and shear type ferro electric crystal accelerometers. 1212 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 manufacture. Sections of conductive coating are deposited on this cylinder as shown. These are used to initially polarize the material, and, in use, to collect the charges produced by stressing the inner and outer fibers of the cylinder. It is interesting to note that these elements can be used to detect, by proper choice of connections, either radial or axial accelerations. Through suitable external instrumentation both directions can also be recorded simultaneously if desired. The above illustrations show but a few applications of this material for detecting accelerating forces. Additional forms will suggest them- selves, each of particular advantage for specific application. The sensitivity of an accelerometer is generally given in coulombs/g or open circuit voltage. Its effective voltage output over a given frequency range is a function of the characteristics of the associated equipment including the connecting cable. Calibration in the lower frequency range up to a few hundred cycles may be performed by vibrating the acceler- ometer on variable frequency vibration machines or resonating spring OUTSIDE SILVER LINED 120* ON OPPOSITE SIDES-INSIDE COMPLETELY s — ^. SILVER LINED FOR LENGTH SHOWN (a) (b) Fio. 6 — (a) Metal cantilever type element and (b) radial and axial miniature cantilever type accelerometer. ACCELERATION EFFECTS ON ELECTRON TUBES 1213 systems at amplitudes that are measurable with the help of a microscope. The peak acceleration in gravitational units for sinusoidal motion is given by: g = .102 (c.p.s.)^ X amplitude (inches) At higher frequencies, this method of calibration becomes increasingly difficult due to the decrease in obtainable amplitudes. A calibration method for a wide frequency range is described in Reference 3. In gen- eral, the sensitivity of an accelerometer is a function of the active ma- terial, its size, and the weight M employed. The useful frequency range increases with decreasing sensitivity. For a given design, a compromise therefore has to be made between these quantities and the over-all per- missible weight of the finished unit. To illustrate the approximate rela- tions of weight, size and sensitivity, units have been constructed in the shape of miniature tubes weighing only 33 per cent more than their prototypes, with sensitivities in the order of 0.005V rms/g and a useful frequency range of 3,000 cycles. Since these accelerometers are calibrated for rectilinear motion, the results obtained in measuring equipment vibrations or testing machines must be carefully interpreted. In many instances the disturbing forces impart a rocking motion to the units so that the position of the active elements in the housings will influence the acceleration magnitudes that are registered. Associated Instrumentation Since the voltage output of self generating accelerometers is small, electronic amplifiers have to be employed to bring the signal to desired levels. Fig. 7 illustrates a typical arrangement of the necessary equip- ment. A cathode ray oscilloscope is shown as the visual indicating means, although other recording instruments can be employed, depending on the nature of the disturbances to be measured and the type of record desired. The prime requirements of the equipment are: (a) phase distortion must be low, so that the disturbance pattern is correctly presented. (b) its transient and frequency response must be adequate for the frequencies to be recorded. In the circuit shown in Fig. 7, the signal is fed into a cathode follower stage having a high input impedance in order to obtain good sensitivity and frequency response. The use of a cathode follower also offers more flexibility in the proper matching of the high impedance pick-ups to suit- able low pass filters or to the following amplifying stage. The generated 1214 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 signal is shunted by the capacity of the connecting cable. Since the volt- age impressed on the grid of the first stage is approximately ^2 = V^ Ci Ci + C2 ■ it is desirable to have Ci large compared to C2. In some types of investi- gations cable whip may result in the introduction of spurious voltages on the signal. These voltages are produced by capacitance changes and static charges on the cable dielectric. To minimize this effect, the cable may be shunted by a padding condenser at the expense of the voltage F2. The National Bureau of Standards recently reported the development of a low noise cable'^ that is reported to overcome the shortcomings of the ordinary shielded cables. Cables with a sufficiently low noise figure have also become available commercially. A timing and calibration voltage can be impressed across the low valued resistor R for comparison with the disturbance signal. If resonant frequency signals of the accelerometers are excited by the dis- turbances being investigated, or the recorded frequency range is to be limited, the signal is fed through low pass filters before being amplified. This prevents possible overloading of the amplifier by the unwanted frequencies, so that maximum gain can be realized for the desired signal. Fig. 7 — Schematic of accelerometer and associated recording circuit. A — Equivalent circuit of acceler- ometer B — Cathode follower stage LP — Low pass filters Ci — Internal capacity of acceler- ometer C2 — Cable capacity C| — Padding condenser D — Voltage amplifier E — Recording instrument F — Audio oscillator 0 — Voltmeter R — Calibrating resistor Vi — • Voltage source Vi — Voltage impressed on first stage ACCELERATION EFFECTS ON ELECTRON TUBES 1215 Particular attention must be paid to the design of the filters since these elements are apt to introduce excessive phase shift and oscillations. A variable frequency electronic filter has been found to be quite serviceable due to its flexibility in the choice of cut-off frequencies. Additional information on shock and vibration instrumentation is given in Reference 5. Since the presentation of mechanical distrubances is, to some extent, limited by the transducers and their associated cir- cuits, there is a trend towards a certain amount of standardization of these components so that results can be compared on an industry wide basis. TUBE DESIGN PROBLEMS IMPOSED BY ENVIRONMENTAL CONDITIONS It follows from the numerous acceleration measurements made at the installation points of equipment, that electron tubes, together with other components, have to withstand a large variety of conditions. Without over-simplifying the problems involved, it is perhaps permissible to divide these disturbances into two classes : ballistic shock, and transi- tory or sustained low g vibrations. Although, as a first approach, the effect of these distrubances on tube elements may be probed by mathe- matical analysis, the final design must be proven in by laboratory tests under controlled conditions. In discussing the influence of shock and vi- bration on tubes, their elements are frequently presented schematically as cantilevers, Fig. 8(a). While this assumption is a close approximation for some of the older tubes, such as the Western Electric No. 349B tube, Fig. 8(b), most tubes of later design. Fig. 8(c), do not lend themselves to such simple analysis due to their more complex structure. The response of elements to even simple shock pulses on tube envelopes are influenced by factors such as the clearances in micas, mica fits in the bulbs and tight- ness of mount assemblies. It is obviously beyond the scope of this paper to analyze the destruc- tive effects of shocks and vibrations on equipment. A few of the many excellent articles and publications on this subject are listed in Reference 6. It is equally impossible to present all of the many problems facing the electron tube engineer in designing tubes that will reliably serve their purpose under adverse conditions. Since tubes must be designed to with- stand distrubances encountered in the field or be adequately protected, the following notes, highlighting some of these problems, will be of in- terest to equipment as well as tube engineers. Influence of High g Shocks In certain applications, tubes are required to withstand occasional high shocks such as those produced in military applications by explosions 1216 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 of some kind. It is generally permissible that, during very severe shocks, electronic equipment is non-operative; therefore, the response of tubes during these disturbances may fall outside of assigned limits. However, since it is highly important that operations resume immediately, the tubes must show no permanent change nor have caused damage to the circuits as a result of temporary faulty operation. Although protective shock mounts are usually employed on equipments exposed to these conditions, damped equipment vibrations excited by the attenuated shock wave are superimposed on the pulse felt by the tube. Brittle or stiff tube components are most susceptible to shock because of their inability to absorb the shock energy by elastic deformations. Failures falling into this category are: (a) glass breakage, which may be brought about by impact due to excessive movement of the tube or adjacent components during the impact. (b) metal to glass seal fractures, produced by shock loads on the tube leads and seals. (c) heater or filament failures and opening of welds. =3r (a) (b) (c) Fig. 8 — Electron tube structures: (a) Simplified anode and cathode structure of a Western Electric No. 349B tube, (b) Western Electric No. 349B tube, and (c) typical miniature tube. ACCELERATION EFFECTS ON ELECTRON TUBES 1217 (d) damage to mica serrations and enlarging of the mica holes which support and space tube elements. The deterioration of the mica is also known to liberate gas, which, in turn, results in a reduction of the vacuum. Permanent damage is also caused by deformation of elements beyond their elastic limit. Bowing of grids and cathodes as a result of shock has been reported in some applications. Although not always realized by design engineers, shocks and vibra- tions of surprising magnitudes may also occur in handling and shipping. If these conditions are not taken into consideration during the design stage, the over-all cost of the product may be adversely affected by the necessary protection that has to be built into the shipping container to assure safe arrival of the packaged article at its destination. While many factors enter into the proper design of shipping containers, such as mois- ture and corrosion protection, the selection of cushioning materials is perhaps the most important. Since it is desirable from a storage and shipping cost standpoint to keep the package bulk to a minimum, and protective packaging cannot always compensate for design weakness, adequate strength must be designed into the tube even though it will not be subjected to severe shocks once it is installed. A rather complete analysis of the dynamics of package cushioning is given in Reference 7. At present, military requirements specify that packaged tubes must safely withstand several three foot drops onto a hard surface. Influence of Low g Disturbances In contrast to the relatively infrequently occurring high peak shock and vibrations, we find that many equipments are often subjected to repetitive shocks and sustained vibrations at lower acceleration levels. These conditions are generally encountered in vehicle, ship and airplane applications. Although these shocks and steady state vibrations can be attenuated through the use of shock and vibration mounts, the effec- tiveness of these mounts may be reduced sharply by a change in disturb- ance frequencies from normal. Tube failures resulting from these conditions are generally caused by fatigue of some tube elements. For instance, the continual hammering of micas against tube walls or chattering of cathodes and grids in the mica, may reduce the value of the micas as supporting and spacing ele- ments, and since tubes are required to function under these conditions, the gradual degradation of the micas will bring about an increase in the tubes' microphonic output. Where microphonism is a factor, the useful 1218 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 life of a tube is, therefore, terminated long before more apparent failures occur in the tube structure. Other points of weakness are heater and filament leads. As may be deduced from the above, microphonism is a frequent cause of complaint when tubes are subjected to low but repetitive shocks or transient vibrations. To illustrate the influence of such disturbances on tube performances, the results of recent investigations of tube micro- phonism found in a certain equipment will be cited. In this case, field reports indicated that certain tubes exhibited exces- sive microphonism in the equipment, although the tubes were found to be within limits when judged by standard factory tests. It was apparent, therefore, that these tests did not simulate actual conditions. Accelera- tion measurements made at the equipment base and at tube sockets revealed that the steep, short duration impact of a blow delivered to the outer case excited resonant vibrations of the chassis on which the tubes were mounted. The magnitudes of these vibrations were only in the order of O.lg, but their lowest frequency (approximately 550 cps) was close to the mount resonances of some of the tubes. Further tests also showed that those tubes having pronounced response, i.e., low damp- ing, to vibratory motion in this frequency range also proved to be micro- phonic in the equipment. (The vibration spectrum of one of the tubes is reproduced in Fig. 9.) It was found that the various modes of vibration of the tube mount produced the high peaks in the range between 500 and 1000 cps. Unfortunately, present factory tests do not include a complete evaluation of tube response over a wide enough frequency spectrum. Observations made on several equipments indicate that structural changes in the chassis or re-positioning of tubes would, in some instances, reduce the effect of mechanical disturbances on tubes. An occasional source of trouble is introduced by equipment motor vibrations, especially after mechanical wear has increased play in the moving parts. The ac- celerations involved in these vibrations are usually very small, but if their frequencies coincide with structural resonances in the tubes, unsatis- factory operation of the equipment may result. The influence of such disturbances on tube operation is not always recognized by equipment designers. Several programs are currently pursued by both military and com- mercial agencies to increase tube reliability. Since the necessary requi- sites that make a tube reliable depend on the type of service to be per- formed and environmental conditions, the requirements stressed in the various programs differ in many respects. Some of these requirements are still in a state of flux, as actual needs are as yet not clearly known. ACCELERATION EFFECTS ON ELECTRON TUBES 1219 20 5 8 t VIBRATION FREQUENCY IN CYCLES PER SECOND Fig. 9 — Frequency response of a tube that exhibited microphonism in equip- ment. The direction of vibration is perpendicular to the major diameter of grids and tube axis. SHOCK AND VIBRATION REQUIREMENTS AND TEST EQUIPMENT General requirements Certain tests have been written into the MIL-E-IB^ and other speci- fications to control the shock and vibration characteristics of tubes, and to check on their behaviour under given mechanical excitation. Long usage has given these specifications some validity in that they control the quality of the product even though actual field conditions may not necessarily be simulated in the tests. At present, one or more of the following three types of tests are called for on tube specifications: (a) Shock tests. These are high acceleration tests to insure that tube structures will withstand occasional shocks of given maximum magni- tudes. Since, in general, tubes are not required to function during high peak shocks, no operating voltages are applied to the tubes for this test. The post shock requirements are that the tube characteristics must not have drifted out of their limits. In many cases, the type of shock tester to be used is specified, because it is difficult to define the shock output of 1220 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 many of these machines in terms of acceleration and frequency content, etc. The response of tube structures may be vastly influenced by shocks of the same nominal acceleration but differing acceleration wave forms. Shock tests are also performed on tubes during their initial development to obtain the degree of cushioning required for safe shipping and handling purposes. (h) Vibration tests. These are low acceleration, fixed frequency tests. They are made on machines with sinusoidal displacement output. Short time duration 25-cycle — 2.5g, or 50-cycle — lOg tests are specified for tubes that are generally not exposed to constant vibrations. Usually no voltages are applied to the tubes under test since their prime purpose is to check for sound tube structures. Electrical post vibration perform- ances are the criteria of tube quality. In this category also fall the long time duration vibration tests made on tubes intended for use in equip- ment that is kno^vn to be subjected to continual vibration, such as shipboard or mobile equipment. Present specifications call for 96-hour tests at 2.5g — 25 cycles. This, therefore, is a fatigue test which deter- mines the capability of tube structures, under prolonged cyclic stresses. (c) Microphonic vibration test. A 25-cycle — ^ 2.5 g test performed with specified voltages on the tubes under test to investigate the influence of low acceleration vibration on the output of tubes. This test is specified on certain tubes, especially those used for audio applications. The per- missible magnitude of spurious signals excited by vibration is limited on the respective tube specifications. (d) Tap tests. These tests are performed for two purposes. One is for the detection of defects such as foreign particles between close spaced adjacent elements, damaged elements, or poor welds. Open and short testers of given sensitivity are used to indicate these defects. The second purpose for tap testing is the investigation of microphonic response of tubes to mechanical disturbances. For these tests the tubes are made to work in Class A amplifier circuits. Acoustic feedback may be used, so that the tube is not only subjected to the mechanical tap, but also to the sound of the tap excited microphonic signal which is reproduced through a loud speaker spaced at a given distance from the tube. Sus- tained microphonism can be produced by these means under certain conditions. The purpose of this type of test is to simulate conditions to which tubes are subjected in some equipments, especially those closely coupled to audio output components. Equipment The following is a brief description of the machines used for the per- formance of the above tests and their output characteristics. ACCELERATION EFFECTS ON ELECTRON TUBES 1221 MIL-E-IB Bump Tester This is one of the earliest devices employed for shock testing of tubes under controlled conditions. In order to assure uniformity of results the MIL specifications give its physical dimensions. Fig. 10 illustrates the tester and its method of use. The magnitude of the shock and its duration is given by tube weight, shape of tube envelope contacted by the ham- mer, resilience of rubber pad on the hammer, and the angle (6) through which the hammer is permitted to swing before striking the tube. Although for the performance of the tests, only the angle (6) is speci- fied, the shock characteristics of this device have been investigated,^*^ so that shock magnitudes and durations for any tube may be computed from the parameters given above. A typical acceleration time curve is sho^vn in Fig. 11. The simple bell shaped outline of the accelerogram is given by the non-linear spring characteristic of the rubber pad and the generally cylindrical shape of the tube envelope. Shock Testing Mechani&m per ASA-C39.3 This mechanism is also used to check and compare the resistance of tubes to mechanical shocks of predetermined magnitude and duration. In this device the sample to be tested is rigidly fastened on a platform. A steel leaf spring supported at both ends is attached underneath this platform. The test on the sample is performed by raising the platform to a certain height, allowing it to fall on a steel anvil, and then catching it on the rebound. The shock magnitude is given by G max = and its duration by 2hk W V 12K9 12Kg where W = tableweight (lbs.) h = height of fall (inches) K = spring constant of leaf spring. A number of leaf springs are available to produce the desired shock characteristics. A full discussion of this mechanism and its performance are covered in Reference 11. A slightly modified version of the tester, used by the Laboratories, together with a typical shock pulse, is sho^vn in Figs. 12 and 13. It can be seen that the pulse is essentially of sinus- oidal shape with higher frequencies superimposed on it. The fundamental frequency is produced by flexing of the leaf spring during its contact 1222 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 Fig. 10 — One of the early devices employed for shock testing of tubes under controlled conditions — the MIL-E-IB bump tester. with the table anvil while the higher frequencies are due to table reso- nances which are excited by the metal-to-metal impact of spring and anvil. High Impact Machine for Electronic Devices This machine was originally intended to test the resistance of tubes to high peak-short duration shocks similar in nature to those encountered by equipment fastened to parts of ships that are likely to be exposed to direct explosion pressures.^^ The shock is produced by a steel hammer pendulum striking a movable steel table on which the tube under test is mounted. The shock magnitude is given by the angle through which the ACCELERATION EFFECTS ON ELECTRON TUBES 1223 pendulum is allowed to swing under the action of gravity before the hammer strikes the table. The forward motion of the table produced by the impact is arrested by two shock absorbers. The impact of the hammer on the anvil of the table produces an abrupt velocity change of the table and excites table resonances, both horizontally and vertically. Because of the structure of the table a very complex acceleration wave form results. In general the accelerations for a given hammer swing vary over the table surface. It is for this reason that, in testing procedures, the position of the tubes on the table and their method of clamping are well defined; and since the acceleration wave shape is the sum of many vibratory frequencies rather than a single shock pulse, the severity of the test is expressed by the angle of hammer swing instead of a shock magnitude and duration. Although some uni- formity in performance is attained by rigid standardization of the struc- ture of the machines and the above mentioned positioning of the tubes, minute differences in these parameters may produce sufficient variations in shock output to reflect on test results. Acceleration-time traces of shocks measured in the shock direction by an accelerometer fastened near the anvil of the table and a second accelerometer clamped to the center of the table, are shown in Figs. 14 and 15. The records were simultane- ously taken through 10,000-cycle low pass filters. Even though the recording of accelerations containing high frequencies of large amplitudes is, to some extent, a function of the measuring equipment, it is seen that significant differences in output exist between the two points of measurements. Fig. 11 — Acceleration-time pulse produced by MIL-E-IB bump tester. 1224 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 Fig. 12 — Shock testing mechanism per ASA-C39.3. Fig. 13 — Typical shock pulse obtained with the tester shown in Fig. 12 (A) acceleration pulse, (B) timing trace, and (C) stress in leaf spring. ACCELERATION EFFECTS ON ELECTRON TUBES 1225 Fig. 14 — High-impact machine for testing electronic devices. Fig. 15 — Acceleration-time pulses produced by machine shown in Fig. 14. Upper trace recorded near anvil, lower trace recorded in center of table. 1226 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 This machine can also be used to produce lower G shocks of longer duration and simpler wave form by interposing a resilient rubber pad between the anvil and hammer. The resulting shock resembles the bell shaped output of the MIL-E-IB bump tester. Its magnitude and dura- tion are given by the hammer swing, resilient characteristics of the pad, and the table and specimen weight. The capacity of the machine permits shock testing of heavier tubes (such as the larger magnetrons) by this method. Attainable shock levels are sufficiently high to cover the require- ments placed on these tubes by conditions encountered during transit. The L.A.B. Package Tester Of particular interest to the packaging engineer, this machine is in- tended to duplicate the destructive vibrations and shocks experienced by a product during transit. It consists of a horizontal table that can be made to vibrate with a circular motion in the vertical plane. Adjust- ments permit variations of this motion to simulate freight car and motor truck movements. Tests performed on this machine, therefore, check on the mechanical strength of the outer shipping container as well as on the adequacy of cushioning materials employed to protect the product. The services are also considering this machine to test equipment designed for use in vehicles. Tests are now in progress by the Signal Corps to determine proper parameters for this application.^ Vibration Machines A number of vibration machines, made by various manufacturers, are employed for vibration testing of tubes. Due to the high hash output of most mechanically driven machines, which contain gears and linkages, great care must be taken in the selection of these machines. Certain tests, especially the microphonic tests on receiving tubes, require machines with good sinusoidal output, in order to obtain comparable results. It is for this reason that the leaf spring vibration machine (Fig. 16), developed several years ago by Bell Telephone Laboratories, has been recommended as a standard for performing vibration tests. This is a fixed frequency, 25-cycle — 2.5 g machine which, due to its construction, has a relatively clean output as shown in Fig. 17. Several types of electronically or motor-generator driven vibration machines are on the market. These are variable frequency and variable amplitude machines especially useful for determining resonance fre- quencies of structures and for performing cycling vibration tests. Ac- cessory equipment has been made available lately to conduct these tests on an automatic basis at either constant acceleration or constant am- ACCELERATION EFFECTS ON ELECTRON TUBES 1227 Fig. 16 — Leaf spring vibration machine developed several years ago by Bell Telephone Laboratories. Fig. 17 — Acceleration output of the leaf spring vibration machine. plitude over a given frequency range. Within the limitations of these machines and their power sources, it is also possible to subject specimens to complex disturbances. For instance, a projected use is the reproduction of recorded equipment vibrations so that the response of tubes to these conditions can be studied in the laboratory. Tube Tappers • The physical construction of tube tappers used in the electron tube industry varies from the simple manually operated cork mallets to the more complicated automatic tappers. Since it has long been recognized that the reproducibility of manual tapping is rather poor, efforts have been made to replace these devices by automatic tappers or other 1228 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 methods of checking microphonism in tubes. The industry, in collabora- tion with the Services, is now engaged in standardizing on automatic tappers. The prime requisites for such tappers are: (a) their shock output must be reproducible and must fall within definable limits, (b) their operation cycle must not adversely affect the time required for perform- ing tap tests, and (c) the results obtained must have some relation to the environmental requirements for the tubes. This last condition is, perhaps, the most difficult to attain considering the diverse conditions encoun- tered by tubes in various equipments. Of the many tappers devised and employed by the industry, the use of only two are at present approved in the MIL-E-IB specifications. One is a manually operated cork mallet. This mallet is still retained in spite of its shortcoming, for lack of better devices. The second is the General Electric Automatic Tapper, specified for checking microphonism of some of the reliable tubes. This is a motor driven tapper which sub- jects the tube under test to damped low "g" vibrations at the rate of 2 taps per second. The tapper and associated circuits are fully described in the MIL specifications. Space does not permit the discussion of other proposed tappers or test methods. Two rather interesting papers are listed in References 13 and 14, which illustrate the problems involved in the development of suitable devices for checking microphonism in tubes and describe some of the various methods of approach. SUMMARY The rapid growth of electronic equipment development during the late war has continued with the ever increasing new applications to both military and civilian purposes. It is realized that this growth has placed more stringent requirements on the mechanical characteristics of electron tubes. In order to define intelligently tube requirements, the nature of the disturbances to which tubes may be subjected must be known. It was pointed out that merely applying equipment requirements to the tubes, is not very realistic, since the shock and vibration patterns may be vastly modified by the equipment structures. With the help of the recent development of light weight accelerometers, it is now possible to investigate the disturbances at the tube sockets. Both the Industry and the Services have begun to utilize these instru- ments to collect this information. The benefits derived from this work will enable the equipment and tube designers to formulate more accurate requirements and to devise test gear that will simulate more closely field conditions to check on tube quality. ACCELERATION AND ITS INFLUENCE ON ELECTRON TUBES 1229 Although, as a result of the trend to miniaturization, the strength of electron tubes has been increased, due to the smaller size and mass of elements employed, much has still to be done to increase tube resistance to ballistic shock. This would result in simplification of shipping con- tainers and reduce the need for protection by shock absorbers in equip- ment. Equally important is the effort now made to reduce the micro- phonic response of the tubes. Here, too, the smaller size of the late tubes is of advantage because element resonant frequencies are increased. And since the higher frequency components of disturbances are largely attenuated by equipment structures, the tube responses have been lessened. With a better understanding of the problems involved in tube pro- tection, it is also quite possible that further improvements can be ef- fected in some instances by structural changes of equipment members, or re-orientation of tubes in the equipment to reduce the effects of shock and vibrations on tubes. BIBLIOGRAPHY 1. Shock and Vibration Instrumentation, NRL report #0-2645. 2. Mechanical vibrations, Den Hartog, McGraw-Hill Book Co. 3. The reciprocity calibration of piezo-electric accelerometers, M. Harrison, A. O. Sykes, and P. G. Marcotte, The David Taylor Model Basin, Report 811. 4. A Noise Free Instrument Cable, National Bureau of Standards, Technical News Bulletin, 36, No. 3. 5. Research and Development Toward Construction of Accelerometers, Gulton Mfg. Co. report. New Techniques for Measuring Forces and Wear, W. P. Mason, B.S.T.J., 31, May, 1952. Proceedings of symposium on Barium Ti- tanate Acelerometers, National Bureau of Standards, Report No. 2654, Aug. 1953. 6. Bureau of Ships Publications, Nav. Ships 900,036, and Nav. Ships 900,052. The David W. Taylor Model Basin, Report 481. Vibration and Shock Isola- tion, C. E. Crede, John Wiley & Sons. Response of Damped Elastic Systems to Transient Disturbances, R. D. Mindlin, F. W. Stubner, H. L. Cooper, Experimental Stress Analysis, 5, No. 2. Measured Aircraft Vibrations as a Guide to Laboratory Testing, D. C. Kennard, Jr., A. F. Technical Report No. 6429. 7. Dynamics of Package Cushioning, R. D Mindlin, B.S.T.J., 24, July — Oct., 1945. 8. Evaluation of Electron Tube Quality on the Tethered Weapons Carrier, Signal Corps Engineering Laboratories, Army Project No. 0302 A. 9. MIL-E-IB, Military Specifications for Electron Tubes. 10. Characteristics of the JAN Bump Tester, R. D. Mindlin, F. W. Stubner and H. A. Lefkowitz, MM-45-2920-4. 11. Shock Testing Mechanism for Electrical Indicating Instruments, ASA-C 39.3 12. Mechanical Shock Characteristics of the High Impact Machines for Electronic Devices, I. Vigness, R. C. Novak and E. W. Kammer, NRL Report No. 0-2838. 13. Proposal for Revised Noise and Microphonic Evaluation for MIL-E-IB Speci- fications, NML Report, Dated July 30, 1951. 14. A Study of Microphonism, Philco Research Report No. 183. Arcing of Electrical Contacts in Telephone Switching Circuits Part I — Theory of the Initiation of the Short Arc By M. M. ATALLA (Manuscript received April 1, 1953) This is a presentation of a theory for the mechanism of the initiation of the short arc commonly observed on the closure and opening of electrical contacts. The theory is based on the experimental evidence that an established arc is generally preceded by a period of local high frequency discharges at the contacts. During this period the circuit current builds up. If and when this current reaches the arc initiation current of the contact a steady arc is es- tablished. It is shown that this initiation period of the arc is directly deter- minable from the circuit conditions and the contact condition. This mecha- nism furnishes rather simple explanations to some complex phenomena commonly observed before the establishment of the arc. The mechanism of the initiation of an individual discharge, however, still remains uncertain. INTRODUCTION In the course of study of arcing phenomena between electrical con- tacts, it has been long established that a condition for sustaining the short arc is to maintain a current through the arc greater than a mini- mum value called the minimum arcing current. This current is generally a characteristic of the contact material and is appreciably affected by surface contaminations. For clean metals the minimum arcing current is usually equal to a few tenths of an ampere. Before establishing the arc, therefore, there must exist a certain mechanism which accounts for a rapid current build-up from zero to a value as high as the mini- mum arcing current. For an inductive circuit, a higher inductance should result in a longer period of current build-up. With a sufficiently high circuit inductance this initiation period may be made long enough to be directly observed and to allow an examination of the mechanism in- volved. Such experiments have been made and observations have in- dicated that the initiation period consisted of a succession of rapid dis- 1231 1232 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 charges at the contacts from the open circuit voltage to a lower voltage. During this process the current in the circuit built up in a discontinuous fashion and the steady arc was established only when the circuit current reached the minimum arcing current of the contact. In this paper is presented our study of this initiation period. The following are the main objectives: (1) to establish the analytic relations governing the performance of a few simple contact circuits during the arc initiation period; (2) to check the analysis by direct measurements; (3) to apply the theory to explain a few arcing phenomena and empirical relations previously reported; and (4) to shed some light on the nature of the rapid local discharges at the contacts and the characteristics of the influential circuitry at the immediate neighborhood of the contact. NOTATION C Main circuit capacitance I Circuit current li Arc initiation current Im Arc termination current or minimum arcing current L Circuit inductance (L) Limit Limiting or maximum inductance above which a steady arc cannot be estabUshed R Circuit resistance V Voltage Vo Initial voltage VcT Main condenser terminal voltage c Local capacitance at the contacts t Local inductance at the contacts n Number of discharges at the contacts r Local resistance at the contacts t Time V Voltage across a steady short arc V Voltage across the contacts at the termination of a single local discharge z Local impedance at the contacts - a Ratio of capacitances c/C « Angular frequency (Lc)"^^'' ANALYSIS In this section a few simple contact circuits are considered. In each case relations are derived for the current and voltage changes in the 1/2 ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1233 circuit during the period of rapid discharges at the contacts preceding the steady arc. The analysis is based on the following simplified model of the mechanism involved: (1) The first local discharge at the contact takes place when the proper separation corresponding to the initial voltage Vo is reached; (2) this discharge time is assumed to be short and negligible in comparison to the following charging time; (3) the local capacitances at the contact recharge from the main circuit until the same initial voltage T^o is reached when a second discharge takes place; (4) this process repeats until a steady arc is established provided that the circuit is capable of building up enough current, — otherwise, the local discharges will continue and finally stop when the main circuit be- comes incapable of charging the local contact capacitances to Vo; and (5) all the local discharges at the contact are terminated at a constant voltage V for any one set of circuit conditions. The nature of v is left to be determined and physically understood from our measurements. — nm^ CONTACTS ::=p C -iiT Vq Fig. 1 — Typical battery-inductance-contacts circuit. Battery Vo, L and Contacts, Fig. 1 Follomng the first discharge at the contact from Vq to v the local contact capacitances will recharge with a current /c V I = {Vo — v) ij j sin 03i, where co = {Lc)"^'^. The voltage at the contact mil reach Fo at ^i = 7r/2a) and the cor- responding current is /. = (n - -.) (if. A second discharge will then take place and recharging wdll proceed with the new boundary conditions: at t = o, the contact voltage is v and the circuit current is h. By following this procedure, a general ex- pression for the nth charging process is obtained: I{n) = (Vo - v) (0''' inf' (a) 1234 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 and (1) (0 . i(n) = cot"^ (n - 1)^^' (b) The Felation between the number of discharges n and time t is given by t =lZ cot"^ in - lf'\ 03 n=l Only an empirical expression for this summation was obtained with a maximum error less than 10 per cent: The current build-up in the circuit is, therefore, expressed as a function of time by the following relation: m =^{Vo- v)t (2) In other words, the current is independent of the contact capacitances and increases hnearly with time at a rate inversely proportional to the circuit inductance. t As soon as the circuit current reaches the arc initiation current /jj, a steady arc is established. The initiation time of the steady arc is therefore given by: U = I- ^. (3) 2 Vo — V For /, = 0.5 ampere, L = 10"^ henry and Fo - i; = 30 volts, U = 2.6 X 10"* second. C, L and Contact, Fig. 2 If the battery is replaced by a condenser C where the ratio a == c/C is much less than 1.0 an analysis similar to the above can be made. In ^ • The assumption that all discharges will take place from the same voltage Vq w only true if: the motion of the contact is negligibly small, the discharges do not cbiinKe the contact geometry to the extent of materially changing the contact separation, and if the effect of the residual ions is negligible. t This is onlv true if the circuit resistance is zero. For a finite circuit resistance H, and ^ Wj much less than 1 ,0, it can be shown that the current approaches the asymptotic value " — . 1 It is shown later by measurement, that the initiation current /» is essentially the same as the arc terminating current I^. ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1235 this case, however, in setting the boundary conditions one must consider the drop in voltage across the main condenser during the previous charging processes. The following are the resulting expressions for the circuit current, main condenser voltage and the charging time, all as functions of n: Kn) = -^^— . I n(l - a(n - 1 + an))\ (4a) V{n) = Fo - oLn{V, - v) (4b) „. tin) = sin- 4, ; T + tan-' T-l^^^^^l'" (4c) Ll - a{n - 1)J L(l + oi)(n - 1)J Only an empirical expression for the summation Z tin) was obtained with an error less than 10 per cent i T o>-t(n) = Unf'\l + an). n=l ^ The current relation indicates that the current increases from zero at n = 0 to a maximum current 1/2 V'(0 Sit an = }/2 then drops back to zero at na = 1.0 when the discharges are terminated. The total discharge time is approximately ^{LCf^ and the terminal voltage on the main condenser is Vct = v. If during the process of current build-up the current reaches a value equal to the arc initiation current a steady arc is established. It is evident that a steady arc cannot be established if the maximum current attainable during the discharges is less than the arc initiation current. This leads to the concepts of a limiting inductance and Hmiting voltage in a circuit that can allow the establishment of a steady arc.^ The limiting inductance is CONTACTS 1 i ^W^ C,Vo Fig. 2 — Typical condenser-inductance -contacts circuit. ^ L. H. Germer, Arcing at Electrical Contacts on Closure, Part I, J. Appl. Phys. 22, p. 955, 1951. 1236 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER, 1953 defined as the largest inductance for a given circuit above which a steady arc cannot be obtained and the limiting voltage is defined as the lowest voltage for a given circuit below which a steady arc cannot be obtained: (L)u»it=J-(^'J (6a) /t\1/2 (Fo-^)Limit=27,-(^y (5b) The assumption made that all charging processes are much longer in time than the discharging times, imposes a Hmitation on the appUca- bility of the above relations. Equation 4c can show that the minimum charging time is (2/o))a'^. The accuracy of the above relations is better the larger this time is compared to the discharge time of the local contact capacitance. Assuming distributed characteristics for the local and rela- tively small contact circuitry* the discharge time is 2{£c)^'^. The limitation involved, therefore, is that (2/o})a'^ must be greater than 2{icf^'^ or L/C > lie. In other words, the impedance of the main circuit must be greater than the impedance of the local circuit at the contact. MEASUREMENTS All the measurements presented were obtained from an inductive type circuit, C-L-Contact. The main circuit, Fig. 3, consisted of a con- denser C in series with a honeycomb inductance which is connected by a short lead, about 2 cms, to a pair of clean palladium contacts operating in laboratory air. The contacts were mounted on a cantilever bar arrange- ment, described by Pearson^ which allows fine adjustments of the sepa- ration between the contacts as well as slow motion of the contacts to avoid physical closure before the end of a transient. Three sets of measurements were made. (1) Contacts voltage measurements: the transients obtained usually needed some correc;tion to compensate for the effects of the measuring OBcilloscope circuit. None of these measurements are presented in this paper. It may be mentioned, however, that they indicated the existence of rapid discharges at the contacts preceding the estabUshment of the steady arc. (2) Circuit current measurements. Fig. 3 (a): these were made by 1 * ^.'L M. J***"®! »J-pti()ri of this paper, measurements were shown to indicate the plaiwibihty of this aaaumption. • G. L. Pearson, Phye. Rev. 66, p. 471, 1939. ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1237 measuring the voltage change across a 10-ohm non-inductive resistor inserted between the main circuit condenser and ground. (3) Main condenser voltage measurements, Fig. 3(b): the oscilloscope plates were shunted by an 1100 X 10"^^ farad capacitor and the com- bination was used as the main circuit condenser. The above sets of measurements 1 and 2 furnished the data necessary for a quantitative estabhshment of the theory. It is evident that the measuring scope cir- cuits did not interfere with the normal behavior of the contact circuit. Development of Circuit Current During Arc Initiation Period The circuit in Fig. 3(a) was used. The contact separation was gradually decreased until the discharges started and the resulting current transients were recorded. The circuit parameters were chosen such that, according to Equation 3 of our analytical results, the initiation time was of the order of microseconds. Fig. 4(A) shows a typical current-time transient where a steady arc was established. The circuit had L = 1100 X 10~^ henry and C — 10"^ farad. The current started from zero and increased during the multiple discharge period until point 1 where the current was 0.2 ampere and a sustained arc was established. The current then increased to a maximum of 1.7 amperes then dropped. At point 3 the arc stopped when the current was 0.24 ampere. In Fig. 4(B) the first L 1 MEG AAA — ' 000 ^"— c^ L CONTACTS > loooii looon -AAA/ — \AAr-i Vo POWER SUPPLY lon^ X OSCILLOSCOPE J^ (a) L 1 MEG — ' 000 ^ — lOOil AAA Vo CONTACTS -.^ POWER SUPPLY c^ -- / J - OSCILLOSCOPE ~~ (b) Fig 3 _ (a) Contacts circuit and circuit current measuring circuit, (b) Con- tacts circuit and main condenser voltage measuring circuit. 1238 THE BELL SYSTEM TECHNICAL JOURNAL SEPTEMBER 1953 TIME, T, IN SECONDS X 10-6 Fig. 4 — Circuit current transients with steady arc established. portion of a similar transient is shown: L = 5100 X 10"^ henry and C = 10"^ farad. The current build-up during the multiple discharge period is shown as an interrupted line 0-1. The steady arc was established at 0.57 ampere. From a group of similar transients, pairs of measurements were made of the initiating current and terminating current of the steady arc. The results are given in Table I. It is concluded from the results that the initialing current and terminating current of an arc are the same. The high-frequency discharges at the contact do not necessarily have to be followed by a sustained arc. A sustained arc is only obtained if the maximum current established during the local discharges at the contact is equal to or greater than the arc initiation current. Fig. 5 (A) shows a cur- rent-time transient without an initiation of the steady arc: Vo = 400 volte, L = 5100 X 10"' henry and C = 1100 X 10"'' farad. The maxi- mum current reached was only 0.13 ampere which was not sufficient for initiating a steady arc. It is of interest to notice that the oscillations superimposed on the zero current line following the transient can allow a calculation of the local capacitances at the contact. Such a calculation gave c = 7.8 X 10""^ farad. Fig. 5 (B) shows a similar transient for the same circuit with Vo = 500 volts. Following the first current build-up and drop, 1-2-3, the main condenser had a residual negative voltage high enough to produce a reversed current build-up and drop, 3-4-5. Voltage Drop in Main Condenser During Arc Initiation Period During the peri(Kl of rapid discharges the current build-up in the circuit is accompanied by a voltage drop at the main condenser. This drop may corre^ipond to one of two phenomena at the contact: (1) multiple discharges leading to a steady arc, and (2) multiple discharges without steady arc, followed by an open circuit. Fig. 6 and 7 are re- ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1239 Table I — Arc Initiation and Arc Termination Currents Palladium Contacts in Air* Arc Initiation Cur- rent 1%: Amps. Arc Termination Current Im'. Amps. 0.12 0.21 0.21 0.150. 250. 19 0.230. 400. 65 0.61 '0.57 0.520. 45 0.13 0.16 0.17 0.20 0.220.24 0.24 0.420.500.520.53 0.58 0.59 * Both numbers given in one column were obtained from the same transient. spective records of the voltage change at the main condenser. Fig. 6 (A) corresponds to the case where the multiple discharge period was short and lead to a steady arc. During this arc the main condenser voltage dropped from point 1 to point 2 when the arc was arrested. This was fol- lowed by striking a second arc in the opposite direction which lasted until point 3. Line 3-4 represented the recharging of the main condenser from the power supply circuit. Superimposed on the same figure is the trace 1-2-3-5 corresponding to a closure at the contact instead of an open circuit. While line 1-2-3 shows two consecutive arcs, it is generally possible to obtain any number of such arcs. An even number of arcs will result in a positive residual voltage at the main condenser, Fig. 6 (A) while an odd number of arcs will result in a negative residual voltage at the main condenser. Fig. 6 (B). Figs. 7 (A) and 7 (B) correspond to the case when the multiple dis- charges did not lead to a steady arc due to insufficient current build-up. The multiple discharges caused a voltage drop 1-2 across the main condenser followed by an open circuit and recharging, 2-3. Discharge of the Local Circuitry at the Contact Fig. 8 (a) represents a plausible representation of the local circuitry at the contact. When the conditions between the contacts are favorable 0 7.4 0 7.4 TIME, T, IN SECONDS X 10"® Fig. 5 — Circuit current transients without a steady arc. 1240 THE BELL SYSTEM TECHNICAL JOURNAL SEPTEMBER 1953 for the initiation of the arc, a small local capacitance c' will furnish the necessary charge, through a small impedance z' ^ according to whatever mechanism that may be involved in the initiation process.* The con- nection from the contact to the main circuit is represented by a short transmission line \vith the distributed characteristics, r, i and c. The drop at the contact from the initial voltage Fo to the arc voltage v will cause a current surge (To — "o) {c/tf^, if r is neglected, for a period 2{(cf'^ corresponding to the time required by the pulse to travel to the end of the line and return to the contact. At this time the arc is ex- tinguished by the reflected pulse and the final voltage at the contact is — (Fo — 2v). Figs. 8(b) and 8(c) are diagramtic representations of the process. For a purely inductive line, therefore, the contact voltage V following one discharge is — (Fo — 2v). For a dissipative line, however, V is algebraically greater. Equation 4 of Germer and Haworth^ was derived to give the voltage following an arc for a similar circuit with lumped characteristics. For r/{t/cf''^ less than 1.0 this equation can be 220 220 10 0 950 TIME, T, IN SECONDS XIO'^ Fig. 6 — Voltage transients across main condenser with steady arc established. 220 70 0 TIME,T, IN SECONDS X 10"^ JPig^7 — Voltage transients across main condenser without steady arc. This mechanism is not clearly understood at the present time. It is the opinion of the writer, however, that the initiation time is only a few times the transit time \r ^T^^ *°" ^^^^^ *^® prevailing conditions. r'r^r^^^^ *"^ ^- ^ Haworth, Erosion of Appl. Phys. 20. p. 1085, 1948. Electrical Contacts on Make, ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1241 approximated by: 2y Fo + ^ • - (7o - t;) (6) where z = (//c)'''. Our measurements were then applied to demonstrate the plausibility of the above description of the local circuitry at the contact. This was done in the foUomng fashion, v was obtained by three methods. (1) Measurements were made of the multiple discharges preceding a steady arc, from records similar to Fig. 4(B), and v was calculated from Equa- tion 3, (2) measurements were made of the maximum current attainable during the multiple discharge period, from records similar to Fig. 5(A), and V was calculated from the expression /max = H^O (a) CONTACTS Jr •'© DISTRIBUTED r,l,c f _r f MAIN CIRCUIT (b) Vo^' ^^ U TIME (c) 2(10)^2 TIME-^^ Fig^ 8 — (a) Representation of local circuitry at contacts, (b) Voltage transi- ent at contacts during one discharge of local circuitry, (c) Current transient at contacts during one discharge of local circuitry. 1242 THE BELL SYSTEM TECHNICAL JOURNAL SEPTEMBER 1953 Table II — Ratio rlz for Local Circuitry at Contacts Initial Voltage r z 100 0.56 150 0.65 200 0.55 230 0.41 250 0.55 300 0.30 360 0.40 400 0.40 460 0.48 and (3) measurements were made of the terminal voltage across the main condenser at the end of the multiple discharges, from records similar to Fig. 7, which, according to our analysis, is equal to v. For these values of v Equation 6 was used to compute r/z. The results are given in Table II. Each value of r/z in this table is the average of 3 to 6 values obtained by the different methods described above. In all cases rjz was between 0.4 and 0.7. This indicates that the local contact cir- cuitry is oscillatory. For our circuit, c was measured at 7.8 X 10~^^ farad and if ^ is assumed to ke 2 X 10~^ henry, the computed resistance r is 20 to 35 ohms. This is about 500 times greater than the dc resistance of the same circuit. With this concept of the local contact circuitry v was well defined and was then possible to perform some checks of the main analytical relations presented in this paper with new measure- ments and with measurements previously published. Comparison of Theory with Measurements: (1) Voltage drop across main condenser: In Fig. 7(B) is shown the voltage drop across the main circuit condenser as a function of time for L = 10"' henry, C = 1100 X 10"'' farad and Vo = 220 volts. Line (a) was measured and Line (b) was computed using Equation 4(b). The value of V used was obtained from Equation 6. Good agreement is in- dicated. (2) Limiting circuit conditions for obtaining a steady arc: It was pointed out in a previous section of this paper that a steady arc can be established only if the maximum current reached during the multiple discharge period is equal to or greater than the arc initiation current. Equations 5(a) and 5(b) are expressions for the limiting circuit induc- tance and the limiting initial voltage respectively. Equation 5(b) was used to compute the hmiting voltage for a set of circuit conditions for which measurements were made and published. Reference 3. In Table IV Column 3 of this reference measured values of the limiting voltage Vo were presented. A comparison of measured and computed Vo are given here as Table III. It may be pointed out that the deviations be- tween measurements and calculations are of the order of the measured ARCING OF CONTACTS IN TELEPHONE SWITCHING CIRCUITS 1243 spread in the minimum arcing current as given in the same reference, Table V. (3) Contact activation and hmiting circuit conditions for arcing: In Reference 3, it was reported that the Hmiting inductance for active contacts was greater by more than 2 orders of magnitude than for clean contacts. By calculation, Equation 5(a), [(^)active/(i')cleanlLimit = 600. It was also reported that for C = 10~^ farad and Fo = 50 volts the limiting inductance observed for active silver was between 10"^ and 10"^ henry. By calculation. Equation 5(a), the limiting inductance for the same conditions is about 4 X 10~^ henry. (4) Contact activation and arc initiation time: According to Equation 3 the initiation time of the steady arc is directly proportional to the arc initiation current. For the same circuit conditions, therefore, active contacts should have a shorter period of arc initiation. This result seems to contradict some published observations* where it was pointed out that the voltage drop into an arc was shorter for clean contacts than for activated contacts. A number of transients, furnished by the authors, were carefully examined. It was observed that in most cases Table III — Computed and Measured Limiting Voltages for Establishing a Steady Arc Between Clean Silver Contacts* Observed (Fo) Limit for first detectable arc Computed (Fo) timit 50 49 66 102 220 211 38 38 85 75 206 153 31 28 64 52 200 102 24.5 23 60 40 30 20 13 16 11.5 14 13 13 25 17 * A few observations were not included in this table. These correspond to the cases where the calculated initiation times of the arc were of the order or greater than the time of physical closure of the contacts. As expected, the observed volt- ages were consistently higher than the computed voltages. ^ L. H. Germer and J. L. Smith, Arcing at Electrical Contacts on Closure, Part, III. J. Appl. Phys., 23, p. 553, 1952. 1244 THE BELL SYSTEM TECHNICAL JOURNAL SEPTEMBER 1953 for both active and clean contacts the transient started with a rapid drop. For clean contacts this drop was directly to the steady arc voltage of the contact metal. For active contacts the drop was to a higher voltage between 15 and 32 volts followed by a gradual and irregular drop in voltage. The time of the first rapid drop was invariably between 2 X 10"^ and 4 X 10"^ second for both clean and active contacts over a range of circuit inductances between 0.1 X 10~^ and 48 X 10~^ henry. This time was just about the time resolution of the scope used indicating that in all cases the initiation time was less than the time resolution of the scope. This was also borne out by our calculations, Equation 3, where the longest initiation time for the conditions studied was only 1.2 X 10~^ second. The higher arcing voltage of active contacts mentioned above has been previously reported in Reference 3. It was pointed out that active contacts have arc voltages comparable with those of carbon, 19 to 30 volts.* The slow drop in voltage following the first rapid drop observed with active contacts is probably a burning off process of the activating substance on the contacts as indicated by a continuous ap- proach of the arc voltage to that of the clean metal. It may be added that for cases where the initial circuit voltage is closer to the carbon arc voltage one should expect a smaller initial drop. This was confirmed by measurements made at 35 volts. In connection with the experimental study of the initiation of the arc, the following concluding remark may be made. Unless the measuring apparatus has a response faster than the individual discharges at the contacts, the recorded transient will essentially be some particular average of a complex contacts transient. It should not, accordingly, be mistaken for the more fundamental and usually much faster formative transient of the arc. The author is indebted to Dr. P. Kisliuk and Dr. L. H. Germer for much valuable discussion. * Recent measurements by the writer on arc lamp carbon have given arc volt- ages as high as 43 volts. Polyethylene Insulated Telephone Cable By. A. S. Windeler (Manuscript received August 25, 1953) The physical properties of polyethylene are such as to make it attractive for many wire insulating applications, particularly in multi-conductor communications cables. This article presents certain factual information relating to new types of multi-conductor cables having extruded polyethylene insulation, and describes briefly their initial installation in working tele- phone plant. The literature is replete with information on the physical and chemical properties and the behavior of polyethylene, so no attempt is made to explore the quality of the material per se. Polyethylene insulation extruded in the form of both solid material and foam to impart certain de- sired electrical properties is discussed. In a brood sense this article may be considered as announcing an important new insulating material for tele- phone cables which may be expected eventually to have very extensive ap- plications in the Bell System plant. From almost the beginning of the art, multiconductor telephone cables have been insulated with paper, applied as a helical tape or laid down directly on the conductor in the form of pulp. Solid paper has a dielectric constant in the order of 2.5 to 3.0 but in the case of either ribbon or pulp insulation a considerable amount of air is included in the electric field surrounding the conductor so that the composite effective dielectric constant is of quite low value, usually about 1.5 to 1.6 in a typical design. In recent years, as various plastics and other polymeric materials have become available, these have been studied as competi- tors of paper, and polyethylene in particular now appears to have an important field of application. Polyethylene appears attractive because of its excellent electrical properties including low dielectric constant and power factor, compared ^vith other useable plastics, and high di- electric strength. It is also highly impermeable to water or water vapor and is available in the desired quantities at a reasonable price. Addi- tionally it is considered probable that the long term price trend will be downward. 1245 1246 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 The various electrical and mechanical characteristics^ • ^ of polyethyl- ene are shown in Table I, along with some of the other materials con- sidered. Among these other materials only polytetrafluoroethylene (Tef- lon) and polystyrene have power factors and dielectric constants in the same low range as polyethylene. There are basic objections to both of these materials. Polytetrafluoroethylene is so expensive as to be un- economical for this application and polystyrene in thick sections is too stiff and brittle to handle in a satisfactory manner. The use of polyethylene in telephone cables is not altogether new but it has heretofore been confined to special types of high-frequency cable. For example, the coaxial cable and video pair have polyethylene disc insulation and strip-and-string insulation respectively (see Fig. 1). These cables were designed for low attenuation in the megacycle range and the use of polyethylene or a similar low power factor material was a necessity. A low power factor is of lesser importance in the carrier sys- tems for which the multipair cables are used and the polyethylene in- sulation on these cables must show other advantages in order to prove in. Polyethylene insulation is applied to the wire by an extrusion proc- ess; the insulation may be either solid or expanded depending on the application. Generally the polyethylene is supplied as granules pre- viously compounded with an antioxidant. In the case of solid poly- ethylene insulation the granules, in which the pigment has been in- corporated, are fed into the extruder and formed on the conductor as a uniform close fitting tube of insulation. Table I — Characteristics of Insulating Materials Density — gms/cc . Tensile strength- psi Elongation % Water absorption % in 24 hrs Diel. strength, RMS volts/mil ^* thickness*. Power factor, 1 300 kc Diel. constant, 1- 300 kc Polyethylene 0.92 1400-2000 600 <0.01 4(X>500 0.0002 2.3 Plasticized Polyvinyl- Chloride 1.2-1.4 1500-3000 200-450 0.4-0.65 300-700 0.09-0.16 3.5-5.0 Polystyrene 1.06 500-9000 2-5 <0.05 500-700 0.0002 2.5-2.6 Polytetra- fluoroethylene (Teflon) 2.2 1500-2500 100-200 Nil 400-500 0.0002 2.0 Poly amide (Nylon) 1.09 7000 100-200 0.4-2.0 400 0.04-0.2 3.5-8 xu-*!^*®*®*^*"®.'**"®."^*" ^^® greater for thinner sections— for example, in 14 mil thicknesses polyethylene has a dielectric strength of approximately 2500 RMS POLYETHYLENE INSULATED TELEPHONE CABLE 1247 Fig. 1 — (a) Polyethylene disc insulated coaxial, (b) Polyethylene ribbon and string insulated video pair. Note expanded polyethylene interstice fillers. The idea of using spongy or foamed hydrocarbons as conductor in- sulation is not new. A British patent issued in 1930 contemplates such a structure and numerous United States patents of more recent dates cover various aspects of cellular hydrocarbon insulation. The problem is one of forming the cylinder of aerated plastic in an extrusion process operating at high speed and producing a closely controlled uniform covering having precise physical and electrical properties. The original development work was carried on by F. B. Lyons of the Western Electric Company in cooperation with the author. This early work demonstrated that material having the desired range of properties could be apphed in a continuous extrusion process and subsequent work has shown that the necessary control of properties and speed of extrusion can be achieved. The expansion of polyethylene is accomplished by methods similar to those employed in the making of many of the numerous polymer and rubber "foams". The process used to produce the cellular polyethylene involves the addition at the extruder of a chemical blowing agent which decomposes under heat and releases nitrogen gas. By proper mixing and process control this nitrogen gas can be entrapped in the poly- ethylene in the form of very small discrete bubbles, thus achieving the cellular structure shown in Fig. 2. It is interesting to note that the foamed plastic tends to form a desirable ''skin" of solid material on the inner surface over the conductor, Fig. 2(a). Various degrees of expansion can be achieved as required by varying the amount of blowing agent and by other means. The degree of ex- pansion, or percent entrapped gas, can be determined readily by weigh- 1248 THE BELL SYSTEM TECHNICAL JOURNAL, SEPTEMBER 1953 Fig. 2(a) — Cross section of expanded polyethylene insulation from 19 gauge conductor — 35 per cent air. Magnified 75 times. ing a sample of insulation, with the conductor removed, on an analytical balance. The inside and outside diameter of the cylinder of insulation and the density of solid polyethylene are required to complete the de- termination. The composite dielectric constant obviously varies with the degree of expansion. In a coaxial configuration this effect is calculable from the formula for the dielectric constant of a mixture,* the relation being given by the following: g --

Jit. H'cstt-rn I'Jrrtrir ('onipan; |,'. |.. tv \ 1' l> K I,. I ic<' I'n's/drnl. hncriinn '/'rlcp/ion^ and 'I't'h'ordf/// ( Oniitanv M h 1. V. I'n^'^idf'nt. Hell 'I^Flejihiim' Lahoratories i: 1) I T () K 1 A L C O M M I T T E E K. I. (; i; K K > - (1} airman A. J. B L! S <; II F. R. T.ACK W. H. D O H K H T Y W. H. N U N N G. D. E D W A H D S H. I. H O M N E S J. B. FISK H. V. SCH M 1 1) T 11. K. H O N A M A N G.N. T H A V E R EDITORIAL STAFF J. D. T E B O, Editor M. E. S T R I E R Y. Mana^iriir Editor R. L. SHE P H E H D, Production Editor THE BELL SYSTEM TECHNICAL JOURNAL is published six times a year by the American Telephone and Telegraph Company, 195 Broadway, New York 7, N. Y. Cleo F. Craig, President; S. Whitney Landon, Secretary; Alexander L. Stott, Treasurer. Subscriptions are accepted at $3.00 per year. Single copies are 75 cents each. The foreign postage is 65 cents per year or 11 cents per copy. Printed in U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XXXII NOVEMBER 1953 number 6 Copyright, 196S, American Telephone and Telegraph Company Design Theory of Junction Transistors By J. M. EARLY (Manuscript received September 8, 1953) ^ The small signal ac transmission characteristics of junction transistors are derived from physical structure and bias conditions. Effects of minority carrier flow and of depletion layer capacitances are analyzed for a one dimensional model. The ohmic spreading resistance of the base region of a three dimensional model is then approximated. Short circuit admittances representing minority carrier flow, depletion layer capacitances, and ohmic base resistance elements are then combined into an equivalent circuit. Theo- retical calculations are compared to observations for two typical designs. LO INTRODUCTION 1 .1 General Junction transistors have been in commercial production for nearly a year. A detailed understanding of their behavior is necessary both for the increasingly exacting requirements of modern circuit engineering and for the wise design of improved types. Design theory, by relating function to structure, can serve both these needs. The principal object of this paper is to develop in logical fashion a design theory for junction transistors. The product of the development is an equivalent circuit, founded on device physics, which predicts the circuit characteristics of junction devices in a simple and intelligible fashion. Although attention is concentrated on small signal transmission performance, some large signal aspects are also examined. 1271 1272 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 1 .2 Method and Assumptions The usefulness of the junction transistor derives primarily from the flow of holes or electrons across two closely-spaced p-n junctions, one of which is biased in the forward or conducting direction while the other is biased in the reverse or non-conducting direction. Development of design theory begins quite properly with analysis of this mechanism, which is considered, for simplicity, as a problem in the flow of holes and electrons in one dimension, at right angles to the p-n junctions. In the analysis, it is assumed that these carriers are controlled largely by the voltages applied to the junctions and that they move principally by diffusion. The dependence of the diffusion currents on the junction volt- ages is reduced to a set of two terminal-pair short-circuit admittances, which form the initial and most important segment of the equivalent circuit model for the junction transistor. Practical transistors have not only the very useful transisting mecha- nism mentioned above, but also passive capacitances across the charge depletion layers which separate the p and n regions at each junction. These capacitances limit the useful frequency range of transistors and must be considered in any practical theory. In the synthesis of the equivalent circuit, these capacitances are placed in parallel with the short-circuit input and output admittances which represent the flow of diffusing holes and electrons. A further limitation on performance is imposed by the ohmic or body spreading resistance of the base region. The base current of the transistor, in flowing from the region between the emitter and collector to the base contact, develops a base contact to emitter voltage which seriously limits the frequency response. Calculation of these effects requires the assumption of flow paths for the base current. The circuit elements rep- resenting base spreading resistance effects appear in series in the base leg of the equivalent circuit. 1 .3 Existing Design Theory W. Shockley's classic paper* announcing the junction transistor also initiated the design theory. Diffusion effects for dc and low frequencies were analyzed, and formulae for depletion layer capacitances were de- veloped. The mechanism of the frequency cutoff of the current trans- mission (alpha) was reported in a subsequent article t, and the effects of * W. Shocklev, The Theory of p-n Junctions in Semiconductors and p-n Junc- tion Transistors, B.ST.J., 28, p. 435. t W. Shockley, M. Sparks, and G. K. Teal, The p-n Junction Transistors, Phys. Rev., 83, p. 151, July, 1951. DESIGN THEORY OF JUNCTION TRANSISTORS 1273 ohmic resistance of the base region were discussed briefly. Still more recently, the dependence of base thickness on collector voltage was used to explain output and feedback effects*. The present paper is both a consolidation and an extension of the earlier works and borrows freely from them. The diffusion current analysis of Appendix A is patterned after Shockley's. 1 4 Scope The design theory developed here is not complete, even for small signal ac transmission. In particular, effects of large carrier emission densities are not considered, nor are the effects of non-parallel junction arrangements. Despite these omissions, it is hoped that the theory developed will be both useful and instructive to those engineers charged with transistor device and transistor circuit design. 2.0 METHODS AND ASSUMPTIONS 2.1 General In developing the design theory, it is convenient to break the transistor down into several internal electronic functions and to consider their dependence on structure and materials individually. These functions are then fitted together and used to predict the terminal electrical char- acteristics. With this approach, it seems proper to describe separately the methods and assumptions used in analyzing each of the functions, 2.2 List of Symbols The symbols listed here are used in the body of the paper. A separate list for Appendix A appears at the end of that section. Emitter and collector currents are assumed to flow inward at the corresponding terminals, in accord with the convention usually used for transistors. a = gradient of (Nd-Na), usually given in atoms/cm . ace = short-circuit forward current transfer constant for theoretical one-dimensional transistor. Cc = collector to base capacitance with emitter open-circuit ac. Cse, Csc = hole storage or diffusion capacitances at emitter and col- lector. These capacitances are directly related to the current trans- * J M Early, Effect of Space-Charge Layer Widening in Junction Transistors, I.R.E., Proc, 40, pp. 1401-1406, Nov., 1952. 1274 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 mission cutoff frequency and may be used as an alternative charac- terization of that quantity. Cre , Ctc = theoretical depletion layer capacitances of emitter and col- lector. D, DpjDn = diffusion constants for minority carriers, usually given in cm /sec. /„ = D/tWo^ = current transmission or alpha cutoff frequency. Qee , Qce , Qec , Occ = low-frequeucy conductance components of ?/'s given below. h^s = set of two terminal-pair parameters, defined by Guillemin, Com- munication Networks, 2, p. 137, John Wiley and Sons. hn = short circuit input impedance. h2i = short circuit forward current transfer ratio. hu = open circuit feedback voltage ratio. hzz = open circuit output admittance. h = average or dc base current. Ipe , Inc , I PC , Inc = holc aud electrou components of average or dc emitter and collector currents. Ipeo = emitter reverse current when collector is also reverse biased. Je = emitter current density in amperes/cm^. k = Boltzmann's constant. kT/q = average thermal energy per carrier, approximately 0.026 elec- tron-volts at 25°C. L, Lpy Ln = diffusion length or average distance a minority carrier will diffuse before recombining; average distance diffused in one life- time (t). Naj Nd = concentration of acceptor and donor atoms in semi-con- ductor, usually in atoms/cm^. n = concentration of electrons/cm^. Ui = electron concentration which would exist in the semi-conductor at thermal equilibrium if donor and acceptor concentrations were zero. Up = thermal equiUbrium concentration of electrons in p-region. p = hole concentration/cm'. DESIGN THEORY OF JUNCTION TRANSISTORS 1275 Pi = hole concentration which would exist in the semi-conductor at thermal equilibrium if donor and acceptor concentrations were zero. Pn = thermal equilibrium concentration of holes in n-region. q = electronic charge, 1.6 X 10~^^ coulombs. q/kT = see kT/q. Tb, ni, rm = ohmic spreading resistances of base region, specifically, the effective base to emitter feedback resistances for diffusion cur- rents and for collector capacitance currents. n , r2 , n = geometrical radii in transistor of Fig. 2(b). T = temperature in °K. Vc = average or dc collector to base voltage. VJ = electrostatic potential across collector depletion region. Vs = electrostatic potential across emitter depletion layer at therma equilibrium (no biases applied). Vc = small signal ac collector to base voltage. w, Wo = base region thickness. Wi , W2 , Wz = base region thicknesses in transistor of Fig. 2(b). Xm = thickness of collector depletion region. yce , Vce , Vec , Vcc = theorctlcal short circuit input, forward transfer, feed- back, and output admittances for one-dimensional transistor. a, oiQ = short-circuit emitter to collector current transfer ratio and its low-frequency value. a*, oo* = collector junction current multipUcation ratio and its low- frequency value. j3, /So = current transport ratio across base region and its low-frequency value. 7, To = current emission ratio at emitter and its low-frequency value. €o = dielectric constant of vacuum, 8.854 X 10" farad/cm. A' = relative dielectric constant, e/eo . P, Pb = resistivity, base region resistivity. (^nc, (Tpc = conductivities produced by electrons and holes in collector region. r, Tn , Tp = lifetimes of minority carriers. 1276 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 fxi,c = constant of feedback generator used to characterize modulation of do base spreading resistance. 03 - 2ir/ = angular frequency in radians. oj^ = 2irfa = 2D/W = alpha or current transmission cutoff frequency in radians. 2.3 Minority Carrier Admittances An admittance representation of minority carrier diffusion is a way of writing the dependence of the diffusion currents on the junction poten- tials. To obtain this dependence analytically, the minority carrier den- sities on both sides of each of the two depletion layers (emitter and collector) are assumed to be exponential functions of the junction volt- ages. This exponential dependence is a result of the normal thermal dis- tribution of hole and electron energies. The carrier diffusion currents are computed directly from the gradients of the minority carrier den- sities at the depletion layer surfaces. Since the gradients of the carrier densities are affected by many conditions besides the junction voltages, additional assumptions are necessary. Their nature and pertinence may be seen from consideration of the normal operation of a junction transistor. The three principal regions of a junction transistor, the emitter, the base or control, and the collector, are indicated in Fig. 1. These regions are separated by transition regions in which the conductivity type changes either gradually or abruptly from p-type to n-type. Roughly coincident with these transition regions are the emitter and collector depletion layers across which the emitter and collector voltages appear EMITTER BASE COLLECTOR 11 n EMITTER DEPLETION LAYER Fig. 1 — p-n-p transistor. DESIGN THEORY OF JUNCTION TRANSISTORS 1277 when the unit is biased. In normal operation for the p-n-p transistor shown, the emitter is biased positive with respect to the base so that a current of holes is injected into the base from the emitter. The collector is biased negative with respect to the base so that the holes dilTusing across the base from the emitter are collected whenever they reach the edge of the collector depletion layer. In the analysis each of the three major regions is assumed to have a uniform resistivity, p; a diffusion constant for minority carriers, I), which is a measure of the speed with which injected carriers will diffuse; and a lifetime for minority carriers, r. This lifetime is the average time which a minority carrier remains free before recombining with a majority carrier. The minority carrier density in each region is assumed to have a thermal equilibrium value in the absence of appHed potentials. The density is increased or decreased exponentially from this value by the applied potentials. The base layer is assumed to have a thickness, w, which is dependent on the collector voltage Vc . An increase of collector voltage increases the collector depletion region thickness, Xm , thus de- creasing the base thickness. The rate at which base thickness changes with collector voltage is determined by the nature of the transition from base to collector. For gradual transitions, the rate of the transition is important, while for abrupt or step transitions the rate of change of base thickness with collector voltage is determined by the base region and collector region resistivities. Determination of the diffusion currents from minority carrier density gradients requires determination of minority carrier densities everywhere in the three principal regions of Fig. 1. These are obtained by solving a continuity equation for carrier flow in each region, subject to the applied junction potentials and other assumptions described above. It must be pointed out that, in normal operation, there may be a significant flow of electrons to the emitter and from the collector in the p-n-p transis- tor of Fig. 1. Small signal ac diffusion currents are determined by as- suming small signal variations of the junction voltages and discarding all but first-order ac terms from the diffusion currents. Results of the analysis are given in Section 3.0, and the analysis appears in Appendix A. 24 Depletion Layer Capacitances In a p-n junction with no bias potential applied, there is a tendency for holes to diffuse into the n-region and for electrons to diffuse into the p-region. This creates a slight unbalance of charge in the two regions 1278 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 and the resulting electrostatic potential keeps each type of carrier in its own region. The potential appears across a thin layer separating the two regions. In this depletion layer, the hole density is lower than in the p-region and the electron density is lower than in the n-region, and there is a net charge density. Acceptor and donor atoms are not neutralized by mobile charge as they are in the p- and n-regions, but instead serve to terminate the field of the electrostatic potential. Application of ex- ternal potential across the junction changes the electrostatic potential, and by exposing more or fewer fixed (donor and acceptor in equal num- ber) charges widens or narrows the depletion layer. The passive capacitance of this region is simply that of a parallel plate condenser having a plate spacing equal to the layer thickness. Calculation of this capacitance is explained in Section 3.0, following the discussion of the minority carrier diffusion admittances. 2.5 Base Spreading Resistance* ImpUcit in the one-dimensional analyses described above is the as- sumption that the base region is everywhere at the same potential. Actually, since the emitter and collector currents are not equal, current must flow through the base region parallel to the junctions. Because the base region has finite, rather than zero resistivity, this current produces transverse voltage drop in the base region. It is assumed that the most important effect of these voltage drops is the feedback produced between the base contact and the emitter junc- tion. In consequence, each of the ohmic base resistances studied is de- fined as the quotient of an average voltage between base contact and emitter junction divided by the current producing it. The need for defining more than one feedback base spreading resistance results from the fact that the base current has two principal ac components. One of these is the difference between the emitter and collector minority carrier diffusion currents. The other is the collector depletion layer capacitance current. The feedback effects of these two currents on the emitter jimc- tion are the same only when the flow paths of the two currents through the base region are the same. Consequently, the representation of base resistance effects is somewhat more complicated in transistors where the flow paths differ than in those where they are identical or nearly so. * The majority carrier resistance of the base region for base current flow- parallel to the junctions. The word "spreading" was suggested by the base con- tact geometry of Fig. 2(a) and readily distinguishes this resistance from the "base resistance" of the familiar tee network, which was long believed to be identical with it. It is not. See Sections 5.2 and 5.3. DESIGN THEORY OF JUNCTION TRANSISTORS 1279 Fig. 2(a) shows a structure for which the flow paths to the base con- tact are substantially the same for all components of the base current. Both the collector capacitance current and the diffusion loss base current enter the base region substantially uniformly over the entire area and follow the same path to the base contact. In Fig. 2(b) these two currents have quite different flowpaths and the associated feedback resistances are likewise very different. The general method of calcula- tion is, however, the same in both cases. Another important effect is associated with modulation of the dc voltage drop in the base region. The base current ordinarily has a dc as well as an ac component, and a dc voltage drop occurs between the base contact and the emitter junction. Since the base region thickness changes when collector voltage changes, the dc resistance of the base region is modulated by the collector voltage, producing a modulation of the dc voltage between base contact and emitter.* This effect is most easily represented by an ac voltage generator in series with the base EMITTER BASE COLLECTOR SECTION A -A EMITTER COLLECTOR Fig. 2 — p-n-p transistor structures. This effect was first pointed out by J. N. Shive. 1280 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 contact. The voltage is computed as the product of the dc base current and the modulation of the dc base resistance. This modulation can be calculated from the base region resistivity and the dependence of base thickness on collector potential. 2.5 Summary of Methods In developing the design theory, simple physical assumptions are made concerning the behavior of the charge carriers in the semicon- ductor. The transistor is studied as a one-dimensional problem and the per unit area electrical characteristics of the one dimensional structure are computed. The effects of current flow within the base region parallel to the junctions are then calculated for a three-dimensional model. Finally, the equivalent circuit representations of these electronic func- tions are combined in structural fashion to give the terminal electrical characteristics of the junction transistor triode. It should be noted that the base region thickness between emitter and collector is assumed uniform, and that design theory has not been ex- tended here to cover the case of non-uniform thickness. Likewise, edge effects at the emitter and surface effects in general are neglected. These omissions were made for mathematical simplicity and are neces- sary omissions in a one-dimensional analysis. The place of surface leakage among the electronic functions is discussed at the end of Section 4.0. Analysis of the effects of sharp discontinuities in base layer thickness requires new solutions to the continuity equation but gradual changes in thickness can be accounted for by averaging over the active area of the transistor the short circuit admittances which are the subject of the next section. 3.0 ONE-DIMENSIONAL TRANSISTOR 3.1 General This section deals mth the small signal transmission electronics of the structure of Fig. 1 . It is assumed that the emitter is biased to provide a flow of carriers into the base and that the collector is reverse biased sufficiently so that no majority carriers can diffuse out of the collector region into the base region (a reverse voltage of 0.5 volts is more than enough to prevent this). The four admittances associated with minority carrier flow and the two depletion layer capacitances are indicated in Fig. 3. In each case, the design expressions are given first in their most exact form and are progressively simplified. For convenience in dis- cussion and comparison, current densities per unit area rather than DESIGN THEORY OF JUNCTION TRANSISTORS 1281 9E C9 j- Yr^V^ Ctc 4B Fig. 3 — Theoretical equivalent circuit for "one-dimensional" transistor. currents are used and the admittances and barrier capacitances are written on a per square centimeter basis. It will be noted that transverse voltage drops resulting from base spreading resistance are ignored in developing the expressions of this section. A number of physical mechanisms are involved in the admittances for minority carrier flow. The forward current across the emitter junction rises as an exponential of the emitter to base region voltage. This is the result of the thermal energy distribution of minority carriers and is com- mon to all thermionic emission. A natural effect of this exponential de- pendence is that a given change in the voltage results in a fixed per cent change in the current, thus producing an ac admittance which is pro- portional to the average or dc emitter current. For several reasons not all of the current which flo\ys through the emitter junction is collected. First, some of the emitter current consists of electrons diffusing into the emitter body and of displacement current through the emitter depletion layer capacitance. These effects are expressed in the emission factor or 7, which is the ratio of the hole current injected into the base region to the total emitter current. Further, some of the injected holes recombine in the base layer. Those which are collected suffer a transit delay which results in a phase difference between emitter and collector currents. These two effects are summed up in the forward current transport factor or (3, which is the ratio of the minority carrier current reaching the collector to that injected by the emitter. Finally, the current of holes entering the collector from the base gives rise to a much smaller flow of electrons from the collector to the base, thus producing a collector multiplication factor or a*, which is the ratio of total carrier current crossing the col- lector junction to the hole current entering it from the base. Although the most important minority carrier flow originates at the emitter, this flow is altered by changes in collector reverse voltage. As the collector reverse potential is increased, the base region becomes 1282 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 narrower because of widening of the collector depletion layer. This re- duction in base thickness permits more emitter current to flow for a fixed emitter to base voltage, so that both emitter and collector currents are increased, thus producing output and feedback admittances. How- ever, since large changes in collector potential are required to produce small changes in base thickness, relatively small changes in the junction currents are produced and the admittances are far smaller than those associated with change of emitter potential. 3.2 Diffusion Current Admittances The short circuit two terminal-pair admittances associated with the diffusion of minority carriers in the structure of Fig. 1 are - JL[(T - T ^ (1 + io^r^;)'" tanh Wo/jDj^Tj;)''' + /n. (1 + io^Tnef'^] „ _ _J_ \(j _ r X fd + ^'^r^)'^' tanh w,/{D^T^f'^^ V 0.995 (3o = 0.993 a* c^ 1.0000 ace > 0.988 dW Xm dVc 27c 2.6 X 10"' 2 X4.5 I = 2.89 X 10"' cm/volt A ^.o Qcc ^^8.15 nmhos The parameters which determine frequency response are: f ^ "^ = -^^ = — = 1.08 mcps •^" 27r irwl 7r(3.6 X 10-^)^ ^ Ct = 5,000 niifd/crn for collector so that Ct ^ 20,000 ^y,fd/cm for emitter Ctc = 22.8 fififd Cre = 22.8 finfd 1296 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 The effective hole storage capacitances of the emitter and collector are: Cse = 3840 nnjd Csc = 0.8 finfd By the equations given in Section 3.0 Tbi = 55 ohms rb2 = 35 ohms The low frequency values of the ''/i" parameters are /i2i o^ -0.988 hu ^ 2.09 X 10"' /i22 ^^ 0.17 fimho hn ^ 26 + (0.012)55 o^ 26.7 These theoretical values are compared with observed values in Table I. The major discrepancy in /121 is charged to surface recombination of in- jected holes, which was ignored in the calculation. It should be noted that /i22 becomes 0.45 X 10~^ mho if the calculated value is corrected by the ratio 0.032/. 012, which is the ratio of the measured and calcu- lated (1 — a)'s or (1 -f /i2i)'s. The difference between computed and measured hn is the sum of a number of effects. First, the actual (1 — a) is greater than the computed value. Next, the junction temperature was probably greater than the assumed 25°C. Finally, the carrier injection level is high enough to modify the emitter diode properties in this direction. The difference between calculated and observed current transmission cutoff is greater than appears from the data, since the theoretical three db response frequency is about 22 per cent higher than the "alpha Table 1 Calculated Measured hi - 0.988 -0.968 hr, 0.17 X 10-« mho 0.48 X 10-6 mho hn 2.09 X 10-4 1.85 X 10-" hn 26.7 ohm 33 ohm Cc 22.8 nnfd 24.7 titijd ^% 1.08 mcps 0.95 mcps n'l 55 ohm 55 ohm n'2 35 ohm 63 ohm DESIGN THEORY OF JUNCTION TRANSISTORS 1297 cutoff" frequency calculated here. The difference is believed to be the result of the fact that holes emitted around the emitter periphery have much longer transit paths than do those emitted into the region directly between the electrodes and consequently reduce significantly the cur- rent cutoff frequency. The serious discrepancy in rb'2 is probably the result of the r^'i cal- culation being very pessimistic because of neglect of peripheral emission effects and of W2 and w^ being somewhat smaller than the assumed values. Again it can be seen that the equivalent tee base resistance n = hn/h^^ is 375 ohms, nearly seven times the high frequency resistance of 55 ohms 5.4 Qualitative Comparison As might be expected, the qualitative agreement between theory and observation is better than the quantitative. For example, Cc , /i22 , and hi2 vary approximately as (Vc)~^^^ in fused junction units. Alpha cutoff frequency increases as collector reverse bias is increased — a natural result of the narrowing of the base region. The qualitative discrepancies that are found are usually associated with large experimental deviations from the assumptions of the analysis. 5.5 Review of Design Calculations Numerical analysis of both drawn junction and fused junction tran- sistors has shown rather good agreement of theory and experiment. It is necessary, however, to modify some of the results empirically because of lack of full understanding of some effects, such as leakage and surface recombination. Qualitative agreement of measurements and theory is, of course, much better than the quantitative correlation. For example, the de- pendence of all of the parameters on emitter current and collector voltage is almost exactly that expected from theory. Some of the dependence on emitter currents involves high carrier injection level theory which has been omitted from this study. 6.0 SUMMARY 6.1 Transmission Theory Design theory of the small signal transmission parameters of junc- tion transistors is relatively complete. A one-dimensional analysis of minority carrier diffusion currents in terms of short circuit admittances has been combined with a similar analysis of depletion layer capacitances and an approximate three- 1298 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 dimensional analysis of ohmic base region spreading resistance. The resulting equivalent circuit has characteristics in good agreement Avith experimental observations. In particular, collector capacitance, ohmic base region spreading resistance and the current transmission three db cutoff frequency may be computed with fair accuracy. The low fre- quency values of the common base hybrid parameters may also be cal- culated, but neglect of surface recombination and surface leakage re- sults in serious errors in the short circuit current transmission factor h^i and the open circuit collector conductance /i22 . The deviations of these two parameters from calculated values are, however, both reasonable and mutually consistent. The ohmic base layer spreading resistance, which is the only base resistance of importance at high frequencies, is very often much smaller than the low frequency base resistance appearing in an equivalent tee network. Qualitative agreement of theory and measurement is excellent. The variation of all parameters with emitter current and collector voltage is within a few per cent that predicted from theory, 6£ State of the Art Since this paper was deliberately limited in scope, it is pertinent both to review its objectives and to point out significant omissions. The principal objective sought was presentation of small signal transmission design theory. No attempt was made to give a simple explanation of the junction transistor, relating both its large signal and its transmis- sion characteristics to simple physical assumptions. While the design theory presented consolidates in one place some already published in- formation, much remains to be done in assembling and integrating such knowledge from its present widely scattered locations. In addition there exists a more detailed understanding of junction transistor characteristics than can be found in the literature. For ex- ample, units are found occasionally with negative hu • This is a result of an easily modulated high resistance between base region and base contact (high jLt6c). Publication of such information can reduce by a few db the amount of head-scratching done by production engineers. Other phenomena for which explanations have been developed are surface recombination and high carrier injection level effects. Despite this, much work remains to be done. ACKNOWLEDGEMENTS The general point of view taken here has been much influenced by discussions with J. A. Morton. Comments and criticisms by R. M. Ryder have been particularly helpful in the preparation of this material. DESIGN THEORY OF JUNCTION TRANSISTORS 1299 Appendix A 1.0 GENERAL This study is an extension of Shockley's analysis of the junction tran- sistor to include high-frequency effects and the voltage dependence of base-layer thickness. Shockley's paper* and the later paper by Shockley, Sparks, and Tealf contain the following of interest here: (a) analysis of the dc steady state of a junction transistor*; (b) analysis of the low-frequency small-signal parameters Ve , a, Cc*t; and (c) analysis of frequency dependence of the transport factor /Sf. In addition to repeating the above, this study gives these new results: (a) analysis of steady-state small-signal ac operation [dc biases pres- ent] ; and (b) the small-signal ac short-circuit admittances yee , ijce , Vee , and Vcc- 1.1 ASSUMPTIONS Semiconductor. The p-n-p type structure assumed is shown in Fig. 6. The emitter, base, and collector regions may each be characterized by a resistivity and a minority carrier diffusion length. The emitter, collector, and base contacts have no effect on the currents which flow at the junctions. Injected carriers pass through the base layer by diffusion and through EMITTER BASE COLLECTOR X X= 0 Fig. 6 — jp-n-p junction transistor. * W. Shocklej^ The Theory of -p-n Junctions in Semiconductors and p-n Junc- tion Transistors, B.S.T.J., 28, p. 435. t W. Shockley, M. Sparks, and G. K. Teal, The p-n Junction Transistors, Phys. Rev., 83, p. 151, July, 1951. 1300 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 the barrier layers* by drift. Base-layer thickness w depends on collector potential Ve through variation of the collector barrier thickness Xm . Variations of emitter barrier thickness are unimportant. The junctions are parallel. Currents and Potentials The currents and potentials studied are those at the collector and emitter barriers. Unless otherwise specified, "potential" implies dif- ference in majority carrier Fermi levels, i.e., externally applied potential, rather than difference in electrostatic potential. The collector reverse potential is assumed to be many multiples of kT/q(e.g. > 0.5 volts), so that the classical p-n diode reverse conductance may be neglected. At 0.5 volt, this is of the order of 10~^^ mhos. It decreases in magnitude one decade per sixty millivolts of bias potential. Base Resistance Majority carrier resistance in the base layer is not considered here. Other Assumptions Surface effects are excluded . from consideration. In addition, several mathematical approximations of little physical consequence appear in the text as needed. 1.2 METHOD The procedure employed is substantially that used by Shockley with some additions. First, minority carrier concentrations on both sides of each barrier are related to the barrier potentials. The dc minority carrier distribution in each of the three transistor regions is then computed from these boundary conditions with the aid of the continuity equation. Next, small-signal ac perturbations of the barrier potentials and of the minority carrier densities at the barriers are used in the same way to find ac distributions of the minority carriers. The effects of voltage dependence of base-layer thickness are found by means of a small- signal ac perturbation of the position of the collector side of the base layer. The resulting ac distribution of minority carriers is computed as before. Finally, dc and ac currents at emitter and collector barriers are com- * I.e., charge depletion layers. DESIGN THEORY OF JUNCTION TRANSISTORS 1301 puted from the gradients of the minority carrier densities at the barriers. In the ac analysis, the forward rotating time function e'^ is used. In general, the first-order solution for small signals is obtained by assum- ing that the disturbance associated with e**^' is small, by neglecting its powers and harmonics, and by using first-order expansions, e.g., The ac magnitudes such as it^i , pei , Ud , represent complex phasors of the form ae^*. A list of S3rmbols is given in Section 1.5. 1 .3 ANALYSIS In the base layer at a^ = 0, P = PeO + Peie = Pne KW at x = w^o , p = Pco-hPcie = Pne " (Id; in which Ve = Veo + Veic"* and Vei « Veo and Vd « Vco so that PeO ^ Pne'^^'^'^^'pcl ^ PeO ^ Vd kT Pel ^ Pco -TTf^ ^cl Pne'^'-'^'Pcl ^ Pco ^ PeO , Pco , ^eo , and Veo 2iVQ avcragc values while pei , Pci , Ve\ , and Vd are ac phasors. The continuity equation for holes for one-dimensional flow is: d'p _ (pn- p\ dp (2) in which r is hole lifetime. A solution to equation (8-18) is: p ^ p^ + Ae^'^ + Be-'^' + C."'^^"' + De'-"^^'^' (3) where L = VD^ and s = (1 + ioiTY'^. Application of the boundary values of equation (la) and (lb) to 1302 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 equation (3) gives hole density in the base layer: p.{t,x) = p. + L 2 sinh (wo/L) J^ _ fpcO - Pn) - (peQ " Pn)e^''^~\ _x/L L 2:sinh (wo/L) J [— swo/^n Pel — Peie sxlL+ioit 2 sinh (swo/L) J Pel - Peie^^n _« (4) -[ 2 sinh {swo/L)j -sx/L+iut 6 Since up to this point w has been assumed constant [w = Wo] , equation (4) does not include effects of voltage dependence of base layer thick- ness. To introduce these, a new set of boundary conditions is used: at X = 0, .^^ p = Peo -^ Peie'"^ (5a) and at x = wq -^ Wi e"^ , . ^ p = Pco + Pcie"" (5b) in which Wi ^ Wq and is a phasor. dw ^j, w, = ^^ F. It can be seen that conditions are as before except that the collector side of the base layer swings about position Wo at angular frequency co. A solution of equation (31) with conditions (5a) and (5b) is given by Pi(t,x) = pQ{t,x) + p(t,x) (6J in which po{t,x) is given by equation (4) and p{t,x) is the perturbation associated with Wie'''^\ If equations (5a) and (5b) are rewritten in terms of p(t,x), they be- come, using first order expansions: pM = 0 (7a) pit,Wo + Wie*"^) = [(peo - Pn) csch (Wo/L) - iPcO - Pn) COth (Wo/L)] ^ 6^"' ^^^^ Since p(t,x) is an ac solution of the continuity equation (2), it has the form : P(t,x) = £-6-/^+*"* + /?Tg— /L+*.* (g^ DESIGN THEORY OF JUNCTION TRANSISTORS 1303 Use of equations (7a) and (7b) leads to: i^ X sinh {sx/L) ,, - (p.o - p„) coth («)o/L)] ^ e*"' (9) The complete solution for hole density in the base layer is: _ \{VcO - Pn) - {peO - Pn)e"°^n _,/^ L 2 sinh (?/;o/L) J L 2 sinh (swo/L) J -[ (10) 2 sinh (st^o/L) J iiJi i„< sinh (sx/L) r, . u / /r\ + L ^ sinh (.Wi^) f(^^" - ^'^^ ^^^^ (^«/^) — (pcO — Pn) COth {w^lVi\ The hole-current density in the base layer is found from equation (10) by the use of the equation for diffusion current /.= -,Z>.| (11) which yields J ^P ( ( \ cosh (x/L) f . /.= - Pn , equations (13a) and (13b) becomes very closely pel r q s tanh (wp/L) ^^ dw sVd 1 ..^m I''' kT tanh (swo/L) ^ '' "^ ^''' dVe' L sinh (swo/L)] ^'"^^^ In equation (13b), Ipeo and Ipco are average emitter and collector hole-current densities. The entire coefficients of Vei and Vd in equation (13b) are the input and the feedback short-circuit admittances associ- ated with hole flow in the transistor. Similarly, forward transfer and output short-circuit admittances as- sociated with hole flow may be found from equation (12) by stubstitu- tion oi X = Wq and use of the approximation pi ^:^ pi — Pn , i.e., pi ^ Pn . In calculating collector current, the sign of equation (12) must be reversed, since equation (12) gives current flow in the x-direction while collector current is assumed to flow in the negative a:-direction. The admittances are given in the summary at the end. Next, there are admittances associated with electron flow in the p-n-p transistor. Flow of electrons from base to emitter gives rise to an input admittance term, while electrons flowing from collector to base give rise to output and forward transfer admittance terms. An outline DESIGN THEORY OF JUNCTION TRANSISTORS 1305 of the derivation of these terms follows. The terms are given in the summary. For electrons in the emitter: at a; = 0, Sit X = — 00 rie = Ueo + Ueie'"' = u^e'"''"' in which, again, Ve = Veo + Feie*"' and first-order expansions are used. X = — oMo; "^ Dor I 2Z)oikf, Inspection of equation (19) shows that ^" = -m2npe'^'' (20) It is then apparent that ni must be of the form ni = npe'^^'fiy) (21) Substitution of equation (2) and equation (21) into (18b) leads to ^ ^^ M (^ -L '^"^O \ ^/ '^^ f ripm2ilne2 (^r,\ A general solution of equation (22) is f{y) = HzH]^ + Be'" + C/^" (23) * W. van Roosbroeck; private communication. DESIGN THEORY OF JUNCTION TRANSISTORS 1307 where '■--K"-5l.)VLK"-£l)M . It may be shown that the boundary conditions on f{y) are : at?/ = 0, / = 0; at?/ = 00, / = 0. Application of these values to equation (23) results in in which r2 is obtained using the negative square root. The electron density in the collector is now given by ^coiKZo The electron current is given by In = n,qDn l-m^e"^'' + e'''' ^-^-' (m^ - m,e''' - r.ene'^A . (26) At the collector junction, the ac component reduces to npqDnfJinm2r2Ei Now, since Ei = ppjpcj , the collector multiplication factor {I pel ~\~ Inclj/Ipcl is a* = l + '-^^^ (28b) (Tpc tcoMo The effect of collector multiplication as given in equations (28a) and (28b) is included in the general admittance expressions given later. Finally, no mention has been made of the admittances associated with barrier capacitances. Since the currents which charge these are majority carrier currents, there are no input-output interactions except those associated with majority carrier resistance of the base layer [an effect not analyzed in this study]. These capacitances add directly to input and output admittances. Shockley* gives methods for calculating these capacitances. * Loc. cit,, vol. 28, page 435. 1308 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 1.4 SUMMARY In paragraph 1.3, derivations were given or outlined for each of the four small-signal short-circuit admittances associated with the hole flow, electron flow, and barrier capacitances in junction transistors. The terms appear in that order in the expressions which follow. yee=U^^^coth {s,Wo/L,) + qu.Dr^Sne/ lJ^ e'"" ^'""^ (29) -{-uaC Te dVc Lp sinh (spWo/Lp) Lp (31) " \dVcLp Sp QPnDp ^/qV^olkT csch (wo/Lp) -f coth (wo/Lp)] tanh (spWo/Lp) Lp csch (wo/Lp) + coth (wo/Lp)] + ic^Cr) (l + ^-^-^fA / \ (TpctOiMo / (32) in which all symbols are defined in Section 1.5. It should be noted that collector multiplication operates on the current to the barrier capacitance since the latter current is a hole current in the collector body. The term dw/dVc is the same for both p-n-p and n-p-n structures. It is: for step junctions and for graded junctions Vc in equations (33) and (34) means dc electrostatic potential dif- ference across the collector barrier. Equations (29) through (32) may be manipulated into many forms. One of these sets which may be employed as a starting point for the approximate forms given in the body of this chapter is: _ q [r Sp tanh (wp/Lp) , 1 , . ^ (on^ DESIGN THEORY OF JUNCTION TRANSISTORS 1309 y-- kT ^-° sinh (s,w,/L,) \} + ^J (^^) _ dw Sp """ " dV. L, sinh is,Wo/L,) ^^ ^^^^ '^^^ " [w; L,UnhX,wo/L,) ^^' + ^^^d L^ ^ SJ (38) The change in signs which occurs in going from equations (31) and (32) to equations (37) and (38) takes place because the current re- placed by I PC had the opposite assigned positive sense. 1.5 SYMBOLS USED IN THE APPENDIX Cc , Ctc = collector barrier capacitance. Cn = emitter barrier capacitance. Dn , Dp = diffusion constants for electrons and holes ^0 = (Pp + np)/(pp/Dn + Up/Dp) Ea = ip = electric field associated with current at thermal equilibrium carrier densities. El , Eo = ac and dc components of Ea he , Ineo , I nei , he , hco , hci = total, average, and ac emitter and col- lector electron currents Ipe , Ipeo , I pel , I pc , I pcQ , I pel = total, average, and ac emitter and col- lector hole currents kT/q = thermal energy of carriers = 0.026 electron-volt Lp , Ln = diffusion lengths for holes and electrons m2 = A/ I ^ ," I 4- ^^^- decay constant for average elec- y \2DoMoJ ^ Dot ^ ^ 2DoMo tron density in the collector Mo = (Pp — hnp)/{pp = Up) rip , Tin = thermal equilibrium electron densities in p and n regions Ueo , riei = dc and ac components of electron density at emitter junction Pn, Pp = thermal equilibrium hole densities in n and p regions Peo , Pel , Pco , Pel = dc aud ac components of hole density at emitter and collector junctions q = electronic charge, 1.6 X 10"^^ coulombs Ve = ac emitter resistance I(-+£|.)V[U-+rI.)]'+S * The factor a* = (1 + a- cApc) is current dependent. At small average collector currents, it is (1 + "o = e ^^^foQO = gQi . (4) /^ = peak value of current flowing in resonant circuit. The relationship of energy stored to rate of change of energy stored is a convenient relationship for computing the value of Qi in this partic- ular case. The ability of a transistor to satisfy the requirements of the resonant circuit for sustaining oscillation as well as building oscillation will be discussed later. Energy which is large in comparison to the power taken by the load is stored in the tuned circuit in order to minimize: (a) The effect on oscillator frequency of small changes in load current phase angle with trunks of different types, connected. (b) The effect on oscillator frequency of different values of inductance in the output windings of the associated oscillator circuit. (c) The effect of surges on the transistor that are transmitted over the connected trunk to the oscillator. (d) The harmonics produced by the method of supplying power to the tuned circuit. This includes the use of an incorrect feedback adjust- ment. The Q for the transformers varied from 60 to 85. The stored energy was then 60 to 85 times the energy dissipated per radian at the oscilla- ing frequency for the particular transformer. By adjustment of the ratio of the turns between windings 1-2 and 1-3 a dissipation in the tuned circuits of from 6 to 10 milliwatts was obtained. In the six-frequency supply three different coil designs are used. One design is used for the two lower frequencies, a second for the two inter- TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY PULSING l!^!!! mediate frequencies and a third for the last two frequencies. Data on these coils are given in Appendix I. The frequency of oscillation for each coil is determined by the capaci- tance used with it and is given approximately by the tntnnila: Sufficient inductance is used in these coils so that mica condensers can be economically used with them. Mica condensers are used because of their low temperature coefficient. Leakage reactance and capacitance between windings make the circuit resonant at more than one frequency. To minimize the current fed back to the emitter under such parasitic conditions, the 4-8 winding was placed next to the core with terminal 8 next to terminal 1 . This placed an ac ground next to the end of the \\'inding connected to the emitter. As a further precaution against parasitic oscillations the resistance be- tween emitter and transformer is kept high. TRANSISTOR OPERATION It is customary to consider the transistor as an amplifier working into a load represented by the tuned circuit. However since the current in the collector and that in the emitter are intimately related during that part of the cycle when power gain is obtained, the collector circuit can be considered equally well as a negative resistance of a value established by the feedback used and the emitter circuit simply as a load. At best either method is only an approximation due to the nonlinearity of the transistor characteristics for large signals. Representation of the transis- tor as a resistance puts the requirements in terms of values readily ob- tained from the static characteristics of a transistor. This form of treat- ment is therefore used. The regions in which positive or negative resistance is obtained is il- lustrated in Fig. 3. The characteristics shown are those for an ideal tran- sistor. That is, the ratio of an incremental change in collector current to an incremental change in emitter current, a, with a constant collector voltage is a uniform value in region 2. Also, that in regions 1 and 3 the slope of the lines in each region is constant. That this is not very different from that obtained from some transistors can be seen by comparison with the actual characteristics shown in Fig. 4. The division line between regions 1 and 2 represents the magnitude of the dc voltage applied between collector and base, Vc-Vi, . On the left hand side of Fig. 3 the phase relationship and magnitude of ac voltage 1320 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 applied to the collector and the emitter current are shown. The condition which limit the ac voltage on the collector to this value will be discussed later. In region 1 the net voltage on the emitter to base circuit is negative and it has no effect on the current which flows in the collector. The cur- rent which flows in the collector will therefore depend purely on the re- verse resistance of the collector acting as a diode and the voltage applied between collector and base. This is a positive resistance but it is not always linear. Since the only voltage active in this region is that obtained from the tank circuit, this resistance acts as a load on the tank circuit. In region 2 the voltage applied to the emitter circuit is positive and the current that flow^s as a result of this voltage exercises control over the collector current. The external resistance that is used in the emitter circuit is sufficiently high so that the small changes that occur in the emitter input resistance as the emitter current is varied are insignificant. The current that flows in the emitter circuit can therefore be considered vary linearly with the ac voltage. Since for each incremental change in collector voltage a proportional change in emitter current will be ob- tained, a plot of this relationship for a given ratio of emitter to collector ac voltage will result in a straight line. The slope will be determined by the amount of feedback (emitter voltage). Since a decrease in the abso- lute voltage on the collector results in an increase in collector current, the line has a negative slope. It therefore represents a negative resistance. In region 3 the emitter current exercises very little control over the collector current. A negative resistance represented by —R2 is therefore obtained from 0 to TT of the ac wave. If a second transistor were added of identical charac- lil 16 II Si > I 30 15 10 5 0 COLLECTOR CURRENT, I^ , IN MILLIAMPERES Fig. 3. — Idealized transistor characteristics with operating regions for oscilla- tor indicated. TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY PULSING 1321 teristics so connected (push-pull circuit) that a negative resistance would be obtained from tt to 27r of the ac wave, the effect of a negative resistance of constant value would be obtained. In order to reduce the transistors to the terms of a two-pole device having a negative resistance the posi- tive resistances represented by the emitter circuits and regions 1 of the transistors would also have to be included. It is more convenient however to combine all elements that produce a loss when using the static characteristics of a transistor to determine if the transistor will satisfy the circuit requirements for oscillation. This method is therefore used when considering the circuit operation. There is one important difference between the characteristics of a transistor and the ideal characteristics shown. That is, for very small values of emitter current bias and a constant E^ , the ratio A/c/A/e, or, a, drops rapidly from three or more to approximately one as the emitter current approaches zero. To eliminate this change in the dynamic nega- tive resistance for very low voltage changes on the collector, a small dc current is supplied in the emitter circuit. The need for this will be discussed later. CIRCUIT OPERATION The negative resistance of the transistor is effectively connected in parallel with the resistances representing the various elements that make up the total loss by transformer action. The requirements for oscillation are therefore met when —R2 for a complete cycle is less than the positive resistance representing all losses. If the length of time that — R2 is ef- fective is only one half cycle the value of —R2 must of course be cut in half.* Since a single transistor could meet the requirement for —R2 only one was used in the working circuit as is shown in Fig. 2. The transistor adopted for this use was the 1729 type, now in produc- tion, which has been given the RTMA designation 2N25. The 1729 type was used because its characteristics are least affected by changes in temperature, and in addition the allowable power dissipation was ap- proximately twice that of other comparable transistors. The various factors that when combined make up the load and their normal variation are given in Table I. All losses are in terms of power into the 1-2 winding. The corresponding value of load resistance is, J? - (^c - n - Fc.)^10()0 ^''^ ~ 2X2XWt * See Appendix II. 1322 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Table I Source of Loss Power in Milliwatts Min. Avg. Max. 6.0 8.0 10.0 4.5 4.5 4.5 2.8 7.0 14.0 4.0 8.5 13.4 12.0 12.0 12.0 29.3 40.0 53.9 Sustaining stored energy in resonant circuit Output load Region 1 of collector operation Loss in emitter Margin for stability of adjustment Total, Wt This would be 1920-ohms, 1400-olims and 1045-ohms respectively. For purpose of illustration the average value is plotted on the characteristics of an average transistor in Fig. 4 along with —R2 for the transistor. It is evident from this that —R2 is lower in value hence the circuit will os- cillate and build up to the required voltage. Minor corrections in the emitter current would normally be required in order to meet test require- ments. The potentiometer that is shown in Fig. 2 provides the means for adjustment. The oscillogram shown in Fig. 5 illustrates the condition described above. The characteristics of the transistor used are shown in Fig. 4. The oscillogram is a multiple exposure from which Rl, —R2 and RS (see Fig. 3) for several values of feedback may be obtained. The normal condition of adjustment is illustrated in Fig. 6 with normal load. Four times normal load is a test condition. When the extra load that is applied during test is removed the output voltage should go up since power input exceeds the power expended. o uj h ■13 -12 -II -10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 COLLECTOR CURRENT, Ic^ IN MILLIAMPERES Fig. 4 — Characteristics of representative 1729 type transistor with negative resistance values plotted for the average condition given in table. A load line is also shown for the average condition. In plotting -R2 a line is drawn from h = 0, 1 414 X 2 Ec = 16 to point corresponding to maximum h . Maximum le ^ WeX ' at Ec ^ 1.5 volts. We is loss in emitter circuit given in table . E., TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY PULSING 1323 AMPLITUDE WITH RESISTANCE TRANSISTOR 1729 TYPE E1459M Figs. 5 and 6 — Oscillograms showing the relationship of collector current to collector voltage during a complete cycle for several operating conditions. Fig. 5 (left) is a multiple exposure made to illustrate the ability to oscillate at a lower output level with decreased feedback. This is due to the much higher alpha ob- tained with low emitter currents. Fig. 6 is for normal operating conditions. This increase in voltage is however very small since the losses will increase as the square of the voltage and the rate at which energy is supplied to the circuit will decrease. The decrease in rate is due to the change from a negative resistance to a positive resistance in region 3. This causes the average negative resistance for the complete cycle to have a higher value. Feedback which is far greater than is required can in some cases cause the peak value of the ac voltage to exceed the dc voltage. Power is drawn from the energy stored in the tank circuit when this occurs. This effec- tively limits any further increase in output voltage. A change will also be introduced in the emitter circuit due to opera- tion in region 3. That is, the voltage feedback introduced in the emitter circuit by collector current flowing in the common base resistance is re- versed in phase in region 3.^ This is due to a reduction in collector current when the voltage applied to the emitter circuit is still rising. This feed- back is sufficient in many cases to cancel the increase in emitter voltage. The emitter current in such a transistor will therefore remain nearly constant in this region. In region 2 however the feedback is in such a direction as to aid the flow of emitter current. The result is that the voltage drop across the emitter resistance is approximately canceled by the voltage across the base resistance. Due to this relationship the tran- 1324 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 sis tor may appear to have zero resistance in the emitter to base circuit or a small positive or negative resistance. It is evident from the foregoing that the abiUty of a transistor to fulfill the requirements for oscillation is therefore dependent upon both the average a along the load Une and, R\^ resistance in region 1. The rela- tionship of these two factors is shown in Fig. 7. Dots on this chart repre- sent the 1729 type transistors tested that met the requirements for the 2N25 transistor. The ambient temperature was +135°F. It is inadvis- able to use transistors having a resistance of much less than 4000-ohms in region 1 since both the average dissipation rate and the peak dissipa- tion rate would exceed allowable limits for continuous operation. This will tend to cause the transistor characteristics to change at a more rapid rate. The effect of different values of feedback on the output of the oscillator is shown in Fig. 8. Fig. 8 also shows the variation in output obtained \vith several different values of load resistance. This was done to illus- trate the use of increased load in determining the proper point for ad- justing the feedback resistance. The proper adjustment is the minimum feedback with which the output changes only approximately 1 db in going from normal load to four times normal load. This degree of margin 10^ 20 O I z \- ^5 lij !=: cr U) . liJ o tr 1^ z o liJ ^ < p f^ ff. 8? I • • • • • •• • • , APPROX LIMIT FOR USE IN OSCILLATOR DETERMINED BY a, BY EFFECT ON FREQUENCY OR BY INTERNAL HEATING (CONTINUOUS OPERATION +135°F AMBIENT TEMP INCLUDED) -v^ \ • • • * < ;r .'•' •1 • m 1 t • « \ • • 0.4 0.8 2.4 2.8 1.2 1.6 2.0 AVERAGE a, ALONG LOAD LINE Fig. 7 — Plot of a versus collector to base resistance for representative group of transistors meeting 2N25 transistor requirements. TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY PULSING 1325 4 TIMES NORMAL LOAD Fig. 8 — Effect of load and value of feedback resistance on the level of the output. permits some deterioration in transistor characteristics before the change in output is sufficient to require readjustments. Variation in the absolute level of output due to variations in Va, (see Fig. 3) between transistors and to differences in coils is taken care of by the use of taps on the output windings. Harmonics of the fundamental frequency are created by the non- linearity of the transistor characteristics. These harmonics are accentu- ated by excessive feedback. The level of the harmonics for a representa- tive transistor are shown in Fig. 9. The effect of variations in feedback on the frequency is shown in Fig. 10. The shift is thought to be due to several factors all small. One is the lack of perfect coupling between the transformer windings. Another is 1050 1000 950 900 850 800 750 700 650 600 550 EXTERNAL FEEDBACK RESISTANCE IN OHMS Fig. 9 — Effect of feedback on harmonics with normal load. 1326 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 100.0 99.8 99.7 1050 1000 950 900 850 800 750 700 650 EXTERNAL FEEDBACK RESISTANCE IN OHMS Fig. 10 — Effect of feedback on frequency. 600 550 the leakage reactance of the input winding. As the time rate of change in current is increased by increased feedback these factors become in- creasingly greater although never very large. Fewer turns are required on the coils used for operation at the higher frequencies hence these effects are reduced. The output level of the oscillator will vary almost directly with the variations in the dc voltage since the amplitude of ac voltage across collector to base is almost equal to the dc voltage applied. Hence, a varia- tion of approximately 0.9 db will be obtained in the output when the central office battery is reduced from 50 volts to 45 volts due to power failure conditions. The over-all output variation from all causes is shown in Fig. 11. This is based on data obtained using the transistors having the distruibtion in characteristics shown in Fig. 7. Decreases in the value of Rl with temperature is normally compen- sated by a corresponding increase in a. However small positive or nega- tive voltage changes that alter the level of output do occur in the cut-off voltage. This is minimized by keeping the dc voltage as high as permissi- ble and still meet the 200 milliwatt dissipation limit for the 2N25 transistor. -2 DBM LEVEL MAX VARIATIONS DUE TO TEMP 40°F TO 135°F ±0.25 DB MARGIN TO TAKE CARE OF POWER FAILURE CONDITIONS 0.9 DB ACCEPTABLE LIMITS AT TRUNK -4 DBM LEVEL Fig. 11 — Output level with effect of various factors that may alter the level indicated. TRANSISTOR OSCILLATOR FOR MULTIFREQl K\(\ IMLSING 1327 The effect of the various factors mentioned before on the frequency of operation are shown in Fig. 12. Since several of the factors causing a shift in frequency were in the negative direction only, the adjustment hmits were set correspondingly higher. The over-all frequency variations could be reduced by reducing adjustment tolerances. The starting condition is important in this type of circuit since energy must be introduced into the oscillatory circuit before the dynamic char- acteristics of the circuit become effective. This means that the build up time is dependent upon the amount of energy introduced into the system at the start. In this application energy is introduced by the current which flows when the dc voltage is apphed to the collector circuit. The value of this current is largely dependent upon the collector to base dc resistance i.oif EFFECT OF TEMP ON COIL AND COND ±0.2% EFFECT OF LOAD -0.1% ACCEPTABLE ADJUSTMENT + 0.7% -0.3% "^NOMINAL EFFECT OF EXCESSIVE FEEDBACK -0.25% ACCEPTABLE LIMITS 0.99f Jl "^(700,900,1100,1300, 1500 OR 1700'V.) Fig. 12 — Frequency of output with effect of various factors that may alter the frequency indicated. which is in turn affected by the ambient temperature. This resistance is between approximately 4,000 and 20,000 ohms. The oscillograms shown in Fig. 13 illustrate this effect. Both are for the same circuit operating under normal conditions of adjustment. Qi for this condition is approximately 18. Oscillogram (a) is for the applica- tion of voltage in the normal fashion to the voltage divider circuit. The closure occurs at the point oscillation starts. Oscillogram (b) shows the build up obtained when no impulse is applied to start oscillation except for minor irregularities in the dc voltage applied. Starting is prevented in this case by a short circuit on 4-5 widing that was removed approxi- mately 2 ms after the start of the trace. The amolitude of oscillation is so low however for the first few cycles that the start is difficult to distin- guish. The build up time for (a) is approximately 27 milliseconds and for (b) it is approximately 37 milliseconds. The exponential build up of ampH- tude is modified greatly by the rapid change in the transistor's a as the voltage approaches the cutoff region. It should be noted also that oscillation would not have started under 1328 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Fig. 13 — Build-up of oscillation in resonant circuit. Normal operating conditions. the condition for oscillogram (b) with some transistors if the small emitter bias current had not been provided. TRANSIENT EFFECTS The trunk conductors are balanced with respect to ground. Voltages set up in these conductors due to electrostatic or magnetic coupling to the source of the interference will cause longitudinal currents to flow. An electrostatic shield in the transformer, shown in Fig. 1, effectively prevents such longitudinal currents from reaching the oscillator circuit. c ^f ■VA ' Fig. 14 — Equivalent resonant circuit. TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY PULSING 1329 If however the voltage becomes sufficient to breakdown one of the pro- tector blocks that are connected between each trunk conductor and ground, a voltage comparable to the breakdown potential of the pro- tector blocks (400 to 600 volts) would then be impressed across the out- put windings of the transformer. The usual cause of this (condition is lightning. However, neither artificially simulated lightning nor transi- ents of longer duration were capable of raising the voltage on the tank circuit to the point where a transistor was damaged. This is due both to R (SERIES) + 0 - SUM (-R) -H (+R) IS NEG SUM (-R) + (+ R) IS POS Fig. 15 — Relationship of shunt to series resistance. the high energy level of the tank circuit and the isolation furnished be- tween repeating coil and oscillator by the series resistors. summary: The transistor oscillator adequately fulfills the requirements for a source of current for multifrequency pulsing over telephone transmission circuits. Adjustments are provided so that the requirements for frequency stability, harmonic level and output level can be met with transistors having a wide range of characteristics. Sufficient margin is provided by 1330 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 design and by initial adjustment so that an appreciable change in transis- tor characteristics can be tolerated before readjustment is required. ACKNOWLEDGMENT The transformer used in the oscillator was designed by H. E. Vaiden and A. M. King. The transistor used was developed by R. J. Kircher and N. J. Herbert. D. J. Houck assisted in testing the circuit. REFERENCES 1. Lepage Seely, General Network Analysis, McGraw Hill, 2. Kurtz and Corcoran, Introduction to Electric Transients, John Wiley and Sons. 3. R. M. Ryder and R. J. Kircher, Some Circuit Aspects of the Transis- tor, B.ST.J., 28, pp. 367-401, July, 1949. Appendix I Transformers moly-permalloy dust core 1.57 O.D. 700- and 900-cycle operation Winding (1-2) = 500 turns Q at 700-cycles, 60 Winding (4-5) = 107 turns Q at 900-cycles, 68 Winding (4r-6) = 115 turns Winding (4-7) = 123 turns Winding (4r-8) = 220 turns Winding (1-3) = approximately 5560 (adjusted to meet inductance require- ments of 5.2H rb 1 per cent) 1100- and 1300-cycle operation Winding (1-2) = 372 turns Q at 1100-cycles, 73 Winding (4-5) = 80 turns Q at 1300-cycles, 75 Winding (4-6) = 86 turns Winding (4-7) = 92 turns Winding (4-8) = 164 turns Winding (1-3) = approximately 4080 (adjusted to meet inductance require- ments of 2.8H ± 1 per cent) 1500- and 1700-cycle operation Winding (1-2) = 305 turns Q at 1500-cycles, 82 Winding (4-5) = 66 turns Q at 1700-cycles, 85 Winding (4-6) = 71 turns Winding (4-7) = 75 turns Winding (4-8) = 135 turns Winding (1-3) = approximately 3360 (adjusted to meet inductance require- ments of 1.9H ± 1 per cent) TRANSISTOR OSCILLATOR FOR MULTIFREQUENCY iHiLSlNG 1331 Appendix II The equivalent circuit for the oscillator is shown in Fig. 14. For stable operation from equation (1), since /o = 1/^, 1 _ g- 2L = g- 2L _ 1^ The negative resistance, { — R), must therefore be equal to the positive resistance (-{-R). If however { — R) is active for only half the time, {—R) must be equal in magnitude to 2(+/2) in order to satisfy the requirements for equality. This assumes that boundary effects are negligible. This assumption was borne out by experiment. The equivalent resistance of the resonant circuit is ,..{^Jn.K'- R where k = (cooL)2. A plot of this relationship when positive and negative resistances are combined, is shown in Fig. 15. In the actual circuit the negative resistance is connected across the 1-2 winding. The 1-3 winding is in the resonant circuit. The equivalent resistance (i?o(i-2) of the resonant circuit across mnding 1-2, is de- termined as follows : _ (turns, winding 1-2)^ ^ ^'''■'' - (turns, winding 1-3)^ ^ ^^' ' Ferrites in Microwave Applications By J. H. ROWEN (Manuscript received May 26, 1953) Since Hogan's* exposition of the extreme usefulness of the microwave Faraday effect numerous other laboratories have begun investigating propa- gation through ferrites and have made significant contributions to the art. In view of the tremendous interest which is being accorded this work this paper has been prepared to summarize some of the observations and develop- ments to date. The plane wave theory is reviewed briefly with special atten- tion being given to the mechanisms by which power is absorbed by the ferrite. The plane wave theory is then modified to describe various waveguide effects. Finally experimental procedures and results are presented to illustrate the theory and to provide general information regarding the design of devices employing these effects. INTRODUCTION The ferromagnetic Faraday effect occurs at microwave frequencies as a direct result of the dispersion in permeabiUty which is associated with ferromagnetic resonance. The resonance can be explained most simply by stating that the total magnetization vector of a magnetized ferromagnetic material has associated with it an angular momentum arising from the angular momenta of all of the spinning electrons con- tributing to the magnetization. Because of this angular momentum (which is directed along the same axis as is the magnetic moment but in the opposite direction) the magnetization vector behaves as a top or gyroscope. If it is displaced from its equilibrium position in a steady magnetic field it will not rotate directly into alignment with the field but will precess about the dc field direction at a frequency determined by the strength of the dc field. In the absence of damping this preces- sion would continue indefinitely, but damping losses are such that the precession will damp out in approximately 10~* sec. * C. L. Hogan, The Ferromagnetic Faraday Effect at Microwave Frequencies and Its Applications — The Microwave Gyrator, B.ST.J., 31, pp. 1-31, Jan., 1952. 1333 1334 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 If an alternating field is applied at right angles to the dc field the magnetization will be driven in precession and when the driving fre- quency coincides with the natural resonance frequency as determined by the strength of the dc field a large amount of power will be absorbed from the driving field. Off resonance the power absorption is small, but the effective permeability seen by the driving field will go through a dispersion such as is exhibited by all resonant systems. With this model in mind we can proceed to discuss the phenomenon of ferromagnetic resonance and the ferromagnetic Faraday effect. INFINITE MEDIUM — LONGITUDINAL FIELD Polder^ has shown that because of the gyroscopic nature of the mag- netization a tensor permeability is required to relate the magnetic flux density and field intensity vectors in a ferromagnetic medium. At low frequencies the off-diagonal components of this tensor are negligible, and the tensor reduces to the ordinary scalar permeability. At frequencies above about 100 mc. these off -diagonal components can become signifi- cant depending upon the magnetic state of the material. When this tensor permeability is introduced into Maxwell's equations and a wave equation is derived for propagation in the direction of the applied mag- netic field we find that the normal mode solutions to the wave equation are two circularly polarized waves rotating in opposite directions. A solution in terms of linear polarizations is, of course, possible; but the result is more readily interpreted in terms of the circularly polarized waves. Furthermore, the propagation constant for either circular wave contains a simple scalar permeability, instead of the tensor required to describe the medium in general. Polder's tensor permeability gives the following relations between b and h when there is a static magnetic field along the positive z axis*. ^y The quantities by — jfloKhx + MO/x/ij, (1) hz = fidhz M = m' - Jm" (2) K = k' - jk" (3) are (complex relative diagonal and off-diagonal components of the tensor permeability. ^ D. Polder, Philosophical Mag., 40, p. 99, Jan. 1949. * Lower case letters are used for RF magnetic quantities. FERRITES IN MICROWAVE APPLICATIONS 1335 Equations giving ^ and k in terms of the applied magnetic field and the fundamental atomic constants are given by Hogan.^ These equations show that both ^ and k have a resonance at a value of effective internal field given by yH = 27r/. If H is in ampere turns per meter and /in mega- cycles the gyromagnetic ratio is 7/27r = 3.51 X 10~l A plane wave travelling in the z direction in an infinite medium de- scribed by equations (1) can be resolved into two counter-rotating cir- cularly polarized plane waves having propagation constants as follows: r+ = icoV/xoeo VeC/x - k) r_ = jojv^o Ve{n + k) (4) where the subscripts zb refer to positive and negative circularly polarized waves.* The terms inside parentheses are effective scalar permeabilities completely describing the medium for a circularly polarized wave. Cal- culated values of these effective permeabilities are plotted as functions of applied field in Fig. 1 for a ferrite operating at three frequencies. -2.0 :i -1.0 \ 1 1 1 ' ' 1 lr:~ — . ■— — 1^2 X 109 1 f-'^ f=~24~X~109 r- rtiT^ '.S^'Z === =r r^S 5S^ 3S= =« ^CHANGE IN SCALE f=8xi09 k24xi09 1 j'f = l2Xl09 1^12X109 L, L^ i ! = 34X,0»| ^^ "*N 1 j 8X)0^ ! u 1 ! 1 100 50 0 -50 , f = 8XI09 f= 12X109 f = 24X10* i I J i_ J i .. A 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 lO.OOO EFFECTIVE MAGNETIC FIELD IN OERSTEDS Fig. 1 — Calculated effective permeabilities for positive and negative circularly polarized plane waves computed for three different wave frequencies. 2 C. L. Hogan, Revs. Mod. Phys., 26, Jan., 1953. * A positive circularly polarized wave is one which rotates in the direction of the positive current producing the dc magnetic field. 1336 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 From these curves we may calculate the rotation per unit path length, the absorption of the positive component, the net insertion loss and the ellipticity of the resultant wave. The rotation per unit length is given by ; = i(^_ - ^+) (5) where ^± are the imaginary parts of the propagation constants, T^. Let us consider the special case in which the dielectric loss is zero. For con- venience we define the complex effective permeabilities seen by the circularly polarized waves as follows: M+ = M — K = /x+ — j/+ t . // /X_ = /X + K = M- — iM- The propagation constants may then be Avritten: (6) r+ = oiVmo \/\ [VI M+ 1 - m; + y Vi M+ 1 + m;] r ^'^ r_ = coViil^o y I [Vl M- I - Mi - iVi M- I + Mil . It is of particular interest to consider what happens to /3+ when /x^. becomes zero or negative. If we rewrite the expression for j8+: ^+ = ^ l/| ^^i4 + Mf) + m; (8) we see that, when ju+' is zero or negative, /?+ depends primarily on the magnitude of \i'\ for wherever ii" is negligible, /3+ is zero. Furthermore, we see that the attenuation constant, a+, given by: "^ = ! Vf ^^4 + Mf - m; (9) becomes dependent primarily upon /x/ when /x+' becomes negative so that we observe a significant attenuation long before \i^" becomes large. In Fig. 2 are shown the rotation of the plane of polarization of the linearly polarized wave and the absorption of the positive circularly polarized component of the wave. The dielectric constant of the ferrite was as- sumed to be 9.0, a typical value for many ferrites. From these curves it is evident that the wave will be elliptically po- larized whenever the effective field is large enough to make /x+' zero or FERRITES IN MICROWAVE APPLICATIONS 1337 N, \ ^, l.^ I \ ! ^3 \7 __---- — ■ — T ^ • • / / J II II ' — 1 1 5^ oz oz si 1 1 II / — , II /I II II M M /I / 1 1 ii / / 1 / «-^r^ £■-1 L 1 _^ '^ _1 1000 -2000 3000 4000 5000 6000 7000 EFFECTIVE MAGNETIC FIELD IN OERSTEDS Fig. 2 — • Rotation of the linearly polarized plane wave and absorption of posi- tive circularly polarized component versus effective static field computed from data of Fig. 1, negative, the amount of ellipticity depending in part upon the distance the wave has traveled in the ferrite medium. < PLANE WAVE, TRANSVERSE FIELD If a wave is propagated in either the x or y direction when the dc field is in the z direction then the wave equation has two orthogonal solutions representing linearly polarized waves. One of these is polarized with the electric vector parallel to the applied dc field and the other has the magnetic vector parallel to the applied dc field. When the magnetic 1338 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 vector of the wave is parallel to the magnetization the torque on the elec- trons is zero and the wave sees an isotropic dielectric medium with relative permeability equal to unity. However^ when the electric vector is parallel to the magnetization the magnetic vector is at right angles to it and can set the electrons into precession. Consider a wave propagating in the x direction in an infinite medium magnetized in the z direction. Let this wave be polarized so that it has components Eg and hy. When this wave enters the magnetized medium, hy exerts a torque on the magnetization vector M causing it to precess about the z axis. This results in both an rriy and an Mx component of alternating magnetization. There is, however, no component of b in the X direction because of internal demagnetizing fields arising from an effective volume distribution of magnetic charge as shown in Fig. 3. Such a volume distribution of magnetic charge arising from the periodic reversal in phase of the driving magnetic field is propagated through the medium at the velocity V = VJS An instantaneous picture of this distribution is shown in Fig. 3. The magnetic poles set up a magnetic field in the x direction throughout the material. This field is commonly called a demagnetizing field and for N S S ^ S S N N NS Ss^^S SN N NN SSS^SSS NN N S S s S s N N ^ ^ S ^ S ^ S ^ ^ N s s s ^NNSS^SSS^SSNN*^ N S s S s S N NN SSS^SSS NN N S ^ ^ S S N N ^ ^ ^ S ^ S ^ ^ " "^ ^ N N S s ^ S 3 S ^ S S N N '^ N S s 3 s S N NN SSSSSSS NN N S S S s s N = s ^ t = s ^ s s s s s N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N '^ N N S 3 S 3 S 3 S 3 S N ^ N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N N Fig. 3 ~ EflFective volume distribution of magnetic poles arising from phase reversals in transversely magnetized infinite medium. FERRITES IN MICROWAVE APPLICATIONS 1339 this particular case it can easily be shown that at every point in the material this field is given by* rtix and by definition Ox = Mo(^z + nix) = 0 In one sense this wave is no longer a plane wave as it has a com- ponent of h in the direction of propagation. However, the electric field, Ej and the magnetic flux density, b, are unchanged and remain the same as in a normal plane wave. The solution to the wave equation for the foregoing case yields a propagation constant 1/ <•■ =^ (10) in which the effective relative permeability of the medium is 2 2 Me£E = (11) The real and imaginary parts of this expression are plotted in Fig. 4 for three frequencies. These curves have the same general shape as those for the positive circular component of the wave propagated along the dc field direction, but here they apply to the entire linearly polarized wave. Again we have the possibility of zero or negative permeabiHty. In the region just above resonance the real part of the permeability takes on large values and maintains these even after the absorption curve is nearly zero. This suggests that it is possible to adjust the permeability to equal the dielectric constant of the material so that the medium matches free space perfectly. The medium then can be used as a switch by changing the field from the point where Mes = 0 to the point where ^^s = €-t In the region between zero applied field and saturation where the curve levels off, the effective permeability changes almost linearly. In the * We follow Stratton in making M an ^-like quantity rather than B-like. t In a waveguide /tefif must be adjusted to satisfy the condition that f^eS 1340 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 <3^ -3 240 200 •40 -I *-- CHANGE 1 "■ 1 1 . IN SCALE 1 1 f=€ JXIO^ f =12X10^ 24 X I09 i 1 f = 24 X109 Vi2XI09 ^ i i Lf V, "> ^ N \ 8X1 A 1 1 1 f = 8X109 f = 12 <109_ 1 I J J 160 120 2000 3000 4000 5000 6000 EFFECTIVE MAGNETIC FIELD IN OERSTEDS Fig. 4 — Effective permeability seen by a plane wave in a transversely mag- netized infinite medium. Real part above and imaginary part below. absence of dielectric loss the propagation constant of the wave is given by: /3 = w- ^'V^ K^) (12) Thus a variable phase delay is obtained by controlling the magnitude of the applied field. This delay is not necessarily accompanied by a change in attenuation so that an ideal phase shifter can be made using this effect. There are numerous other applications which can be made of the longitudinal and transverse applied field phenomena. Many of these will be discussed later in this article. LOSS MECHANISMS IN FERRITES AT MICROWAVE FREQUENCIES Neglecting dielectric losses, the plane wave theory predicts almost no loss at all for a negative circularly polarized wave and a single absorption line for a positive C.P. wave. In practice a more complicated behavior is observed, and to facilitate the discussion we show in Fig. 5 typical loss characteristics superposed upon the theoretical loss curve of Fig. I. We will enumerate the main points of interest before proceeding with the discussion. The specific differences in behavior are: FERRITES IN MK.'UOWAVE APPLICATIONS 1341 Curve A. Broad resonance absorption line. Curve B. A loss which disappears when the material is maRnotizod called ''Low Field Loss". Curve C. Loss which goes to zero for one component and rises for the other. Curve D. A loss which appears to be independent of magnetic Held over a wide range and can be related to the dielectric lo.ss tangent of the material, hence called dielectric loss. Curve E. Higher order modes causing erratic v^ariation- in l(».s.s. Curve F. Double peaks due to "Cavity Resonances". Qualitative and semiquantitative explanations have been developed to explain all of these phenomena. Some of them follow from a simple extension of the plane wave theory and the rest are based upon con- siderations of the special case of a partially filled waveguide. Curve A J Fig. 5 Associated with the precessional resonance there is a damping term by which power is dissipated in the lattice. The exact nature of this damping term is not fully understood, and measured line widths are always greater than those predicted by present theory. Nevertheless we have at our disposal empirical damping constants which can be used to predict resonance absorption losses as was done in the calculation of the curves of Figs. 1 and 4. These apply, however, only to small ellipsoidal samples which are ground from single crystal ferrites. In polycrystalline ferrites the ab- sorption line is generally broader for three reasons, namely; crystalline anisotropy, strain anisotropy and varying internal demagnetizing fields due to the variety of shapes of the constituent crystallites. Many ferrites have a high crystalline anisotropy which behaves in -CIRCULAR POLARIZATION ^CIRCULAR POLARIZATION MAGNETIC FIELD Fig 5 — Typical loss characteristics encountered in the measurement of van- ous ferrite samples in cylindrical waveguides with longitudinal static field. 1342 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 N N N N N N s s s s s s WITH FIELD APPLIED NO FIELD APPLIED Fig. 6 — A typical domain wall pattern showing the movement of the wall in response to the application of an external magnetic field. many ways like an internal field tending to keep the magnetization of the constituent crystallites along one of the axes of easy magnetization. In most ferrites there are four such axes and since the magnetization can be in either direction along any one of these, there are eight directions of easy magnetization. When a field is applied the effective internal field is roughly the vector sum of the applied field and the anisotropy field associated with the nearest axis. In a polycrystalline ferrite composed of many randomly oriented crystallites it is evident, therefore, that the internal field varies from point to point in the body so that the resonance absorption line is broadened by an amount proportional to the mag- nitude of the anisotropy field. A similar broadening can arise from magnetostriction due to fields arising from varying strains throughout the polycrystalline ferrite, and non-uniform internal demagnetizing fields due to the shape of the constituent particles or crystallites can like^vise broaden the resonance line. Since a broad resonance line results in el- lipticity of the transmitted wave it is desirable to use a ferrite having low anisotropy. Curve By Fig. 5 Frequently a loss is observed at low fields as indicated in Fig. 5 by Curve B. Neglecting waveguide effects this hump is S3nnmetricaf so that it evidently depends on a phenomenon which affects both circular components equally. It is generally agreed that it depends upon the existence of domain walls within the material since it usually disappears as soon as the body is magnetized. However, there is some question as to the specific mechanism involved. • Fox and Weiss, Revs. Mod. Phys., 26, p. 262, Jan., 1953. FERRITES IN MICROWAVK APIM.K ATIoNS 1343 A ferromagnetic crystal consists entirely of regions called domains which are completely magnetized along one of the directions of easy magnetization. In general the direction of magnetization of these do- mains is varied in an orderly manner as shown in Fig. 6 so that the energy of the crystal as a whole is a minimum. In the region Ix^tween adjacent domains there is a (usually) narrow wall in which the mag- netization goes through a gradual change in direction from that of one domain to that of the other. When an external field is applied the mag- netization of the crystal is increased by the growth of some domains at the expense of their neighbors. When the crystal is saturated substan- tially all of the walls have disappeared and the material behaves as a single large domain. There are currently two proposed mechanisms by which these domain walls could cause a loss at low fields. Becker and Doring* have shown that there can be associated with the motion of a domain wall either relaxation or resonance frequencies. Galt^ has measured relaxations in a single crystal of magnetite at 3,000 cps and in a single crystal of nickel ferrite at approximately 2.5 mc and has presented a rather convincing argument that these are due to domain wall motion. In general the relaxation frequency would be expected to occur far below the micro- wave frequencies, but resonances could conceivably occur at micro- wave frequencies and could be quite broad. Until recently no other theory had been advanced which would explain the losses so often observed at low fields, and these were, therefore, attributed to a high frequency domain wall resonance. There is a more satisfactory explanation which has recently been stated in different ways by Rado* and by Smit and Polder.' Rado has observed a resonance absorption in the microwave region with zero applied field and has shown from temperature dependence that the fre- quency of this resonance depends upon the saturation magnetization and the crystalline anisotropy of the ferrite. Smit and Polder have presented a model by which we can see how both of these quantities can enter to produce a loss at low fields.' We consider an ellipsoidal crystallite as shown in Fig. 7. The domain structure shown is one which could exist in some crystallites in a polycrystallme ferrite. The magnetization in domains numbered 1 will respond to right cir- cular polarization and the others to left circular. In other words a wave rotating clockwise is positive circularly polarized in domains one while a 4 Becker and Doring, Ferromagnetismus, Sringer, Berlin, 1939. 6 J. K. Gait, Phys. Rev., 86, Feb 15, 1952 • G. T. Rado, R. W. Wright, et al., Phys. Rev Nov^2. ' D. Polder and J. Smit, Revs. Mod. Phys., 26, pp. 89-90, Jan. 1963. 1344 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 wave rotating counterclockwise is positive circularly polarized in domains numbered 2. In the absence of any other effects the resonance fre- quency of all of these domains would be determined by the anisotropy field. However, if we excite both circular polarizations simultaneously and if the relative phase of the two circular polarizations is as shown in Fig. 7(a) poles will be set up at the domain walls as indicated in the figure. The demagnetizing fields associated with these will cause the resonance for both circular components to occur at a frequency given by: / = T^cff ^ yM, (13) On the other hand, if the phase of the two circular polarizations is as shown in Fig. 7(b), no poles will be set up on the walls and the re- sonance will be determined primarily by the anisotropy field. / = T^eff ^ yH, (14) These two examples of the relative phase of the circular waves cor- respond to linear polarizations in the x and y directions respectively. This simplified derivation gives the maximum and minimum fre- quencies at which resonances can occur. In a material containing a large number of randomly shaped and randomly oriented crystallites, res- onances can occur at all frequencies between these limits. Most ferrites have values of M^ between 80,000 and 240,000 and anisotropy fields which probably range from 8000 to 80,000 amp. turns/meter with per- (a) POLES ON WALL ^LINEAR POLARIZATION^ V IN X DIRECTION , (b) NO POLES ON WALL ^LINEAR POLARIZATION' \ IN y DIRECTION Fig. 7 — Model used by Smit and Polder to illustrate their theory for the "low- field loss." FERRITES IN MICROWAVE APPLICATIONS 1315 haps a few which are higher still. Therefore, we have as typical fre- quencies /i = T^^anis ^ 300 to 3,000 mc (15) and /2 = yM, ^ 3,000 to 9,000 mc. (16) It is evident that at 9,000 mc only that loss associated with M, will contribute to the "Low Field Loss" and since the mechanism depends upon the existence of domain walls it will disappear when the material is saturated in agreement with our observations. Curve C, Fig. 5 Some ferrites exhibit a low-field loss which disappears at saturation for only the negative component and increases for the positive com- ponent. This is thought to be due to an effective anisotropy field in the material. In order for such behavior to be present at 9,000 mc, however, the internal field must be of the order of 240,000 ampere turns per meter. While such a value of crystalline anisotropy might be found in cobalt or other high anisotropy ferrites it appears to be somewhat too high for a nickel-zinc ferrite such as that in which this characteristic was first ob- served. However, high internal fields could result from demagnetizing effects similar to those discussed under item B but differing in that the poles are set up on nonmagnetic grain boundaries instead of domain walls. These, of course, would persist when the body is saturated, but there would be loss for only one circular component inasmuch as all of the crystallites are then magnetized in the same direction. Such a loss characteristic can be quite useful where one wishes to absorb one circular component selectively without the necessity for applying a large dc field. Curve D, Fig. 5 Dielectric losses are present in all of the ferrites which have been made to date, although in some materials this loss is very low. Low dc con- ductivity in itself is not a sufficient criterion of the dielectric properties of a material as some ferrites appear to consist of conducting regions surrounded by an insulating matrix, and these have fairly high loss tangents at microwave frequencies. The major mechanism of dielectric loss involves the exchange of electrons between ions in the crystal lattice. It has been found that the presence of ions of the same metal in different valence states on the same 1346 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 lattice site gives rise to high conductivity and hence high dielectric loss. Conversely, when a ferrite is carefully prepared so that all of the con- stituents are present in exactly stoichiometric porportion, and when the possibility of multiple valence states is eliminated the conductivity is very low. To illustrate this point a series of measurements is reported in which the iron content of theferriteswas carefully varied about stoichiom- etry in a nickel-zinc ferrite. These measurments are discussed at a later point in this paper. A ferrite is made by reacting a mixture of metallic oxides at a tempera- tures below the melting point of these oxides. As the oxides react a new crystal structure is evolved in which the metallic ions occupy positions in the interstices of a close-packed oxygen lattice. There is a very well authenticated theory due to Neel^ explaining the way in which the spin orientations of the ions are distributed in the two types of lattice site which exist in the Spinel oxygen lattice. Whenever metal ions in more than one valence state occupy the same type of site, e.g., the octahedral position, there is a possibility for the easy transfer of an electron from one to the other since the crystal structure is unchanged by the trans- fer.* In the case of nickel ferrite which has the composition NiOFe203 an excess of iron will tend to replace some nickel atoms by entering the lattice in the divalent state. Since the remainder of the iron is trivalent, comparatively high conductivity is observed. The problem of producing ferrites with extremely low dielectric losses appears to be fairly well understood and is progressing satisfactorily. By choosing the proper set of metal ions to insure the absence of multiple valence states and by maintaining the proper oxygen stoichiometry one may be able to achieve loss tangents as low as 0.001. The subject of dielectric losses is well covered in the literature.^' ^° Curves E, F and G, Fig. 6 The loss mechanisms indicated in Fig. 5 by Curves E and F all arise from the particular behavior of ferrites in waveguides as differentiated from the plane wave theory. For example, the erratic behavior indicated by Curve E has been shown to be due to the presence of higher order modes in the ferrite region in a waveguide, and the subsidiary hump on the absorption Curve F has been shown to be a "cavity resonance" which is strongly dependent • L. Neel, Physica, 16, pp. 350-53, 1950, and Zeit. Anorg. Chem., 262, pp. 175- 184, 1950. » E. J. W. Vcrwey and J. II. DeBoer, Rec. des Travaus Chemiques des Pays- Bas, 56, pp. 531-54, 1936. 10 E. J. W. Vcrwey et al., Phillips Res. Reps., 5, pp. 173-187, 1950. FERRITES IN MICROWAVE APPLICATIONS 1347 upon the diameter of the ferrite cylinder and upon the guide wavelengths but not upon the length of the cylinder. Other waveguide effects causing anomalous loss behavior, such as shown by curve G, have been discussed by Fox and Weiss" and \vill be treated by them in greater detail in a forthcoming publication. In order to discuss these effects more fully we must examine the modi- fications of the plane wave theory which must be made to explain the behavior of a ferrite in a waveguide. WAVEGUIDE THEORY, LONGITUDINAL FIELD When a piece of ferrite is placed in a waveguide and magnetized it is necessary to modify the foregoing plane wave theory to describe the behavior of a wave passing through the ferrite. Because of the anisotropic nature of the magnetized ferrite it is necessary to obtain a solution to the specific problem of the waveguide containing the ferrite. When the mag- netization of the ferrite precesses about the applied dc field it sets up components of h which do not exist in any of the classical modes, and unless one can deal with small perturbations the solution becomes quite involved. The modes which can exist in the ferrite will often resemble the clas- sical modes so that for convenience we will refer to them as modified TE or TM modes. Suhl and Walker^^ have obtained solutions for the case of a cylindrical waveguide completely filled with a magnetized ferrite, and they have shown that the modified dominant TEn mode behaves much like the plane wave in the region of small fields but that the behavior of the TM modes cannot be approximated by a simple extension of the plane wave theory. A waveguide large enough to sup- port the dominant mode when filled with air will, when filled with ferrite, support three or four higher order modes including some of the modified TM modes. In this case, it is possible to have several present at the same time with the result that observations of rotation, loss and ellipticity are almost impossible to interpret. Accordingly, we should reduce the size of the waveguide in the ferrite-filled region, and this involves the creation of discontinuities in the waveguide. This is not always necessary, however, because higher order modes will not always be set up in the ferrite even though the waveguide is large enough to propagate them. If care is taken to avoid geometries which favor a given mode the prob- " A. G. Fox and M. T. Weiss, Revs. Mod. Phys., 26, p. 262, Jan., 1953. 12 A preliminary report of this work has been published in the form of a Letter to the Editor by H. Suhl and L. R. Walker in Phys. Rev., 86, p. 122, 1952. A more detailed account is scheduled for publication in the J. Appl. Phys. 1348 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 ability of its occurrence will be greatly reduced. In particular it has been found that a flat-ended cylinder completely filling the waveguide can be introduced into the full-sized waveguide without mode complications, but Fox and Weiss^^ have shown that putting conical tapers on the ends will favor the establishment of the modified TMn mode. In most ap- plications of the Faraday effect the ferrite element is in the form of a very thin pencil at the center of the waveguide so that the mode problem is greatly simplified, but in order to obtain quantitative fundamental information about ferrites themselves it is often necessary to work with a completely filled waveguide. In such cases considerable care must be taken to insure the validity of the measurements. One method of making impedance measurements which has been used successfully by H. Suhl is to cut a shallow longitudinal slot in the cylinder of ferrite and to make standing wave measurements directly in the medium. Because the slotted section is filled with ferrite the size of the waveguide can be reduced to insure the presence of a single mode. This is restricted to unmagnetized materials as rotation of the plane of polari- zation would result in radiation by the slot. Another suggested procedure is to make a transformer from full-size rectangular waveguide to circular waveguide of diameter equal to c?/\/e where d is the diameter of a dominant-mode air filled pipe and e is the relative dielectric constant of the ferrite. This transformer can be treated as a four terminal impedance transformer and its network impedances can be determined by measurement. Impedances measured in the air filled guide will have to be transformed through this network to obtain the true impedances of the ferrite-filled guide, but this can be done if the need for the measurements warrants such effort. An exact solution of the partially filled waveguide is considerably more difficult to obtain than the solution for a completely filled waveguide. Yet this geometry is the one usually used in most practical applications of the Faraday effect. In the absence of an exact solution one must develop simple physical explanations based upon plane wave theory plus intuition for numerous observed phenomena. A theory for the partially filled longitudinally magnetized waveguide can easily be de- veloped from two simple observations. First, we consider the circular components of the wave separately and observe that each sees an effec- tive scalar permeability which is a weighted average of the permeability of the pencil for that component and that of the surrounding medium, and second we postulate that a small enough pencil will not act as a dielectric rod waveguide and will merely create a small perturbation of »» A. O. Fox .irid M.T. Weiss, Revs. Mod. Fhys., 25, p. 262, Jan., 1958. FERRITES IN MICROWAVE APPLICATIONS 1349 the original mode. When the above assumptions are valid the plane wave theory can be extended easily to explain loaded waveguide behavior. In the plane wave case it was shown that a negative effective permeability results in attenuation of the positive circularly polarized wave. Quite a different result is observed in waveguides containing very small cylinders of ferrite. It appears that if the cylinder is small enough n(.i lo art as a dielectric waveguide then the negative permeability inside the rod simply is averaged with the permeability of the surrounding rof-ion so that \\w rotation curve (which depends upon the difference between the s(|iiare roots of the permeabilities seen by the two circular components of the wave) follows the dispersion curve of the permeability of the pu.siiive circularly polarized wave, even following the pattern of the permeability when it is negative. In Fig. 8 are shown measured curves of the rotation of the wave and the absorption of the positive circularly polarized component of the wave as functions of applied dc field for comparison with Figs. 1 and 2. To amplify our arguments we point out that the propagation constant in a waveguide containing a very small pencil of ferrite is of the form: where /3e is the propagation constant of the empty guide ri is the radius of the ferrite cylinder ro is the radius of the waveguide € is the dielectric constant of the ferrite and juj. are the effective complex permeabilities (n db k) A is a constant = 3.2 We see that the expression within the brackets is finite for all values of M± except iLi4. = — 1 + jO. Accordingly if the damping parameter (or line width) is large enough to insure that fi+ can never take on this value, there will always be a cylinder radius, r', for which the perturbation term is small relative to /?/. However, the cylinder diameter does not have to be very large before the above arguments fail and the rotation and loss behavior become quite different. The cutoff wavelength of the TEn mode in round guide is Xe = 0.1708 d Vm^ where d is the diameter in centimeters of the waveguide and ^ and € are the effective relative permeability and dielectric constant of the medium contained therein. When a cylinder of ferrite is placed at the center of 1350 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 20 16 ^ 12 UJ oi a. „ o 8 UJ o ? 4 z (b) 1 1 / \ / I / v b i K N ^^^ b ( (a) J^ / > k ^^ ^^ , , — < . = r ■ r 1 l_f3— ^ ^^ J /I ly [^ / ^ 1260 1680 2100 FIELD CURRENT IN OERSTEDS 2520 2940 3360 Fig. 8 — Rotation of the plane of polarization and absorption of the positive component as functions of applied dc field for a wave propagating through a waveguide containing a very small cylinder of ferrite. The inflection of 2,400 oersteds is due to a "cavity resonance" as discussed later. the waveguide the cutoff wavelength is increased because the effective value of € is increased. If the ferrite cylinder is quite small this alteration will be unimportant, but if the diameter of the rod exceeds about one- quarter that of the air filled dominant mode guide this mechanism can lead to the existence of higher order modes for the negative component. When this happens the plane wave theory obviously cannot be expected to apply, for the presence of multiple modes in the propagation of either component will introduce an additional variable not present in the plane FERRITES IN MICROWAVE APPLICATION'S 1351 wave theory. Experimentally one will observe very erratic and fre- quency-dependent behavior under these conditions. WAVEGUIDE THEORY — TRANSVERSE FIELD A waveguide, either round or rectangular, filled with ferrite and mag- netized by a field parallel to the electnc vector of the dominant mode will exhibit a behavior qualitatively the same as described in the plane wave theory of the transverse field. In fact, it has been shown that in a rectangular guide all of the TEo^ modes can exist with only slight modi- fication.^^ That this result is probable may be seen from the fact that the precessing magnetization vector sets up components of h in the x and y directions when the applied field is in the z direction, and both of these components normally exist in the TEon modes. The primary modification of the mode arises from RF demagnetizing fields in the ferrite. Because of this modification it is extremely difficult to match the boundary con- ditions for normal incidence at an interface between the ferrite and air in the waveguide. An infinite series of modes is actually required, but in practice the mismatch due to magnetic effects is usually not very large. If one matches the dielectric constant by means of tapered dielectric horns the remaining mismatch is slight except where ti^u approaches zero and at resonance. While the completely filled waveguide magnetized by a transverse magnetic field parallel to the electric vector of the wave will exhibit a reciprocal behavior in respect to phase change and attenuation, an in- teresting and potentially useful modification of these effects occurs when a small piece of ferrite is located asymmetrically in a waveguide. Chait and Sakiotis^^ of the Naval Research Laboratory and Turner of the Holmdel laboratory of Bell Telephone Laboratories have independently observed a phase shift which is dependent upon the direction of propaga- tion of the wave, and a simple explanation of this effect has been made by Turner^ ^ and by Kales. ^^ Suhl and Kales^^ have shown the theoretical validity of this explanation. The idea can be demonstrated by considera- tion of the field configuration shown in Fig. 9. An observer at the point P will see an h field which is elliptically po- larized in a plane normal to the direction of Ha- The sense of the rotation of the larger circular component of the ellipse will depend upon the direction of propagation of the wave. Thus for one direction the major ^''A A. van Trier, Paper presented orally at meeting of Amer. Phys. Soc., Washington, D. C, April, 1952. 16 M, L. Kales, H. N. Chait and N. G. Sakiotis, Letter to the Editor, J. Appl. Phys., June, 1953. 16 E. H. Turner, Letter to the Editor, Proc. I. R. E., 41, p. 937, 1963. 1352 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 part will be a positive circular polarization with respect to the applied Ha and will experience a decrease in /*' and for the other direction of propagation the h field will be primarily negative circularly polarized and the wave will experience an increase in /x . Since the h field is linearly polarized in the transverse plane at the center of the guide and is linear in the longitudinal direction at the edge of the guide it is evident that there is a point in between where the differential phase shift is maximum. Suhl has shown that this point always occurs halfway between the guide wall and the center regardless of the proximity of cut-off. Obviously one has merely to adjust the length of the sample and the strength of the magnetic field so that the differential phase shift is 180° and he will have a gyrator. Then it is an easy matter to design a cir- culator, isolator, or any of the numerous devices depending upon the gyrator action/^ EXPERIMENTAL PROCEDURES AND RESULTS In order to verify the theory and to determine the optimum per- formance obtainable in microwave devices employing the Faraday Ro- tation an extensive measurement program has been set up. Information of both a fundamental and of a practical nature is obtained through a variety of measurements. From a practical point of view we are interested fp- 1 f ^ Ji 9 ~ I^ec \ |IS 70 ^ I CO 60 ai (D O 55 S50 <0 11145 o z h- S 35 -30 0 u. 20 15 10 5 \ sis \ 1 \ \ 1 \ i \l \ j\ |\ K, S 0 ^ ^ ■0 -0.01 0 0.01 0.02 0.03 IRON EXCESS OR DEFICIENCY IN ATOMS PER MOLECULE Fio. 19 — Factor of merit versus iron content of NiZn ferrite. FERRITES IN MICROWAVK A IMM.K \ ri()\> 13()5 netized. Since these latter effects are quite new two measurements of this type are shown to illustrate the sort of performance obtainable. In the first experiment a' slab of ferrite 20 mils thick was placed successively in several positions in a rectangular w^aveguide, and the phase shift as a function of field was measured for both directions of propagation. When the plate of ferrite is centrally located the phase shift is the same in both directions but when the ferrite is placed half way between the center and the edge of the guide a large difference in phase shift is observed. Finally when the slab of ferrite is located at the edge of the guide there is only a small difference between the positive and negative phase characteris- tics. This small difference will vanish as the thickness of the ferrite plate goes to zero. In Fig. 21 (a) we show the phase shift versus applied field for three positions of the ferrite slab and in Fig. 21 (b), the differential phase shift at a constant value of applied field is plotted against the posi- tion of the ferrite plate. The solid line curve taken at 9,500 mc indicates that maximum differential phase shift is obtained when the ferrite is located approximately 0.100'' from the guide wall. Suhl's prediction is that the position at w^hich maximum differential phase shift is observed should be independent of frequency. The dotted curve in Fig. 21 (b) taken at 8,200 mc verifies this part of the prediction in that the maximum again occurs where the ferrite plate is 0.100" from the guide wall even though the point at which h is circularly polarized has been shifted sig- nificantly by the change in frequency. The second measurement was designed to measure the non-reciprocal absorption which is obtained when the strength of the dc magnetic field is adjusted so that the ferrite is at ferromagnetic resonance. In order to obtain the minimum forward loss and maximum reverse loss it is essential that the ferrite be located precisely at the point where the transverse and longitudinal components of the h field of the wave are equal, i.e., where the h field is circularly polarized in a plane perpendicular to the magnetic field. Since this condition exists at only one point in the half- waveguide the ferrite slab must be made very thin. Measurements were made in the 6000-7000 mc band in RG50 w^aveguide. A thin plate of "Ferramic G" was cut so as to extend from one broad wall to the other Its length w^as approximately 1-}^'' and its thickness was originally 0.050" and was subsequently reduced to 0.025" and finally to 0.009". In the last case the ferrite was so fragile that it was necessary to support it by cementing it to a }i-mch plate of polystyrene. For convenience the ferrite plate was fastened securely in place at a point calculated to be the point where the h vector is circularly polarized at the center of the band, and frequency was varied about this center frequency. At each 1366 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 80 •100 ■120 -140 ■160 200 175 150 Z 125 100 75 50 25 200 300 400 500 600 700 APPLIED FIELD IN OERSTEDS (b) ->\ |<-t= 0.020" Pv 1 k, Xk- \^ \ \ \ \ it N N^- 6200 MC 9500 MC- K \ \ / X \ ^ / \ ^ ia^ 0.100 0.200 0.300 0.400 SPACING, X, FROM GUIDE WALL IN INCHES Fig. 20 — Phase shift versus applied field and differential phase shift versus position of the ferrite plate. FERRITES IN MICROWAVE APPLICATIONS 1367 frequency the dc magnetic field was adjusted so that absorption was a maximum for the wave for which the absorption is the greater. The field was then reversed and the absorption was again measured. Since reversmg the field is entirely equivalent to reversing the direction of propagation we obtain in this way the insertion loss for both directions of propagation. Data taken in this manner are shown in Fig. 22 along with a sketch of the geometry employed. The VSWll was measured in the case of the thinnest plate and was found to be less than 1 05 to 1 over the band. The variations in the dotted curve, however, are probably due to reflections from the ends of the sample. One must bear in mind that as the frequency is changed, the field required for ferromagnetic resonance is changed so that these data do not give a true index of the bandwidth of the device. This must be measured at a constant value of field. However, if we had used a ferrite 36 32 28 § '6 CD < H^t t = 0.050" C >-— — — o 1 1 t = 0.009" c 1 ^ 13 /e"U o- .o-'" ^--o "~~.»^ — -o '''' ,- — o- ^N D -o- ' -H O" 0.009" PLATE SUPPORTED BY Vs" POLYSTYRENE SLAB '^. "-.. "^'rc 5RWAF ^^— o — — o o '--..^^ ^^x^ o 1 0 u ,u n 6200 6400 6600 6800 7000 FREQUENCY IN MEGACYCLES PER SECOND Fig. 21 — Absorption of forward and reverse waves at the optimum value of field as functions of frequency. Two sample thickness are shown to illustrate the effect of sample thickness. 1368 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 having a loss characteristic such as is shown in Fig. 5, Curve C, in which the loss differential becomes great as soon as the material is saturated, then the bandwidth would be as shown in Fig. 21. Of all the materials which have been measured in the past two years only one sample has shown this effect. At the time such behavior was regarded as the an- tithesis of the ideal, and no further investigation was made. Now that there is a use for such a material, effort is being directed toward maximiz- ing the effect. Since the effect is thought quite definitely to be due to a very high effective anisotropy field arising either from crystalline aniso- tropy or demagnetizing effects, the problem of creating the proper ma- terial should be quite straightforward. At this point we should consider the relative merits of the transverse field non-reciprocal devices and those employing the Faraday rotation. We have seen that it is possible to construct a simple isolator using the non-reciprocal absorption in rectangular waveguide. Such an isolator has a minimum forward loss of more than 0.5 db for 20 db reverse loss where an isolator employing the Faraday Rotation can be made with less than 0.1 db forward loss for 30 db return loss. On the other hand the transverse field isolator is much simpler, more compact and easier to .FERRITE 'DC -FERRITE LOSS FORWARD RETURN LOSS FORWARD RETURN 10,000 6000 4000 0.7 DB 0.8 DB 22 DB 22 DB 10,000 6000 4000 0.1 DB 0.2 DB 25-30 DB 25-30 DB ISOLATORS ^FERRITE CIRCULATORS Fig. 22 — Comparison of two basic non-reciprocal elements of the Faraday eflfect and transverse field effect types. FERRTTES IN MICROWAVE APPLK ATIONS 1369 match impedance-wise. To illustrate the comparison between compar- able devices using each principle we have prepared a figure showing the relative structural complexity and performance standards of each. This comparison is shown in Fig. 22. In general the transverse field devices which depend on differential phase shift have about the same insertion loss as those employing the Faraday effect, while the transverse field devices which depend upon differential absorption are somewhat more lossy but simpler to construct than the corresponding Faraday Ro- tacion devices. SUMMARY This paper has reviewed the plane wave theory, extended it to discuss waveguide effects, analyzed the various loss mechanisms present in fer- rites at microwaves, and discussed numerous measurement techniques and results. It is known that there are original papers in preparation in Bell Telephone Laboratories, and possibly elsewhere, which make this review incomplete at the time of writing. However, it is hoped that the information summarized herein will be of assistance to those who are seeking orientation in this new and rapidly expanding field. In conclusion the author wishes to thank A. G. Fox, M. T. Weiss, J. P. Schafer, H. Suhl, A. D. Perry and L. R. Walker for permission to discuss herein some of their work and ideas which have not previously been published. The cooperation of F. J. Schnettler and L. G. Van Uitert in providing us \vith a mde variety of excellent ferrites and in suggesting useful variations in ferrite properties has been of tremendous help in the advancement of the art. The advice given by C. L. Hogan, J. K. Gait and H. Suhl in discussions has proved invaluable and finally the author wishes to thank J. L. Davis for the able assistance he rendered . Cold Cathode Tubes for Transmission of Audio Frequency Signals By M. A. TOWNSEND and W. A. DEPP (Manuscript received August 13, 1953) Cold cathode gas filled tubes have been extensively applied as electronic switching elements in the telephone system. In general, these applications have been limited to control circuits. The usefulness of these tubes can be further extended by making them capable of carrying voice frequency signals. The transmission properties that are required of the tube for this use are considered. It is shown that troublesome oscillatory noise can be eliminated and that the insertion loss of the tube can be reduced to a low value. Fur- thermore, by a special design of cathode a stable insertion gain of a few db may be realized. Other requirements on bandwidth, power and distortion are satisfactorily met. Thus, these tubes are potentially useful in coordinate type switches in which voice frequency signals must be rapidly switched. INTRODUCTION Cold cathode glow discharge tubes have found increasing usefulness as two-valued switching elements in telephone and other automatic con- trol systems. In many applications the tube functions as a simple control element which either does or does not pass current, depending on the control signals applied. The output signal is in the form of a voltage or a current which can be used to trigger other tubes or operate relays.^ In other types of circuits the glow discharge tube may be used as a switch in series with a transmission path for audio frequency signals. When used for this purpose, the tube not only must fulfill the SNNitching requirements but also must meet an additional set of requirements which may be realized by controlling the dynamic properties of the dis- charge. This paper describes some of the characteristics of cold cathode tubes which have been developed for use as switching elements in series with voice frequency telephone transmission circuits. TRANSMISSION REQUIREMENTS OF AN ELECTRONIC SWITCH When an electronic switch is used as a substitute for a pair of metallic contacts, a number of requirements must be met in order that voice 1371 1372 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 frequency currents be faithfully transmitted. These are as follows: 1. Gain or Loss — The impedance level of telephone voice frequency circuits is usually of the order of several hundred ohms. The impedance of any device inserted in this circuit should be sufficiently small com- pared to this value in order to restrict the insertion loss to less than 1 or 2 db. Of course, the impedance levels of the circuits may be raised to higher values. Aside from the extra cost of transformers, the problems of noise pickup and crosstalk become bothersome when this is done. Since no amplification of the signal from the telephone set is normally used in local transmission circuits, any value of loss is highly undesirable. In fact, since the use of the electronic switch may require the introduc- tion of other circuit elements which introduce loss, it would be desirable that the electronic switch provide a small amount of gain. The above discussion has considered the gain or loss of the switch in its *'closed" condition. Between two busy circuits, there may exist paths consisting of one or more ''open" switches. Each of these paths may con- tribute crosstalk into the circuits. Therefore, impedance required of an individual switch must be high. For a large coordinate switch, there will be a very large number of these undesired paths. This requires that the open impedance of the individual switch be of the order of hundreds of megohms. 2. Bandwidth — ^It is desirable that the electronic switch transmit faithfully frequencies of 300 to 3,500 cycles per second. 3. Power Output and Distortion — ^ Since the impedance of the elec- tronic switch may vary with the current passing through it, distortion may be introduced when the current savings are large. On the other hand, without the proper current swing, insufficient power Avill be delivered to the load impedance. Telephone circuits need to handle powers of the order of a few milliwatts with a harmonic distortion less than a few per cent. 4. Noise — ^The noise introduced into a telephone switching system by an electronic switch should not be noticeable to a subscriber. This means it should be below 10~^ or 10"^ watts. 5. Stability — The properties that have been considered above must be highly stable with time. In central office use, such devices might be used for periods of ten to forty years. STATIC CHARACTERISTICS A common form of glow discharge tube comprises a pair of metal electrodes in a glass envelope which is carefully evacuated and filled with COLD CATHODE TUBES FOR A TDK) FREQUENCY SIGNALS 1373 a chemically inert gas to a pressure ranging from 1 to 100 mm of mercury. The negative electrode, called the cathode, is given a special processing which permits it to emit electrons readily when bombarded with positive ions of the gas. The positive electrode, the anode, serves to collect elec- trons emitted by the cathode as well as those produced in the gas by ionization. A typical voltage-current characteristic is shown in Fig. I . For discus- sion purposes we may divide the curve into several current ranges. In current range I a small residual current flows even at low voltages be- cause of ionization resulting from cosmic rays or radio-active material placed in the tube. The two curves in range I are for different residual currents. At higher voltages this residual current is amplified as a result of additional ions and electrons formed in the gas but it is still extremely small. If the tube voltage is increased still more, the current increases very rapidly until in current range II, a self-sustaining discharge is es- tablished. Each electron released from the cathode gains enough energy on the way to the anode so that it produces a large number of positive ions, excited atoms, and photons in the gas. When these particles, arriv- ing back at the cathode, on the average, release another electron, the 260 240 220 200 lU CD ^ 180 <0 IT) I 1 n 1 m 1 1 1 1 m 1 ¥ 1 1 ^BREAKDOWN. -Ur- 1 1 - 1 1 1 / ^ 1 \ 1 / 1 / 1 1 \ / 1 \ i OJ 1 \ O 140 ^ 120 1 1 1 \ 1 / \ / 100 NOR GLC MAL )W ABNO 1 GLC RMAL 5W 80 60 — 1 1 1 10- 10-2 10"® io-« 10-* TUBE CURRENT IN AMPERES Fig. 1 — Volt-ampere characteristic of cold cathode diode. 1374 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 self-sustaining condition has been achieved. The value of voltage at which the discharge becomes self-sustaining is usually referred to as the breakdown voltage. A detailed discussion of current ranges I and II has been presented by Druyvesteyn and Penning.^ Because the tube is a good insulator in current range I at low voltage and suddenly becomes a good conductor in range II, we often think of a gas tube as a voltage controlled device and use it as such in switching circuits. Actually, however, in current range II the tube can be con- trolled only by regulating the current. For this reason in the remainder of this paper, the current is considered as the independent variable. In current range III the voltage falls rapidly with increasing current. A space charge of positive ions begins to develop close to the cathode. By the beginning of current range IV most of the voltage drop in the tube appears across this space charge layer. This region of nearly constant voltage with increasing current is called the normal glow discharge. The space charge layer immediately in front of the cathode is conamonly called the cathode dark space. Electrons emitted from the cathode as a result of positive ion bombardment and other processes are accelerated through the high field of this cathode dark space and produce an adjoin- ing layer of intense ionization and excitation called the negative glow. In tubes of the type considered here, this negative glow is the most luminous part of the tube. Beyond the negative glow toward the anode is the so-called Faraday dark space in which no new ionization or ex- citations are produced. Electrons from the negative glow can readily flow through this region to the anode because their space charge is almost completely cancelled by positive ions diffusing from the negative glow. Over current range IV the cathode current density is nearly constant. This means that the cross-sectional area of the discharge increases in proportion to the current. This is evidenced by the familiar spreading of the negative glow with increasing current until it covers the entire cathode area. In current range V the cathode is completely covered with the negative glow and the current density must increase in direct proportion to the total current. This range of currents is called the abnormal glow dis- charge. The current-voltage characteristic of a cold cathode for current ranges IV and V may be modified by changing the geometry from a single plane cathode to a hollow cathode.' A hollow cathode is one in which there is an overlapping of the regions of cathode fall and negative glow from two portions of the cathode. This overlapping can occur on the inside of a cylindrical or spherical surface or between two plane cathodes more or COLD CATHODE TUBES FOR AUDIO FREQUENCY SIGNALS 1375 less paraUel to each other and closely spaced. The optimum dimensions are a function of the density and kind of filling gas used. Electrons and ions generated in the cathode fall and negative glow regions of one cathode can be more efficiently used in producing new electrons and ions if they can enter another cathode fall region instead of diffusing outward into the Faraday dark space. This effect is particularly noticeable in the abnormal glow range of currents. This is illustrated by the curves of Fig. 2 which were taken on a tube containing two parallel plane cathodes with an adjustable spacing be- tween them. Because Fig. 2 is plotted to a linear current scale, current ranges I and II of Fig. 1 are compressed to the left-hand axis. The upper curve is obtained when the cathodes are far apart so that there is no in- teraction. The lower curve is obtained when the spacing is close enough to give a hollow cathode effect. The sustaining voltage of a normal glow discharge is dependent upon the anode to cathode distance. To investigate this we can arrange a parallel plane cathode-anode structure with the anode attached to a piece 122 120 ^ / y^ y 'XATHODE SPACING 118 UJ i? CURRENT range: / 0.25 INCHES m m 3z: / O 116 1 / 1 / ? 114 < 1— 1 1 Y 1 1 . 1 / ^- t ' / 110 A ) y \ J /^ i. / VI 3^ 'XATHODE SPACING 108 v^ y y 0.025 INCHES ^ ^ i \ ^A^o^ r A*3»*WV 106 16 16 20 0 2 4 6 8 10 12 14 CURRENT IN MILLIAh/PERES Fig. 2 — Volt-ampere characteristic of parallel plane cathodes at two different spacing. 1376 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 122 120 118 116 114 112 O Z 110 z H 108 ^ 106 104 102 100 Iqq = 10 milliamperes ^ AT p = 100 MM OF Hg ^ ^ ^ r - i 1 1 OB Dl STRUCTED I SCHARGE ANODE IN FARADAY DARK SPACE SPACE CHARGE NEAR ANODE <\ i 1 / \ 1 / K 1 1 1 0.04 0.06 0.08 0.10 0.12 0.14 ANODE CATHODE SPACING IN INCHES 0.16 0.18 Fig. 3 — Sustaining voltage as a function of anode-cathode spacing. of magnetic material which can be moved by a magnet external to the tube. Fig. 3 shows a typical set of data obtained in this manner with an operating current in the normal glow range of current. At very close spacings of the order of the negative glow distance, the voltage is higher than normal because some of the electrons released from the cathode are able to strike the anode before dissipating their energy in producing ions and excited particles in the gas. This is referred to as an "obstructed discharge".^ The voltage must, therefore, be higher because electrons which do produce excitation or ionization must be given extra energy in order to maintain the current. As the anode is moved away from the cathode through the Faraday dark space, the tube sustaining voltage stays at a fairly constant minimum value. This is possible because at close spacings more than a sufficient number of electrons and ions are present in the Faraday dark space to carry the current. As the anode is moved away the Faraday dark space is lengthened. Ions and electrons needed to carry the current diffuse from the cathode region. At suffi- ciently large distances, however, the ionization density decreases so that not enough ions are present to cancel space charge near the anode. A space charge sheath builds up in the anode region and the sustaining voltage rises with increasing distance. When the voltage has increased by about 10 to 18 volts depending upon the gas filling, it begins to level off with increase in distance. At about this point, an anode glow may appear in front of the anode. This is due COLD CATHODE TUBES FOR AUDIO FREQUENCY SIGNALS 1377 to the fact that some of the electrons have gained sufifcient energy for excitation. A slight further increase in anode-cathode distance usually results in the anode glow changing from a uniform layer to a "ball'' of glow. When this occurs, oscHlations of several volts amplitude appear across the tube terminals. These oscillations result from a sequence of events which is initiated when the electrons gain enough energy in passing through the anode fall region so that they may ionize. A small number of ions generated in this region will, because of their relatively low velocity, enable a large electron current to flow without developing space charge! This then mil reduce the voltage appearing across the anode fall and greatly reduce the number of ionizations taking place. As soon as the recently produced ions leave this region the voltage drop across this region increases causing the ionization to build up again. This alternate building up and decaying of ion density results in the observed oscilla- tion which is ordinarily in the frequency range from 0.5 to 20 kilocycles per second. This oscillation usually cannot be prevented by external circuit means. However, by proper choice of anode-cathode spacing, type and density of the gas filling, and to lesser extent the geometry of the anode, a tube can be made which is free from anode oscillations. The main restriction that this puts on tube design is the limitation of breakdown voltage. IMPEDANCE From the previous discussion of transmission requirements it is clear that one of the most important properties of a gas tube is the impedance presented to small ac signal currents superimposed on the steady dc op- erating current. At low frequencies, these signals cause the voltage across the tube to vary in accordance with the static characteristic. The im- pedance of the tube to these signals is almost entirely resistive and is equal to the slope of the static characteristic. At higher frequencies, how- ever, there is a lag in the adjustment of the voltage across the discharge to the changes in current. Hence, at these frequencies, the impedance of the tube may have both resistive and reactive components. This is illus- trated in Fig. 4. The small superimposed signals result in current- voltage loci which are ellipses. In current range HI at 200 cps, the position of the ellipse corre- sponds to a negative resistance in series \nth an inductance. The negative resistance changes rapidly with frequency and as shown at 2,000 cps, it may be positive. Because of the rapid variation of impedance, both \nth frequency and current, this range of currents is not generally useful for dependable transmission of voice frequency signals. 1378 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 In the higher current range IV and V the impedance can be represented by a positive resistance in series with an inductive reactance. The re- sistance increases slowly with frequency as indicated by the elhptical current-voltage loci at 200 and 2,000 cps. The impedance of a cold cathode tube also varies with anode-to- cathode spacing. This may be studied by means of the same movable- anode parallel-plane tube used to obtain the data of Fig. 3. We again operate the tube in the current range of the normal glow (IV) and, for illustrative purposes, choose a measuring frequency of 1,000 cycles per second. The results are shown in Fig. 5. The resistive and reactive com- ponents of tube impedance are independent of anode spacing throughout the Faraday dark space and well into the obstructed discharge region. At large distances where the electron space charge sheath begins to build up in front of the anode, the impedance increases rapidly with distance. Some useful conclusions about the design of transmission tubes can be drawn from the data of Fig. 5. We can consider the total tube impedance as being made up of the sum of the impedances introduced by the various regions of the discharge. Since large variations in the length of the Faraday dark space do not affect tube impedance, it is concluded that this region has negligible impedance. This means that, so long as the anode-to-cathode distance is short enough to avoid a space charge sheath at the anode, we can concentrate our attention on the cathode fall region. The following detailed discussions of impedance are consequently re- stricted to the cathode portions of the glow discharge. 106 m 105 5 104 i ? 103 1 «^ 102 IE W 2 U 2000 CPS (^"aoo ^ \ y y 101 \ s 2000 CPS~ y 100 1 1 & rn c ) ' > 2 CURF ?ENT 1 X N ^ ^IL JAMPE 5 RES ' t ) 9 Fig. 4 — Static and dynamic volt-ampere characteristics COLD CATHODE TUBES FOR AUDIO FKEgUEXCV SU.NAUS 1379 3000 2800 Idc = ^o'^"-liamperes f = 1000CPS GAS FILLING-NEON AT p= 100 MM OF Hg 1 2600 2400 1 CO 2 2200 t I O 2 2000 •J 1800 r REACTAN 8 8 OBSTRUCTED DISCHARGE ANODE IN FARADAY DARK SPACE CHANGE NEAR 4- Ah 40DE Q Z ji > O C S 8 § 3DNV1S 1; 1 (0 ""^ UJ a. 600 400 200 i 1 «/ !(oL V 1 ^-^^. -^i — ^^- . _L 0 0.02 0.04 0.06 0.08 0.10 ANODE CATHODE SPACING IN INCHES Fig. 5 — Resistive and inductive components of impedance anode-cathode spacing. as a function of IMPEDANCE OF PLANE CATHODE TUBES Let US consider first the impedance of a tube with a plane cathode or of a cathode with a radius of curvature large compared to the combined thicknesses of the cathode dark space and the negative glow. An example is the Western Electric 313-C which has a gas filling of 99 per cent neon, 1 per cent argon and a cathode surface of barium and strontium oxides. At a fixed steady current of 25 ma flowing through the starter gap, it is found that the resistive component of the impedance varies with fre- quency as sho^Ti in Fig. 6. In the middle of the voice band it has a value of about 200 ohms. There is also an inductive reactance of about 65 ohms at this frequency. At a fixed frequency, it is found that the resistive component of this type of tube decreases with current approximately inversely as the square root of the current. This is shown in Fig. 7. The 1380 THE BELL SYSTEM TECHNICAL JOUENAL, NOVEMBER 1953 800 1200 1600 2000 FREQUENCY IN CYCLES PER SECOND 2400 2800 Fig. 6 — Resistive and inductive components of 313-C starter gap impedance as a function of frequency. 7dc = 25 mm. inductive reactance is related to the current approximately as follows : wL = 5 + f- . (1) B and C are functions of frequency. If we wish to use tubes of this type in central office switching circuits which pass voice frequency currents, we find that even one tube con- nected between a 600-ohm source and a 600-ohm load gives a prohibi- tively high insertion loss. The impedance level of the voice frequency circuit may, of course, be raised up to 4,000 or 5,000 ohms. But practical switching circuits might require as many as four tubes in series. Hence, we see that the impedance of tubes of this type severely limits their usefulness. From Fig. 7 we might argue that to get low resistive components we might continue to increase the direct current flowing through the tube. However, in this range of currents, the life of the tube varies approxi- mately as \/l\c while the resistive component varies as l/Idi^. Over the range of currents plotted in Fig. 6, the cathode is fully covered with glow. As noted above, the resistive component is (2) COLD CATHODE TUBES FOR AUDIO FREQUENCY SICWI- 1381 If instead of passing a given direct current through a single tube we pass the same current through a parallel combination of n tubes, the resistive component of the combination would be R comb. = '©' n . (3) = 4..R. 1/2 The resistive component of the combination is — - times that of single tube passing the same total current. This assumes, of course, that the cathodes of the tubes are covered with glow. Obviously this is not a practical method of attaining a low impedance because of the instability of paralleled individual tubes. It does suggest, however, that by increas- ing the cathode area of a single tube until the glow just covers the full area, a lower impedance is obtained. This has been done experimentally. 350 250 200 z < 150 100 50 A \ p_ 1080 Vloc \ \ V N k. \ X ^, N V -^ ^J k ■o 10 15 20 25 30 DC CURRENT IN MILLIAMPERES 35 40 Fig. 7 — Resistive and inductive components of 313-C starter gap impedance as a function of direct current. 1382 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Fig. 8 shows the data for a tube of this type which has a cathode area about six times that of the 313-C. This change in cathode area along with the necessary change in anode geometry led to an impedance reduc- tion of about five times. This type of design, of course, gives a large increase in life of the tube but it is undesirable from the standpoint of tube size. As pointed out above, low impedance can be attained by an increase of the tube current and within limits by an increase of the cathode area. A third parameter which may be varied is the density of the gas filling. At a constant current and with a given cathode area, the resistive com- ponent of the impedance decreases with increase in density or the pres- sure, p, at a fixed temperature approximately in accordance with the relation (4) The curve in Fig. 9 was obtained over a limited range of argon fillings with a barium strontium oxide cathode. Since the current density in- creases approximately in proportion to p^, the effective cathode area will tend to be reduced unless the total current is kept high enough to cover the cathode fully. Therefore, with a fixed total current there is a limit to either increase in pressure or in cathode area. In further search for low impedance, a number of different gas fiUings have been tried. They have included all the rare gases as well as several mixtures of them. No significant advantages were obtained by the use of other than the more common neon or argon gas fillings. We therefore see that the abihty to control impedance properties of 60 50 O ^ 40 lUl 20 U « ^ ss-::: — — o— R — _,,--^ "aJiT £r^ ^ ^-''^ 400 800 1200 1600 2000 FREQUENCY IN CYCLES PER SECOND 2400 2800 Fig. 8 — Resistive and inductive components of impedance for a large area cathode tube as a function of frequency, /dc = 25 mm. COLD CATHODE TUBES FOR AUDIO FREQUENCY SIGNALS I.3K3 400 350 (0 I 300 O 250 O 200 O 150 100 50 Ioc = 0015 AMPERES ARGON GAS TILLING P = PRESSURE IN MM OF Hg AT 25'C > / y / /" X / a/ / ? 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 AO 4.5 5.0 -^XI03 Fig. 9 — Resistive component of impedance as a function of density (pressure at 25°C) of gas filling. the plane cathode through changes in tube current, cathode area, density and type of gas filling is limited. A reduction by a factor of five or ten times over the values present in commercial tubes is certainly possible. But the possibility of obtaining values of less than +10 ohms or negative values seemed remote. Consequently, development effort was concen- trated on the hollow cathode tube described below. HOLLOW CATHODE TUBES The static voltage-current characteristic of a hollow cathode was shown in Fig. 2 to be below that of a plane cathode of the same area when operating at currents in the abnormal glow region. It has been found that by proper choice of the cathode dimensions and the kind and density of the filling gas, desirable transmission characteristics can be achieved. The follomng discussion illustrates the manner in which some of the variables are interrelated. We ^vill consider a "U" shaped cathode which has been formed by folding a piece of molybdenum sheet in the form illustrated in Fig. 10. The choice of dimensions of the hollow portion will be discussed lat<»r. The anode is a cylindrical rod placed in front of the cathode. The anode- to-cathode spacing is selected so that the anode is always within the Faraday dark space and hence does not influence the impedance ap- preciably. It is assumed that the structure of Fig. 10 is sealed in a bulb 1384 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 which has been carefully evacuated and filled to a fraction of an atmos- phere with a rare gas such as neon. A typical static volt-ampere characteristic of the structure of Fig. 10 is shown in Fig. 11. This curve was taken with a cathode having a hollow portion 3^" long and 3^6" deep with a cathode gap of 0.023''. A neon filling pressure of 58 mm of mercury was used . It can be seen that there is the usual low-current negative slope associated with the transition from breakdown (current range III of Fig. 1). A new characteristic of interest is the second region of negative slope in the abnormal glow range of currents. It has been found that this second region can be made stable with time in a given tube and reproducible from tube to tube. It is also tzA DEPTH OF HOLLOW / / / / 7 CATHODE GAP - CATHODE ANODE-CATHODE SPACING LENGTH OF HOLLOW ANODE Fig. 10 — Electrode geometry of a hollow cathode tube. found that the tube impedance has a negative resistance component over the voice frequency range. Thus, this second region of negative resistance offers attractive possibilities as a transmission element. The impedance of this tube at 300 and 3,000 cps is shown in Fig. 12 for the same range of operating current as Fig. 11. The optimum current for negative resistance and the value of negative resistance are functions of the cathode gap, but so long as the other cathode dimensions are constant the optimum current is relatively independent of the density of the filUng gas. A useful way of studying the interrelation of cathode gap and filling pressure is shown by Figs. 13 and 14. For these data the length and depth of the hollow portion were kept constant at the values of ]/^" and \{^" respectively. Fig. 13 shows the resistive component of impedance as a function of frequency for different filling pressures ^vith a fixed cathode gap. COLD CATHODE TUBES FOR AUDIO FREQUENCY SIGNALS 1385 110 S 109 < 108 106 x-^ u-- 1 T » 4 8 12 16 20 24 28 DIRECT OPERATING CURRENT IN MILLIAMPERES Fig. 11 — Static volt-ampere characteristic of hollow cathode tube. 15 \ \ f=3 KC INDUCTANCE 300X ^ <> ^ ^ 26 32 6 8 10 12 14 16 18 20 22 24 26 DIRECT OPERATING CURRENT IN MILLIAMPERES jTjg 12 — Resistive and inductive components of impedance of a hollow cathode tube. 1386 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 40 32MM " 20 0 / ' 1 ^ y 1 1 «««, // -160 -180 1 1 1 1 1 1 / 1 0.2 0.4 0.6 0.8 1 2 4 6 8 10 FREQUENCY IN CYCLES PER SECOND 20 30 X103 Fig. 13 — Resistive component of impedance as a function of frequency for different neon filling pressures. Cathode gap = 0.024 in. Fig. 14 shows the variation of the resistive component of tube im- pedance as a function of frequency for several cathode gaps and at a fixed density of filhng gas. For both Figs. 13 and 14, the current was adjusted for optimum negative resistance. It can be seen from Fig. 13 that the choice of a neon filHng gas at a pressure near 60 mm and from Fig. 14 that a cathode gap near 0.024 inch could be expected to yield a negative resistive component of impedance which is reasonably insensitive to filling pressure and which is also constant in value over the voice frequency range. This justifies the choice of cathode gap and filling pressure used in the tube on which the data of Figs. 11 and 12 were taken. One other parameter of interest is the limit of anode-to-cathode dis- tance. This too is a function of cathode gap and gas density. A typical curve taken for the same cathode geometry and gas filling as used for Figs. 11 and 12 is shown in Fig. 15. It is seen that the negative resistive component is essentially independent of anode distance for a distance of approximately 0.050 inch. This means that the breakdown voltage of COLD CATHODE TUBES FOR AUDIO FRKgUENCY SIGNALS 1387 60 40 20 0 -20 -40 -100 -140 0 050 " CATHODE GAP^^ ^ 1 0 OAsX. / \ \ 1 \ V. J ao. t> N S / \, / ( ).024" -^ N \ / -v J 1 1 1 \ 1 / 1 ^ 0.2 0.4 30 XIO- 0.6 0.8 1 2 4 6 8 10 FREQUENCY IN CYCLES PER SECOND Fig. 14 — Resistive component of impedance as a function of frequency for different values of cathode gap. Neon filling pressure = 50 mm of Hg. -20 -40 -60 -80 -100 ■120 ■160 0.06 0 0.01 0.02 0.03 0.04 0.05 ANODE-TO-CATHODE DISTANCE IN INCHES Fig. 15 — Resistive component of impedance as a function of anode-to-cathode distance. 1388 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 a switching tube can be adjusted by changing this spacing so long as the upper limit is not exceeded. Although the above discussion is limited to only one adjustable cathode dimension, the cathode gap, other variations are of course possible. The length and depth of the hollow should be at least a few times the width of the cathode gap in order that an efficient hollow be formed. Increasing the hollow length or depth requires larger currents and, if carried too far, the glow may not completely fill the hollow at the optimum current for negative resistance. This is undesirable because the unused portion of the cathode may change its properties with time and produce unstable characteristics. The entire geometry may be scaled to a larger size if the density of the filling gas is reduced by approximately the same factor. The choice of cathode material is restricted by the high current densities of hollow cathodes. Coatings of alkafine earth oxides or similar materials have too short a Ufe. Pure molybdenum has found to give a satisfactorily low sustaining voltage together with long life and stable operating characteristics. Life tests have shown that tubes can be made which will operate satisfactorily for the equivalent of 20 to 40 years in central office service. It is seen from the above that by changing the cathode geometry and density of the filling gas a variety of impedance properties can be ob- tained. The final choice must be determined by the overall transmission requirements. As an example, the transmission performance of a typical tube will now be discussed. TRANSMISSION PERFORMANCE OF A TYPICAL NEGATIVE RESISTANCE DIODE The circuit of Fig. 16(a) shows a cold cathode switching tube in series with a transmission path. The voltage across the load resistor Rl under conditions of Fig. 16(a), divided by the voltage across Rl with no tube in the circuit. Fig. 16(b), is one measure of the transmission performance. This ratio, called the insertion voltage gain, is given by IVG = ^"^ ^J^ (^^ Rs + Rl +Rt + jc^L, ' ^ ^ The derivation of this expression assumes that the transformers are ideal and that the reactance of the condenser C is negligible. Maximum gain occurs at low frequency where the reactive component of tube impedance, jc^Li , can be neglected. If the resistive component of tube impedance, Rl , is negative, the gain will be greater than unity. The gain approaches an infinite value as the unstable condition is approached where the COLD CATHODE TUBES FOR Al Dlo KIIEQUENCY SIGNALS 1389 negative resistance of the tube equals the sum of the source and load resistance. Large values of gain are not practical because this imposes undesirable restrictions on the constancy of the circuit and tube im- pedances. Additional restrictions on gain arise from bandwidth and distortion considerations. If it is assumed that both Rt and Lt are independent of frequency over the voice band, it is possible to use the above equation for an approximate calculation of bandwidth. The half-power point occurs at the frequency /, at which the reactive term equals the sum of the other three terms in the denominator, or fc = Ra + Rl + Rt 2tLi (6) Since the gain does not fall off at low frequency, the upper cut-off frequency fc is a measure of the bandwidth. Substituting Ra and Rl from equation (5) into equation (6) gives /c (1 — Low frequency I.V.G.) = Rt 2irLt (7) Thus for a given tube an increase in gain is accompanied by a decrease in bandwidth. As shown in Fig. 12 the impedance of a negative resistance tube is dependent on the current passing through it. This will cause some dis- tortion as the signal current swings above and below the average direct current value. The distortion is small so long as the non-linear tube resistance is small compared to the total circuit impedance. It can be seen from the above discussion that for a given tube the gain, bandwidth, and distortion are all dependent on the source and load impedances. Fig. 17 shows the experimental performance of a typical tube for the special case where source and load impedances were equal. Insertion gain has been converted to power gain in db. The distortion Rt+ja;Lt (a) (b) Fig. 16 — Transmission circuits with ami withoiit tube. 1390 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 6 ^ 1.50 o z o o UJ «0 lU 0. irt UJ _l u ^ 3 o _J z o z 5 1 1.25 1.00 IMN-^l^-^- 0.75 2 3 0.50 ^ 0.25 O I ^ // ^ \ INSERTION \ GAIN 7 - m 1 > h TYPICAL VALUE OF y' REQUIRED POWER UNSTABLE REGION w TYPICAL -REQUIREC BANDWID- / — ^ -y y r;;:; N ^ ^ iiii;;;:Hi;H S > xfi 1 llM;:HH ;iii BANDWIDTH^ / H- ^ WM i > V / 1 1 --^ v» ■i^-Vi"---i' 'M^ ^ Y y^OWER 1 1 1 •.. : wm > V 1 1 1 12 8 UJ O z z < 6 O cr UJ V) z Fig. 17 diode. 0 25 50 75 100 125 150 175 200 225 250 275 300 325 IMPEDANCE LEVEL IN OHMS (SOURCE AND LOAD RESISTANCE EQUAL) Performance curves for experimental cold cathode negative resistance has been related to the power handUng capacity by measuring at each value of source and load impedance, the maximum power output at which the total harmonic distortion is 30 db below the signal. The noise introduced by a tube affects its usefulness as a transmission element. By designing the tube so that the anode is in the Faraday dark space, anode oscillations are avoided. The remaining noise is at a low level, typical values being 10 decibels above the noise reference level of 10"'' watts. CONCLUSIONS Cold cathode glow discharge tubes can be made with stable and re- producible impedance characteristics. By proper choice of anode-cathode spacing and pressure of filling gas, it is possible to eliminate oscillation noise associated with the anode region. By properly choosing the density of filling gas and the area of a plane cathode, it is possible to obtain a low positive resistance component of impedance. By proper correlation of cathode geometry and filling gas density, hollow cathodes can be used to obtain a negative resistance component of impedance. Bandwidth, power, distortion, and noise requirements of voice frequency transmis- COLD CATHODE TUBES FOR AUDIO FREQUENCY SIGNAL^ 1301 sion circuits can be satisfied without sacrificing switching characteristics. Cold cathode glow discharge tubes are, therefore, a potentially useful electronic switch for use in series with voice frequency transmission circuits. ACKNOWLEDGMENT The authors wish to acknowledge the contributions of F. T. Andrews in analyzing and measuring the amplifier performance of negative resis- tance gas tube diodes. REFERENCES 1. Ingram, S. B., Elec. Eng. (A.I.E.E. Transactions), 68, pp. 342-346, 1939. 2. Depp, W. A., and W. H. T. Holden, Elec. Mfg., 44, pp. 92-97, 1949. 3. Druyvesteyn, M. J. and F. M. Penning, Revs. Mod. Phys., 12, pp. 87-174, 1940. Balanced Polar Mercury Contact Relay By J. T. L. BROWN and C. E. POLLARD (Manuscript received June 19, 1953) A new type of relay making use of solid contacts maintained continuously wet with mercury has been developed. It has a symmetrical polar structure, resulting in improved sensitivity and speed compared with the existing neu- tral structure with similar contacts. It is also particularly well adapted to switching of high frequency circuits. Two magnets are used for polarization, and the relay is adjusted after assembly to desired values of sensitivity for operation in both forward and reverse directions by selective adjustment of the magnet strengths. INTRODUCTION In a previous paper^ a mercury contact relay is described in which the contact elements are maintained wet with mercury through a capillary path to mercury reservoir below the contacts. The present paper de- scribes a new design making use of this same mercury contact principle, but with a S3mametrical polar structure which gives improvement in sensitivity and speed capabilities over the previous neutral structure. VERTICAL RELAY Fig. 1 shows one design of the relay, adapted for general use in a vertical position. It provides a single pole double throw magnetic switch in a sealed glass tube, enclosed along vnth an operating coil and polariz- ing magnets in a steel can with a medium octal plug base similar to the previous type relay (275 type). Fig. 2 shows the glass enclosed magnetic switch element. The armature is a tapered reed welded to a tubular stem which is sealed in the glass at the lower end. Mercury, and gas under pressure are introduced through this tube, which is then welded closed. The magnetic working gaps are formed between the armature and fixed pole-pieces which are sealed in 1 J T L. Brown and C. E. Pollard, Mercury Contact Relays, Eloc. Enj?., 66, Nov., 1947. 1393 1394 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Fig, 1 — (a) Switch element, (b) Magnets, spool, and side plates added, (c) Coil added, (d) The complete relay. BALANCED POLAR MERCURY CONTACT RELAY 1395 the glass at the upper end. The fixed contacts are made from small platinum alloy balls which are welded to the pole-pieces. The armature strikes the fixed contacts at a point which is close to a node for one of its principal modes of vibration. This limits the bounce on impact to an amphtude which does not open the mercury bridge which is formed. Mercury in a pool at the bottom of the switch is fed to the contact area through grooves rolled in the armature surface. Except for the platinum alloy contacts, the surfaces of the pole-pieces inside the switch have an oxide coating which prevents them from being wet by the mercury. This limits the mercury bridge which is formed between the armature and pole-piece to the area of the platinum alloy contact. A ceramic detail is inserted between the pole-pieces at the top of the switch. The ceramic is specially resistant to wetting by the mercury and thus prevents mercury from collecting between the pole-pieces. The polarizing magnets are soldered to the pole-piece terminals out- side the switch. Permalloy plates are soldered to the outer poles of the magnets and extend down on the outside of the coil, forming a return path to the lower end of the armature. The coupling at the lower end is made relatively loose in order to limit magnetic soak effects. The steel can provides a magnetic shield. STATIC MERCURY CONFIGURATION The static configuration of the mercury surfaces in the switch depends upon the shape and contact angle to mercury of the solid surfaces with which it comes in contact, and the curvature of the free mercury surfaces. This curvature depends on the surface tension of the mercury surface T and the pressure difference Ap between the inside and outside of the mercury at the point under consideration. That is, where Ri and R2 are principal radii of curvature of the surface (radii taken in planes cutting the surface at right angles to each other). In this switch the pressure difference is a function of the height of the point under consideration above the surface of the reservoir: Ap = pgh. (2) Substituting the values p =13.6 g/cc for mercury, g = 980 cm/sec^ and T = 450 dynes/cm for mercury, 1396 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 CONTACT ARMATURE GROOVES MERCURY WELD Fig. 2 — • Vertical switch. BALANCED POLAR MERCURY CONTACT RELAY 1397 we obtain h = 0.0338 \Ri Ri) cm. (3) Fig. 3 shows a cross section of the armature and pole-piece at the contact, indicating the configuration of the mercury in the grooves and around a closed contact, as determined from equation (3). In the grooves, it will be noted, the surface is concave cylindrical, with a radius of -0.0163 cm, providing a path from the reservoir at this height with nearly the full capacity of the groove. The contact angles to the armatuie R2 = -0.0163 CM |Rl = 00 I R, = 0X)263CM DEPTH OF WEAR MERCURY (2.07 CM ABOVE RESERVOIR) Fig. 3 — Cross-section showing configuration of mercury surface in grooves and around closed contact. and platinum alloy contact surfaces are all zero, as these show complete wetting. The mercury fillet around the contact has positive curvature in the plane of the armature {ft\ = +0.0178 cm) and negative curvature (i?2 = —0.0085 cm) in planes through the axis of the contact. The depth of wear indicated in Fig. 3 is brought to a stable value by an aging process in production. DYNAMIC MERCURY CONFIGURATION Fig. 4 shows flash photographs of the contacts at various instants during operation at 60 cps. It will be noted that, as the armature moves away from the fixed contact on the right, Figs. 4(a), (b) and (c), a bridge is drawn out which breaks in two places, leaving a free drop which falls out, Fig. 4(c). 1398 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 r^^^^'f/'^;'. Fig. 4 — Dynamics of mercury contact, (a) Armature against contact on right. Note spherical shape of mercury on left contact, (b) Armature moves to left, drawing out mercury bridge, (c) Armature further to left; bridge about to break, (d) Armature reaches contact on left. Mercury bridge has broken in two places, leaving a tiny ball which later drops away. BALANCED POLAR MEHCLKY ('()NT.\( r ICKLAY 1399 When the ball falls out as the result of an operation, the loss of mer- cury results m a small, temporary constriction of the fillets in the grooves near the contact that was just opened. That is, the curvature of the surfaces of these fillets becomes more negative. As shown by equation (1), this produces a local decrease in liquid pressure. Mercury therefore flows up the armature from the reservoir to restore the normal static balance between surface tension and negative head. This is the fun- damental behavior of an ordinary wick. The ball drops into the reser- voir unless it happens to hit the armature on the way down. Thus, for repeated operations, there is a continuous circulation of mercury. As indicated in Fig. 3, the fixed contact is a ball, with a flat surface where contact is made to the armature. In Fig. 4(c), taken directly after the breaking of the mercury bridge, the mercury remaining at the fixed contact on the right has been thrown back on the contact by surface tension forces, laying bare the flat surface. After several flow oscillations that are not shown, it comes to rest with the spherical contour shown by the contact at the left in Fig. 4(a). That is, being disconnected from the reservoir and having a limited wet surface to spread over, it assumes a positive head corresponding to a positive spherical radius about equal to that of the contact. This provides a mercury ''cushion" in the form of a segment of a sphere, to which contact is made when the relay operates. ADJUSTMENT OF SENSITIVITY For various combinations of MMF's (magnetomotive forces) in the two magnets, various corresponding pairs of sensitivity values exist for operation in the two directions. A theoretical analysis of this relationship is given in the attached appendix. The adjustment of magnet strengths to obtain a specified pair of sensitivity values is made on the completed relay, using two electromagnets placed outside the can opposite the relay magnets. An automatic circuit is provided for this purpose that makes a complete adjustment in about 15 seconds, the time being de- pendent upon the precision required and the uniformity of the product being adjusted. The procedure used and the basis for it are discussed in the Appendix. All of the adjustments used are of the type for which the armature moves all the way from one contact to the other when an operate? current is applied. This represents the condition for the minimum differential ampere turns obtainable between the two operate values l)ecause it makes use of armature momentum. 1400 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 11 z a. PiO Fig. 5 ^ o 10 20 100 110 120 30 40 50 60 70 80 90 OPERATE FROM BACK- AMPERE -TURNS Minimum differential ampere turns as function of magnetic bias. BIAS ADJUSTMENTS As bias adjustments are made magnetically, rather than with the spring used in some polar designs, the armature flux tends to approach the same value at the operating points as it has in a balanced adjust- ment with the same differential ampere turn value. Such difference as does enter mth increase of bias is due to differences in armature flux distribution between MMF introduced by the coil and that introduced by the magnets. The effect of bias is illustrated in Fig. 5, which shows the minimum differential ampere turn value obtainable in a relay as a function of the higher operate value. The minimum differential value of 3.6 NX with a balanced adjustment is about the same as is obtainable with other polar relay designs. For operate values of about 100 ampere turns a "release to operate ratio" of about 90 per cent is obtained. Larger differential ampere turn values, both with and without bias, are of course possible. The amount of bias obtainable in combination with larger differentials is somewhat reduced because of the greater magnet strength required. EFFECT OF MAGNETIC SOAK ON SENSITIVITY The effect of "soak" on sensitivity is illustrated in Fig. 6. It shows the change which takes place when the ampere turns to operate are BALANCED POLAR MERCURY CONTACT RELAY 1401 measured after applying a given ampere turn soak in each of two polari- ties. SPEED OF OPERATION Fig. 7 shows typical operate times of the relay for one balanced adjust- ment and one biased adjustment. The ordinates correspond to time to effect closure at the opposite contact after input is applied to the coil. The abscissae are shown both in terms of power in the full relay winding and the corresponding number of ampere turns. Two circuit conditions are indicated, one in which the voltage was applied directly to the full winding and the other in which the voltage was appUed through a re- sistance equal to the coil resistance. The winding used in this case is one which is specially designed for speed rather than sensitivity, being shorter in length, with the working gap near the middle. Its time constant (inductance-resistance ratio) is about 0.0006 second. Fig. 8 shows "release" time measurements, where the relay, with various biased adjustments, was operated by opening the circuit. The curve is shown plotted against the "operate" setting of the relay. In this form it is relatively independent of the differential ampere turn adjust- ment. The time required to open the contact from which the armature moves is typically about 0.0002 second longer than the closure time, as the 2.25 1.50 1.25 1.00 0.75 0.50 0.25 ^^m ^X""*^ O X 40 80 120 160 200 240 N.I. SOAK VALUES PLUS AND MINUS 280 320 Fig. 6 — Change in NI. To operate versus plus and minus NI soak. 1402 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 AMPERE-TURNS 20.4 204 10 ID 2 H 2 10- i 1 ^ iL *^ 5n ^ ^ ^ A V^ ^ ^^ ^ ^ ^ ^ :£& ~ — ; : — ' — . 8 _p 10 2 POWER IN COIL IN WATTS 1.0 Fig. 7 — Time to close opposite contact after voltage is applied to coil, (A) ±5 NX adjustment, voltage directly on coil. (B) Same as (A), except with resist- ance equal to coil in series. (C) Operate +20.4 NX, release +9.9 NX voltage directly on coil. (D) Same as (C), except with resistance equal to coil in series. mercury bridge does not ordinarily break until after contact is made on the opposite side. SXhe natural frequency of the free armature, wet with mercury, is about 220 cps. The relays have been used at frequencies up to about 350 cps where the drive conditions were individually selected with respect to the phase of impact transients. Fairly well controlled operation with- out such selection can be obtained up to about 100 cps. CONTACTS The relay has been designed with telephone, rather than power appli- cations in mind. The contacts are smaller than those in the previous neutral design, and this appears to be associated with less capabihty for closure of very high current circuits. The capacity required across the contact to prevent noticeable arcing appears to be about the same as that required in the earlier type : C = — microfarads. but in most cases, larger values than this will be needed to hold peak BALANCED POLAR MERCURY CONTACT RELAY 1403 voltages to safe values. A small resistance in series with the condenser to limit the closure current is usually necessary. A considerable amount of very satisfactory experience has been had with inductive loads of 0.5 ampere at 50 volts, with 0.5 mf in series with 10 ohms across the con- tacts. The contact closure shows no chatter for time intervals of 0.1 micro- second or more. LIFE Tests of relays with protected contacts indicate that the only change is a sensitivity change due to wear. Under conditions producing fairly high velocity of contact impact this change is of the order of 5 ampere turns increase in differential ampere turns for a billion operations, the change being roughly proportional to the logarithm of the number of operations. HORIZONTAL TYPE RELAY Most of the apparatus in the telephone plant is mounted on vertical panels. The substantially vertical mounting requirement for the relay shown in Fig. 1 tends therefore to be uneconomical from the space stand- point. Fig. 9 shows a modification of the glass enclosed switch which avoids this limitation by being designed to operate with its axis hori- zontal. The switch modification consists in adding to the switch a special 20 30 40 50 60 70 80 90 100 110 OPERATE AMPERE-TURNS Pig 8 _ Release time versus operate ampere turn adjustment. 1404 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 o N o w I s BALANCED POLAR MERCURY C()\TA( T KKI.W | K).", capillary reservoir which establishes a negative head in the nuMrnry equal to that obtanied by the height of the contacts above the mercury reservoir ni the switch shown in Fig. 2. To provide for the return of free mercury to the reservoir, a wet strip of metal is placed along the glass wall underneath the contacts. The capillary reservoir is essentially a bundle of tubes with walls wet by the mercury. The amount of mercury in the switch is such that the tubes are about half full. The mercury meniscii in these tubes are tangent to the walls and therefore have a spherical surface with a radius equal to that of the tubes. The proper tube radius (Ri = R. = 0.033 cm) for Fig. 10 — Horizontal relay. the desired negative head {h = 2.07 cm) is obtained from equation (3). As the tubes have a uniform bore, the negative head which they estab- lish is not critically dependent on the amount of mercury in them. The capillary reservoir is made from thin strip, formed and then rolled into a cylinder. It surrounds the metal tube and is welded to the base of the armature. The drain element is sealed into the glass at one end, and, at the other end, is held in contact with the capillary element by the surface tension and negative head of the mercury. The horizontal switch is an experimental design and has not been put into production. It can be assembled in the housing shown in Fig. 1 for plug connection to a vertical panel. Fig. 10 shows a preliminary model with the new type terminals for wrapped connections.^ It is adapted for 2 Solderless Wrapped Connections, B.S.T.J., May, 1953. 1406 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 mounting on a vertical panel interchangeably with existing polar types. The terminals, which may be up to eight in number, are also brought out at the rear for test purposes. HIGH-FREQUENCY SWITCHING The small size, sjnnmetry, and simplicity of the vertical switch struc- ture has been found to be particularly well adapted for incorporation into coaxial structures for high-frequency switching and it is being used for a variety of apphcations of this kind. USES Small scale production of the relays has been started for a few special uses. Designs for larger scale uses are still in a preliminary stage. In gen- eral, it is expected that the relay will be well adapted for applications where a transfer contact in combination with low inductance, high speed, high sensitivity, low contact resistance, freedom from chatter, good high frequency switching characteristics, stability, long life, or any combina- tion of these, is required. Appendix magnet strength versus sensitivity A simplified representation of the magnetic system of a relay of this general type is shown in Fig. 11. Here the armature is shown working between two magnetic poles A^ and aS^, each of which has an area A . The armature position is indicated as a deviation ^ toward N from the mid position, and is hmited in its motion by stops at / = zb^i from moving through the full magnetic gap range, defined by the positions / = d=/2 . M, and Mn are MMF's in ampere turns, introduced by the magnets on either side, with positive values as indicated by the arrows. Similarly, Me represents the MMF introduced by the operating coil. These values are not those of the magnets and coil in the actual relay. Instead they are assumed to be "Thevenin" equivalent open circuit values of MMF looking away from the working gaps on either side. The gaps are assumed to have the simple geometric dimensions of length li — t and li + /, and area A. The reluctances looking away from the working gaps are as- sumed to be represented by the fixed gaps of area A and length I2 — l\ on either side. BALANCED POLAR MERCURY CONTACT RELAY i4o; The magnetic pull on the armature in a gap of this kind is ^ 980 ^"^ '''^ g^ (124.9) ^^ ^*^ where / is the force in grams g is the gap length and M is the MMF in jDractical ampere turns across the gap. In the structure shown, therefore, the net force on the armature from the gaps on both sides is J m — (124.9)2 L (A - ly % + ^y J' (Ms- (2) where the values of M are expressed in ampere turns. If Mn = Ms this can be converted to fml L('-^)" (-01 (3) where /„; \124.9/ Fiff. 12 shows curves of ^ versus - for various values of —=^ . Only ^ fml h Mn ' one quadrant is shown as the system is symmetrical. It provides a con- venient means for analysis of this type of relay. MnJ + S N Q + O 1 1^ 1— 4 1 — i 1 Mc * r^"""''- -la ♦ s N Ms| + mm^ 1 Fig. 11 — Simplified representation of relay. 1408 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 15 f _ { MnV A II ^f".- \\2A.9j 12 ' y" \\\ '1 7 // \ //I ' i // 9 8 7 6 5 4 3 2 /// 1 /// // / /// 7/ / // /// / y^ ' 1 // // //' / '/, // /; / \^ J-^ 5^ y/y^l // y. / _^ ^ ^ y / X <> V /^/ '// A/ ' /o ^ /> // 0 \^ ^ X. y^ y "'O- ^ -? ^^^ ^ 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 \_ l2 Fig. 12 — Balanced gap relay chart. BALANCED POLAR MERCURY CONTACT RELAY MOO Also shown on the figure is a typical spring characteristic fml ti (4) If the stops li and A are placed at points of intersect iofi of the spring characteristic with the magnetic characteristic for M = 0, (in this case dXlJli = 0.48), we have a condition that corresponds to a hypothetical relay of infinite sensitivity. That is, the armature can be released from contact with either side with an infinitesmal coil input, and, if there are no magnetic or mechanical losses in its travel, it \vi\\ just swing to the opposite side without any change in the total energy of the system. I>et us define the magnet strength for this condition as Mo and let us assume that Mc = 0 for operation in either direction with this magnet strength. Assume now that the magnet strengths on each side are increased equally to the values Mni = Mn = Mo -h A. (5) For values of A/A near 1 this change can be balanced out by a coil input of about the same amount, as practically all of the pull on the armaimc would be from the nearer pole. For values of /i/A near 0 the effect of a coil input will be equal and aiding in both gaps. The coil inputs required to just operate from the N and *S poles, defined as Mcn\ and Mc,\ , re- spectively for this particular case, will be Mcsi = - Mcni = pA = piMni - 3fo) = p(M.i - 3/o), (6) where p is a value between 1 and 0.5. Values for individual cases can Ix; worked out with reference to the curves for various values of MJMn in Fig. 12. For the case shown, where (Jti = 0.48, p is about 0.8. An adjustment in accordance with (6) is thus a balanced one with a spread of 2pA between the two sensitivity values, the amount of spread increasing with increase in the strength of the two equal magnets M^i and Msi above the value Mo . A general type of adjustment, including all possible combinations of the two sensitivity values, can be obtained by adding a suitable value B to a balanced pair of sensitivity values in accordance with equation (6), These general sensitivity values would then be Mc, = Me,l + B, /yx Men = Mcnl + B. This is the type of change that would be produced by a bias oi -B 1410 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 ampere turns in an auxiliary coil. The equivalent of such a bias would be obtained by decreasing the strength of the A^ magnet and increasing the strength of the S magnet, each by the value B. The general magnet strength values would then be Mn = Mn, - B, Ms = Msi + B. (8) Combining equations (5), (6), (7) and (8) to eliminate Mcsi , Mcni , Mni , Msi and B, we obtain the general relations between sensitivity and magnet strength as - M, 2 - M„ (9) (10) 160 120 5 60 m cr. lU 40 Q. < 20 -40 -60 i / \ RELEASE FROM N / / V ' / \ TEST VALUES \ 1 V -\ f/L^ h ^ \ \ 1 \ k— — y — M ^ ^ 1 1 7 \ K PO- p / RELEASE FROM S 10 01 2345678" ADJUSTMENT SEQUENCE Fig. 13 — Sequence of sensitivity values obtained in adjustment. BALANCED POLAR MERCURY CONTACT RELAY 1411 ADJUSTMENT OF SENSITIVITY The procedure used to adjust the relay to any desired pair of sensi- tivity values is as follows: L Magnetize the two magnets fully by a flux directly across them. 2. With the armature starting from one side, progressively decrease the magnetization on that side in small steps until the armature just operates to the other side on the final current value desired for this iluw- tion of operation. 3. With the armature starting on the other side, progressively decrease the magnetization on that side in small steps until the armature just op- erates from that side on the final current value desired for this second direction of operation. 4. Repeat 2 and 3 alternately until the relay just operates in both directions on the two current values desired. Within limitations such as the size of demagnetizing increments used per step, this procedure results in an adjustment to the two test values used. Let us consider why this result is obtained. If we should substitute p = 1 in equations 9 and 10 we would obtain Mc, = M, — Mo, Men = M, - Mn . That is, in such a case, the two adjustments would be independent of each other and the above process would require only one adjustment of each magnet. In the more general case, where the force on the armature is affected by both poles, each adjustment on one side results in a greater magnet strength at the pole being adjusted than is required for the final adjustment, because of the pull from the opposite pole, the amount of which is greater than normal because that pole has not been reduced to its final magnet strength. Thus the final adjustment is normally reached through a number of alternations between the two sides, which progres- sively approach the final pair of sensitivity values. The sequence of sensitivity values obtained in a typical adjustment of this kind is shown in Fig. 13. Dynamic Measurements on Electromagnetic Devices By M. A. LOGAN (Manuscript received July 6, 1953) A sampling switch with adjustable make and fixed break times can be used to obtain dynamic measurements of reciprocating phenomena. A test set has been developed using this principle to measure the flux-time, current- time, displacement-time, and velocity-time response of electromagnets and similar devices. The tested device is steadily cycled. A dc instrument is switched in by the synchronous control at any preselected instant in the cycle, and out when a steady reference condition has been reached. The steady reading of the instrument is proportional to the value, at the closure instant, of the variable being measured. The instrument switching is con- trolled by an electronic timing system. This system operates mercury cmitact relays, which do the actual switching. For the displcu^ement-time and velocity- time measurements, an optical transducer with associated dc amplifiers is added. The design of these devices is described. The results of an investiga- tion of dynamic flux rise and decay in solid core electromagnets are reported. Modified first approximation equations are developed to give a better repre- sentation of eddy current effects. Introduction The application of common control methods to telephone switching systems has led to the widespread use of high-speed relays. The actua- tion time of these relays is affected by many parameters such as the power supplied, how far the armature has to move, the mechanical work the armature has to perform during its motion, the winding design, the magnetic structure, and eddy currents introduced in the magnetic mem- bers caused by the application of current to the winding. The eddy cur- rents act to oppose a magnetic flux change and hence retard a building or decay of flux. This causes the actuation time to l)e increased compared to a relay without such effects. An analytical determination of the de- velopment and effects of eddy currents can l)e made for simple sym- 1413 1414 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 metrical magnetic structures having an infinitely long or torus shaped round or rectangular cross-section, assuming linear magnetization char- acteristics. However, for relay-like structures having air gaps, leakage flux which only partly completes its circuit through the magnetic mate- rial, varying cross-section so that boundary conditions become compli- cated, and non-linear magnetic properties, an analytical approach be- comes unmanageable. For a fundamental study and direct measurement of eddy current effects, a test set has been developed to measure the dynamic flux rise and decay characteristics of relays, and similar structures. This test set is electronically operated on a synchronous switching principle. It displays on an ordinary dc instrument, as a steady reading, the instantaneous flux obtaining at any selected time after energization or de-energization of the magnet. The application of the test set is restricted to devices which operate under conditions of suddenly applied or removed dc voltage and which can be cycled between these two conditions until the dc instrument reaches its steady state reading. This article wiU present the theory of the basic synchronous switching circuit and the relation applying between the dc instrument reading and the instantaneous flux linkages, a description of the electronic control circuits, and measurements of dynamic flux rise and decay using solid core electromagnets. Supplementary additions to the basic circuit are available using the same principles, to measure current-time, displacement-time, and ve- locity-time curves of reciprocating devices. The first fundamental meas- uring set using the switching principle, was built by E. L. Norton^ in 1938. The earlier set used a synchronous, motor driven, phase adjustable commutator, to perform the switching. A limitation of the earlier set in how fast it could be operated, combined with brush wear and chatter troubles led to the development of the new electronically controlled set employing sealed mercury contact relays to perform the switching. DYNAMIC MEASUREMENTS ON ELErTHOM \c;\KTI(' DEVICES 1415 Part I — Description of Fluxmeter System BASIC MEASURING CIRCUIT A schematic of the basic measuring circuit is shown in Fig. 1. A battery switching contact indicated as A, applies and removes voltage to the magnetizing winding of the electromagnetic device under test with a 50 per cent duty cycle. The time of one complete cycle is indicated as T, The electromagnetic device is equipped with a search coil of A^ turns, having dc resistance, including wiring, of R^ ohms. A B contact, syn- chronously switched at instants to be described later, connects a dc microammeter to the search coil. A damping contact (7 connects the instrument to a resistance, preset to the same value as the search coil, when the instrument is not connected to the latter, thus providing the same instrument damping at all times. The A timing cycle is shown schematically for one interval T. Next below is shown the cycle for the B contact, followed by that for the C contact. The B contact always opens and the C contact closes just before the A contact closes. The point of closure of the B contact, indicated as ti may be adjusted to occur at any time during the cycle T. The cycle time T is chosen sufficiently long so that the flux in the electromagnet has sufficient time to reach a substantially steady state during both the closed and the open intervals. The flux rise and decay have the general characteristics shown as the

'^ MICRO- AMMETER (a) C CLOSED (b) OPEN CLOSED OPEN CLOSED (c) (d) OPEN K= (e) (f) r E=Jiixio-8:=: I (? (Rc+Rm) (g) EQUIVALENT LINEARIZED CIRCUIT Fig. 1 — Fundamental operation of dynamic measuring set. DYNAMIC MEASUREMENTS ON ELEfTROM \(;\!;tI( Di \i( i;s 1 117 The dc instrument will read the average vahie of current for the cycle. As the period of the cycle is T, and ti is the chosen delay for .-losure of the B contact measured from the time of the A closure, tlw diK . t (-ur- rent mdicated by the mstrument will be: (3) {Rn. + Rc)f • Now the product LJ at ^i when the B contact is closed must be zero. Furthermore, if the period T is long in comparison with the time of rise and decay of the flux, and with the time constant L/R of the measuring circuit, the product Lmi is also negligible at ^ = T. We then have: ^t, - ^T = ?^ X 10' Maxwells, (4) where R is now the total resistance of the measuring circuit . The factor 10^ has been added to convert from practical units to c.g.s. units. The flux $r is the residual flux, so that if this is taken as a reference value, the value of the flux at the time of closing the B contact is simply $ = ?:IIl X 10' Maxwells. (5) It may be objected that the flux in equation (2) is made up not only of the flux to be measured but also of a component due to the current in the measuring circuit. This is true, and the flux linking the search coil changes differently during the time the B contact is closed compared to that without the instrument circuit. Note, however, that we are not concerned with how the flux varies between the times h and T but only with its value at the limits. Since the flux at ti is that which has been established with the search coil circuit open, the one requirement is that the constants be such that the current in the measuring circuit he sub- stantially zero just before T. This can be verified through the meas- urements themselves by noting whether there is any change in measured flux in this interval. If there is, then the cycle time T can be increased until this requirement is met. It will be convenient later to regard the switched meter as a linearized circuit composed of an equivalent dc voltage E = ^X 10-* vdis (6) in series with the resistor R, causing a dc current h to flow, Fig. 1 (g). 1418 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 MEASUREMENT OF CURRENT The instantaneous value of current in a circuit being cycled by the A contact is measured by inserting the primary of a toroidal air core transformer in series with the circuit as shown in Fig. 2. The secondary is connected to the meter through the B contact. The best design consists of a secondary occupying half the winding volume and wound to the same resistance as the meter. The primary can be wound in two sections half inside and half outside of the secondary to provide maximum mutua inductance, of a wire coarse enough to provide the number of turns for i convenient meter deflection. In an air core transformer, the magnetic circuit is Hnear and the flux exactly proportional to the current, a situa tion that does not exist when magnetic materials are present. Because of this linearity, the analysis can be made in terms of inductances anc currents rather than flux linkages and currents. By definition, the voltage induced in the secondary of a linear trans former is related to the primary current, through the mutual inductance M by the relation: which corresponds exactly to (1) with M substituted for N, and i for (p Hence, the instantaneous current at time ti is given by the relation where the other symbols have the same significance as before. This rela tionship makes the calibration of the mutual inductance easy by setting up a battery and resistor for the primary circuit and measuring the steady state current with an ammeter. Then when the A contact i; being switched and the instrument current /o is read, corresponding te the known final primary current i, relation (8) can be solved for M. MEASUREMENT OF FLUX USING A BRIDGE CIRCUIT The search coU method of measurement just described is preferrec because of the absence of any dc voltage in the meter circuit. However AIR-CORE TRANSFORMER ^TEST B C WINDING L J M Fig. 2 — Circuit for the measurement of current. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1 110 sometimes the use of a search coil is iiicoiiveiiient, or for other reiwons it is desired to measure the flux hiikages in the magnetizing winding itst>lf . The test set includes a bridge circuit shown in Fig. 3 for such measure- ments. With the contacts .1 and B ch)sed and the bridge in the flu.x rise arrangement, the bridge is first bahuiced for no current in the instrument, by adjustment of Rd. The same instrument is used for setting the balance as later used for the flux measurement. The dampinj^ resistance is next set to the value facing the instrument, namely, Re = j^^ {R. + R.i\ (0) If then the contacts are operated with the period 7\ the instniment will read a current /o, and the flux-turns at time h during rise will be N^ X 10-' = RmhT Tl + A' ^ -f |i -f ^S\ - />,{, (10) where A'' is the number of turns of the winding, ^ is the average flux per turn, and the other quantities are indicated on the drawing. The term Ll^■ is a correction term, usually negligible, involving the self inductance of the primary of the air core current transformer and i the instantaneous KR E=riii xio-« + L,i)± Rd (b) E = (Mx,o-*L,i)i (d) K(Ro+Rm) Fig 3 _ Bridge circuit for the measurement of flux— (a) and (b), rise; (c) and (d), decay. 1420 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 /y wTrTnrnirini»r\-r=^ DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1421 current at the time h; iis determined by a separate measurement de- scribed above. If its value is not needed, the air core coil can be omitted from the circuit. For flux decay measurements made after the A contact has opened, the bridge circuit has to be modified to remove the resistance Rq and KRo from being across the electromagnet and affecting its flux decay. They are transferred to the battery side of the A contact as shown in Fig. 3(c), to maintain the same battery drain. This avoids any error due to the internal resistance of the battery, which would othenvise cause the final flux at the end of a rise test to differ from the initial flux at the beginning of a decay test. The added resistor K(Ro + Rm) is connected in series with the meter to make the expression for the flux-turn linkages the same as before. For decay measurements the damping resistor is set at the value Re = K{Ro + ie„, + Ra). (11) In Figs. 3(b) and 3(d) are shown the linearized equivalent circuits for rise and decay respectively, from which equation (10) above can be derived conveniently. MEASUREMENT OF DISPLACEMENT AND VELOCITY An optical probe is provided, in which the amount of light falling on a photocell is controlled by the relative position of two flags, one cemented on each of the relatively moving parts to be studied, such as an electro- magnet. The change in output current from the photocell thus is pro- portional to the displacement of the armature with respect to the core, one flag being on each. A block diagram of the system is shown in Fig. 4. Amplifiers effective from dc up to a frequency determined by the resolu- tion required, with substantially no phase shift, deliver a current into the primary of an air core transformer proportional to the instantaneous displacement of the armature. By virtue of the linearity of this trans- former, the flux developed is proportional to the displacement. Thus by operating the electromagnet with the A contact and connecting the secondary of the transformer to the B contact, the instmment gives a dc reading proportional to the armature displacement at the time the B contact closes. The total displacement can be measured statically. The instrument reading at a time after complete operation corresponds to this known displacement, permitting the scale to be converted to a dis- placement scale. By changing one amplifier input resistor to a capacitor, the amplifier 1422 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 becomes a differentiator, in which the output current is proportional t rate of change of input voltage. With this one change, the instrumen reads the instantaneous velocity of the armature. Under these condition; the second amplifier differentiates the input voltage once, the outpu transformer differentiates a second time, and the dc instrument integrate once, leaving a net result of one differentiation. Description of System Components fluxmeter General In the following descriptions of the several components of the systen the general characteristics required are considered, the specific devig tions from ideal are determined, and an evaluation of the measuremen errors is made. The description starts with the dc instrument followe by the associated vacuum tube filter. Then the heart of the system, th contact switching circuit itself is considered. Following this is the timin impulse generating circuit, the counting rings, the time selector, coinci dence circuits, memory, and relay control circuits. These elements mak up the fluxmeter proper. The auxiliary circuits for displacement-time and velocity-time meas urements are the concern of the next part of the paper. The component are the optical system, the photocell amplifier, the amplifier-differentiate and finally the output amplifier. The last section first shows typical measurements on a telephone typ relay. Then a description of a more fundamental study of dynamic flu rise and decay in solid core electromagnets is given. This study has le^ to two new first approximation equations for dynamic flux rise and deca> as will be seen. Dc Instrument and Effect of Damping Resistance The dc instrument used for a majority of the measurements is heavily damped 50.0-ohm, 200-microampere full scale instrument, wii a }/2. P^r cent of full scale accuracy. The open circuit decay time constan is about 2 seconds, and with a shorted winding about 4 seconds. On Fi^ 5 is shown a plot of the decay time constant referred to a short circuit versus the damping resistance referred to the meter resistance. An evaluation of the small error introduced by not providing th damping resistor is as follows. Consider a square wave flux pattern pro duced by an ideal electromagnet, with zero winding time constant o DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1423 0.8 \ 0.6 -^^— — - TO 00 ~ 0.4 - 0.2 - 0 1 1 1 I 1 1 1 1 1 2 4 6 8 10 12 14 16 ^d/^ METER Fig. 5 — Decay time constant of DC instrument. 20 eddy currents. The induced voltage in the search coil then consists of two equal and opposite impulses occurring respectively at 0 and T/2. To measure the constant flux, the switching epoch can be anywhere between these two values. The switching omits the impulse at zero time. For limit- ing cases ^1 can approach either 0 or T/2. The instrument reading of course should be independent of the choice, as the flux remains at the constant maximum throughout this interval. The instrument receiv^es only the decay impulse, and thus there will be an average dc component. For the case of U approaching zero, the instrument is always connected to the search coil, which usually is of negligible resistance and the short circuit meter decay time constant applies. Just before a succeeding im- pulse, the meter will have decayed to a relative value of: r^"'^ ^ 1 - (12) where T is the cycle time as before and Ti is the short circuit decay time constant. The pointer next abruptly rises to its original maximum value, followed again by another equal decay in a cyclic manner. For the case of h approaching T/2, the instrument is connected to the the search coil only half the time, being open circuited for the remainder of the time, if no damping resistor is provided. It thus decays at two different rates following the impulse. The first half is the same as before but the second half is under the condition of open circuit. For small errors the relative decay will be: -TI2T^^-TI2T, _ J -21A+A]' '''' where T2 is the open circuit decay time constant. 1424 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 The search coil impulse is the same in each case, and for small errors, the instrument will rise the same amount for either condition followed by equal decays. The two maximum values Mi and M2 are not equal but are related by: Ti Ml M, K'+S) (15) The ratio of the two instnunent readings thus is a function of the ratio of the two instrument decay time constants. For the case of Ti = 4 sec, T2 = 2 sec, the limiting error is 33 per cent. This shows that without a damping resistor, the error can be many times the 3^ per cent instrument error, but that the timing for switching the meter damping is not critical. That is, small time gaps with the damper off will introduce an insignificant error. In a sense, the omission of the damper is equivalent to using an electronic integrator with finite in- ternal gain. Vacuum Tube Filter Circuit^ The above discussion brings out the fact that visible motion of th instrument pointer without a filter occurs, increasing as the cycle time i^ lengthened. From equation (12) and using 7" = 0.1 sec, and Ti = 4 sec^ the amplitude would be about 2}4 per cent of full scale. Reading the' center of the vibration is easy for cycle times T less than 0.1 second and for such measurements the filter is not used. Many measurements, however, require a slow pulse rate and even a sluggish instrument will follow the pulses to such an extent that an accurate reading is impossible. Fig. 6 is a schematic of a vacuum tube circuit which serves as a very efficient low pass filter. This filter must fulfill a variety of very stringent requirements, chief of which may be fisted : (a) Low loss to direct current. (b) High loss, probably exceeding 30 db to frequencies above one cycle per second. (c) A constant resistance input. (d) AbiUty to suppress peaks of relatively high voltage of the order of several hundred or thousand times the dc voltage being measured. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1425 (e) No effect other than one which may be accurately calculated on the dc reading. (f) The circuit should not greatly increase the time required for the instrument to reach a steady reading. Requirement (c) may call for some comment. It was pointed out earlier that the method requires the time constant of the measuring circuit to be small in comparison with the period of the pulses. A Hlter built in the ordinary way to have high suppression to pulses of period T would not have a time constant small in comparison with this figure. The recjuire- Fig. 6 — Vacuum tube low pass filter circuit. ment of constant resistance input is a convenient way of expressing the necessity of fulfilling this condition. The original work on the system used was done by R. F. Wick, and the featm-es to be described are due to him and E. L. Norton. From left to right in Fig. 6, the elements are as follows: a balanced impedance bridge containing the ammeter in one arm, a blocking condenser, a con- stant current high impedance power supply for the power tube, and a two stage amplifier with an interstage phase adjusting circuit. The three- point double-pole key when in the normal position removes the meter (50 ohms) and substitutes a 50-ohm resistor. When off-normal, the meter is connected in either polarity. The operation of the circuit is best understood by assuming the re- actance elements to be omitted from the bridge and the contact on the potentiometer forming one diagonal to be at the lower left. The input to the amplifier is then directly across the line and any feedback is elimi- 1426 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 nated, since the bridge is balanced. Any alternating component applied will be amplified and sent back through the meter in the opposite polarity to that coming directly from the hne. If then the gain in the amplifier is correct and its phase shift nearly zero, the alternating component through the meter will be greatly reduced. The bridge provides the constant resistance feature at the input, since the output of the amplifier can have no effect on the input impedance. It does however necessitate a loss from both line and amplifier to the measuring instrument. Practically, the difficulty in design is in securing an amplifier with enough peak output capacity and negligible phase shift at frequencies from one cycle up. This has been accomplished by several methods. Most of the phase shift is introduced by the plate condenser in the last stage. The effect of this is reduced by the constant current power supply. The gain of the amplifier is controlled by feedback secured by the potentiometer in the diagonal arm of the bridge. The gain is sufficient so that a substantial amount of feedback may be used with a consequent further reduction in phase shift. The final phase compensation is secured by the interstage potentiome- ter. The effect of this is illustrated in Fig. 7. The lower curve is the net phase shift of the amplifier without the interstage circuit. The upper curve is the phase shift which may be introduced at the grid of the sec- ond stage. By proper adjustment, these may be made to compensate each other down to a frequency of one or two cycles. The circuit constants are such that the final adjustment of low frequency phase may be made with a negligible effect on high-frequency gain. It may be of interest that if the bridge is unbalanced by shorting the lower 50-ohm resistance, an error of 20 per cent or more is introduced in the dc reading due to the extremely large effective reactance of the feedback amplifier. If the bridge is unbalanced by shorting the instrument the amplifier is quite likely to "motor-boat" and blow the fuse used for protection. It is for this reason that the instrument key may be set to replace the instrument by a 50-ohm resistor. When in this position the fuse is also shorted to avoid needlessly blowing it due to the high surges which frequently occur when the amplifier is turned on and the condensers start to charge. The filter as described, omitting the reactance elements from the bridge, would be entirely satisfactory within the power capacity of the amplifier. With certain measurements, notably those of velocity, the peaks of the wave as applied to the filter, which in this case would be proportional to acceleration, are so high that no reasonable amplifier would be able to handle them. These knifelike peaks however are so sharp that they may DYNAMIC MEASUREMENTS ON ELECTROMAONKTIC DEVICh^S 1427 easily be removed by reactance elements added to the bridge as Hhown. It will be noted that the bridge is still of constant resistance, and is still balanced at all frequencies. Three adjustments are provided on the amplifier: one controls the feedback, a second the phase correction, and the third the plate current. The amount of suppression is controlled by a three position key which may be used to cut down the sizes of the capacitors in the circuit in a ratio of two and four to one. In cases where one of the higher pulse rates is used, adequate suppression and somewhat faster readings may Ik» secured by using smaller capacitors. With the maximum suppression, two cycles may be reduced to such an extent that it has no detectable S 0 FREQUENCY FREQUENCY »• Fig. 7 — Illustration of method for phase compensation of amplifier. effect on the meter pointer. Accurate measurements at one cycle may be made, although there is a slight motion of the pointer. Analytical studies have indicated that the transient response of the circuit depends to a large extent on the ratio of the interstage coupling capacitor to the output capacitor and moreover that there is an optimum ratio of the phase compensating capacitor to the other two. In altering the amount of suppression therefore, the ratios of the three capacitors are held constant. The Switching Circuit Requirements. The basic feature of this measuring system is the switch- ing circuit, consisting of the A, B and C contacts. The re(iuirements for these contacts are: (a) Negligible dc resistance, (b) No contact chatter, (c) Contact potential less than 10 microvolts, and (d) Stability of opera- tion of 20 microseconds. Consideration of these facters led to the selec- tion of mercury contact relays as the basic switching element.s. The choice 1428 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 of relays leads to additional requirements: (e) Substantially equal op- erate and release times, (f) A make and a break contact, and (g) Sub- stantially zero transfer or bridging time when operated or released. The first requirement is to avoid correction terms in the test set equations. The second is to make the test set operation conform to the differential equation applying. The third depends upon the fact that meters having dc resistances of the order of 50 ohms are used, and no error deflection due to contact potential effects should be present. The stability requirement results from an objective of studying flux phenom- ena in intervals as smafl as 25 microseconds. The added relay require- ments will be developed during the switching circuit description. All except the last can be met by individual relays. The last one has not been met by individual relays, but by using contacts of two relays actu- ated simultaneously, and adjusting their relative timing by winding shunts, the contact switching interval can be reduced to a few micro- seconds. Contact Operation. The switching functions required were shown in Figs. 1 (b), (c) and (d) where the interval ti can be as low as 25 microseconds or almost as long as T. The A switch in unaffected in duty cycle, but the B and C switches vary from 0 to 100 per cent duty cycle. The operate and release times of relays are affected by their duty cycle and for low or high values become very erratic. To avoid this effect all relays are operated with a 50 per cent duty cycle, and the switching is accomplished by com- binations of relay contacts. The epoch of the B relay cycle can be adjusted by manual selection, anywhere between 0 and T/2. This is shown sche- matically in Fig. 8 for the make contacts. The break contacts of course show exactly the reverse behavior. The C make contact is fixed in phase and is set to open just before the A contact closes. To produce a B cycle for flux rise shown in Fig. 1 (c) where ^i is less than T/2, a parallel connection of the B and C make contacts is used. To pro- 0 1 1 5 0 1C )0 I t<-t,-* Kt,. _r' 1 °~ 48 1 50- 1 -98 1 el 9 a 1 4 ii Fig. 8 L -----T J The operation of the switching make contacts. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1429 duce a B cycle for flux decay where h is greater than T/2, a break contact on the B relay is used in series with the C make contact. Note that in all conditions the meter circuit closure is performed by a B contact and the meter circuit open is performed by the C contact, in fixed phase relation- ship to the A contact. The addition of the switched meter damping resistor is accomplished by other contacts on the same relays. A minimum contact circuit^ to perform all these switchings is shown in Fig. 9 but was found to be unsatisfactory because of failure to meet requirement (g) above. Only a single transfer switch is necessary to change from flux rise to decay, by transferring the wire marked "x" from the search coil to the damping resistance. The relays are shown in the released position with the break contacts closed. For the flux rise circuit, contact B can have a momentary open interval when it operates. However, it must not bridge or the damping resistor will momentarily be across the instrument at the initial connec- tion, causing an error. But after contact C operates and then B releases at {T/2 + ^i), an open interval at B will momentarily disconnect the instrument. This would cut out part of the current which must be inte- grated, causing an error. Thus there are conflicting requirements for flux rise measurements, and transfer contacts of a single relay cannot meet both simultaneously. For flux decay, the B contact cannot bridge because when it releases to connect the instrument, a bridging would momen- tarily connect the damping resistor across the instrument at the instant of greatest interest, causing an error. The requirements for the C contact are of the same type but less stringent, as the flux is never changing rapidly whenever it is switched. The foregoing discussion was directed toward establishing the basis for the requirements (f) and (g) above. The former is because of the need for phase reversals in contact operation to perform the necessary sc ♦ \*- Rd: RISE ^ DEC^\ . . Fig. 9 — Unsatisfactory minimum contact circuit because of transit time requirements. 1430 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 instrument and damping resistor switching. The latter is to eUminate errors due to momentary circuit configurations not in accordance with the basic equation (3). The final circuit developed is shown in Fig. 10, which uses two B relays, called Bl and B2, a single C relay, and compatible timing require- ments for the two B relays whose windings are connected in series. In all cases Bl shall be faster than B2. The former controls the damping re- sistor while the latter switches the instrument. A composite circuit is drawn, with the cross marks indicating a closed contact for the function being measured, open otherwise. The change from rise to decay in this circuit requires four switch contacts compared to two for the minimum contact circuit. As before, the relay contacts are all shown in the un- operated position. A preliminary sorting of the Bl and B2 relays is required and is based on operate and release time measurements. The faster relay is assigned as Bl, the other as B2. B2 is further slowed down selectively to match the A relay as to operate and release time by the shunts and diode shown in Fig. 10. B2 can be lined up with the A contact both for simultaneous closure of the make contacts when ^i = 0 and for simultaneous closure of the ADJUST RISE AND DECAY + VA- © © 1; % ADJUST RISE r •^ (b) Fig. 10 — Final schematic of basic contact circuit. DYNAMIC MEASUREMENTS ON ELECTliOMAGNETIC DEVICES 1431 break contact of Bl with the opening of tlie make contact of A. The A relay winding has a somewhat similar But fixed set of Hhimts across it, to slow it down with respect to B2 and to equalize operate and release time. The adjustment of the two potentiometers prot-eeds as follows. The time ^i is set at 0, flux decay is chosen, and potentiometer 1 is ad- justed to the point where the meter just starts to drop from the full reading. Then flux rise is chosen and the potentiometer 2 is adjusted to the point where the meter just starts to rise from zero. A fine check on these settings is to record the meter readings for a lew ((lual small time steps and observe that the successive differences are consistent. For applying an on and off battery condition as shown in Fig. 9 for the A contact, only one contact on one relay is used. The A contact circuit actually is made of two pairs of relays operated in a push-pull arrangement. The circuit may be changed to connect the four sets of transfer contacts into a lattice configuration to supply battery reversals to the apparatus under test. This is used in testing core materials to eliminate residual effects and for polar relays and similar structures. The electronic control circuits which apply or stop the relay winding currents will be described later. Timing Control System The timing control system consists of a frequency source, a pulse forming circuit, three binary-quinary ring counters in tandem, a time selection circuit, coincidence circuits, memory circuits, and three control circuits, one for each of the A, BI-B2 and C relay circuits. A block diagram of the system is shown in Fig. 1 1 . Oscillator. The frequency source is a bridge stabilized oscillator. This is compared to the Bell System standard frequency for calibration. The accuracy of measurement, both for magnitude and the time scale depends directly upon this oscillator. The magnitude error enters through equa- tion (5) where the cycle time T is exactly the interval for 1000 cycles of the oscillator frequency by virtue of the 1000 discrete steps in the three decades. The time scale is determined by these same steps. Hence the accuracy and stability of the oscillator enters directly; 0.1 per cent can be realized easily. This is as good as necessary as it exceeds the dc instru- ment accuracy. Pulse Shaper— The pulse shaper is shown in Fig. 12. It consists of an initial cathode coupled amplifier followed by three more direct coupled stages. The square wave output of the third stage is differentiate by a small condenser for voltage spire production. It is a Schmidt* type circuit designed by I. E. Wood. 1432 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 _i _j -J o o o ^ ir tr ££ cDh (Ji- <^- ■z 2 2 o O O o o O +J + f 1- It o ^ t-|(\J d >;i- >-i- 5^ 20 § UL U. O 2 o U- t tu UJ o o Zl- Zh CD — O liJ^ 0£r ^Oa: 2o Zo O o o o o 1 O o o 00 o 15 1 § tr ,,,£ i'" t-_i UJ CO 1 t t f 1 — I— I ir a: II ii -^ —► ! "^ o xo > 3| ^ si u. H 03 o O bC DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1433 The differentiated output appears as a positive voltage impulse across the cathode resistor of a coupling tube, impressed on top of its normal dc bias due to current from the following quinary ring. The coupling circuit has the appearance of a cathode follower, but except during the short intervals of impulse transmission, the plate does not conduct. This lends itself to supplementary use of the rings as counter circuits. An on-off gate can be connected, as shown, to enable or disable the grid bias of this tube, without introducing any false starting or stopping counts. Quinary-Binary Counters. The unit decade counter is shown in Fig. 13. It is a modification of Weissman's quinary-binary circuit* and was chosen because it provides simple two wire selection for the coincidence circuits. The principal modification consists of applying the shift pulses to the odd cathode lead rather than to a grid multiple, and using a direct coupled positive impulse for shifting. The shaded tubes are non-conducting at 0 time and can be set in this condition by momentarily opening and then closing the reset key. The plate current from the right half of the zero tube passes through the cathode resistor of the output tube in the pulse shaper described above, and biases both tubes, the latter beyond cutoff. The plate currents from the other four tubes pass through the second cathode resistor having 3^ the resistance, developing the same bias. A positive cathode impulse applied by the pulse shaper shifts the (0) tube from right to left half conduction. In shifting, the negative pulse from the prior non-conducting half of the zero tube cuts off the conduct- ing half of the (1) tube, causing it also to shift. The shift of the (1) tube puts an aiding backward pulse into the (0) tube and a fonvard pulse into the (2) tube to keep it unchanged. Successive cathode impulses continue the ring stepping, the ring closing on itself after the fifth count. Thus the ability to shift properly depends solely upon the impulse wave- form from the pulse shaper, which can be controlled independently of the ring circuit. At the beginning of the fifth impulse, a square negative impulse is passed by the conducting half of the twin coupling diode to the (5) binary tube, causing it to shift. On the next round, the other half of the diode conducts, restoring the (5) tube. The twin diode behaves as an infinite capacitance in a circuit having zero time constant. For each of ten successive impulses there is a different configuration of conducting tubes, and two wires, one to the quinary ring and the other to the binary tube, can identify when a selected state is reached by a coincidence circuit noting that a low voltage is present on each wire. 1434 DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1435 For the other nine configurations, one or the other of the wires will carry nearly the full battery voltage. The leads to the time selector switches are indicated by the arrows marked from 0 to 4, oil and 5L, inclusive. When the (5) tube reverts after the tenth impulse, its step plate voltage rise is differentiated by a small capacitor. This drives another pseudo cathode follower to shift the tens counter exactly as has been described for the units counter. For this tens ring, the take off leads to the time selector are identified as 00 to 50. In a similar manner the hundreds counter leads are identified as 000 to 500. Thus by exactly six wires, two for each decade, any cycle in 1000 can be selected. Note that for 0 and 500, needed for the A relay the (500) tube itself provides complete information. This eliminates the need for a coincidence and memory circuit for the A relay drive tube. Also by choosing 480 and 980 for the C relay, an abbreviated coincidence circuit can be used, as access to the units counter is not needed. Neon lamps are provided as indicators for each ring to aid in circuit checking, trouble shooting, and as the counter indicator when used with the gate circuit. The separation between successive time intervals which can be selected is one/one thousandths of one complete closure and open cycle, being 25 microseconds for a cycle time of 25 milliseconds, 100 microseconds for a cycle time of 100 milliseconds, etc. This is controlled by the discrete states of the counting rings used to generate the switching signals. This relation between the cycle time and the successive a^'ailable time inter- vals is not a handicap because if the device is slow and a long cycle time has to be employed, it is slow because the flux buildup or decay is slow and hence closely spaced measurements are superfluous. The maximum speed is 40 on-off cycles a second obtained with a 40-kc oscillator. The lowest speed is limited only by the ability of the vacuum tube filter to suppress instrument pointer vibration. The counting ring system with its discrete steps precludes an auto- matically recording de\'ice as would be possible with a motor driven commutator and a gear driven take-off brush. However, the elimination of brush troubles is considered to be worth this sacrifice. Time Selector. The time selection is controlled by two two-gang decade switches for the units and tens selections and one five position switch for the hundreds selection. The schematic is shown in Fig. 14. The dial positions are marked directly in time for a lO-kc oscillator, that is the units selector indicates one-tenth milliseconds, ihc tens milliseconds, and 1 1436 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 the hundreds tens of milliseconds. For other oscillator frequencies the times indicated are scaled in proportion. Thus for a 40-kc oscillator, the indicated times are divided by 4 and the fine steps become 25 micro- seconds. The start of the A cycle is marked by L500 and the stop by R500, the 500 tube itself serving as the memory circuit, and no coincidence circuit is needed. The C cycle is wired permanently with a phase to close just before 500 and open just before 0. By choosing 480 and 980, access to the units decade is unnecessary and only four coincidence circuits are provided. The start of the B cycle is marked by six coincidences, the end by five of these and the alternate connection to the 500 tube. The hundreds switch is simplified because the B relay always operates between 0 and 500 and releases between 500 and 1000. Therefore, the start and stop coincidence circuits can be wired permanently to the 500 tube and only COUNTER LAMP DISPLAY TO TIME SELECTOR SWITCHES (SEE FIG. 14) RESET Fig. 13 — Inte DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1437 five switched wires are necessary. The change from flux rise to flux de- cay by another switch as sho^vn in Fig. 10, converts the contact switch- ing to provide the proper instrument closing time, with the time dials now reading time from the beginning of the open of the A circuit. Coincidence Circuits. Each coincidence circuit consists of triodes, one for each marked wire. The circuit for the B cycle is shown in Fig. 15. Each selector switch wire connects to a grid, but all the cathodes of a set of coincidence tubes use a common cathode resistor. Except for the instant chosen, at least one coincidence tube has a high potential on its grid and the plate current through that tube holds the cathode resistor at a high potential. Only at the chosen time are all plates of the counter tubes con- ducting and therefore at a low potential. This abruptly drops the cathode potential of the coincidence circuit for one cycle of the oscillator, recur- ring every 1000 cycles. 500 cycles after each start pulse a similar stop 100K. lOOl ^<> ^ /jlF 2> >100K 3INARY _L er schematic. B COUNTER 1438 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 pulse occurs in a duplicate set of coincidence tubes. These alternate pulses control the state of the memory circuit. The C cycle coincidence circuits are similar except only four triodes are used for the start, and four for the stop pulses. Memory and Relay Control Circuits. The memory and control circuits are shown in Fig. 16. The function of the bi-stable memory circuits is to accept the alternate start and stop pulses from the coincidence circuits, switch to the state representing the imposed condition, and hold it until the next successive imposed condition. The function of the relay control circuits is to operate or release the relays under control of the memory circuits. The relay control circuits are direct coupled to the memory circuits and one or the other plate is cut-off by the grid bias condition imposed. The relays are in the plate circuits of the control tubes and either have full or no current applied. The operate and release times of the relays add a delay in the contact ROO- R400- R300-' R200-" R100-- R000-- -O 20 O 40 30 HUNDREDS SELECTOR R30' R50 R400' L500' L500' R500 B RELAYS START COINCIDENCE CIRCUITS STOP COINCIDENCE CIRCUITS C RELAYS START COINCIDENCE CIRCUITS STOP COINCIDENCE CIRCUITS A RELAYS • START ■STOP Fig. 14 — Time selector schematic. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1439 actuation compared to the state of the memory tube. By making these small and equal, through selection of the relay design, the constant time lags do not alter the relative preset switching instants. The relays are special mercury contact, single t ransfer, Western Elec- tric 291-type. They have 38 gauge 1500 ohms, 14,()00 turns windings and are adjusted to operate with 3.5 milliamperes and release with 2.0 milliamperes. The 5687 tubes switch from 0 current to 10 milliamixres through the relays. The release and operate times of the relays are about 1.3 milliseconds but as described earlier, are selectively adjusted to 1.5 milliseconds where necessary, by shunts such as those shown in Fig. 10. The bi-stable memory circuits are switched at the grids through the tw^in diodes coupling the coincidence and memory circuits. While waiting for a shift pulse, the grid potential of the memory tube is lower than the cathode potential of the coincidence circuit, and the diode circuit there- fore does not conduct. This is necessary because small voltage variations occur in the coincidence circuit cathode potential as the several tubes folloAV the decade counters. These small variations w^ould shift the memory circuit falsely if directly connected. When the shift coincidence occurs, the cathode voltage drop is to a potential lower than t hai of t lie memory circuit grid and the conduction of the diode connects the two circuits. This negative impulse shifts the memory circuit. This drives the start grid to a new potential low^er than that of the coincidence tubes cathodes, again cutting off the connection through the diode. When the coincidence cathode rises at the end of the shift pulse, the diode is even further cut off. The stop diode behaves in an exactly similar manner due to the symmetry of the circuits. The diodes thus afford a means of only momentarily making connections at the required instants, at all other times isolating the memory circuit from the remainder of the counting system. Summary of Dynamic Flvxmeter Description Up to this point, the circuit description has been concerned only \\ ith the dynamic fluxmeter. This circuit was designed and built first, aiul put into operation for dynamic current and flux studies of which one will be described later. About fifty standard type tubes in all are used, with good segregation of the circuit functions. This aids in localizing circuit troubles. Particularly useful is the neon indicator lamp di.splay, also employed when the decade system, with the gate circuit, is adapted as a counter or time interval circuit. 1440 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 I -AW-i^V\A/- rVW-1 [-AA/V i[f" i-WV-* *-AW -VW LVvV-| pAAAr £ I — LvwJ L^ oiit 2^ VvV 62 I o , _ O (T O O I 10 -^i- o o 9 I \ U) -^ "^ -J O I I -I >n o _i o o § O pi bib I I I I cr o 0 9 in if) 1 cc DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1441 •♦•250V START (JO ? lljt- — STOP V y I 12AL5 B OR C MEMORY AND RELAY CONTROL CIRCUIT R 500 L 500 — < 47 910K>^^p.-p 5687 TO A3 AND A4 '^-- Ai AND A2 RELAYS /V 47 10 K 620 K A RELAY CONTROL CIRCUIT Fig. 16 — Memory and relay control circuits. 1442 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Part II — Displacement and Velocity Measuring System The Optical System} A block diagram of the displacement and velocity measuring system was shown as Fig. 4. As part of this figure, a geometric schematic of the optical part of the system is shown. Most of the features of the optical system are due to the advice and assistance of associates in these lab- oratories who were connected with the sound motion picture develop- ment. The fundamental design is similar to the film reproducing system with certain necessary changes in dimensions. Referring to Fig. 4, the elements are the lamp, condensing lens, a slit, objective lenses, a vane on the part whose motion is to be studied, and a photocell. The lamp and the condensing lens are the same as are used in the film reproducer. The slit is the same except that its width has been increased to 0.005'' and a feature added permitting its length to be ad- justed from zero to about 0.1''. The objective lenses are inexpensive Bausch and Lomb achromatic lenses. The second objective lens is inter- changeable for different focal lengths, and may be moved back and forth in the supporting tube for precisely focusing the image of the slit on the moving vane. The three lenses and slit are mounted in a tube fastened to the lamp housing which in turn is supported by adjustable supports mounted on a vertical stand. To the lamp housing support is also fastened the photo- cell container with its first amplifier tube. A permalloy shield surrounds the photocell except for an opening for access of light which passes by the moving vane. The shield is provided to prevent the changing stray magnetic field from the nearby electro- magnet under test from affecting the photocell current. This system can be moved in any of three directions for lining up with the vanes of the device being tested. The relay or structure being measured is not fastened to the optical system support, but is secured to another appropriate stand resting on the same laboratory bench. This system may seem more complicated than necessary at first sight, but its advantage is that it provides a rectangular beam of light of con- stant width and adjusta})le length at the point being measured. The necressity of numerous light shields and screens is done away with, al- though it is advisable to throw a black cloth over the whole apparatus when once adjusted to keep out the overhead illumination. The reason for this last is of interest. The fluxmeter timing is controlled by a stable oscillator. The 60-cycle power for the overhead lights is also DYNAMIC MEASUREMENTS ON ELECTliOMAGNKTIC DEVICES 1 I 13 well controlled. The two systems therefore operate in substantially a synchronized fashion when convenient cycle times are chosen, such as 10 cycles per second. The overhead fluorescent lamps fluctuate in light intensity at 120 cycles per second, so that the contacts in the fluxmeter operate in synchronism with the fluctuations in light intensity. Any measurements made with overhead illumination getting into the photo- cell will therefore have a superimposed 120-cycle ripple. For the same reason extreme care has to be taken to avoid any 60-cycle pickup in the apparatus. The lamp and the heaters on the first amplifier tul)es are supplied with well filtered dc and cannot be operated by ac. Noise not in synchronism with the contacts is of little importance, as the averaging of the meter cancels it. A comment on the linearity of the photocell should be made at this point. This system operates on a variable width, rather than a variable density basis. Consequently the linearity of the system depends upon the uniformity of emission of the photocell surface. This can be verified by moving a vane with a micrometer and record- ing the output of the amplifier with a precision dc voltmeter. If the rela- tionship is not linear other photocells can be substituted until sufficient linearity is achieved. Amplifier System General Fig. 4 showed the three dc amplifiers in block diagram form. These amplifiers have each been designed to operate with full internal gain from dc to 10,000 cycles. The external transfer characteristic of each is controlled by its input and feedback networks. The design of these networks proiddes ideal transfer characteristics from dc up to 10,000 cycles. At 10 kc, the frequency response deviates by less than 3 db from the ideal. This same frequency corresponds to a time constant of 16 microseconds. On a small signal basis, the fidelity of measurement therefore will extend to events occurring in times of the order of 16 microseconds. A more serious limitation to the accuracy is the finite plate voltage available for the output amplifier which results in amplifier over loading on large peaks. As will be shown, this can delay the response of the meter to sudden velocity discontinuities. The amplifiers have been designed to operate from two voltage sup- plies, plus and minus 250 volts. The —250 volts is a series regulated throe stage circuit with an output impedance of less than 0.8 ohm at all fre- quencies. It also ser\^es as the reference ^'oltage for a three stage shunt regulated +250-volt supply. Keeping the magnitudes equal minimizes errors due to power supply voltage variations. 1444 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 The photocell amplifier converts the current from the high impedance photocell into a proportional voltage, having an internal impedance of 10 ohms. The following amplifier-differentiator has an internal gain of 80 db from dc to 10 kc. When used as a differentiator the rising external gain characteristic reaches 67 db at 10 kc. The output amplifier also has an internal gain of 80 db from dc to 10 kc. The feedback network includes an equalizer to produce current, and hence flux, in the air core output transformer proportional to input volt- age from dc to 10 kc. The inductance of the transformer, with constant applied ac voltage, would cause a 6 db per octave decrease in current above the frequency where its Q is unity. This is counteracted by design- ing the external amplifier gain to increase at 6 db per octave, starting at the same frequency. The output amplifier thus has the same characteris- tics as a differentiator at high frequencies and its external gain is 64 db at 10 kc. The overall gain of the system, then, is 131 db at 10 kc. Stability has been obtained, even with this much overall amplification without resort to compartmentation. Each tube stage utilizes shielded turret construc- tion, exposed interstage leads are very short and only the input grid leads are shielded because they connect to controls on the front panel. A box shield surrounds the gain selector and the displacement-velocity switch of the middle amplifier, the lowest level point on the main panel. This system uses an electronic differentiator. Ordinarily in analogue computers, these are avoided because of the high-frequency noise intro- duced as a result of the attendant large amplification. In the present system, the averaging of the succession of impulses by the dc instrument minimizes this effect. With each cycle a discontinuous section of the noise is sampled, containing a dc component, but as these are random in sign, their average gives rise to no error. The system is dc coupled up to the output transformer. Slow drifts, due to grid currents or other reasons, do not result in instrument current errors because of the differentiating action of the transformer. No actual dc source appears in the meter circuit itself. The only concern here is to maintain the amplifiers somewhere near their best operating point. Actu- ally the main source of dc drift is the temperature coefficient of the photo- cell and the voltage stability of the lamp power supply. Calibration. The calibration of the system starts with a static measure- ment of the total displac^ement of the reciprocating motion to be studied. This can be done using thickness gauges or a tool maker's microscope. Then the device is brought into alignment with the light beam and cycled DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1445 by the A contact. The cycle time is chosen for complete operation and the time selector is set for a time near the end of the operate interval. At this time the tested device is known to be at its maximum displace- ment, and the instrument reading, using the displacement connection, corresponds. A convenient instrument indication is obtained by adjust- ment of the amplifier gain controls. For instance if the displacement were 0.040'', an instrument reading of 200 microamperes could be used. At any other time, when the parts are in relative motion, the instantane- ous displacement is read directly from the instrument using the same scale conversion factor. The calibration for velocity measurements depends upon the displace- ment calibration and relationship between the input differentiating ca- pacitor and the resistor it replaces. Once a displacement gain setting has been chosen, it must not be altered during the associated velocity meas- urements except by calibrated steps. The differentiating capacitor has been chosen so that an instrument displacement deflection, correspond- ing to a given number of thousandths of an inch, represents the same number in inches per second. For the example given above, a 200-micro- ampere instrument reading would represent an instantaneous velocity of 40 inches per second, with a linear calibration for intermediate readings. A plot of measured displacement and velocity for a fast wire spring relay shown in Fig. 21, will be described when system errors are con- sidered. Photocell and Impedance Transformer Amplifier. A schematic of the photocell and dc impendance transformer is shown in Fig. 17. The high vacuum photocell has an impedance of thousands of megohms and acts essentially as a constant current device, the current depending upon the instantaneous illumination. This current is connected to the grid of a series stabilized^ twin triode amplifier tube, to which grid also is con- nected the current through the feedback resistor and a balancing current from the — 250-volt powder supply. The zero adjustment is made under quiescent conditions for the desired dc output voltage, there being, of course, essentially no dc current in the grid. The Western Electric low grid current 420A input tube is stabilized both for heater voltage and plate potential. It provides a voltage ampli- fication of exactly fji/2 or 35, by virtue of the upper tube ha\ing an im- pedance exactly equal to that of the plate of the lower tube. This input tube is mounted immediately behind the photocell in the same shield. The output from the first amplifier tube is connected through a double shielded cable to the output tube mounted on the main chassis. The 5687 twin triode output tube uses both halves as cathode followers, one 1446 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 vw -A^V-t--VVV- I UJ UJ in z o u 1 OtO 1 cro ^2 CZJ < I t^ r-t bi) fe DYNAMIC MEASUREMENTS ON ELECTROMAGNETK l)l.\ h i > 1 I |7 to pro\ade the output voltage and the other as a driver for the inMilairc! inner shield to reduce the effective capacity to ground of the intersta^ir cable connection. The cathode potential of the output tube is operated at exactly +125 volts, set by means of the zero adjustment previously described. The input networks of the succeeding amplifier-differentiator are shown as part of this circuit as they provide the path for the output tube current and enter directly into the fre(nu'ii< \ response and loop gain cutoff design of this feedback amplifier. The dc grid voltage of the suc- ceeding amplifier is 0, and this potential is found one-third the way down the cathode resistor which connects to the —250 volts. This point then provides the input to the following grid for displacement measurements and the input resistance for gain computations of the following amplifier is only this one-third part of the total cathode resistance or 10,000 ohms. For velocity measurements, the succeeding grid is switched to the polystyrene differentiating capacitor whose grid side also is kept in readiness at 0 voltage by a grounded 1 -megohm resistor. For a change in current from the photocell due to a change in light, the feedback acts to supply an equal but opposing current from the output to keep the input grid nearly at its virtual ground potential. Thus the output voltage change E, is given to a good approximation by the simple relation E, = I^^-I,Rf. (16) 1 - /ijS where Is = change in photocell current, Rf = feedback resistance, and MiS = loop gain. A word about the differentiator connection is in order here. For this use, the cathode load approaches 160 ohms at high fre(iuencies because the input grid of the next amplifier is a virtual ground point, and the load becomes equal to the phase shift controlling resist. )i- in scries with the 0.1 mf capacitor, the value of which will he (hs.ussed later. The impedance of the cathode follower without feedback is 250 ohms but this does not mean that it can be connected to a 250 ohm load and op- erated at the normal power output rating. The basic limitation for any tube is the allowable plate current change, regardless of the extenial load, consistent with never drawing grid current or l)eing cutoff. For instance, in a cathode follower, if a load equal to l/Clm is used, the small signal output voltage is only half of the applied grid voltage, a (i db loss. Lower load resistances result in corresponding greater losses. In the present circuit, about an 8 db reduction in loop gain occurs at 10 kc ^AA vvw— AAA ' < o — ?<l in I LX^ No 1452 DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVrCES 1453 360 320 280 240 200 160 120 80 40 0 -40 / +0.5 VOLT INPUT j / -N +025 1 jl (a) RISE TIME 1 \\ \\ (b) DECAY TIME \\ \ ^ \ N 1 1 1 1 1 1 ! . 1 1.1 1 I 1 0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 TIME IN MILLISECONDS Fig. 20 — Output amplifier square wave response. followed by (2) an oscillation caused by the frequency response of the amplifier deviating from the ideal for frequencies above 10 kc. The for- mer effect is caused by the finite power supply voltages available in the final stage of the output amplifier. The current in the primary of the output transformer represents the function being measured. If a step discontinuity occurs, as during ve- locity measurements, then the current suddenly has to be changed to a different value. The rate of change of an increase in current which can be developed is proportional to the power supply voltage, and hence is finite. For a decrease, the distributed capacitance of the winding and the shunting resistors delay the current decay. These result in a delayed transition from one condition to the other, occuring, from Fig. 20, in about 0.1 millisecond. A discussion of the operating conditions of the system leads to a method of evaluating these forms of error and determining what limitations are imposed upon the use of the measured data. It will be shown that only immediately following velocity discontinuities are the data not usable. Displacement data are never in difficulty from an overload standpoint and fortunately we ordinarily are not too interested in velocities after impacts. Discontinuity Errors Dm to Overloading The arbitrary gain setting for the output amplifier is chosen to provide about the full scale of 200 microamperes to represent the full displace- 1454 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 S3HDNI-nilA| Nl 'x'3DNVlSia o D ^ fVJ O 00 <0 'T f QNOOaS b3d S3H0NI Nl 'X 'AlDOiaA 9,M M 03 o3 0) OO S^ 7 22 r- III o3 oj n ^ U o as. ^s ro (0 ^.2 G ^ Mo n ^ 2-5b nj o LU ^.2^ In ^---S 2 1 (0 2.i^ - _i Q z o ? <^B o '^ .. o III 1^ a s CO 02 >— 1 fVJ 1 ' 1 O 5^ 02 "? ^ m^ « ' •7 -i2 C 03 d O 4J III 5; u C o til 5 g C3^ -^ i^j o> to o O) (UXJ CO u 1- ty m eleas ase, o Jg£ r- u ir ;o III C -O 03-13 LL C .C < c« C 03 /) O - (n 0) OJ.S Q 7 •. \ MEASURED FLUX DECAY < ■|-= 0.601 e-''^AeN \ V. \ \ ^>.^ \ \^ '•^-, \ \ v \, \ s \ \ \ 0.2 0.4 0.6 o.e I.O .2 1.4 1.6 1.8 2.0 2.2 2.4 2.G 2.8 TIME t/te Fig. 23 — Comparison of measured open-circuit flux decay with one-term exponential equations. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVICES 1461 The true decay cun-e has an initial steep slope followed by continuous curvature. A better approximation is to stipulate an initial jump dis- continuity. A good fit to the flux range in which release of «>U'.t inmagnets occurs, shown by the lower straight line, is I = .601 e"'^'" t ^ 0. (20) This discontinuity of flux in the first approximation is just the reverse of the continuity of flux concept usually used. Of course, no actual dis- continuity occurs, as shown by the true decay curve. The failure of any single exponential equation to represent the true curve merely makes clear the fact that the behavior of the core is not that of a single coupled turn, but rather is that of an infinite line. Bozorth^ gives the intercept as 0.691 for the linear case, which does not represent the continuously curving decay which actually occurs. The chosen intercept of the first approximation curve at / = 0 is ad- mittedly somewhat arbitrary. It was arrived at in a broader study in- cluding flux rise curves. Accepting this ec^uation, a convenient determina- tion of te can be made similar to that for exponentials. If t is set eciual to te, then = 0.221. (21) Thus, after measuring a dynamic flux decay curve, the time can be determined for which the above ratio obtains. This directly is te. From linear circuit theory and Lenz's Law, the inductance for one turn is: i. = r/. (22) whence Ge can be determined. Now because of magnetic saturation and the shape of the hysteresis loop, the values will depend upon the particular final ampere turns (NX) used in the experiment. For uniqueness and uniformity in rating electro- magnets for comparison purposes, the particular set chosen is that for which Li is a maximum. For comprehensive operating studies, measure- ments of course have to be made under the actual conditions of interest. Except for rating purposes, t^ is not ordinarily split up into components. Thus, while the rated value of Ge for a particular electromagnet has the dimensions of conductance, it includes other factors as well. Some are: (a) Core material conductivity, (b) Magnetic non-linearity, (c) Shape of the hysteresis loop, (d) Non-uniform flux distribution, (e) Eflfect of pole 1462 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 face and other air gaps, and (f) Decay of leakage field. It, therefore, also has the nature of a mop-up factor in which is included other effects not explicitly covered in the elementary analysis. The above discussion has been directed toward establishing an easily determined core time constant having characteristics useful for rating purposes, and demonstrating that the distributed nature of the core eddy currents precludes an accurate representation of the core as a single coupled turn. Open circuit flux decay measurements of round and rectangular solid core electromagnets of magnetic iron, 1 per cent silicon iron, and 45 per cent permalloy all are accurately represented by the measured flux decay curve of Fig. 23, after establishing appropriate values for Ge and Li in each case. DYNAMIC FLUX RISE Dynamic flux rise curves, measured with the relay armatures blocked open, are shown in Fig. 24. A family of curves, is needed because of the added variable of the winding. The value of te is determined by the method just described, except that the armature is held open, resulting primarily in a new and lower value of Li. The winding may be charac- terized by its coil constant: Go = N^/R, where N = number of turns, and R = dc resistance of winding circuit. These curves are a composite of measurements on many structures and fit all data to within a few per cent. An empirical closed form expression has been determined which fits these data, and in fact was used to compute these curves. However, once the curves have been plotted there is no further need for the rather complicated expression. Any particular rise curve desired can be inter- polated from the drawing. These curves again show, with windings of relatively low coil constants, a divergence of the dynamic flux rise from being a straight line. The particular curve Gc/Ge = 0 actually is a decay curve because such a rise measurement would involve the use of both an infinite voltage battery and an infinite resistance winding. Now for this case, as well as for all the others, the rise and decay curves differ because of the shape of the hysteresis loop, the decay curve persisting longer. This effect is most evident for the last few per cent of the flux change. However, the differ- ences are small on a full range basis. The curves shown near Gc = 0 are DYNAMIC MEASUREMENTS ON V! »•,•',??<»>! \. . \ i i k DKVK K.s llCii a compromise for this effect. Actually the rise curves do not cross, but they do approach each other. Coil to core constant ratios now in use range from 2 upwards, the case of 2 being the best condition for 25 watts power. An examination of this particular curve shows some curvature for small times. This effect of an initial increased rate of rise reduces the operate time of an electromag- net, and is automatically taken advantage of in the experimental design of windings. With a winding as part of the system, an elementary consideration shows why this curvature exists. This follows from N^^E-iR, (23) where: iV = number of winding turns, E — applied battery voltage, R = resistance of winding, i = instantaneous current, and ^ = in- stantaneous flux. This equation is exact and holds whether or not there are eddy currents. At the instant of circuit closure there are no winding or eddy currents (because of leakage inductances if for no other reason) and the initial rate of flux rise is dependent only upon the number of turns and the applied voltage. After a transition interval, the eddy cur- rents become effective and slow down the rate of flux rise. For any given total flux, the time required, therefore, is less than that given by the well known equation: ^ = 1 - e-'/^i(«.+<'.). (24) However, for large coil constants where Gc predominates the effect diminishes and the above equation is an excellent representation and desirable for its simplicity. NEW FIRST APPROXIMATION FLUX RISE EQUATION Returning to Fig. 24, while the effect we have been discussing is all important for open circuit flux decay, practical coil constants at present do not have ratios to the core constant much below 2. These curves do not diverge greatly from straight lines on this plot. Can a minor correc- tion term be made a part of the simple equation heretofore used which will retain its simplicity and extend its accuracy to windings now used? One such equation is ^ = 1 - c-'i^ii<'c+o.*-<''''''\ (25) 1464 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 where $ is the final flux. This may also be written in integral form as . = (a + 0.— )f^/^-^,. (26) The change is the introduction of the exponential modifying the core constant Ge. Fig. 25 shows comparisons of curves for four different coil to core constant ratios. In each case, the actual flux rise, the older equation, and the new first approximation are shown. For large coils (as an example on Fig. 25, Gc/Ge = 10) the actual flux rise is well represented by either expression, their accuracy being within 1 per cent. For very small coils (on Fig. 25, Gc/Ge = 0.5) it is clear that the older representation is never a good approximation of actual flux rise. The new approximation represents the start of the dynamic flux rise quite well, but is not accurate for the second half. However no single exponential 0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 5.5 6.0 6 t/te Fig. 24 — Flux rise curves. 5 7.0 7.5 8.0 8.5 9.0 DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DEVK Ks I \(\r> equation can represent the actual rise when the plot is as curved as this one IS. The best that can be done is to approximate the more important initial part with a straight line, and in this sense the new approximation shows a good fit. A typical speed relay will have a value of Gc/G. around 2 and the range from 1 to 10 covers all relays in this class. For a ratio of 2, the new representation of flux rise is within 2 per cent of the actual rise, as opposed to about 5 per cent with the earlier equation. For the ratio of 1, the accuracies are 4 per cent and 11 per cent respectively. ACTUAL FLUX RISE NEW FIRST APPROXIMATION EQUATION 0.85 *^~ " CORE CONSIDERED AS A COMPLETE TURN ___ ./ / 1.0 / / / 0 80 / / > /■- ^^^ .V 7 -^x t / <7 / \ / 0 75 / / / / / ''// / / / 1 1 / / / >/ / 0.70 I 1 1 / / '-/ f y / y. / f / / <-V / / // / 2.0. y 0.65 / / J / >7 / ^ X 1 / / A 7 f y f / / /^ / /^ y / // / y / 1 f 7 1 A / V * / t / y / / // / / / ^ 4 / // ^ y // 9 *- 10 *- \ 0.2 ^/// V ,01* >m0^ I/a '4y ^•^ i^** 0.1 0 *3r — — - 0.5 1.0 1.5 2.0 TIME,t/tt 2.5 3j0 3.5 Fig. 25 — Comparisons of equations for flux rise. 1466 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Fig. 26 — The dynamic fluxmeter. Fig. 27 — The optical probe and associated dc amplifier system. DYNAMIC MEASUREMENTS ON ELECTROMAGNETIC DLVlCLb 1 107 Thus we have shown the new expression more accurately represents the eddy current effect on the initial flux rise for speed coils, and for most relays the overall accuracy will be within 2 per cent, and all relays should be within 4 per cent, an error reduction of about two-f birds compared to the older approximation. EQUIPMENT The dynamic fluxmeter is sho^v^l in Fig. 26. Two panels, each with a subchassis attached, comprise the set. The dc instrument and a supple- mentary voltmeter and ammeter are on the bench beside the cabinet. The upper panel contains the timing system, with the time selection con- trols. The lower panel includes the power supply and the test relay circuit controls, including the mercury contact relays. The control keys arrange the test relay circuit for any test condition. The optical probe, a test relay, and the dc amplifier system are shown from left to right in Fig. 27. The photocell is in the shielded container above the relay, with the lamp and lens system below. A right angle prism turns the light beam into the vertical, to pass between the vanes on the relay. The upper panel of the bench rack is the —250- volt supply. The lower panel includes the +250-volt supply, the three dc amplifiers and the magnetically shielded air-core output transformer. Its secondary is con- nected to the fluxmeter through a shielded cable. The controls are for zero setting the dc amplifiers, for selecting the amplifier gain and whether a displacement or velocity measurement is to be made. REFERENCES 1 E L Norton, Dynamic Measurements on Electromagnetic Devices, A.I.E.E. Trans.,64, p. 151, April, 1945. „ . ,. ^. . ^„ .. 2. Keister, Ritchie and Washburn, The Design of Switching Circuits, D. Van Noe- strand, 1951. ^ ^, „. ,^«« 3 Otto Schmidt, Thermionic Trigger, J. Scientific Instr., 16, p. 24, 1938. 4. Richard Weissman, Stable Ten-Light Decade Scaler, Electronics, 22, May, 1949. 5. M. Artzt, Survey of DC Amplifiers, Electronics, Aug. 1945. 6. Bode, Network Analysis and Feedback Amplifier Design, I). \ an Xostrand 7. R. M.Bozorth, Ferromagnetism, D. Van Nostrand, 1951, Chapter 17. Selenium Rectifiers — Factors in Their Application By J. GRAMELS (Manuscript received July 1, 1953) Selection of the proper selenium rectifier stacks for best results in the de- sign of dc power supplies involves consideration of characteristics not ordi- narily found in published data. This paper describes the data required for the selection of selenium cell sizes and cell combinations, shows typical voltage-current characteristics, and gives the results of extensive life test data necessary for evaluating the life expectancy of the product. The life test data indicate that there are substantial differences in the life expectancy of selenium stacks as manufactured by various companies in this country. Shorter life can be anticipated as the rms cell voltage ratings are increased. In addition, the life is affected by the current density and the temperature at which the selenium cells are operated. INTRODUCTION Since their introduction in this country about 1939, selenium rectifier stacks have proved to be a useful means of converting ac power to dc power for Bell System applications. These applications vary from 2 to 2,500 volts and in power sizes from a few watts to 10 kilowatts. Properly designed, the selenium rectifier has a relatively long-life expectancy and requires a minimum amount of maintenance. For these reasons, selenium rectifier stacks are widely used in telephone plants for battery charging, relay operation, plate and filament supply for vacuum-tube amplifiers, bias supplies, telegraph and teletypewriter circuits. Up to 1952, about 245,000 rectifiers of all types have been manufac- tured for the Bell System. These include tungar, copper-oxide, vacuum tubes, thyratrons and selenium types. Of this total, about 25 per cent are of the selenium type. Although Bell Laboratories studies of selenium rectifier stacks date from 1939, rectifiers using such stacks did not enter the U^lephone plant until 1945. From 1940 to 1945, however, selenium stacks were designed 1469 1470 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 mdely into communication systems for World War II military projects. Since 1945, selenium rectifier power supplies have increased rapidly. For example, in 1945, out of a total of 9,000 rectifiers of all types, about 1,000 or 11 per cent used selenium. In 1951, about 30,000 out of 45,000 rectifiers, or 67 per cent, were of the selenium type. There are many companies in this country who manufacture selenium stacks but the quality and behavior of the various manufacturers' products show considerable variation, particularly in regard to life ex- pectancy. For many years the Laboratories has carried on an extensive testing program to evaluate properly the various suppliers' rectifiers. As a result of these continuing investigations, the Laboratories is in a position to select cell sizes and combinations of selenium rectifier stacks for use in new power applications. Specifications then are written on one or more suppliers. These specifications have a three fold purpose: (1) They are used as a purchasing and inspection document, (2) they cover the electrical and mechanical requirements necessary for proper elec- trical design and equipment layout, and (3) they are useful in maintain ing records of the electrical and mechanical characteristics. CELL MANUFACTURE AND STACK ASSEMBLY A selenium rectifier cell is an elementary rectifying device having one positive electrode, one negative electrode and one rectifying junction. Cells are made by coating a chemically treated base plate, usually alum- inum, with a thin layer of purified selenium to which a halogen element (chlorine, iodine or bromine) has been added. This mixture is applied to the base plate by one of the following methods: 1. Sprinkled or dusted on and subjected to heat and pressure. 2. Deposited by an evaporation process in an evacuated chamber. 3. Dipped in molten selenium and spun. The selenium then is converted to the desired crystaUine structure by heat treatments. During the heat treatment, a blocking or barrier layer is formed on the exposed surface of the selenium. This layer is further developed by various chemical means, which are "trade secrets" with each manufacturer. A thin layer of low melting point alloy, the front electrode, is then sprayed on the selenium. The manufacturing process is completed by electrically ''forming" the cells by applying a pulsating dc voltage in the non-rectifying direction for a specified time interval. Selenium cells are assembled on an insulated bolt or stud, mth in- dividual cells separated by metal spacer washers to allow free passage of air for cooling the assembly. Contact terminals are brought out in various SELENIUM RECTIFIER APPLICATION CONSIDERATIONS 1471 arrangements for series, parallel or series-parallel connection of the ceUfl as required. The completed assembly is designated as a rectifier stack. A rectifier stack is a single structure of one or more rectifier (!ells. Small size cells (less than 1") have no center hole for mounting. These cells are assembled without spacer washers in glass, metal, phenol fibre or bakelite tubing. Terminal leads extend outside these enclosures. SYMBOLIC NOTATION The combination of cells on a stack is described by a sequence of four symbols written a-b-c-d with the following significances:* (a) number of rectifying elements; (b) number of cells in series in each rectifying ele- ment; (c) number of cells in parallel in each rectifying element; and (d) symbol designating circuit or stack connections. The symbols for the more common types of stack assemblies are shown schematically in Fig. 1. A basic selenium stack is defined as a stack having a single selenium cell in each rectif3dng element. For instance, a 4-1- IB stack is a basic full-wave single-phase bridge rectifier stack with one selenium cell in each of the four rectifying elements. The total number of selenium cells on any stack is the product of the three numbers indicated. For example, a single-phase full-wave bridge stack with three cells in series and two cells in parallel per rectif3dng ele- ment would be designated as a 4-3-2B stack assembled with 4X3X2 or 24 cells. "H" HALF-WAVE STACK "D" DOUBLER STACK "C" CENTER TAP STACK ONE RECTIFYING ELEMENT TWO RECTIFYING ELEMENTS TWO RECTIFYING ELEMENTS 1-1-lH 2-1-lD 2-1-IC OR AC AC 'B" SINGLE-PHASE FULL-WAVE STACK "B" 30 FULL- WAVE STACK FOUR RECTIFYING ELEMENTS SIX RECTIFYING ELEMENTS 4-1-1B 6-1-IB - AC + AC - I AC AC I AC + Fig. 1 — Stack assembly sjTnbols. Color coding of terminals (AIEE and NEMA Standard) : yellow for ac, red for plus dc output, and black for negative dc output. * This method of specifying stack assemblies has been standardized by both the National Electric Manufacturer's Association and the American Institute of Electrical Engineers. 1472 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 DESIGN CONSIDERATIONS The proper selection of selenium stacks for use in dc power supplies involves a number of important factors that must be carefully consid- ered. (1) Circuit requirements must be carefully analyzed so that alllowance may be made for the normal variations in the voltage-current charac- teristics of each manufacturer's product as well as variations that exist for the same stacks processed by different manufacturers. For a fixed ac input voltage, differences in the forward voltage drop may vary the dc output voltage at least zb 3 per cent from the mean value. If this cannot be tolerated, selenium cells have to be carefully graded and selected to obtain uniformity or other circuit adjustments have to be provided. Special selection of cells, obviously, will increase the cost of the product. (2) The engineer must take into account the magnitude of the changes in the voltage-current characteristics of the rectifier stacks over the specified temperature range of his project. At very low temperatures, output voltages may be 5 to 10 per cent lower than at normal room tem- peratures. For high temperature operation, the stacks must be properly derated for both current and voltage to prevent overheating and rapid failure. (3) Selenium rectifiers age with time. Compensation for this aging should be provided if load requirements warrant. The project engineer should determine what life he expects or requires of the application. For military applications life requirements may vary from minutes to thou- sands of hours. On other applications, such as telephone and elevator in- stallations, it is desirable to design selenium rectifiers for life expectancies of ten to twenty years or more. (4) The equipment engineer must anticipate the differences in me- chanical details of the same stack assembled by different suppliers. There is no standardization in the selenium industry regarding cell sizes or mechanical details such as the overall length and height of the stack and particularly the type of mounting. However, a committee for the National Electrical Manufacturers' Association is attempting to stand- ardize these mechanical details so that stacks assembled by different suppliers will be mechanically interchangeable. (5) Unless otherwise specified, rectifying stacks are coated with vari- ous types of paints and varnishes for protection against moisture in normal conditions of humidity. For military projects and other applica- tions where selenium rectifiers may be exposed to high humidities, fun- gus, salt or other corrosive atmospheres, the rectifier stacks must be SELENIUM RECTIFIER APPLICATION CON8IDKH \Tff •Sk M73 provided with a more suitable type of protective coating or iuMi. 'Hiese finishes are available from most manufacturers. (6) When selenium stacks are mounted in cabineU or housing >nth other heat-generating devices, they should he arranged in such a manner that heat from the other components docs not roach the rectifier stacks. The stacks should be mounted below the oilin ...mponents, and in such position that the free flow of air through the net ificr stack is not impeded. The stack should be mounted with the ass('inl)ly stud in the horizontal, not vertical, position. When two or more stacks are moimicl m i h. -mic housing, they should be in a horizontal plane with each other, or projx^rly staggered. Cabinets should be provided with louvers to dissipate the heat within the enclosure. ELECTRICAL RATINGS The electrical ratings of selenium rectifier stacks are based on their voltage, current and thermal characteristics. All three must \ye consid- ered carefully for initial design purposes, as any one can affect life ex- pectancy. Voltage Ratings The voltage rating usually is expressed as the "reverse voltage rat- ing." It is the maximum rms sinewave voltage above which an excessive reverse current would flow and overheat the cell, causing breakdown. However, the important consideration establishing the rating is the peak voltage applied to the cell. If the applied voltage differs significantly from a sine-wave, it is important that the applied peak voltage shall not exceed 1.41 times the rated rms voltage. When selenium cells originally were manufactured in this country, their rms reverse voltage ratings ranged from 14 to 18 volt«. Ratings later increased to 26 volts. One manufacturer already has successfully produced 33-volt cells for several years; and another supplier announced recently a 40-volt cell. Cells rated at higher voltage have been produced in the laboratory. For non-critical applications, such as low-cost radio and television sets, some selenium manufacturers make cells rated at 45 volts rms. Generally, in such applications, a high reliability product is not required and the units may have a relatively short life. The nominal dc output voltages obtained from various common basic rectifier stacks assembled with cells rated at 18-, 26- and 33-volt rms are listed in Table I. These ratings apply for stacks operating at mUsl dc output current into a resistance load. 1474 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Current Ratings The current rating of selenium cells is based upon a normal current density of 0.32 ampere for each square inch of active rectifying surface for a single-phase full-wave bridge rectifier stack (4-1-lB) operating into a resistance load. The rating applies to stacks in which the cells are sep- arated by spacer washers so that the cells are cooled by convection air currents. Small size cells mounted without spacer washers are rated at a lower current density. The dc output current ratings for various selenium cell sizes are listed in Table II. These ratings are based upon continuous operation in ambient tempera- tures up to -f 35°C with unrestricted ventilation for convection cooling of the stacks. For battery or condenser loads on single phase circuits, the above current ratings usually are derated to 80 per cent of the above values. Thermal Ratings Voltage and current ratings usually are based upon operation of the rectifier stack in a normal ambient temperature of +35°C. It is im- portant to note, however, that for selenium cell application, ambient temperature is defined as the temperature immediately surrounding the stack within the equipment enclosure, not the temperature area where equipment is installed. Stacks operating at rated current and voltage into a resistance load have temperature rises ranging from 15 to 30°C when measured with a thermocouple in direct contact with the cell. The temperature rise depends upon the manufacturing techniques used by the various companies as well as the spacing of the selenium cells on the assembled stack. For long life expectancy, the actual cell tempera- ture should not exceed +75°C. Table I — Nominal DC Output Voltages of Rectifier Stacks Basic Stack Cell Rating, Volts rms. Single Phase Half -wave Full -wave Center tap .... Bridge Three-Phase Full -wave bridge 1-1-lH 2-1-lC 4-1-lB 6-1-lB DC Output Volts 18 6 12 19 26 9 18 30 33 12 12 24 38 SELENIUM RECTIFIER APPLICATION >ii)i;katioN8 I i For operation in ambient temperatures above -f 35°C, the voltage or current ratings, or both must be reduced to limit the cell temiKTuture (ambient plus heat rise) to -f-75°C. The de-rating factors for voltage and current vary somewhat from supplier to supplier, but Fig. 2 shows typical de-rating curves for operation of selenium stacks above a +35*'C am- bient. Selenium rectifiers can be operated at cell temperatures above TS^C, but aging accelerates with temperature and ex(^essive temperatures will result in rapid failure. A special problem today is the application of selenium rectifiers for high temperature military uses. Many military projects are requesting designs to operate at ambient temperatures of +70 to -}-90°C. Unfor- tunately very little data about rectifier operation and life expectancy under these conditions have been obtained up to now. VOLTAGE CURRENT CHARACTERISTICS To demonstrate the non-linear characteristics of selenium rectifier cells, Fig. 3 illustrates representative dynamic voltage-current curves showing the open circuit and short circuit characteristics of bridge con- nected stacks composed of various cell sizes rated at 26 volts, rms. This rating was chosen since all manufacturers presently are producing cells of this type. Cells with higher voltage ratings generally have a slightly greater for- ward voltage drop. Table II — DC Output Ratings CeU Size y2'' V X 2" X V X 5'' X h" X Dia. . . Dia. . . X IW- xVAr 2" .. .. z\. . Dia. . h" x6''.. 6*.. .. Area Square Inches* 0.049 0.196 0.56 1.1 1.7 3.1 7.0 12.5 21.2 21.2 26.2 Continuous DC Amps at +3S*C Single-Phase ' Three- Phaae HaU 0.006 0.025 0.090 0.175 0.270 0.500 1.10 2.00 3.40 3.40 4.20 Iridce CJ. 0.012 0.050 0.18 0.35 0.54 1.0 2.2 4.0 6.8 6.8 8.4 Approximate active rectifying area (varies with suppliers) Bridfe 0.018 0.075 0.270 0.525 0.810 1.50 3.30 6.00 10.2 10.2 12.6 1476 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 The data shown in Fig. 3 and subsequent data are plotted for basic 4-1-lB stacks. Tests were made with 60-cycle sinusoidal supply voltage. The * 'forward voltage drop" is expressed as the rms volts required to produce a specified current in a moving coil dc ammeter connected di- rectly to (short circuiting) the output terminals of the rectifier stack. For stacks w4th more than one cell per element, the voltage drop is ob- tained by multiplying the observed drop in Fig. 3 by the number of cells in series per rectifying element. The reverse current is measured by applying a specified rms voltage to the ac terminals with the dc terminals open circuited and noting the rms input current after the current has stabihzed. When selenium stacks have been ''off voltage" for some time, a relatively high reverse current is obtained for the first few seconds. The current then decays approxi- mately exponentially. Usually, the current will stabiUze after voltage has been applied for 5 to 10 minutes. It should be emphasized that these curves are only typical characteris- tics. Selenium cell manufacture requires individual testing of each cell before assembling into a stack. Cells are graded by their electrical char- acteristics. Large variations exist between the lowest and highest grade. Each manufacturer sets up his own standards regarding the variations lOU 90 80 1 ™ 1- < .60 O 50 h- 40 Z lU u a. 20 10 0 N V >v R MS INPUT VOLTAGE \ ^ . \ DC OUTPUT .CURRENT u! \ Zj l Dl.i. \ l lu.\.^ 1177 of characteristics in a given grade. More than one grade of cell may Ix; assembled in a rectifier stack. The art of manufacturing selenium cells is such that, considering production over a yearly period, diflFerences in the forward voltage drop at rated current may vary as much as ±30 to 50 per cent from the mean value. In the reverse direction, a larger per cent spread exists in the reverse current, particularly in cells or stacks made by different suppliers. Fig. 4 shows dynamic characteristics plotted on a linear scale to illus- trate variations of the forward voltage and reverse current character- istics of selenium rectifier stacks that are processed by two different suppliers. Variation of this magnitude exist, not only from supplier to supplier, but also may occur in a particular suppliers' product. Selenium rectifier stacks in common with other semi-conductor recti- fiers, have a negative temperature coefficient of resistance in the fonvard direction. The forward voltage drop at a specified current decreases as the ambient temperature increases (see Fig. 5). In the reverse direction, the reverse current decreases as the temperature is lowered to approxi- mately — 20°C. Below this temperature, there is no apparent change in the current except at the higher voltages. At the higher voltages, the current again tends to increase. STACK DESIGN The dc output voltage-current characteristics for various ac input voltages of basic rectifier stacks (one cell per rectifying element) are represented in Figs. 6, 7 and 8. The data were obtained by maintaining a constant 60-cycle rms voltage at the stack input terminals and do not take into account transformer regulation or other regulating devices that may be used. Fig. 6 shows the single phase full-wave bridge characteristics for re- sistance loads. Fig. 7 shows the single-phase full-wave bridge characteri-stics ior battery loads. It vnW be observed that, with single-phase circuits for a given dc output voltage, lower ac input voltage is required for a battery load than for a resistance load. Capacitor loading is somewhat similar in output characteristics to battery loading, but the value of output voltage is de- pendent upon the magnitude of the capacity, the quality of the capacitor, and the current drawn by the load. For batt^^ry loading, the output volt- age is dependent upon the type of battery, the condition of the battery, the battery voltage and the charging current rate re(|uire § 1 s; I ^*v\! ^^ t < 14 ^- 18 Z 111 (- _l o ' ■ — . 16 ^ ■ 14 -* 10 1— D Q. ^ 12 5 8 O ■~^"~~ ^ ■ — 10 6 ^- , - 8 4 2 0 _6Er 0 25 50 75 100 125 150 LOAD AS PER CENT OF RATED CURRENT Fig. 6 — Typical output voltage-current characteristics. 60-cycle single-phaae full-wave bridge circuit. Basic stack 4-1-lB. Resistance load. 1482 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 V TEMPERATURE 25°C \ \ ^ v ■ — — ^ ■ — 32_E RMS \ ^"^ ^ ^^ ■" \ ^ ^ ■ — _30_ V ^ . _28_ \, ^^ "" ^ -— 26 V — ■ --- — V "^ ^ . ' _24__ \ ^. ^ ^ — ■ __22_ • . \j ^^ .^ ■ , ^ __20_ \, "\ ---- " — — ^18 _ 16 ^ ^ ■-^ , \ ^ .^^ "~~~~- \, ^^ — _ 14 . S, -^ — — 12 ^ — "^ ■— — 10 ^ "^ ^ -^ __ ^ ^ ■ — ■ — 6 ErmS 1 20 160 40 60 80 100 120 LOAD AS PER CENT OF RATED CURRENT Fig. 7 — Typical output voltage-current characteristics. 60-cycle single-phase full-wave bridge circuit. Basic stack 4-1-lB. Battery load. SELENIUM RECTIFIER APPLICATION « mn>ii,kh \ lloNS 1483 <2> 40 60 80 100 120 LOAD AS PER CENT OF RATED CURRENT Fig 8 — Typical output voltage -current characteristics. 60-cycle three full-wave circuit. Basic stack 6-1-lB. Resistance and battery loads. t«o phase 1484 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 selected would now be rated at 0.50 X 1 or 0.5 ampere. Therefore, for the specified 1.0 ampere load, two cells in parallel are required or the next larger cell size, (3" square) could be selected. The 3'' square cell would be rated at 0.50 X 2.2 or 1.1 ampere. At 60°C ambient, the input voltage for new or aged stacks should not exceed 92 per cent of the normal rms voltage rating. An alternative method of calculating the input voltage required for any stack cell combination may be obtained by the use of the formulas shown in Table IV. Since the selenium rectifier stack is a non-linear device whose charac- teristics vary depending upon the circuit and operating conditions, the values given in Table IV are not precise but they are reasonably accurate for most design purposes. To illustrate the formulas method, let us determine the input voltage for the previous example, that is, a single-phase full-wave bridge stack to supply 1 .0 ampere dc at 48 volts into a resistance load . As previously determined, the stack cell combination would be a 4-2- IB circuit. The Table III — Stack Design Cell Rating Volts rms. Cell Combination Total Cells 33 26 18 4-2-lB 4-3-lB 4-4-lB 8 12 16 Table IV — Calculating the Input Voltage Required for Any Stack Cell Combination Formulae K Rectifier Circuit Resistance Load Battery or Capacity Load Single-phase Half -Wave Single-phase Full-Wave, Cen- ter TAP* Single-phase Full-Wave Bridge Three-phase Full -Wave Bridge Eac = KEdc + nDV Eac = KEdc -f nDV Eac = KEdc -1- 2nDV Eac = KEdc + 2nDV 2.3 1.15 1.15 0.74 1.0 0.80 0.80 0.74 Eac = Stack input volts, rms. Edc = Average dc output volts. K = Circuit form factor. n = Number of cells in series in each rectifying element. DV = RMS forward voltage drop at specified dc output current (see Fig. 9). * For center tap circuits, Eac is the voltage to the mid-tap on the transformer. SELENIUM RECTIFIER APPLICATION CONSIDERATIONS 1485 0.8 0.6 0.4 0.2 ^ ^ ^ 10 -HALF WAVE^ BATTERY AND 10 -BRIDGE - CONDENSER 10 -CENTER TAPJ LOADS ^ ^ ^ ,^ ^ y^ ^ 10 -HALF WAVE ^ opc.cTAKirc 10 -BRIDGE -"^J^ISSLh 1 0 - CENTER TAP J LO^OS...-^ ^ ^ y ^ ^ ^^^ --- y / ^ ^ .^ ^ 30 -BRIDGE -ALL LOADS A / ^ ^^ ^ ^ 0 1 D 2 0 3 0 4 0 5 PE 0 6 R CEN 0 7 T OF D 8 RATE 0 9 D CUF 0 IC JRENT K) 11 0 120 130 140 150 Fig. 9 — Dynamic forward voltage drop per cell. forward rms voltage drop per cell (DV) obtained from Fig. 9 is 1.10 volts at 100 per cent normal rated current. From the formulas in Table III, the input voltage is: Eac = l.loEdc + 2nDV, Eac = 1.15 X 48 + 2 X 2 X 1.1 = 59.6 volts rms. This input voltage is required for a new or unaged stack. Since the stack ages under service operating conditions, additional input voltage will be required to maintain the original dc output. It is considered good practice to design for a 100 per cent increase in the initial forward voltage drop (DV) of the stack. The aged stack, then would require Eac = 1.15^^c + 2(2nDF), Eac = 1.15 X 48 + 2(2 X 2 X 1.1) = 63 volts rms. The amount by which an increase of 100 per cent in the forward volt- age drop or forward resistance will change the output is dependent upon the design of the circuit. It depends upon the ratio in per cent of the for- ward rectifier stack resistance to the total circuit resistance including 1486 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 the transformer, ballast, load, etc. For most practical design considera- tions, one or more aging taps on the input transformer to provide 5 to 10 per cent additional voltage, will compensate for the forward aging. It should be emphasized that the curves in Figs. 6,7,8 and 9 are based upon empirical data obtained on new rectifier stacks and the charac- teristics mil vary slightly depending upon normal manufacturing varia- tions in the voltage current characteristics mentioned earlier in this article. AGING Selenium rectifier stacks are subject to aging. Aging is defined as any persistent change (except failure) which takes place for any reason in either the forward or reverse resistance characteristics. The important factor in selenium rectifier aging is the increase in the forward voltage drop which results in a decreased dc output. For normal rectifier applica- tions aging of the reverse current is not critical. For design purposes, a selenium stack is considered to have reached the end of its useful life when the stack input voltage required to main- tain rated output voltage exceeds the rms reverse voltage rating assigned by the manufacturer. Operation beyond this limit will result in over- heating of the selenium cells and rapid failure of the stack. The extent and rate of aging of selenium stacks cannot be predicted mathematically. Aging characteristics must be determined by actual test involving lengthy time consuming projects. To determine whether a given rectifier stack will give satisfactory performance for five years, as an example, tests would have to be conducted for this period. Aging can be accelerated by operation at high temperatures or at load currents above normal, but no accepted correlation exists between this type of aging and that obtainable under normal operating conditions. Aging data were obtained on sample rectifier stacks obtained from different manufacturers in this country. The samples were set up on life test racks as single phase full wave rectifiers and were operated continu- ously at normal room ambient temperatures. For the duration of the tests, the 60-cycle ac input voltage to the stacks was kept constant at approximately 10 per cent below the maximum rms voltage rating. The stacks operated into a resistance load. The resistance was selected so that the rectifiers operated at rated dc load currents. The resistance was not changed thereafter. Long term forward aging characteristics of selenium rectifier stacks SELENIUM RECTIFIER APPLICATION CONSIDERATIONS 1487 100 Q. o Q Q 90 / / ^^^ J / _^ .^^ i/ ^ / ^^^ / aA^-^ ^ ► MFR A 1 ^ .,.-*- r -MFR B '^''' '> y \ ,,.'-' ^ .,>' 0 123456789 YEARS Fig. 10 — Forward aging characteristics of 18-volt cells. 10 assembled with cells rated at 18 volts rms are shown in Fig. 10.* After 10 years (approximately 90,000 hours) continuous operation, the forward drop on one supplier's product increased approximately 90 per cent. A second supplier's product increased 100 per cent in 3 years under similar operating conditions. Further development programs by each manu- facturer resulted in improved aging qualities as indicated by tests made on stacks obtained at a later date. By introducing changes in their manufacturing techniques, the in- dustry eventually increased the voltage ratings to 26 volts rms per cell. This change in processing, however, raises the question — how has the operatinglifebeenaffected? Fig. 11* clearly illustrates that although many manufacturers are commercially producing 26-volt cells, the life ex- pectancy is vastly different. In order to save space, weight and critical materials, manufacturers are attempting to produce cells with still higher voltage ratings. Most companies have had a 33-volt cell development program under way for some tune, but again the problem of aging must be considered. Fig. 12* compares the forward aging of 26- and 33-volt cells after 5,000 hours operation. It is evident that in all cases except one, the 33-volt cells age substantially faster than the 26-volt cells. The exception is manufacturer "E" who has not produced 26-volt cells but has commercially processed * The designations A, B, C, etc. shown on all aging curves are not *« be in- terpreted to represent the same manufacturer's product on each of the figures. 1488 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 50 O 40 O < a. O 35 IL 30 25 20 15 y / J / MAN UFACTURERX /" 1 / / A> y , / y y 1 / / A ^ ^ y^ 1 / / / / ^ ^ y^ // / y / ^ ^ ^ - ^ ^ ■ A 0 4000 8000 12,000 16,000 20,000 24,000 HOURS OF OPERATON Fig. 11 — • Forward aging characteristics of various suppliers 26-volt cells. 33-volt cells for several years. Again, there is a wide difference in the rate of aging in the various manufacturers' products. Continued aging tests on stacks made by manufacturer "E" show that after 21,000 hours of continuous operation, the forward voltage drop has increased only 8 per cent. Extrapolation of these data indicate a life expectancy of thirty years. It should be emphasized that the data on the 33-volt cells, with the one exception, were taken on experimental stacks obtained from various companies while they still were in the research and development stages of processing the 33-volt cell. Undoubtedly, further development pro- grams on this type of cell will result in improved aging qualities. Fig. 13 shows clearly what happens to the aging rate as manufacturers attempt to increase still further the voltage rating. These data are based upon 26-, 33- and 40- volt cells made by one supplier who is commercially produ(;ing cells with these ratings. The aging rate is accelerated greatly as the voltage rating is increased. All the foregoing aging data were taken on stacks operated at rated load currents. A longer operating life may be expected if selenium stacks are selected to operate at lower load currents. As shown in Fig. 14, the aging SELENIUM RECTI FIKK Al'lLK ATION CON'SI DERATIONS 1489 at 50 per cent normal current rating is about one liall tliat atrakd lur- rent. Above normal rating, the aging is arreloratod to a greater degree. These data were obtained on a particul.ti Mippli.i -^ m.i. k. with 26-volt cells, but similar behavior has been observed on other supphers' produ(?t«, including stacks assembled with 33- and 40-volt cells. The behavior of the reverse current during these aging studies indi- cates that the reverse current generally decreases slightly during the first few months of operation and then remains substantially constant there- after. A few manufacturers' rectifiers, however, show the oppasite effect. The current rises slightly before leveling off. For normal dc power apph- cations, these changes are insignificant. As previously discussed, selenium stacks are rated for operation at +35°C ambient temperatures and should be derated at higher ambients for normal life expectancy. However, the problem frequently arises — what life can be expected at high ambient temperatures and why cannot aging qualities be predicted over a short time interval under accelerated conditions, such as high ambient temperatures? Fig. 15 shows data ob- tained on four different suppliers' product when operated at a -|-80®C ambient temperature. The temperature of the selenium cells under these conditions (ambient plus temperature rise) is about lOCC. For comparison purposes, data are included for similar stacks aged at normal MANUFACTURERS Fig. 12 — Forward aging characteristics. Comparison of 26- and 33-volt cell ratings. 1490 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 90 a" O 0 80 o a. 1 70 a. o u. _! 60 5 O ^40 z UJ O 30 20 10 / / 40 V OLTS/ f / 33/ J / / / / / V / _ 26 _ ■ ^ 4000 6000 HOURS OF OPERATION 8000 10,000 Fig. 13 — Forward aging versus cell voltage rating. room temperatures. It is readily evident that at 4-80°C ambient (1) the forward aging rate is increased (2) the different manufacturers' stacks age at different rates and (3) there is no definite correlation of the aging rate with normal operation at room temperature. As previously men- tioned under room temperature operation, longer life can be expected if stacks are operated at reduced current loads. At 80°C ambient however, three of the four manufacturers' stacks aged as much at half load as at full load. CONCLUSIONS The development and use of selenium rectifier stacks in this country has been quite rapid in the last ten years. The voltage ratings per cell have increased from 14 volts rms to 26 and 33 volts on a commercial basis. The trend in the selenium industry seems to indicate that cell voltage ratings will be further increased. The proper selection and procurement of selenium rectifier stacks and their application to dc power supplies for large scale industrial or military projects usually require considerably more information than can be ob- tained from published data in manufacturers' catalogs. Considerable differences have been observed in the performance of similar stacks SELENIUM RECTIFIER APPLICATION CON8IDBRATION8 1491 35 30 / / r / r— 1 / ■ 1509^ / y y / NORMAL RATING^ ^ ^ y . ' 50%^ ^ ^ — = ^ 1 .-"'^ 2000 4000 6000 HOURS OF OPERATION 8000 Fig. 14 — Efifect of load current on forward aging. 100 80 Q. 60 O •-S^ 7Q 40 ojn ^rr f.t 20 n (T O u Zli. iii-i 100 (n< itit RO (T^ O Z5 -o 60 a. MFR A .--" -^ / ^ / _„-.- --""" MFR C — ■^80•C AMBIENT ROOM AMBIENT -^ ■ / ^_ / „^— -.— •— • 20 MFR B .^ "" ^ ^ ^ ^: MFR D ^ ^- ■— .^ ,—-• ,.i><»< f^- ^^- ' 1000 2000 3000 4000 0 1000 HOURS OF OPERATION 2000 3000 4000 Fig. 15 — Forward aging characteristics. Comparison of room and +80**C ambient temperature operation. 26-volt cells, resistance load, rated dc ampe. input voltage — 80 per cent normal for +80*'C ambient and 90 per cent normal for room ambient. 1492 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 furnished by different manufacturers. To appraise properly the qualities of different manufacturers' stacks, the stability of the life characteristics should be considered more important than the initial characteristics. There is very little information published by the selenium manufac- turers regarding the life expectancy of their product. Aging appears to be directly related to the individual manufacturing techniques used by each supplier. The life of selenium rectifier stacks seem to decrease as the cell voltage rating is increased. Longer life can be expected if stacks are operated at load currents below the present ratings given in the manu- facturers' literature. Selenium rectifier stacks properly designed and conservatively rated can be expected to give satisfactory performance for 10 to 20 years. Careful consideration of the rate and extent of aging must be evaluated so that proper allowance may be made in the circuit design to obtain maximum life expectancy. REFERENCES E. A. Harty, Characteristics and Applications of Selenium Rectifier Cells, Elec. Eng., Oct., 1943. J. H. Hall, Transformer Calculations for Selenium Rectifier Applications, Elec. Eng., Feb., 1946. Glen Ramsey, The Selenium Rectifier, Elec. Eng., Dec, 1944. J. Gramels, Problems to Consider in Approving Selenium Rectifiers, A.I.E.E. Technical Paper No. 53-215. Also published in Communications and Elec- tronics, Sept., 1953. W. F. Bonner, Advanced Developments in Metallic Rectifiers, Electrical Manu- facturing, Oct., 1951. I. T. Cataldo, Development of 40-volt Selenium Rectifier Plates, Electrical Manufacturing, May, 1952. S. Niciejenski, Selenium Rectifiers, Radio and Television News, Oct., 1952. H. K. Henisch, Metal Rectifiers, Oxford, Clarendon Press, 1949. E. A. Richards, The Characteristics and Applications of the Selenium Rectifier, J. Inst. Elec. Engrs. (London), 88, Part III, Dec, 1941. Arcing of Electrical Contacts in Telephone Switching Circuits Part II — Characteristics of the Sliort Arc By M. M. ATALLA (Manuscript received May 27, 1953) Results are presented of an experimental study of the characteristics of the short arc in air which is the major cause of contact erosion in telephone switching circuits. Measurements were made of the arc initiation voltage, the voltage drop across the arc and the minimum arcing current. The following are the main conclusions: (1) For ''normal" contacts in air, the arc is initiated at a constant field strength of a few million volts/cm up to separa- tions of about 2-3 mean free paths of an electron in air. At larger separa- tions the arc is initiated at the well knoum spark breakdown potentials of air. In vacuum the linear relation holds for larger separations followed by a transition into a square root relation Vai = K{df^. {2) For "clean^* con- tacts in air, no constant field strength line is obtained for separations as low as 1600 A. Instead, the arc is initiated at the spark breakdown potentials of air, possibly due to adsorbed air molecules or due to breakdown along a longer path at the Paschen^s minimum potential. In vacuum, it is specu- lated that the above square root relation will hold. (3) For ''activated" con- tacts and small separations the arc is initiated at a constant field strength of about 0.6 X 10^ volts/ cm. (4) For "normal" contacts the minimum arcing current increases with an increase in the maxiynum current during the arc due to surface contaminations and the arc cleaning action. (3) For arc cur- rents above 1 .5 ampere and energies of the order of thousands of ergs the cathode determines the arc characteristics. INTRODUCTION The electrical erosion of contacts presents an important problem in the design of telephone switching apparatus. There are several physical phenomena that occur between contacts and contribute to their erosion. The short arc,* which may occur on both make and break of a contact, * The short arc is characterized by its constant voltngo, independent of the current, which is of the order of the ionizing potential of the contact material. 1493 1494 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 is generally considered to be the major contributor. For illustration, a palladium contact, 10~* cm^ in volume, will last for more than 10^ opera- tions* only if the arc energy per operation is less than 2.5 ergs. This is based on an erosion rate of 4 X 10~^^ cm^ per erg.^ Furthermore, a short arc with a half ampere current, lasting for only one microsecond, will dissipate as much energy as 70 ergs. Contact erosion may also take place, though at much lower rates for the usual ranges of current and voltage in switching circuits, due to molten bridges^ on contact break and due to glow discharge.^ In Part I of this series, was discussed the mechanism of the initiation of the short arc as determined by contact and circuit conditions. Three characteristics of the arc were used in the presentation without elabora- tion as to their nature: (1) the arc initiation voltage, (2) the voltage drop across the arc, and (3) the arc initiation and the arc termination currents. These characteristics have been the subject of a recent study to which this part of the series is mainly devoted. In the course of this study, it was found that there should be some repetition of previous work to isolate effects of certain pertinent parame- ters that were not previously given due consideration. No attempt is made here to give a complete survey of the related studies in the literature. Only a few publications are referred to as typical references to the subjects discussed. NOTATION C Capacitance Ea Energy dissipated in the arc y F Gross field strength between the contacts: — I Current li Arc initiation current Imax Maximum current in the arc Im Minimum arcing current or arc termination current * This is the actual life requirement of some contacts in existing switching circuits. 1 L. H. Germer and F. E. Haworth, Erosion of Electrical Contacts on Make, J. App. Phys. 20, p. 1085, 1949. 2 See for example: J. J. Lander and L. H. Germer, The Bridge Erosion of Electrical Contacts, J. App.^ Phys. 19, p. 910, 1948. 3 F. E. Haworth, Electrode Reactions in Glow Discharge, J. App. Phys. 22, p. 606, 1951. * M. M. Atalla, ''Arching of Electrical Contacts in Telephone Switching Cir- cuits. Part I— Theory of the Initiation of the Short Arc," B.S.T.J., 32, pp. 1231- 1244, Sept., 1953. ARCING OF CONTACTS IN TELEPHONE SWITCHING SYSTEMS 1 lO") K Constant in the relation Vai = Kidf^ R Resistance V Voltage Vai Arc initiation voltage d Minimum separation between contacts h Height of a metal bridge formed during one arc t Time ta Arc duration V Constant voltage drop across the short arc ARC INITIATION VOLTAGE Consider the simple contact circuit in Fig. 1 comprising a pair of con- tacts in series with a resistor R and a variable voltage power supply. By fixing the separation between the contacts and gradually increasing the R R, AMr CONTACTS -L_ Q i AAA. 0 VARIABLE POWER SUPPLY i Fig. 1 — Contact Circuit. voltage, an arc is usually initiated when the voltage reaches a certain value ''Vai' called the arc initiation voltage. In general Vai is a function of: (1) separation between the contacts, (2) geometry of the contact surfaces, (3) the surrounding atmosphere, and (4) contact material and its surface. Our experiments were limited to contacts in atmospheric air with special emphasis on separations of the order of and less than the mean free path of an electron in air. For larger separations the arc initiation voltage follows the well known curve of the sparking potential of air/ For the smaller separations where the presence of air molecules would not be expected to affect the arc initiation voltage, it was pre\iously re- ported that breakdowns occurred at some constant field between 0.6 X 10' and 16 X lO' volts/cm.'* '• ' In our experiments a cantilever bar, described by Pearson,* was used. The setting for zero separation was determined by a 0.1 volt source, a ^ See for instance J. D. Cobine, Gaseous Conductors, McGraw-Hill, N. Y., 162, 1941. 6 G. L. Pearson, Phys. Rev. 56, p. 471, 1939. ' L. H. Germer and J. L. Smith, J. App. Phys. 23. p. 553, 1952. 1496 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 10,000 ohms resistor and a cathode relay oscilloscope. The zero setting could be repeated with a precision of =b 500A. All the reported results were obtained by fixing the contact separation, raising the voltage and observing the breakdown on a cathode ray oscilloscope. Contacts tested were given one of three different surface treatments: (1) The contact surface was polished with fine emery paper, washed with methyl alcohol, then exposed to the laboratory atmosphere for a few hours. After about 5 arc discharges, readings of arc initiation voltage seemed to vary at random. Contacts thus treated are referred to as ''normal" contacts. (2) Contacts were subjected to arcing for about 5 minutes at the rate of 15 arcs per second. Arcing was produced by the discharge of a half microfarad condenser at 500 volts through a 10-ohm resistor. Measure- ments of the arc initiation voltage followed immediately after this treat- ment. These contacts are referred to as "clean" contacts. Their behaviour usually changed to that of ''normal" contacts after a short exposure to the laboratory atmosphere. (3) Contacts were subjected to arcing at the rate of 3 arcs per second for about one hour in air saturated with d-limonene. The arc was pro- duced by discharging a 0.1 -microfarad condenser at 50 volts through a 100-ohm resistor. These contacts are referred to as "activated" con- tacts.* Fig. 2 shows the results obtained with "normal" palladium contacts. Each point represents the average of five readings. The maximum spread was 40 per cent of the average. For separations less than 10,000 A, about two mean free paths of an electron in air at normal conditions, a constant gross field strength line of 3 X 10^ volts/cm was obtained. At larger separations the measured arc initiation voltages were essentially the well known sparking potentials of air. Fig. 3 shows the corresponding results obtained with "normal" carbon contacts in air. Below a separation of 15,000 A, the arc was initiated at a constant gross field strength of 2.4 X 10 volts/cm. The maximum spread of the individual points was only 15 per cent of the average. In the absence of air, it is expected that the constant field strength lines will hold for higher separations.* In Table I, Column 2 are given the measured values of the gross field strengths at which the arc was initiated for a group of "normal" contact materials. * Recent unpuhlinhod measurements by Dr. P. Kisliuk on similar contacts in vacuum have indicated that the constant field strength relation Wi = F((i) holds initially for larger separations and is followed by a gradual transition into a 8(juare root relation Vai — K(d)^l^ as proposed by Cranberg.^ » L. H. Germer, Arching at Electrical Contacts on Closure —Part I, J. Appl. Phys. 22. p. 955, 1951. ARCING OF COXTACTS IX TELEPHOXE SWITCHING SYSTEMS 1497 2000 1000 800 600 0) 400 < g 200 z ^ 100 ^ 80 60 40 20 "normal" Pd CONTACTS ^ ^-^ ^ .*f ---'''' .v^^i ['?o^ ^/' '^>\^ ^o / / ^^Ai 2-fr^ ^-o- .-C / ^ X / / 4^^^ ^f: /= Fig. 2 2 ^ « « ,04 2 ^ « « 10^ 2 CONTACT SEPARATION IN ANGSTROMS Arc initiation voltages for "Normal" palladium contacts. Due to the spread of the above measurements and their observed de- pendence on surface exposure and treatment it was suspected that the constant field strength characteristic obtained was due to surface con- tamination. This was verified by testing contacts cleaned by the heavy arcing process explained above. The results are shown in Fig. 4. The familiar constant field strength line was not obtained for separations as low as 1500 A. Instead, the arc was initiated at voltages comparable to the sparking potentials of air. Since the separations were too small, the smallest being three times less than the mean free path of an electron in air, it was thought that the effect was due to some mechanism invohdng the adsorbed air molecules or due to breakdown along a longer path at the Paschen's minimum potential.^ In the absence of air, it is, therefore, expected that higher voltages and higher field strengths in excess of 20 X 10^ volts/cm, as obtained at 1500A, will be needed to initiate the arc. It is possible that Cranberg's relationk^ V = K{df''^, will hold for separa- tions as low as a few thousand angstroms. In Fig. 4, this relation, with K = 10' volt, cm"''', is plotted. 9 L. Cranberg, The Initiation of Electrical Breakdowns in Vacuum, J. Appl. Phys. 23, p. 518, 1952. 1498 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 1000 ........ 800 IN AIR TIATION VOLTAGE i i /o — ^ 1 — J y / ARC INI S^^ ■^f o -y 20 / 103 10^ CONTACT SEPARATION IN ANGSTROMS Fig. 3 — Arc initiation voltages for carbon contacts. For contacts activated in organic vapors, constant field strength lines were obtained. In Fig. 5 results are shown for palladium contacts acti- vated by d-limonene. The average field strength for arc initiation is only 0.6 X 10* volts/cm with a spread of as much as 100 per cent of the average. At separations greater than 50,000A the arc was initiated by the familiar spark breakdown of air. In vacuum the constant field strength line should hold for larger separations possibly until it intersects Cranberg's line.^ Breakdowns at low fields were also observed for metals with inorganic films. For instance Gleichauf^^ obtained a constant field strength line of 0.24 X 10* volts/cm for copper electrodes in vacuum at separations of the order of millimeters. Our measurements on copper contacts at sepa- rations less than 10,000A have shown breakdowns at fields as low as 0.7 X 10* volts/cm. It is concluded that the presence of organic or inorganic films on a contact usually leads to a reduced gross field strength at which the breakdown will occur. The reduction can be by as much as two orders of magnitude. It is possible that this reduction was only apparent and the electrons actually came from the underlying metal by extraction in an intense field set up by the positive ions lying on the surface of the film." '0 P. H. Gleichauf, Electrical Breakdown Over Insulators in High Vacuum, J. Appl. Phys. 22, p. 766, 1951. 1^ F. L. Jones, Electrical Discharges in Gases, Nature, 170, p. 601, 1952. ARCING OF CONTACTS IN TELEPHONE SWITCHING SYSTEMS 1499 Table I. — Arcing Characteristics of Contact Materials (1) Contact Material (2) (3) Carbon . . . Nickel .... Palladium. Silver Tungsten . i Field strength to 1 initiate the arc for Short arc Voltage normal contacts. 10« Volts/cm. 2.4 20-43 4.2 12-13 3.0 14-15 2.0 11-13 4.9 12-13 (4) Minimum arcing current for "clean"* contacts. Amps. 0.03 0.4 1.1 0.8 0.7 * For "normal" contacts, the minimum arcing current is less by 50 per cent or more. The results of the above section are summarized in Fig. 6. The solid lines were actually measured for palladium contacts under different surface conditions. The broken lines are only speculative. For a certain separation and surface condition the arc will be initiated at a voltage as given by the lowest corresponding line in the figure. VOLTAGE DROP ACROSS A SHORT ARC The short arc may be defined as a discharge of electricity between electrodes with a voltage drop of the order of the minimum ionizing po- tential of the atoms of the electrodes.* Furthermore, due to the small separation between the contacts and the local high pressure metal vapor, the characteristics of the established arc are independent of a surround- ing atmosphere at normal or low pressures. The short arc is characterized by its constant voltage for currents above a minimum value called the minimum arcing current of the contact. In contrast to the short arc, the long arc between contacts at a fixed separation has a voltage drop which decreases with an increase in current .^^ Most arcs occurring between con- tacts on both make and break of telephone switching circuits, are short arcs. In Table I, Column 3, are given our measured values of the short arc voltage for a few materials. ARC initiation AND TERMINATION CURRENTS The arc termination current or minimum arcing current is defined as the lowest current at which the arc can be sustained. The arc is extin- * The arc voltage is about 50 per cent higher except for the carbon arc which has a much higher voltage; see Table I, Column 3. ^2 See for instance : K. Gaulrapp, Untersuchung der elaktrisehen Eigenschaften des Abreisbogens, Ann. Phvsik. 25, p. 705, 1936. 1500 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 2000 O 1000 > 2 SOO O 400 100 ^^CLEAN contacts" D Pd o Ag ,-' A Pt X W + NL >. ..--' ^1 ^^'' ^'' ^.-P^x. — TJ- AIR ? 10^ 10^ CONTACT SEPARATION IN ANGSTROMS Fig. 4 — Arc initiation voltages for ''Clean" metals. guished when the circuit current drops to this value. To initiate the arc a minimum current must be furnished by the circuit called the arc initiation current. It was previously shown that the arc initiation and termination currents are essentially the same numerically. The existence of these limiting currents as such, rather than current densities, is not understood from a fundamental standpoint. The limiting currents are a function of the contact material and are appreciably affected by the surface condition of the contacts. Surface contaminations generally reduce the limiting currents of the contacts. The results presented here were obtained by measuring the residual volt- age in an R-C circuit following an arc. This voltage is equal to ImR-, from which Im was determined. Our measurements are given in Table I, Column 4 for ''clean" contacts. The maximum spread is 20 per cent of the average. For normal contacts, however, surface contamination causes a wide variation in the results. Furthermore, the maximum intensity of the arc, or the maximum current furnished by the measuring circuit, was found to have an appreciable affect on the measurements. This ap- pears to be due to the surface cleaning action of the arc. This effect is demonstrated in Fig. 7 for ''normal" palladium contacts. Each point represents an individual measurement of I^ plotted against the cor- responding maximum current during the arc. An R-C contact circuit was used. While the measurements show a considerable spread, they indicate a definite trend of an increase in I^ with increasing I max- ARCING OF CONTACTS IN TELEPHONE SWITCHING SYSTEMS 1501 1000 800 600 Pd contacts IN AIR 400 1 "clean" Pd • "^ 1 ^7 \RC INITIATION VOLTAGE 5 8 8 8 4 / .^^)> ^9 7 ^b/' d^/ /^ f\ cT 40 20 -f A ^ o / / 103 10^ CONTACT SEPARATION IN ANGSTROMS Fig. 5 — Arc initiation voltages for "Activated" palladium contacts. 10,000 8000 6000 4000 800 < 400 o a. 200 < 100 80 60 1 ^ ,S-^"> ^ -< sj^- r .■^ '/ \' 4 a-^^^ #- ,f/^ /^ 1' ^0^ ^ «< ,v-^' yr rOti^ ^ ^v^^^-^^' ^o^?:^ y ^^ ^ ;i>^'" ^v*- ^^CLEAN" CONTACTS ^T^^^JORMALlSllSi:^ 1 vj AIR 1 J t^ / CO NT/ ^ A ^ > f O.b y y • \ • 0.4 /' 1 • < 0.J 0? • • 0.4 0.5 0.6 0.8 1.0 1.5 2 3 4 MAXIMUM CURRENT IN ARC IN AMPERES 7 8 9 10 Fig. 7 — Dependence of minimum arcing current on maximum arc current. In the first part of this paper it was shown that contact activation by- organic vapors reduces the arc initiation voltage for a fixed separation. In other words, for a pair of closing contacts the arc will be initiated at a wider separation and a longer arcing time is obtained. In addition, activation tends to decrease the minimum arcing current. Germer , meas- ured a minimum arcing current of only 0.027 to 0.037 ampere for active silver. Our measurements for active palladium gave a minimum arcing current of 0.1 ampere. This substantial decrease in the minimum arcing current of contacts due to surface activation usually causes a further increase in the arcing time. Contact activation, therefore, enhances arcing between closing contacts in two ways; first, by initiating the arc at wider separations, and second by maintaining the arc at much smaller currents. The following results are presented to indicate the quantitative significance of contact activation. A pair of ''normal" palladium contacts were operated in air saturated with d-limonene at 3 cps. The contacts closed a circuit consisting of a 0.5-microfarad condenser, charged to 50 volts, in series with a 100-ohm resistor. The transient on make was observed on a cathode ray oscilloscope to determine the arcing time. The arc energy Ea was calculated and the ratio EJCv {Vq — v) was plotted against the number of operations. Fig. 8. The denomenator is the maxi- mum arc energy, which is only attained if the arc is maintained until the current reaches zero. The results indicate a rapid increase of the arc energy corresponding to an increase in surface activation. When the contacts become fully active, the arc energy was about two orders of ARCING OF CONTACTS IN TELEPHONE SWITCHING SYSTEMS 1503 magnitude greater than the energy for inactive contacts. Contacts may also be activated by inorganic films. Experiments on palladium and silver contacts have shown* that glow discharge between contacts operat- ing in air produced a second type of activation. Nitrides were formed on the contact surfaces and the rainimum arcing current dropped to about 0.1 ampere for silver and 0.2 ampere for palladium. This effect was more pronounced with silver contacts. In carrying out the measur\^ments of the minimum arcing current it was observed that the arc wks generally interrupted by one of three causes (1) the minimum arcing current was reached, (2) physical closure of the contacts, and (3) shorting of the arc by a metal bridge formed during the arc. A study of the bridge formation has shown that the height of the bridge was a function of the arc energy. The bridge height was measured by setting the zero separation point before and after the arc. The difference gave the height of the bridge. This was plotted in Fig. 9 against the measured arc energy. The height of the bridge in- creased roughly with the cubic root of the arc energy up to energies of 1.0 0.8 0.6 0.4 ^ 0.10 ^ 0.08 0.04 0.010 0.008 ^^ : ^^^:^ o ^ 10 12 14 16 18 20 22 NUMBER OF CONTACT OPERATIONS 0 2 4 6 Fig. 8 — Increase in arc energy due to contact "Activation" 24 26 28 30, x103 * This information was obtained from unpublished work of F. E. Haworth. See also Reference 8 for a discussion of the effects of insulating films on arc initi- ation. 1504 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 about 800 ergs. This was followed by a rapid transition into metal loss instead of a bridge. This loss increased with increase in energy. DISSIMILAR CONTACTS EFFECT OF POLARITY ON ARC CHARACTERISTICS An experiment was carried out to find the contributions of the anode and cathode in determining the characteristics of the arc. Use was made of the fact that carbon contacts are unique in having low minimum arcing current and high arc voltage compared to most metals; see Table I. Furthermore, carbon contacts always gave a constant field strength line for arc initiation at small separations even when they were cleaned by the heavy arcing process. Tests were carried out with a variety of con- tact metals against carbon. All the results obtained were qualitatively the same. For illustration only the experiments with palladium-carbon contacts are reported here. Test specimens were carefully prepared in the following fashion. Pal- ladium to palladium contacts, mounted on the cantilever bar set-up, were cleaned by the heavy arcing process. One contact was removed and replaced by a carbon contact with its surface polished and freed from HEIGHT OF BRIDGE OR LOSS IN ANGSTROMS O -1- - o o o o o o o o o \ • , , • \ • • \ ► 1 h • ' 1 • ,- — 1 < \ • • • •• .•- • — •— « • "■» \ \ \ 5" w o ■-i • ( • m* _ Q» =r m o / L - l^ * Fig, 9 — Bridge formation during the arc. ARCING OF CONTACTS IN TELEPHONE SWITCHING SYSTEMS 1505 Table II — Effect of Polarity on the Characteristics of the Arc — Palladium-Carbon Contacts (1) Contact Configuration C+, Pd- Pd+, Pd C-, Pd+ C-, c+. . (2) Arc initiation, voltage at 6000A separation 300 320 130 120 (3) Arc voltage 13-15 14-15 20-30 20-43 Minimum arcing current. Im amps. 0.2-0.5 1.1 0.2-0.3 0.03 loose particles. The separation between the contacts was set at 6000A. By gradually raising the voltage across the contacts and observing it on a cathode ray oscilloscope, the arc initiation voltage was determined. The same measurement was repeated several times. Preceding each mea- surement, the contacts were recleaned by the same process explained above. The polarity was then reversed and a new set of measurements was taken. The results are shown in Table II, Column 2. They indicate that the arc initiation voltage is determined by the cathode. This furnishes support to the postulate that field emission is the first step of the mechanism of arc initiation. By recording the voltage across the contacts during the arc, measurements were made of the arc voltage and the minimum arcing current. The results are given in Table II, Columns 3 and 4. The arc voltage measurements indicate rather conclusively that the arc voltage is determined by the cathode. The minimum arcing cur- rent measurements, however, were only slightly, yet consistently, higher with a palladium cathode. It is thought that during a single short arc, particularly with the high intensity arcs used, there is a certain amount of exchange of materials between the electrodes. This exchange is possibly responsible for the observed influence of the anode on the arc character- istics. It is concluded that the arc initiation voltage as well as the arc voltage are characteristics of the cathode while the minimum arcing current seems to be influenced by both electrodes with stronger inclination towards the cathode characteristics.* The following reservation, how- * Early experiments by H. E. Ives^' have led to the conclusion that the arc voltage is a characteristic of the anode. In his experiments, however, currents of the order of one ampere were established in the circuit while the contacts were closed. Arc measurements were made during the subsequent break of the contacts. It is thought, therefore, that metal bridges must have formed during the break, transferring metal from the anode to the cathode. ^ The arc produced must have been influenced accordingly. In our experiments this difficulty was entirely eliminated. 13 H. E. Ives, Minimal Length Arc Characteristics, J. Franklin Inst. 198, No. 4, 1924. 1506 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 ever, has to be made. All these measurements, although made at a voltage range between 50 and 400 volts corresponding to a contact separation range between 1500 and 100,000A, had high maximum cur- rents above 1.5 amperes. The energy dissipated in each arc was of the order of thousands of ergs. This reservation is made because recent studies of metal transfer have indicated a reversal in the direction of transfer between the anode and the cathode depending on the rate of energy dissipation in the arc. ACKNOWLEDGEMENTS I am indebted to L. H. Germer, P. Kisliuk and F. H. Haworth for much valuable discussion and to A. S. Timms for assistance with many of the experiments reported here. Abstracts of Bell System Technical Papers* Not Published in this Journal Anderson, J. R/ Electrical Delay Lines for Digital Computer Applications, I.R.E., Trans. P.G.E.C., 2, pp. 5-13, June, 1953. A survey of existing lumped parameter and distributed parameter delay lines has shown that their maximum storage capacity is about 23 pulses and 15 pulses respectively regardless of total delay time. An analysis of pulse trans- mission through distributed delay Unes indicates that dissipation in the in- ductive elements is the chief factor limiting storage capacity. A method is proposed for decreasing this dissipation through the use of high Q nickel zinc ferrites around straight conductors for inductive elements. Armstrong, C. A.^ Communications for Civil Defense, A.I.E.E., Trans., Commun. & Electronics, 7, pp. 315-326, July, 1953. Barotta, P. J., see S . P. Gentile. BiRDSALL , H. A. , see D. A. McLe^ BOGERT, B. P.' On the Band Width of Vowel Formants, Letter to the Editor, J. Acoust. Soc. Am., 25, pp. 791-792, July, 1953. Measurements were made on a sample of vowel utterances, by male talkers, of the band widths of the first three formants. It was found that the band * Certain of these papers are available as Bell System Monographs and may be obtained on request to the Publication Department, Bell Telephone Laboratories, Inc., 463 West Street, New York 14, N. Y. For papers available in this form, the monograph number is given in parentheses following the date of publication, and this number should be given in all requests. 1 Bell Telephone Laboratories. 2 American Telephone and Telegraph Company. 1507 1508 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 width was essentially constant and independent of the particular vowel. The mean values for bars 1, 2, and 3 were 130, 100, and 185 cps, respectively. Ten per cent of the 300 band widths measured were less than 90 cps and ten per cent greater than 260 cps. Brangaccio, D. J., see C. C. Cutler. Brattain, W. H., see A. M. Portis. BULLINGTON, K.^ Frequency Economy in Mobile Radio Bands, I.R.E., Trans., P.G.V.C., 3, pp. 4-27, June, 1953 (Monograph 2109). Calbick, C. J., see D. A. McLean. Campbell, R. D.^ Path Testing for Microwave Radio Routes, Elec. Eng., 72, pp. 571- 577, July, 1953. Campbell, W. E.^ Solid Lubricants, Lubrication Eng., 9, pp. 195-200, Aug., 1953. CiCCOLELLA, D. F.^ AND L. J. LaBRIE^ High Frequency Crystal Units for Use in Selective Networks and Their Proposed Application in Filters Suitable for Mobile Radio Channel Selection, I.R.E., Trans, P.G.V.C. 3, pp. 118-128, June, 1953. Conwell, E. M.^ High Field Mobility in Germanium With Impurity Scattering Domi- nant, Phys. Rev., 90, pp. 769-772, June 1, 1953. Experimental measurements show a variation of mobility with electric field intensity of electrons in n type germanium which differs at 20 degrees K from that observed in the same specimen at 77 degrees K and higher temperatures. This difference can be accounted for by scattering by ionized impurities. A crude quantitative treatment is carried out along the lines of Shockley's treatment for the case of lattice scattering. As in that case, the resulting theory fits the data well if the rate of energy loss is taken several times higher than that given by the theory assuming that the surfaces of constant energy are spherical. 1 Bell Telephone Laboratories, Inc. 2 American Telephone and Telegraph Company. ^ Formerly Hell Toloi)h()nc Laboratories. ABSTRACTS OF TECHNICAL ARTICLES 1509 Cutler, C. C/ and D. J. Brangaccio^ Factors Affecting Traveling Wave Tube Power Capacity, I.R.E., Trans., P.G.E.D., 3, pp. 9-23, June, 1953. Dacey, G. C' Space-Charge Limited Hole Current in Germanium, Phys. Rev., 90, pp. 759-763, June 1, 1953. A situation can arise in semi-conductors similar to the space-charge limited emission of electrons in vacuum. The theory of Shockley and Prim for this phenomenon has been extended to the high field case using the approximation that the drift velocity of the carriers is y = ti{EEoy''^, where n is the low field mobility, E the electric field, and Eo the "critical field." For this approxima- tion the current density analogous to Child's law for a plane parallel diode is J = {ys)(y3y"Kf.Eo'^Wa"vw''\ where Va is the potential across a diode of thickness w and K is the dielectric constant in mks units. Good agreement between theory and experiment for hole flow in germanium at liquid air temperature has been obtained, using values of m and ^o obtained independently by Ryder. EsHELBY, J. D.^ Read, W. T} and W. Shockley^ Anisotropic Elasticity with Applications to Dislocation Theory, Acta Metallurgica, 1, pp. 251-259, May, 1953. The general solution of the elastic equations for an arbitrary homogeneous anisotropic solid is found for the case where the elastic state is independent of one (say Xs) of the three Cartesian coordinates Xi , X2 , Xs . Three complex variables 2 (^^) = a:i + 'p{l)x2{( = 1, 2, 3) are introduced, the/)(^) being complex parameters determined by the elastic constants. The components of the displacement {ux , U2 , Uz) can be expressed as linear combinations of three analytic functions, one of 2(i) , and of 0(2) , and one of 0(3) . The particular form of solution which gives a dislocation along the Xa-axis with arbitrary Burgers vector (ai , a^ , az) is found. (The solution for a uniform distribution of body force along the a^s-axis appears as a by-product.) As is well known, for isotropy we have Ws = 0 for an edge dislocation and Ux = 0, f/2 = 0 for a screw dislocation. This is not true in the anisotropic case unless the XxX^ plane is a plane of symmetry. Two cases are discussed in detail, a screw dislocation running perpendicular to a symmetry plane of an otherwise arbitrary crystal, and an edge dislocation running parallel to a fourfold axis of a cubic crystal. Fuller, C. S.^ and J. A. Ditzenberger^ Diffusion of Lithium Into Germanium and Silicon, Letter to the Editor, Phys. Rev., 91, p. 193, July 1, 1953. 1 Bell Telephone Laboratories, Inc. 8 University of Illinois. 1510 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 Gentile, S. P.^ and P. J. Barotta* Transistor Physics Simplified, Radio and Telev. News, 50, pp. 44-46, 100-102, July, 1953. HiNES, M. E.' Traveling-Wave Tube, Radio and Telev. News, 49, pp. 12-14, 26, June, 1953. KoLB, E. D., see W. P. Slighter. Labrie, L. J., see D. F. Ciccolella. LiNVILL, J. G.^ Transistor Negative -Impedance Converters, I.R.E., Proc, 41, pp. 725-729, June, 1953. May, a. S.' Microwave System Test Equipment, Commun. Eng., 13, pp. 24-25, 44-45, May-June, 1953. McKay, K. G.' Bombardment Conductivity, Ind. Diamond Rev., 13, pp. 127-130, June, 1953. McLean, D. A.,^ Birdsall, H. A.^ and C. J. Calbick^ Microstructure of Capacitor Paper, Ind. and Eng. Chem., 45, pp. 1509-1515, July, 1953 (Monograph 2142). McMillan, B.^ Basic Theorems of Information Theory, Ann. Math. Stat., 24, pp. 196-219, June, 1953 (Monograph 2124). This paper describes briefly the current mathematical models upon which communication theory is based, and presents in some detail an exposition and partial critique of C. E. Shannon's treatment of one such model. It then presents a general limit theorem in the theory of discrete stochastic processes, suggested by a result of Shannon's. 1 Bell Telephone Laboratories, Inc. * Hudson Technical Institute. ABSTRACTS OF TECHNICAL ARTICLES 1511 Mertz, P/ Influence of Echoes on Television Transmission, J.S.M.P.T.E., 60, pp. 572-596, May, 1953 (Monograph 2144). Peterson, G. E.^ Basic Physical Systems for Communication Between Two Individuals, J. Speech and Hearing Disorders, 18, pp. 116-120, June, 1953 (Mono- graph 2135). Pierce, J. R.^ Transistors, Radio-Electronics, 24, pp. 42Ht4, June, 1953. PoRTis, A. M.,^ A. F. Kip,^ C. KiTTEL,^ AND W. H. Brattain^ Electron Spin Resonance in a Silicon Semi-Conductor, Letter to the Editor, Phys. Rev., 90, pp. 988-989, June 1, 1953. Prim, R. C, see W. Shockley. QUARLES, D. A.^ Progress and Problems, Elec. Eng., 72, pp. 667-669, August, 1953. QUARLES, D. A.^ Report to the Membership, Elec. Eng., 72, pp. 477-479, June, 1953. Read, W. T., see J. D. Eshelby. Ryder, E. J.^ MobiUty of Holes and Electrons in High Electric Fields, Phys. Rev., 90, pp. 766-769, June 1, 1953. The field dependence of mobility has been determined for electrons and holes in both germanium and siUcon. The observed critical field at 298 degrees K beyond which /x varies as E~^'^ is 900 volts/cm for n-type germanium, 1400 volts /cm for p-type germanium, 2500 volts /cm for n-type siUcon, and 7500 volts/cm for ??-type siUcon. These values of critical field are between two to four times those calculated on the basis of spherical constant energy surfaces in the Brillouin zone. A saturation drift velocity of 6(10)^ cm/sec is observed in germanium which is in good agreement with predictions based on scattering 1 Bell Telephone Laboratories, Inc. ^ Sandia Corporation. 7 University of Cahfornia, Berkeley. 1512 THE BELL SYSTEM TECHNICAL JOURNAL, NOVEMBER 1953 by the optical modes. Data on n-type germanium at 20 degrees K show a range over which impurity scattering decreases and the mobihty increases with field until lattice scattering dominates as at the higher temperatures. Shockley, W., see J. D. Eshelby. Shockley, W.^ AND R. C. Prim^ Space-Charge Limited Emission in Semi-Conductors, Phys. Rev., 90, pp. 753-758, June 1, 1953. A situation analogous to thermionic emission into vacuum can occur in semi- conductors. A semi-conductor analog for a plane parallel vacuum diode may consist of two layers of n type semi-conductor bounding a plane parallel slab of pure semi-conductor. The current density analogous to Child's law is J = O^eo^F^/STF^, where k = dielectric constant, eo = mks permittivity, n = mobility, V = applied voltage, and W = thickness of pure region. The condition prevailing at the space-charge maximum is analyzed taking into account diffusion due to random thermal motion. Brief discussions are given of the effect of fixed space charge, the dependence of mobilit}^ upon electric field strength and the role of space-charge limited emission in a new class of unipolar transistors. Slichter, W. p.' and E. D. Kolb' Solute Distribution in Germanium Crystals, Letter to the Editor, Phys. Rev., 90, pp. 987-988, June 1, 1953. Townsend, J. R.^ A Dynamic Program for Conversion, Metal Progress, 63, pp. 79-81, June, 1953. Turner, E. H.' New Non-Reciprocal Waveguide Medium Using Ferrites, Letter to the Editor, I.R.E., Proc, 41, p. 937, July, 1953. Wannier, G. H.' Threshold Law for Single Ionization of Atoms or Ions by Electrons, Phys. Rev., 90, pp. 817-825, June 1, 1953. When an electron hits an atom or ion, it may knock off an electron. This process is fundamental in almost all types of gas discharge. The reaction is endothermic; hence there is a threshold value in the electron energy below Bell Telephone Laboratories, Inc. ABSTRACTS OF TECHNICAL ARTICLES 1513 which it does not occur. In this paper, the dependence of the yield on the energy just above this threshold is derived. The derivation is not rigorous because it circumvents some of the difficulties of the three-body problem by applying ergodicity, albeit in a weakened form. The result is that, for atoms, the yield rises as the 1 . 1 27th power of the energy excess. For ions the exponent Hes between this number and unity. WiLLARD, G. W.' Ultrasonically Induced Cavitation in Water : A Step-by-Step Process, J. Acoust. Soc. Am., 25, pp. 669-686, July, 1953. A 2.5-mc, barium-titanate, spherically focusing radiator was used to produce cavitation in both degassed and aerated water entirely within the restricted, high intensity focal region, remote from the water boundaries. The sonic intensity rises to 1.8 kw/cm^ and the pressure amplitude to ±70 atmospheres at the focus. High-intensity illumination and an unusual high speed photo- graphic technique permit observation and timing of the step-bj'-step process of cavitation development. Feather-shaped cavitation bursts are sporadically produced, being initiated in the insignificant quill portion nearest the radiator, then abruptly expanding to form the catastrophic plume portion. The plume is believed to be formed by myriads of microcavities, too small and close for individual observation. These two fundamental steps are identically produced, and with equal ease, both in degassed and aerated water. The whole action is over in several milli- seconds, except that in the case of aerated water a third bubble step is pro- duced. In aerated water, non-collapsing gas bubbles are generated by and concurrently with, the catastrophic step. These bubbles remain after collapse of the burst, to be blown off down stream by the sonically induced liquid streaming. The bubble step is not generated without the presence of the catastrophic step. The latter is generated only if the initiation step reaches a definite degree of development (not always attained). This requires sonic activation for increasing lengths of time for decreasingly smaller sonic intensities. Origina- tion of the initiation step, and hence of the whole cavitation phenomena, is believed to occur whenever a stray nucleus (weak spot) streams into the high intensity sonic field. Young, W. R.' Comparison of Mobile Radio Transmission at 150, 450, 900 and 3700 Mc, I.R.E., Trans., P.G.V.C. 3, pp. 71-83, June, 1953. Bell Telephone Laboratories, Inc. Contributors to this Issue M. M. Atalla, B.S., Caiio University, 1945; M.S., Purdue University, 1947; Ph.D., Purdue University, 1949; Studies at Purdue undertaken as the result of a scholarship from Cairo University for four years of graduate work. Bell Telephone Laboratories, 1950-. For the past three years he has been a member of the Switching Apparatus Development Department, in which he is supervising a group doing fundamental research work on contact physics and engineering. Current projects include fundamental studies of gas discharge phenomena between con- tacts, their mechanisms, and their physical effects on contact behavior; also fundamental studies of contact opens and resistance. In 1950, an article by him was awarded first prize in the junior member category of the A.S.M.E. He is a member of Sigma Xi, Sigma Pi Sigma, and Pi Tau Sigma, and a junior member of the A.S.M.E. Frank E. Blount, B.S. in E.E., Oregon State College, 1928. Bell Telephone Laboratories 1928-. Mr. Blount tested panel system circuits for a year and then transferred to development work on special switching and signaling circuits. His interest in circuit design has also included circuits for automatic toll ticketing for the step-by-step system, for radar testing, and more recently for No. 5 crossbar system. Member of Tau Beta Pi, Eta Kappa Nu and Phi Kappa Phi. J. T. Lindsay Brown, B.S. College of the City of New York, 1915. Western Electric Company, 1915-1925. Bell Telephone Laboratories 1925-. For twenty-five years Mr. Brown was concerned with the develop- ment and testing of telephone instruments and such allied apparatus as loudspeakers, microphones, and earphones. In 1940 he transferred to work on the development of glass sealed magnetic switches. He is cur- rently in charge of a group developing mercury contact relays. Member of the A.I.E.E., I.R.E. and Acoustical Society of America. Wallace A. Depp, B.S. and M.S. in E.E., University of Illinois, 1936 and 1937. Bell Telephone Laboratories, 1937-. In his early laboratories 1515 1516 THE BELL SYSTEM TECHNICAL JOURNAL NOVEMBER, 1953 association Mr. Depp was concerned with thoriated tungsten and tan- talum emitters and later with cold cathode tubes. During World War II he worked on pulsing thyratrons and fixed spark gap tubes for radar, and miniature thy rations used in the proximity fuse. He was subse- quently in charge of the basic development of all types of gas-filled tubes. Recently he transferred to Transmission Systems Development with responsibility for broad band carrier terminal equipment, N and 0 carrier systems and automatic switching for the L3 coaxial cable system. Member of the A.I.E.E., Eta Kappa Nu, Tau Beta Pi, Phi Kappa Phi and Sigma Xi. Senior member of the I.R.E. James M. Early, B.S. cum laude, New York State College Of Forestry, 1943; M.S. and Ph.D. Ohio State University, 1948 and 1951. Bell Tele- phone Laboratories 1951-. After teaching Electrical Engineering at Ohio State University for five years while studying for his Master's and Ph.D. degrees, Dr. Early joined an electronic apparatus development group, participating in the development of the junction transistor. At present he is doing theoretical as well as development work on high fre- quency junction transistors. Member of the I.R.E. and Eta Kappa Nu. Associate of Sigma Xi. Joseph Gramels, B.S. in E.E., New York University, 1936. Bell Telephone Laboratories 1925-. Mr. Gramels was first occupied with testing and development work in transmission, including handsets, recording apparatus and 33 rpm records. In 1937 and 1938 he woiked on electrolytic condensers and silicon carbide varistors. Since 1938 he has been concerned with investigations of selenium rectifier cells both for Bell System applications and for military use. Member of the A.LE.E. Mason A. Logan, B.S. in Physics and Engineering, California Insti- tute of Technology, 1927; M.A. in Physics, Columbia University, 1933- Carnegie Institute of California, Seismological Laboratory, 1926-1927- Bell Telephone Laboratories, 1927-. His early Laboratories' projects were concerned with wire transmission problems particularly those of losses, noise and cross induction in local, manual and dial telephone circuits. This was followed by circuit research on alternating current methods of signal- ing including the use of non-linear elements and electronic terminal equip- ment. From 1941 to 1948 he worked on military projects, including a mine fire control system, anti-aircraft gun director, magnetic proxim- ity fuses, and guided missiles. For the past five years he has been a CONTRIBUTORS TO THIS ISSUE 1517 member of the Switching Apparatus Development Department in which he is supervising a group concerned with static and dynamic behavior of new electromagnets and relays. He is also engaged in investigations of the performance of electrical contacts on telephone relays. C. E. Pollard, Jr., Polytechnic Institute of Brooklyn, 1927-1931. Bell Telephone Laboratories, 1925-. Mr. Pollard first worked on voice recording and reproducing equipment and spent some time on the development of telephone carbon microphones before becoming inter- ested in mercury contact relays. In this field he has been concerned with a wide variety of relays, and during World War II concentrated on their application to military projects. He is currently engaged in development work on mercury contact relays. John H. Rowen, B.E.E., Ohio State University, 1948; M.S., Ohio State University, 1951. U.S.N.R., maintenance of Air Force radar equipment, 1944-1946; Ohio State University, Research Foundation, Antenna Laboratory, 1948-1951; Bell Telephone Laboratories, 1951-. Concerned with the practical application of fundamental research, he has spent his two years at Bell Laboratories on studies of the microwave behavior of ferrites. While at Ohio State University, Mr. Rowen worked on several methods of measuring the radiation efficiency of small aper- ture antennas. Member of the I.R.E. and Eta Kappa Nu. Mark A. Townsend, B.S. in E.E., Texas Technological College, 1936; S.M. in E.E., Massachusetts Institute of Technology, 1937. General Electric Company, 1937-1943; Massachusetts Institute of Technology, Radar School, 1943-1945. Bell Telephone Laboratoiies, 1945-. Mr. Townsend has been concerned with the basic development of gas-filled tubes. His projects have included work on the voltage reference tube, cold cathode stepping tubes, and the development of tubes for use in transmission and switching. At present Mr. Townsend is in charge of a group responsible for basic development of gas-filled tubes. Member of the A.I.E.E., Tau Beta Pi and Sigma Xi. THE BELL SYSTEM Jecnnical ournal DEVOTED TO THE SC I EN TIKIC^^^ AND ENGINEERING ASPECTS OF ELECTRICAL COMMUNICATION ADVISORY BOARD S. Bracken F. R. Kappel M. J. Kelly EDITORIAL COMMITTEE E. I. Green, Chairman A. J. BuscH F. R. Lack W. H. DOHERTY W. H. NUNN G. D. Edwards H. L Romnes J. B. FisK H. V. Schmidt R. K. Honaman G. N. Thayer EDITORIAL STAFF J. D. Tebo, Editor M. E. Strieby, Managing Editor R. L. Shepherd, Production Editor INDEX VOLUME XXXII 1953 AMERICAN TELEPHONE AND TELEGRAPH COMPANY NEW YORK LIST OF ISSUES IN VOLUME XXXII No. 1 January Pages 1-264 " 2 March 265-522 " 3 May 523-778 " ^ July 779-1018 " 5 September 1019-1270 " 6 November 1271-1518 Index to Volume XXXII ADP See Ammonium dihydrogen phos- phate 'AIEE See American Institute of Elec- trical Engineers AM See Amplitude modulation AMA See Automatic message account- ing Abstract(s) 255-60, 506-18, 767-74, 1007- 14, 1257-65, 1507-13 "Acceleration Effects on Electron Tubes" (F. W. Stubner) 1203-29 ''Acoustic Gyrator" (W. E. Kock) abstract 1261 Ahearn, A. J. "Formative of Negative Ions of Sulfur Hexafluoride" abstract 1007 "Ahnlichkeit zwischen Vokalformanten und Formanten von Musikinstru- menten" (W. E. Kock) 1261 "AIEE Progress" (D. A. Quarles) abstract 1013 Alley,Reuben,E. J., Jr. biographical material 1267 "Review of New Magnetic Phe- nomena" 1155-72 Alloy (s) copper, photometric determination of silicon 772 ferrous, silicon, photometric determi- nation 772 magnetic properties 1155, 1258 Mishima, physical and magnetic struc- ture 1262 nickel-carbon, reaction rates with barium oxide 771 silicon p-n junction diodes 512 zone-melting 513 Alsberg, D. A. "Principles and Applications of Con- verters for High-Frequency Meas- urements" abstract 506 American Institute of Electrical Engineers Quarles address. Phoenix meeting 514 telephone invention 768 American Society of Civil Engineers AND Architects centennial 768 professional engineering 514 Ammonium Dihydrogen Phosphate dielectric properties 510 elastic properties 510 piezoelectric properties 510 polymorphism 257 properties interpretation 510 Amplifier (s) L3 coaxial system 879-914 regenerative, for digital computers 508 transistor — cutoff frequency 516 traveling-wave-type, broad-band in- terdigital circuit for use in 256 vacuum tube electrometer 1261 Amplitude Modulation land mobile service 62 "Analysis of Measurements on Magnetic Ferrites" (CD. Owens) abstract 1012 Anderson, A. E. "Transistors in Switching Circuits" abstract 506-7 Anderson, F. B. "Gain and Phase Angle Measuring Set" abstract 1007 Anderson, J. R. "Electrical Delay Lines for Digital Computer Applications" abstract 1507 "Ferroelectric Storage Elements for THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Anderson, J. R. — continued: Digital Computers and Switching Systems" abstract 506 Anderson, P. W. ''Concept of Spin-Lattice Relaxation in Ferromagnetic Materials" 767 "Exchange Narrowing in Paramag- netic Resonance" abstract 1257-8 Andrus, J. "Motion of Domain Walls in Ferrite Crystals" 1260 Anglo-American Committee on Tech- nical Terminology report 773 "Anisotropic Elasticity with Applica- cations to Dislocation Theory" (Eshelby, Read and Shockley) abstract 1509 Answering Services See Telephone answering services Antenna(s) ground-plane, matching coax line 258 microwave relay stations 770 -reflector problem, theoretical study 770 Anti-Ferromagnetic Cobalt Discil- CIDE domain structure evidence 508 Antimony photometric determination in lead 1261 "Application of Information Theory to Research in Experimental Pho- netics" (G. E. Peterson) abstract 513 "Approximating the Mode From Weight- ed Sample Values" (H. L. Jones) abstract 1010 Arc, short, characteristics 1493-1506 "Arcing of Electrical Contacts in Tele- phone Switching Circuits" (M. M. Atalla) Part I — Theory of the Initiation of the Short Arc 1231-44 Part II — Characteristics of the Short Arc 1493-1506 "Arithmetic Processes for Digital Com- puters" (J. H. Felker) abstract 1009 Armstrong, C. A. "Communications for Civil Defense" 1507 abstract 1007 "Telephone Industry in National De- fense" abstract 507 Arnold, S. M. "Growth of Metal Whiskers" 1261 "ASTM Standards — Their Effect on Plastics Technology" (R. Burns) abstract 255 Atalla, M. M. "Arcing of Electrical Contacts in Telephone Switching Circuits" Part I — "Theory of the Initiation of the Short Arc" 1231-44 Part II — • "Characteristics of the Short Arc" 1493-1506 biographical material 1267 Atom(s) ionization, single, by electrons, thresh- old law 1512-13 magnetic resonance 74, 384 "Attenuation Equalizers" (F. R. Bies) abstract 1258 Audio Frequency Signals transmission, cold cathode tubes for 1371-91 "Auditory Tests with Synthetic Vowels" (R. L. Miller) abstract 1011 "Automatic Line Insulation Test Equip- ment for Local Crossbar Systems" (Burns and Dehn) 627-46 Automatic Message Accounting tape-to-card conversion 1260 toll messages billing, 1264 "Automatic Recognition of Spoken Digits" (Davis, Biddulph and Bala- shek) abstract 768-9 . "Automatic Switching for Nation-Wide Telephone Service" (Clark and Osborne) abstract 507 INDEX "Automatic Toll Switching Systems" (F. F. Shipley) 514 B Babcock, Wallace C. biographical material 261 "Intermodulation Interference in Radio Systems" 63-73 Bakelite (photoelastic) connections, solderless, wrapped 557 Balanced Polar Mercury Contact Re- lay" (Brown and Pollard) 1393- 1411 Balashek, S. "Automatic Recognition of Spoken Digits" abstract 768-9 Band(s) radio, mobile, frequenc}^ economy 42- 62 Bardeen, John biographical material 261 "Surface Properties of Germanium" 1-41 Barium Oxide nickel-carbon alloys, reaction rates 771 Barium Titanate domain properties 511 Barotta, P. J. "Transistor Ph3^sics Simplified" 1510 "Basic Physical Systems for Communi- cation Between Two Individuals" (G. E. Peterson) 1511 "Basic Theorems of Information Theory" (B. McMillan) abstract 1510 Battery (ies) lead-acid, stationary 257 Beach, A. L. "Solubility and Diffusion Coefficient of Carbon in Nickel: Reaction Rates of Nickel-Carbon Alloys with Barium Oxide" abstract 771 Beck, A. C. "Low-Loss Waveguide Transmission" abstract 1011 Becker, J. A. "Field Emission Microscope and Flash Filament Techniques for the Study of Structure and Adsorption on Metal Surfaces" abstract 1008 "Behavior of Magndic M.iicrijils" (R, M. Bozorth) abstract 1259 Bell Telephone Laboratories Tran- sistor Teachers Summer School 767 BeND(8) circular electric wave, transmission around 511 Benedict, T. S. "Microwave Observation of the Col- lision Frequency of Electrons in Germanium" 1258 Bennett, A. F. biographical material 775 "Improved Circuit for the Telephone Set" 611-26 Bennett, William R. biographical material 1267-8 "Correlatograph — A Machine for Continuous Display of Short Term Correlation" 1173-85 Biddulph, R. "Automatic Recognition of Spoken Digits" abstract 768-9 Bies, F. R. "Attenuation Equalizers" abstract 1258 BioAssAY Test 152, 154 Birdsall, H. A. ''Microstructure of Capacitor Paper" 1510 Black, H. S. "Experimental Verification of the Theory of Laminated Conductors" abstract 255 Blount, Frank E. biographical material 1514 "Transistor Oscillator for Use in Multifrequency Pulsing Current Supply" 1313-31 6 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Bodle, D. W. "Lightning Protection for Mobile Radio Fixed Stations" abstract 257 Bogert, B. P. ''Band Width of Vowel Formants" abstract 1507-8 BOLTZMANN EQUATION 170, 193 ''Bombardment Conductivity" (K. G. McKay) 1510 Boolean Functions sj^mmetry types, on n variables 1264 Bozorth, R. M. "Behavior of Magnetic Materials" abstract 1259 "Magnetic Crystal Anisotropy and Magnetostriction of Iron-Nickel Alloys" abstract 1008-9 "Magnetic Study of Low Temperature Transformation in Magnetite" 1264 "Measurement of Magnetostriction in Single Crystals" abstract ,1008 "Permalloy Problem" abstract 1258-9 Brangaccio, D. J. "Factors Affecting Traveling Wave Tube Power Capacity" 1509 Brattain, Walter H. biographical material 261 "Electron Spin Resonance in a Silicon Semi -Conductor" 1511 "Surface Properties of Germanium" 1-41 Bridgman, D, C. "College Graduates and the Country's Telephone Industry" 767 Briggs, H. B. "Infrared Absorption in High Purity Germanium" abstract 507 "New Infrared Absorption Bands in p-Type Germanium" abstract 507 "Broad-Band Interdigital Circuit for Use in Traveling-Wave-Type Ampli- fiers" (R. C. Fletcher) abstract 256 "Broad-Band Matching with a Direc- tional Coupler" (W. C. Jakes) abstract 509-10 Brown, J. T. Lindsay biographical material 1514 "Balanced Polar Mercury Contact Relay" 1393-1411 Buehler, E. "Single-Crystal Germanium" abstract 257 Bullington, Kenneth biographical material 262 "Frequency Economy in Mobile Radio Bands" 42-62 "Frequency Economy in Mobile Radio Bands," IRE Trans. 1508 "Radio Transmission Beyond the Horizon in the 40- to 4,000-Mc Band" abstract 767-8 Burns, R. "ASTM Standards — Their Effect on Plastics Technology" abstract 255 Burns, R. M. "Science and Scientists in Telecom- munications" 1259 Burns, R. W. "Automatic Line Insulation Test Equipment for Local Crossbar Sys- tems" 627-46 biographical material 775 Cable (s) Clogston 695/ coaxial: Clogston Cable compared with 703- 5 See also carrier: L3 multi-conductor 1245 polyethylene insulation 1245 Calbick, C. J. "Microstructure of Capacitor Paper" 1510 "Calibration of the Rolling Ball Vis- cometer" (H. W. Lewis) 1010 Call(s) INDEX delay curves for calls served at ran- dom, Riordan 100-19, 1266 delayed exponential, served in random order, working curves for, Wilkinson 360-83 queueing 367 Camera (s) three-phase power from single-phase source 256 Campbell, R. D. "Path Testing for Microwave Radio Routes" 1508 Campbell, W. E. "Solid Lubricants" 1508 Capacitor Dielectrics 1264-5 Capacitor Paper microstructure 1510 Carbon (in) nickel, solubility and diffusion coefficient 771 Carbon Monoxide ions, drift velocity 1013 Card Translator See Translator "Card Translator for Nationwide Dial- ing" (Hampton and Newsom) 1037- 98 Carrier (s). Carrier Systems L3: amplifiers 879-914 equalization 833-878, 943 manufacture 969 quality control 943-967, 969 regulation 833-878 system design 779-832 television terminals 915-942 Type-N 1259 Type-0 1259 Cause (assignable) type detection 512 Cavitation water, ultrasonically induced 1513 "Charge Transfer and the Mobility of Rare Gas Ions" (J. A. Hornbeck) abstract 509 Ciccolella, D. F. "High Frequency Crystal Units for Use in Selective Networks and Their Proposed Application in Filters Suitable for Mobile Radio Channel Selection" 1508 Circuit (s) telephone set (500 type) 611-26 Civil Defense communication 1007, 1507 telephone S3'stem 258 Clark, A. B. "Automatic Switching for Nation- wide Telephone Service" abstract 507 "Telephony Development in the United States" abstract 768 Clarke, K. B. "Western Electric 's Service with Standards" abstract 507 Clogston Type Conductors See Con- ductor Clos, Charles biographical material 519 "Non-Blocking Switching Networks Study" 406-24 Coaxial Carrier System L3 See Car- rier Cobalt Disilicide superconducting properties 1011 Cobalt Oxide (CoO) domain structure evidence 508 magnetomechanical effects 1260 Cobalt-Silicon System superconductivity 256 Cold pressure connection 533 Cold-Cathode Stepping Tube See Electron tube "Cold Cathode Tubes for Transmission of Audio Frequency Signals" (Town- send and Depp) 1371-91 "College Graduates and the Country's Telephone Industry" (D. C. Bridg- man) 767 Colley, Reginald H. biographical material 262 "Evaluation of Wood Preservatives" 120-69, 42&-505 "Review of American Standard Fiber Stresses of Wood Poles" 1262 8 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Communication between two individuals, physical systems 1511 "Communications for Civil Defense" (C. A. Armstrong) 1507 abstract 1007 "Comparison of Mobile Radio Trans- mission at 150, 450, 900 and 3700 Mc" (W. R. Young) 1513 "Comparison of Recording Processes" (J. G. Frayne) abstract 509 Computer (s) binary, functions 769 digital: delay lines for 1507 ferroelectric storage elements 506 regenerative amplifier 508 transistor, 255, 508, 769 historical development 772 "Computers — ■ Past, Present, and Fu- ture" (W. H. MacWilliams, Jr.) abstract 772 Conductivity bombardment 1510 electric fields, high, mobility in 768 Conductor (s) Clogston type, laminated, transmis- sion properties 695-713 electroformed for drop wire 1099-1135 laminated 695 experimental verification of the theory of 255 Connection (s) permanent, conditions for obtaining 551-91 pressure 525, 526 quality 532 solder 525 solderless, wrapped 523-610 "Connection Formulas Between the Solutions of Mathieu's Equation" (G. H. Wannier) abstract 1014 Contact (s) See Electrical contacts Contact Erosion switching circuits 1493 "Controlled Gas Leak" (J. Morrison) 1262 Converter (s) high-frequency measurements 506 transistor, negative-impedance 1510 "Conveyorization Avoids Handling Headaches" (P. T. Mathy) 1262 Conwell, Esther M. "High Field Mobility in Germanium With Impurity Scattering Domi- nant" abstract 1508 "Mobility in High Electric Fields" abstract 768 "Mobility of Electrons in Germanium" abstract 507 "Properties of Silicon and Ger- manium" abstract 507 CoO See Cobalt oxide Copper Alloys silicon, photometric determination 772 "Copper as an Acceptor Element in Germanium" (Fuller and Struthers) abstract 256 Copper Oxide varistors 514 Copper-Steel Wire See Wire "Correlatograph — ■ A Machine for Con- tinuous Display of Short Term Cor- relation" (W. R. Bennett) 1173-85 Corrosion connections, solderless, wrapped 598 Counter (s) crystal conduction 1262 reversible binary 516 "Coupled Resonator Reflex Klystron" (E.D.Reed) 715-66 Coupler (directional) broad band matching 509 multi-element 258, 511 Coy, J. A. "Type-0 Carrier Telephone" abstract 1259 Creosote as wood preservative 458, 461, 469, 473, 474, 479, 489-90 Crossbar System No. 1 automatic line insulation testing 627- 46 INDEX 9 Crossbar System No. 5 automatic line insulation testing 627- 46 throwdown machine 292-359 traffic -carrying characteristics study 292-359 Crosstalk interchannel 514 Crystal (s) antiferroelectric 510 conduction counter 1262 dislocated, geometrical relations 1012 electrons, e/m values, interpretation 515 ferrites, motion of domain walls 1260 germanium: single, purification and segregations prevention 1012 solute distribution 1512 high-frequency units for primary fre- quency standards 260 high-frequency units for use in selec- tive networks and proposed applica- tion in filters for mobile radio channel selection 1508 magnetic resonance 74 "Crystal Conduction Counter" (K. G. McKay) abstract 1262 Cutler, C. C. "Factors Affecting Traveling Wave Tube Power Capacity" 1509 Dacey, G. C. "Space-Charge Limited Hole Current in Germanium" abstract 1509 Dalton, A. G. "Practice of Quality Control" abstract 1009 Darrow, Karl K. biographical material 262, 519-20 "Magnetic Resonance of Electrons" 384-405 "Nuclear Magnetic Resonance" 74-99 Davis, K. H. "Automatic Recognition of Spoken Digits" abstract 768-9 "DC Field Distribution in a 'Swept Intrinsic' Semiconductor Configura- tion" (R. C. Prim) 665-94 Debye, P. P. "Mobility of Electrons in Germanium" abstract 507 DeCamp, R. T. "Matching Coax Line to the Ground- Plane Antenna" abstract 258 Decibel 6 in search of the missing 512 Defense See Civil defense Dehn, J. W. "Automatic Line Insulation Test Equipment for Local Crossbar Sys- tems" 627-46 biographical material 775 Delay Calls See Call "Delay Curves for Calls Served at Ran- dom" (John Riordan) 100-19 addendum to 1266 Delay Formula See Erlang Delay Lines digital computers 1507 storage capacity 1507 Depp, Wallace A. biographical material 1514-15 "Cold Cathode Tubes for Transmis- sion of Audio Frequency Signals" 1371-91 "Design Theory of Junction Trans- istors" (J. M. Early) 1271-1312 "Development and Manufacture of Elec- troformed Conductor for Telephone Drop Wire" (Gray and Murray) 1099-1135 Dial Telephone, Dialing card translator for nationwide 1037 exchanges, traffic engineering design 257, 516 numbering system, nation-wide, need for 512 Dickinson, Frank R. biographical material 1015 "L3 Coaxial System — Amplifiers" 879-914 10 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Dickten, E. "High -Frequency Transistor Tetrode" abstract 774 "Junction Transistor Tetrode for High -Frequency Use" abstract 517 Dielectric (s) capacitor 1264-5 ND4D2PO4 , properties 510 "Diffusion of Lithium Into Germanium and Silicon" (Fuller and Ditzen- berger) 1509 Digit(s) register 508-9 spoken, automatic recognition 768-9 Digital Computers See Computer Dimensional Resonance See Reso- nance Diode (s) semiconductor gates 1137-54 silicon p-n junction alloy 512 Directional Coupler See Coupler Dislocation Theory anisotropic elasticity 1509 Ditzenberger, J. A. "Diffusion of Lithium Into Germanium and Silicon" 1509 Dodge, H. F. biographical material 1015 "L3 Coaxial System — Quality Con- trol Requirements" 943-67 "Domain Properties in BaTiOa" (W. J. Merz) abstract 511 Domain Structure anti -ferromagnetic CoO 508 Domain Walls ferrite crystals 1260 motion 1162, 1260 "Dominant Wave Transmission Char- acteristics of a Multimode Round Waveguide" (A. P. King) abstract 256 "Drift Velocities of Ions in Krypton and Xenon" (R. N. Varney) abstract 517 "Drift Velocity of Ions in Oxygen, Nitro- gen, and Carbon Monoxide" (R. N. Varney) abstract 1013 Drop Wire See Wire Drvostep, J. J. "Standardization of Rigid Coaxial Transmission Lines" 1009 "Dual Photomagnetic Intermediate Studio Recording" (Frayne and Livadary) abstract 769 "Dynamic Measurements on Electro- magnetic Devices" (M. A, Logan) 1413-67 "Dynamic Program for Conversion" (J. R. Townsend) 1512 "Dynamic Spectrograms of Speech" (Kock and Miller) 771 "Dynamics of Transistor Negative-Re- sistance Circuits" (B. G. Farley) abstract 508 E Ear Canal sound pressure 512 Early, James M. biographical material 1515 "Design Theory of Junction Trans- istors" 1271-1312 "Effects of Space-Charge Layer Widening in Junction Transistors" abstract 507-8 Ebers, J. J. "Four-Terminal p-n-p-n Transistors" abstract 508 ECASS See Electronically controlled automatic switching system EcHo(es) television transmission 1511 "Economics of High-Speed Photogra- phy" (A. C. Keller) abstract 771 "Effect of Electrode Spacing on the Equivalent Base Resistance of Point-Contact Transistors" (L. B. Valdes) abstract 516 "Effects of Space-Charge Layer Widen- INDEX 11 ing in Junction Transistors" (J. M. Early) abstract 507-8 Ehrbar, R. D. biographical material 1015 "L3 Coaxial System — Sj^stem De- sign" 781-832 Elasticity anisotropic, applications to disloca- tion theory 1509 antiferromagnetic CoO, evidence of domain structure, measurements for 508 ND4D2PO4, properties 510 "Elasticity and Thermal Expansion of Germanium between —195 and 275° C" (M. E. Fine) abstract 1260 Electric Fields (high) mobility 768 Electric Wave circular, transmitting around bends 511 Electrical Contacts switching circuits, arcing of 1493- 1506 "Electrical Delay Lines for Digital Com- puter Applications" (J. R. Ander- son) abstract 1507 Electrical Stability connections, solderless, wrapped 591 Electrode Spacing equivalent base resistance of 516 transistors, point-contact 516 "Electroformed Copper-steel Wire De- velopment" (A. N. Gray) abstract 1260 Electrolysis Switch improved 259 Electromagnetic Devices dynamic measurements 1413-67 Electrometer Amplifier vacuum tube 1261 Electron (s) crystals, e/m values, interpretation 515 ejection from Mo by He+, He"*"^, and He2+ 769 electric fields, high, mobility in 768 germanium: collision frequency, microwave ob- servation 1258 mobility in 507 magnetic resonance 74, 384-405 mean free paths in evaporated metal films 514 microscopy, of solids 770 mobility in high electric fields 1511-12 recombination statistics 259 resonance. See Electron — magnetic resonance spin resonance in a silicon semi-con- ductor 1511 streams, microwave noise calculation 773 threshold law for single ionization of atoms or ions 1512-13 Electron (s) and Holes diffusion constant and mobility 767 mobility in high electric fields 151 1-12 recombination statistics 259 "Electron Ejection from Mo by He"*", He++, and He2+" (H. D. Hagstrum) abstract 769-70 "Electron Spin Resonance in a Silicon Semi-Conductor" (Portis, Kip, Kit- tel and Brattain) 1511 Electron Tube(s) acceleration effects 1203-29 cold cathode, for transmission of audio frequency signals 1371-91 computer designs 769 electrometer amplifier 1261 shock measurement 1203, 1215, 1219- 21 stepping, cold-cathode, 10-stage 773 traveling wave 1510 factors affecting power capacity 1509 Electronically Controlled Auto- matic Switching System 406 Electronics solid state phj^sics 257 Elmendorf, C. H. biographical material 1016 "L3 Coaxial System — System De- sign" 781-832 12 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Encoder optical position 508 Engineering Profession organization 514 Equalizer (s) attenuation 1258 Equation (s) transistor 1013 Erlang, A. K. 100 Erlang Formula 100, 1266 Erosion See Contact Erosion Eshelby, J. D. "Anisotropic Elasticity with Applica- tions to Dislocation Theory" abstract 1509 "Evaluation of Wood Preservatives" (Reginald H. Colley) 120-69, 425-505 "Evidence for Domain Structure in An- tiferromagnetic CoO From Elasticity Measurements" (M. E. Fine) abstract 508 "Exchange Narrowing in Paramagnetic Resonance" (Anderson and Weiss) abstract 1257-8 "Experimental Verification of the Rela- tionship Between Diffusion Constant and Mobility of Electrons and Holes" 767 abstract "Experimental Verification of the The- ory of Laminated Conductors" (Black, Mallinckrodt and Morgan) abstract 255 FM See Frequency modulation "Factors Afi'ecting Traveling Wave Tube Power Capacity" (Cutler and Bran- gaccio) 1509 Fading radio signals, super-high frequency 1187-1202 Faraday Effect ferromagnetic 1260, 1333 microwave 1166 Farley, B. G. "Dynamics of Transistor Negative- Resistance Circuits" abstract 508 Felker, J. H. "Arithmetic Processes for Digital Computers" abstract 1009 "Regenerative Amplifier for Digital Computer Applications" abstract 508 "Typical Block Diagrams for a Tran- sistor Digital Computer" abstract 255, 769 FEMF (foreign e.m.f .) Test 636 Ferrite(s) crystals, motion of domain walls 1260 description 1156 gyrator 1261 inductor cores 265-91 initial permeability and related losses 256 magnetic, measurements analysis 1012 magnetic phenomena 1155^ waveguide medium, new non -reciprocal 1512 "Ferrite Core Inductors" (H. A. Stone, Jr.) 265-91 "Ferrites in Microwave Applications" (J. H. Rowen) 1333-69 "Ferroelectric Storage Elements for Digital Computers and Switching Systems" (J. R Anderson) abstract 506 Ferromagnet spherical model 510 Ferromagnetic Alloys silicon, photometric determination 772 "Ferromagnetic Faraday Effect at Micro- wave Frequencies and Its Applica- tions" (C. L. Hogan) 1260 Ferromagnetic Resonance 1164-6 Ferromagnetic Substances magnetic resonance 74, 398 relaxation, spin-lattice 767 Ferrous Alloys silicon, photometric determination 772 Fields See Electric Fields; Magnetic Field "Field Emission Microscope and Flash INDEX 13 Filament Techniques for the Study of Structure and Adsorption on Metal Surfaces" (Becker and Hart- man) abstract 1008 Filamentary Transistors See Tran- sistor Film(s), metal, evaporated electron mean free paths 514 Film Pulling Mechanisms three-phase power from single -phase source 256 Filter (s) mobile radio channel selection, high frequency crystal units 1508 Finch, Tudor R. biographical material 1016 "L3 Coaxial System — Equalization and Regulation" 833-78 Fine, M. E. ''Elasticity and Thermal Expansion of Germanium between —195 and 275°C" abstract 1260 "Evidence for Domain Structure in Antiferromagnetic CoO From Elasticity Measurements" abstract 508 ''Magnetomechanical Effects in an An- tiferromagnet, CoO" 1260 Fletcher, R. C. "Broad-Band Interdigital Circuit for Use in Traveling-Wave-Type Am- plifiers" abstract 256 "New Infrared Absorption Bands in p-Type Germanium" abstract 507 Follingstad, H. G. "Optical Position Encoder and Digit Register" abstract 508-9 "Formative of Negative Ions of Sulfur Hexafluoride" (Ahearn and Hannay) abstract 1007 "Four-Terminal p-n-p-n Transistors" (J. J. Ebers) abstract 508 Fox, A. G. "Magnetic Double Refraction at Mi- crowave Frequencies" 517 Frayne, J. G. "Comparison of Recording Processes" abstract 509 "Dual Photomagnetic Intermediate Studio Recording" abstract 769 "Frequency Economy in Mobile Radio Bands" (Kenneth Bullington) 42- 62, 1508 Frequency Modulation land mobile service 62 "Frequency Response and Stability of Point-Contact Transistors" (B, N Slade) abstract 515 Frost, George R. biographical material 520 "Throwdown Machine for Telephone Traffic Studies" 292-359 Fuller, C. S. "Copper as an Acceptor Element in Germanium" abstract 256 "Diffusion of Lithium Into Germanium and Silicon" 1509 "Properties of Thermally Produced Acceptors in Germanium" abstract 256 "Gain and Phase Angle Measuring Set" (Anderson) abstract 1007 Gait, J. K. "Initial Permeability and Related Losses in Ferrites" abstract 256 "Motion of Domain Walls in Ferrite Crystals" 1260 Garrett, Robert F. biographical material 1016 "L3 Coaxial System — Application of Quality Control Requirements in the Manufacture of Components" 969-1005 GAs(es), rare 14 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Gas(es) — continued: ion charge transfer and mobility 509 ion-atom collisions 509 Gas Leak controlled 1262 Gaseous Ions motion in strong electric fields 170- 254, 259 Gaseous Kinetics charged particles moving through a gas under the influence of a static, uniform electric field 170 equations 178 glossary 178 Gentile, S. P. "Transistor Physics Simplified" 1510 "Geometrical Relations in Dislocated Crystals" (J. F. Nye) abstract 1012 Germanium acceptors, thermally produced, proper- ties 256 carrier density and mobility 507 copper as an acceptor element 256 crystals: single 257 solute distribution 1512 uses for the Hall effect 1010 elasticity 1260 electrical properties 507 electron mobility 507 electrons, collision frequency, micro- wave observation 1258 electrons and holes: diffusion constant and mobility, re- lationship 767 field dependence of mobility 1511 filament, theory of magnetic effects on the noise in a 647-64 high field mobility in, with impurity scattering dominant 1508 high purity, infrared absorption 507 infrared absorption 507 lithium diffusion 1509 minority carrier lifetime, measurement 517 phototransistors 1263 properties 507 p-type, infrared absorption bands 507 space-charge limited hole current 1509 surface properties 1-41 temporary traps 1260 thermal conversion, impurity effects 257 thermal expansion 1260 varistors 514 Goucher, F. S. "Interpretation of a-values in p-n Junction Transistors" abstract 1009 Graham, R. Shiels biographical material 1016 "L3 Coaxial System — Television Terminals" 915-42 Gramels, Joseph biographical material 1515 "Selenium Rectifiers — Factors in Their Application" 1469-92 Gray, A. N. biographical material 1268 "Development and Manufacture of Electroformed Conductor for Tele- phone Drop Wire " 1099-1 135 "Development of Electroformed Cop- per-Steel Wire" abstract 1260 Green, E. I. biographical material 1016-17 "L3 Coaxial System — Foreword" 779-80 Greensalt evaluation 487 Grossman, Alexander J, biographical material 1017 "L3 Coaxial System — System De- sign" 781-832 Groth, W. B. "Principles of Tape-to-Card Conver- sion in the AMA System" 1260 "Growth of Metal Whiskers" (Koonce and Arnold) 1261 Gyrator(s) acoustic waves 1261 employing magnetic field independent orientations in germanium 1010 ferrite 1261 INDEX 15 Hagstrum, H. D. "Electron Ejection from Mo by He+ He++, and He2+" abstract 769-70 "Hall Effect Modulators and 'Gyrators' Employing Magnetic Field Inde- pendent Orientations in German- ium" (Mason, Hewitt, and Wick) abstract 1010 Hamming, R. W. "Measurement of Magnetostriction in Single Crystals" abstract 1008 Hampton, L. N. biographical material 1268 "Card Translator for Nationwide Dialing" 1037-98 Hannay, H. B. "Formative of Negative Ions of Sulfur Hexafluoride" abstract 1007 "Hard Rubber" (H. Peters) abstract 513 Harris, J. R. "Transistor Shift Register and Serial Adder" abstract 509 Hartman, CD, "Field Emission Microscope and Flash Filament Techniques for the Study of Structure and Adsorp- tion on Metal Surfaces" abstract 1008 Haynes, J. R. "Temporary Traps in Silicon and Ger- manium" 1260 Heat pressure connection 533 wood sterilization 456 Heidenreich, R. D. "Methods in Electron Microscopy of solids" abstract 770 "Physical and Magnetic Structure of the Mishima Alloys" 1262 Helium Isotopes vapor pressures 772 Hewitt, W. H. "Hall Effect Modulators and 'Gyra- tors' Employing Magnetic Field Independent Orientations in Ger- manium" abstract 1010 "High Field Mobility in Germanium With Impurity Scattering Dom- inant" (E. M. Conwell) abstract 1508 High-Frequency measurements, converters for 506 "High-Frequency Crystal Units for Primary Frequency Standards" (A. W, Warner) abstract 260 "High-Frequency Crystal Units for Use in Selective Networks and Their Proposed Application in Filters Suit- able for Mobile Radio Channel Se- lection" (Ciccolella and Labrie) 1508 "High-Frequency Transistor Tetrode" (Wallace, Schimpf and Dickten) abstract 774 High-Speed Photography See Photog- raphy "Higher Frequencies for Ground-Air Communications" (S. B. Wright) 1265 Hines, M. E. "Traveling-Wave Tube" 1510 Hogan, C. L, "Ferromagnetic Farada}^ Effect at Microwave Frequencies and Its Ap- plications" 1260 Holcomb, A. L. "Three-Phase Power From Single- Phase Source" abstract 256 Hole (s) and Electrons See Electron (s) and holes Hopper, A. L. "Xonsynchronous Time Division with Holding and with Random Sam- pling" abstract 258, 513 16 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Hopper, H. G. "Motion of Domain Walls in Ferrite Crystals" 1260 Hornbeck, J. A. ''Charge Transfer and the Mobility of Rare Gas Ions" abstract 509 "Temporary Traps in Silicon and Ger- manium" 1260 "How to Conceal Telephone Wires, Keep Desks Neat" (A. R. Hutchinson) 770 "How To Detect the Type of an Assign- able Cause" (P. S. Olmstead) abstract 512 Hughes, W. T. "Vacuum Tube Electrometer Ampli- fier" 1261 Hulm, J. K. "Superconducting Properties of Co- balt Disilicide" abstract 1011 Humidity connections, solderless, wrapped 598 Huntley, H. R. biographical material 1268 "Transmission Design of Intertoll Telephone Trunks" 1019-36 Hussey, Luther W. biographical material 1269 "Semiconductor Diode Gates" 1137- 54 Hutchinson, A. R. "How to Conceal Telephone Wires, Keep Desks Neat" 770 Hyperfine Structure of Electron Resonance 396, 405 "Improved Circuit for the Telephone Set" (A. F. Bennett) 611-26 "Improved Electrolysis Switch" (V. B. Pike) abstract 259 "Impurity Effects in the Thermal Con- version of Germanium" (Slichter and Kolb) abstract 257 "In Search of the Missing 6 Db" Mun- son and Wiener) abstract 512 Indium Antimony (InSb), magneto- resistance effect in 1263 Inductor (s) ferrite materials as cores 265-91 "Influence of Echoes on Television Transmission" (P. Mertz) 1511 Information Theory basic theorems 1510 "Information-Bearing Elements of Speech" (G. E. Peterson) abstract 773 "Infrared Absorption in High Purity Germanium" (H. B, Briggs) abstract 507 "Initial Permeability and Related Losses in Ferrites" (J. K. Gait) abstract 256 InSb See Indium antimony Insulation crossbar systems, local, automatic line test equipment 627-46 leakage paths 627, 628 moisture 627 polyethylene 1245-56 stress relaxation 517-18 subscribers' lines 627 "Interesting Property of Certain Con- ductive Rubbers" (L. G. Kersta) 1261 Interference co-channel 42, 45 intermodulation 63-73 "Nonsynchronous Time Division with Holding and with Random Sam- pling" (Pierce and Hopper) abstract 258, 513 "Intermodulation Interference in Radio Systems" (Wallace C. Babcock) 63-73 "Interpretation of a-values in p-n Junc- tion Transistors" (Goucher and Prince) abstract 1009 "Interpretation of e/m Values for Elec- trons in Crystals" (W. Shockley) 515 INDEX 17 Ion(s) carbon monoxide, drift velocity 1013 gaseous, motion in strong electric fields 259, 170-254 gases, rare, charge transfer and mo- bility 509 ionization, single, by electrons, thresh- old law 1512-13 krypton, drift velocity 517 nitrogen, drift velocity 1013 oxygen, drift velocity 1013 sulfur hexafiuoride, negative 1007 xenon, drift velocities 517 Iron-Nickel Alloys magnetic properties 1258 Jakes, W. C, Jr. "Broad Band Matching with a Direc- tional Coupler" abstract 509-10 "Theoretical Study of an Antenna-Re- flector Problem" abstract 770 Joining Methods See Connection Jones, H. L. "Approximating the Mode From Weighted Sample Values" abstract 1010 Junction Transistor (s) See Transis- tor "Junction Transistor" (M. Sparks) abstract 516 "Junction Transistor Tetrode for High- Frequency Use" (Wallace, Schimpf and Dickten) abstract 517 Kaplan, E. L. "Tensor Notation and the Sampling Cumulants of A;-Statistics" abstract 770 Kaylor, Robert L. biographical material 1269 "Statistical Study of Selective Fading of Super-High Frequency Radio Sig- nals" 1187-1202 Keister, William biographical material 520 "Throwdown Machine for Telephone Traffic Studies" 292-359 Keller, A. C. "Economics of High-Speed Photog- raphy" abstract 771 "New General -Purpose Relay for Telephone Switching Systems" abstract 510 Kern, H. E. "Solubility and Diffusion Coefficient of Carbon in Nickel: Reaction Rates of Nickel-Carbon Alloys with Barium Oxide" abstract 771 Kersta, L. G. "Interesting Property of Certain Con- ductive Rubbers" 1261 Ketchledge, R. W. biographical material 1017 "L3 Coaxial System — Equalization and Regulation" 833-78 Kinetic Energy electrons 769 Kinetics See Gaseous kinetics King, A. P. "Dominant Wave Transmission Char- acteristics of a Multimode Round Waveguide" abstract 256 Kinsburg, Boris J. biographical material 1017 "L3 Coaxial Sj^stem — Quality Con- trol Requirements" 943-67 Kip, A. F. "Electron Spin Resonance in a Silicon Semi-Conductor" 1511 Kittel, C. "Electron Spin Resonance in a Silicon Semi-Conductor" 1511 Klie, Robert H. biographical material 1017 "L3 Coaxial System — System De- sign" 781-832 Klystron reflex, coupled resonator (tube) 715- 66 18 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Kock, W. E. "Acoustic Gyrator" abstract 1261 ''Xhnlichkeit zwischen Vokalforman- teii und Formanten von Musikin- strumenten" 1261 "Dynamic Spectrograms of Speech" 771 "Selective Voice Control Problems" abstract 771 "Similarity Between Vowel Formants and the Formants of Musical Instru- ments" 1261 Kohman, G. T. "Polyethylene Terephthalate — Its Use as a Capacitor Dielectric" abstract 1264-5 Kolb, E. D. "Impurity Effects in the Thermal Con- version of Germanium" abstract 257 "Solute Distribution in Germanium Crystals" 1512 Koonce, Miss S. E. "Growth of Metal Whiskers" 1261 Kruger, M. K. biographical material 1017-18 "L3 Coaxial System — Quality Con- trol Requirements" 943-67 Krypton ions, drift velocity 517 A;-Statistics tensor notation and sampling cumu- lants 770 Labrie, L. J. "High Frequency Crystal Units for Use in Selective Networks and Their Proposed Application in Filters Suitable for Mobile Radio Channel Selection" 1508 Laminated Conductors See Conduc- tor Lander, J. J. "Solubility and Diffusion Coefficient of Carbon in Nickel: Reaction Rates of Nickel-Carbon Alloys with Barium Oxide" abstract 771 "Vacuum Tube Electrometer Ampli- fier" 1261 "Large Current Amplifications in Fila- mentary Transistors" (W. Van Roos- broeck) 774 "Larmor Frequency" 85 "Larmor Precession" 85 Latex Ebonite 513 "Lead-Acid Stationary Batteries" (U. B. Thomas) abstract 257. Lebert, A. W. "Standardization of Rigid Coaxial Transmission Lines" 1009 Lentinus Lepideus Mad. 534 (test fungus) 154 Lenzites Trabea Mad. 617 (test fungus) 155 Lewis, H, W. "Calibration of the Rolling Ball Vis- cometer" 1010 "Multiple Meson Production in Nuc- leon-Nucleon Collisions" 771 "Spherical Model of a Ferromagnet" abstract 510 "Lightning Protection for Mobile Radio Fixed Stations" (D. W. Bodle) abstract 257 Line Insulation See Automatic Line Insulation Test Linvill, J. G. "Transistor Negative-Impedance Con- verters" 1510 Liquid (s) magnetic resonance 384 Lithium diffusion into germanium and silicon 1509 Livadary, J. P. "Dual Photomagnetic Intermediate Studio Recording" abstract 769 Logan, Mason A. biographical material 1515 "Dynamic Measurements on Electro- magnetic Devices" 1413-67 Lovell, G. H. biographical material 1018 "L13 Coaxial System — Amplifiers" 879-914 INDEX 19 "Low-Drain Transistor Audio Oscilla- tor" (D. E. Thomas) abstract 516 "Low-Loss Waveguide Transmission" (Miller and Beck) abstract 1011 L3 Coaxial System Amplifiers (Morris, Lovell and Dickin- son) 879-914 Application of Quality Control Re- quirements in the Manufacture of Components (Garrett, Tuffnell and Waddell) 969-1005 Equalization and Regulation (Ketch- ledge and Finch) 833-78 Foreword (E.I. Green) 779-80 Quality Control Requirements (Dodge, Kinsburg and Kruger) 943-67 System Design (Elmendorf, Ehrb^r, Klie and Grossman) 781-832 Television Terminals (Rieke and Gra- ham) 915-42 Lubricant (s) solid 1508 Luke, C. L. "Photometric Determination of Anti- mony in Lead Using the Rhodamine B Method" 1261 "Photometric Determination of Silicon in Ferrous, Ferromagnetic, Nickel, and Copper Alloys — A Molyb- denum Blue Method" abstract 772 Lumsden, G. Q. "Quarter Century of Evaluating Pole Preservatives" 772 "Review of American Standard Fiber Stresses of Wood Poles" 1262 M McKay, K. G. "Bombardment Conductivity" 1510 "Crystal Conduction Counter" abstract 1262 McLean, D. A. "Microstructure of Capacitor Paper" 1510 McMahon, W. "Polyethylene Terephthalate — Its Use as a Capacitor Dielectric" abstract 1264-5 McMillan, B. "Basic Theorems of Information The- ory" abstract 1510 McRae, J. W. biographical material 776 "Solderless Wrapped Connections — Introduction" 523-4 "Transistors in Our Civilian Econ- omy" abstract 510 Mac Williams, W. H., Jr. "Computers — Past, Present, and Fu- ture" abstract 772 Magnetic Alloys 1155, 1258, 1262 "Magnetic Crystal Anisotropy and Mag- netostriction of Iron-Nickel Alloys" (Bozorth and Walker) abstract 1008-9 "Magnetic Double Refraction at Micro- wave Frequencies" (Weiss and Fox) 517 Magnetic Field noise in a germanium filament 647 Magnetic Materials behavior 1259 classical concepts 1155 phenomena, new 1155-72 Magnetic Resonance discovery and importance 74 equation, general 80 nuclear 74-99 practical value 74 "Magnetic Resonance of Electrons" (Karl K. Darrow) 384r405 Magnetic Structure Mishima alloys 1262 "Magnetic Study of Low Temperature Transformation in Magnetite" (Wil- liams and Bozorth) 1264 Magnetite low temperature transformation, mag- netic study 1264 "Magnetomechanical Effects in an Anti- ferromagnet CoO" (M. E. Fine) 1260 20 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 "Magnetoresistance Effect in InSb" (Pearson and Tanenbaum) 1263 Mallina, R. F. biographical material 775-6 "Solderless Wrapped Connections — Part I — Structure and Tools" 525-55 Mallinckrodt, C. O. "Experimental Verification of the The- ory of Laminated Conductors" abstract 255 Mapes, C. M. "Telephone System in National De- fense" abstract 258 Mason, W. P. biographical material 776 "Hall Effect Modulators and 'Gyra- tors' Employing Magnetic Field Independent Orientations in Ger- manium" abstract 1010 "Piezoelectric, Dielectric, and Elastic Properties of ND4D2PO4" abstract 510 "Rotational Relaxation in Nickel at High Frequencies" abstract 1262 "Solderless Wrapped Connections — Part II — Necessary Conditions for Obtaining a Permanent Connection" 557-90 "Matching Coaxial Line to the Ground- Plane Antenna" (R. T. DeCamp) abstract 258 Mathieu's Equation connection formulas between the solu- tions of 1014 Mathy, P. T. "Conveyorization Avoids Handling Headaches" 1262 Matthias, B. T. "Piezoelectric, Dielectric, and Elastic Properties of ND4D2PO4" abstract 510 "Polymorphism of ND4D2PO4" abstract 257 "Superconducting Properties of Cobalt Disilicide" abstract 1011 "Superconductivity in the Cobalt-Sili- con System" abstract 256 May, A. S. "Microwave System Test Equipment" 1510 "Mean Free Paths of Electrons in Evap- orated Metal Films" (Reynolds and Stilwell) 514 "Measurement of Magnetostriction in Single Crystals" (Bozorth and Ham- ming) abstract 1008 "Measurement of Minority Carrier Life- time in Germanium" (L. B. Valdes) abstract 517 Mechanical Stability connections, solderless, wrapped 591 "Mechanized Billing of AMA Toll Mes- sages" (F. D. Slade) 1264 Mellor Approximation 100, 111-13 Mertz, P. "Influence of Echoes on Television Transmission" 1511 Merz, W. J. "Domain Properties in BaTiOs" abstract 511 "Polymorphism of ND4D2PO4" abstract 257 Meson Production in nucleon-nucleon collisions 771 Message Accounting See Automatic Message Accounting Metal whiskers, growth 1261 zone-melting 513 Metallurgy solid state physics 257 "Methods in Electron Microscopy of Solids" (R. D. Heidenreich) abstract 770 Microscopy electron, of solids, methods 770 "Microstructure of Capacitor Paper" (McLean, Birdsall and Calbick) 1510 INDEX 21 Microwave Frequencies ferrites 1333-69 ferromagnetic Faraday effect 1260 magnetic double refraction 517 Microwave Noise in electron streams, calculation 773 "Microwave Observation of the Collision Frequency of Electrons in German- ium" (Benedict and Shockley) 1258 Microwave Radio routes, path testing 1508 Microwave Relay Stations antenna 770 "Microwave System Test Equipment" (A. S. May) 1510 "Microwaves from Coast-to-Coast" (J. R. Rae) 1013 Miller, E. S. "Notes on Methods of Transmitting the Circular Electric Wave Around Bends" abstract 511 Miller, R. L. "Auditory Tests with Synthetic Vow- els" abstract 1011 "Dynamic Spectrograms of Speech" 771 Miller, S. E. "Low-Loss Waveguide Transmission" abstract 1011 "Multi -Element Directional Cou- plers" abstract 258 MisHiMA Alloys 1262 Mobile Radio See Radio "Mobility in High Electric Fields" (E. M. Con well) abstract 768 "Mobility of Electrons in Germanium" (Debye and Conwell) abstract 507 "Mobility of Holes and Electrons in High Electric Fields" (E. J. Ryder) abstract 1511-12 Modulator (s) employing magnetic field independent orientations in germanium 1010 Moisture insulation of subscriber lines 627 Molybdenum Blue Method 772 Montgomery, H. C. "Transistor Noise in Circuit Applica- tions" abstract 511-12 Morgan, S. P. "Experimental Verification of the The- ory of Laminated Conductors" abstract 255 Morris, Lester H. biographical material 1018 "L3 Coaxial System — Amplifiers" 879-914 Morrison, J. "Controlled Gas Leak" 1262 "Motion of Domain Walls in Ferrite Crystals" (Gait, Andrus and Hop- per) 1260 "Motion of Gaseous Ions in Strong Elec- tronic Fields" (Gregory H. Wannier) 170-254 abstract 259 Motion Pictures three-phase power from single -phase source 256 "Multi -Element Directional Couplers" (Miller and Mumford) abstract 258, 511 "Multiple Meson Production in Nucleon- Nucleon Collisions" (H. W. Lewis) 771 Mumford, W. W. "Multi -Element Directional Cou- plers" abstract 258, 511 "Optimum Piston Position for Wide- Band Coaxial-to-Waveguide Transducers" abstract 772 Munson, W. A. "In Search of the Missing 6 Db" abstract 512 Murray, G. E. biographical material 1269 "Development and Manufacture of Electro-formed Conductor for Tele- phone Drop Wire" 1099-1135 22 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 N National Defense See Civil defense National Political Conventions (Chi- cago) television coverage 1263 "Nation-Wide Numbering Plan" (W. H. Nunn) abstract 512 "Nature of Solids" (G. H. Wannier) abstract 517 ND4D2PO4 See Ammonium dihydrogen phosphate Nelson, R. A. "Vapor Pressure of He^ a He^ Mix- tures" 772 Nesbitt, E. A. "Physical and Magnetic Structure of of the Mishima Alloys" 1262 "New General-Purpose Relay for Tele- phone Switching Systems" (A. C. Keller) abstract 510 "New Infrared Absorption Bands in p- Type Germanium" (Briggs and Flet- cher) abstract 507 "New Method of Calculating Microwave Noise in Electron Streams" (J. R. Pierce) abstract 773 "New Non-Reciprocal Waveguide Me- dium Using Ferrites" (E. H. Turner) 1512 Newsom, James B. biographical material 1269-70 "Card Translator for Nationwide Dial- ing" 1037-98 Nickel carbon in, solubility and diffusion co- efficient 771 rotational relaxation at high frequen- cies 1262 Nickel Alloys silicon, photometric determination 772 Nickel-Carbon Alloys barium oxide, reaction rates with 771 Nitrogen ions, drift velocity 1013 Noise germanium filament, theory of mag- netic effect on 647-64 in a temperature-limited electron beam in a drift space 773 microwave, in electron streams, calcu- lation 773 oscillatory 1371 "Non-Blocking Switching Networks Study" (Charles Clos) 406-24 "Nonsynchronous Time Division with Holding and with Random Sam- pling" (Pierce and Hopper) abstract 258, 513 "Notes on Methods of Transmitting the Circular Electric Wave Around Bends" (E. S. Miller) abstract 511 N-P-N Junction Phototransistors See Phototransistor Nuclear Magnetic Resonance discovery 99 "Nuclear Magnetic Resonance" (Karl K. Darrow) 74-99 Nucleon-Nucleon Collisions multiple Meson production 771 Nucleus, nuclei magnetic moments 74 Number (s), numbering See Telephone numbers Nunn, W. H. "Nation-Wide Numbering Plan" abstract 512 Nye, J. F. "Geometrical Relations in Dislocated Crystals" abstract 1012 O'Connor, S. F. "Plating Room Waste Water Disposal " 1012 Oil-Type Preservatives 160 Olmstead, P. S. "How to Detect the Type of an As- signable Cause" abstract 512 Olsen, K. M. "Purification and Prevention of Segre- INDEX 23 gation in Single Crystals of German- ium" 1012 ''On the Band Width of Vowel Formants" (B. P. Bogert) abstract 1507-8 "On the Number of Symmetry Types of Boolean Functions on n Variables" (D. Slepian) 1264 "Optical Position Encoder and Digit Register" (Follingstad, Shive and Yaeger) abstract 508-9 "Optimum Piston Position for Wide- Band Coaxial-to-Waveguide Trans- ducers" (W. W. Mumford) abstract 772 "Organization of the Engineering Pro- fession" (D. A. Quarles) abstract 514 Osborne, H. S. "Automatic Switching for Nation- wide Telephone Service" abstract 507 "Rose by Any Other Name" 773 Oscillator low-drain transistor audio 516 transistor, use in multifrequency puls- ing current supply 1313-31 Osmer, T. F. biographical material 776 "Solderless Wrapped Connections — Part II — Necessary Conditions for Obtaining a Permanent Connection" 557-90 Overseas Telephony single-sideband system 514 Owens, C. D. "Analysis of Measurements on Mag- netic Ferrites" abstract 1012 Oxygen ions, drift velocity 1013 Paramagnetic Resonance exchange narrowing 1257 Paramagnetic Salts magnetic resonance 74 Paramagnetic Solids electron resonance 393 "Path Testing for Microwave Radio Routes" (R. D. Campbell) 1508 Pearson, G. L. "Magnetoresistance Effect in InSb" 1263 "Silicon p-n Junction Alloy Diodes" abstract 512-13 Peck, D. S. "Ten-Stage Cold-Cathode Stepping Tube" abstract 773 Pentachlorophenol evaluation 488 "Permalloy Problem" (R. M. Bozorth) abstract 1258-9 Peters, H. "Hard Rubber" abstract 513 Peterson, G. E. "Application of Information Theory to Research in Experimental Phonetics" abstract 513 "Basic Physical Systems for Communi- cation Between Two Individuals" 1511 "Information-Bearing Elements of Speech" abstract 773 Pfann, W. G. "Principles of Zone-Melting" abstract 513 "Purification and Prevention of Segre- gation in Single Crystals of German- ium" 1012 "Segregation of Two Solutes, with Particular Reference to Semicon- ductors" abstract 258 Phonetics information theory application 513 Photocell (s) germanium, single-crystal 257 M-1740 p-n junction 514 Photography high-speed, economics 771 24 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Photomagnetic Intermediate Studio Recording 769 "Photometric Determination of Anti- mony in Lead Using the Rhodamine B Method" (C. L. Luke) 1261 "Photometric Determination of Silicon in Ferrous, Ferromagnetic, Nickel, and Copper Alloys — A Molyb- denum Blue Method" (C. L. Luke) abstract 772 Phototransistor (s) germanium 1263 n-p-n junction multiplier 1263 p-n junction 1263 point contact 1263 See also Transistor "Physical and Magnetic Structure of the Mishima Alloys" (Nesbitt and Hei- denreich) 1262 Pierce, J. R. "New Method of Calculating Micro- wave Noise in Electron Streams" abstract 773 "Nonsynchronous Time Division with Holding and with Random Sam- pling" abstract 258, 513 "Transistors" 1511 "Piezoelectric, Dielectric, and Elastic Properties of ND4D2PO4" (Mason and Matthias) abstract 510 Pike, V. B. "Improved Electrolysis Switch" abstract 259 Pilliod, J. J. "Fundamental Plans for Toll Tele- phone Plant" 513 Pine (wood) See Southern pine Piston transducers, wide-band coaxial-to- waveguide, optimum position 772 Plane Wave Theory 1333 "Plans for Toll Telephone Plant, Funda- mental" (J. J. Pilloid) 513 Plastics stress relation 517-18 "Plating Room Waste Water Disposal" (S. F. O'Connor) 1012 P-N Junction Phototransistors See Phototransistor P-N Junction Silicon Diode 512 P-N-P-N Transistors See Transistor Point Contact Phototransistors See Phototransistor Pole(s) See Telephone poles Pollard, C. E., Jr. "Balanced Polar Mercury Contact Re- lay" 1393-1411 biographical material 1516 Polyethylene connections, solderless, wrapped 557 "Polyethylene Insulated Telephone Ca- ble" (A. S. Windeler) 1245-56 "Polyethylene Terephthalate — Its Use as a Capacitor Dielectric" (Wooley, Kohman and McMahon) abstract 1264-5 "Polymorphism of ND4D2PO4" (Wood, Merz and Matthias) abstract 257 Portis, A. M. "Electron Spin Resonance in a Silicon Semi-Conductor" 1511 "Practice of Quality Control" (A. G. Dalton) abstract 1009 Precession 84 Prim, R. C. biographical material 776-7 "DC Field Distribution in a 'Swept Intrinsic' Semiconductor Configura- tion" 665-94 "Space-Charge Limited Emission in Semi-Conductors" abstract 1512 Prince, M. B. "Interpretation of cc-values in p-n Junction Transistors" abstract 1009 "Principles and Applications of Con- verters for High-Frequency Meas- urements" (D. A. Alsberg) abstract 506 "Principles of Tape-to-Card Conversion in the AMA System" (W. B. Groth) 1260 INDEX 25 ''Principles of Zone-Melting" (W. G. Pfann) abstract 513 i*rogress and Problems" (D. A. Quarles) 1511 "Properties of Germanium Phototransis- tors" (J. N. Shive) abstract 1263 "Properties of Junction Transistors (K. D. Smith) 774 "Properties of M-1740 p-n Junction Pho- tocells' (J. N. Shive) abstract 514 "Properties of Silicon and Germanium" (Esther M. Conwell) abstract 507 "Properties of Thermally Produced Ac- ceptors in Germanium" (Fuller, Theuerer and Van Roosbroeck) abstract 256 Proton Resonance Absorption 75 Pugh, S. G. "Southern Bell Switches to Chemical Brush Control" abstract 1263 "Purification and Prevention of Segrega- tion in Single Crystals of German- ium" (Pfann and Olsen) 1012 Q Quality Control L3 coaxial system 943, 969 Quarles, Donald A. "AIEE Progress" abstract 1013 "Organization of the Engineering Pro- fession" abstract 514 "Progress and Problems" 1511 "Report to the Membership" 1511 "We Begin a New Institute Year" abstract 514 "Quarter Century of Evaluating Pole Preservatives" (G. Q. Lumsden) 772 Radiation out-of-band Radio (s) 514 connections, solderless, wrapped 525 Jf field strength fluctuations far beyond the horizon 773 frequency spectrum, use 42 lightning protection 257 microwave, routes, path testing 1508 mobile : bands, frequency economy 42-62, 1508 channel selection, filters suitable for 1508 channels usability 42-62 fixed stations, lightning protection 257 intermodulation interference 63-73 path loss between antennas 43 transmission, comparison at 150, 450, 900 and 3700 Mc 1513 super-high frequency, signal fading, study 1187-1202 "Radio Transmission Beyond the Hori- zon in the 40- to 4,000-Mc Band" (K. Bullington) abstract 767-8 Rae, J. B. "Microwaves from Coast-to-Coast" 1013 Ralston, R. W. "Television Coverage of the National Political Conventions" abstract 1263 Random Service calls, delayed exponential, working curves 360-83 delay curves 100-19 Rapid Line Insulation Testing switching systems 627, 631-3 Read, W. T. "Anisotropic Elasticity with Applica- tions to Dislocation Theory" abstract 1509 "Statistics of the Recombinations of Holes and Electrons" abstract 259 Recording See Sound recording Rectifier (s) germanium, single-crystal 257 selenium 1469-92 26 THE BELL SYSTEM TECHxVICAL JOURNAL, 1953 Reed, E. D. biographical material 777 "Coupled Resonator Reflex Klystron" 715-66 Reflector plane, antenna used with 770 Regenerative Amplifier See Ampli- fier "Regenerative Amplifier for Digital Computer Applications" (J. H. Felker) abstract 508 Relaxation electric fields, high, mobility in 768 ferromagnetic materials 767 nickel at high frequencies, rotational 1262 nuclear magnetic resonance 92-9 Relay AFtype 510 AG 510 AJ 510 electromagnetic, general purpose 510 mercury contact, balanced polar 1393-1411 "Report to the Membership" (D. A. Quarles) 1511 Resonance 83 dimensional 1163-4 electron spin, in a silicon semi-conduc- tor 1511 ferromagnetic 1164-6 See also Magnetic resonance; Nuclear magnetic resonance; Paramagnetic resonance Resonator coupled reflex klystron 715-66 "Review of American Standard Fiber Stresses of Wood Poles" (Lumsden and Colley) 1262 "Review of New Magnetic Phenomena" (R. E. Alley, Jr.) 1155-72 Reynolds, F. W. "Mean Free Paths of Electrons in Evaporated Metal Films" 514 Rhodamine B Method photometric determination of anti- mony in lead 1261 Rice, S. O. "Statistical Fluctuations of Radio Field Strength Far Beyond the Horizon" abstract 773 Rieke, John W. biographical material 1018 "L3 Coaxial System — Television Ter- minals" 915-42 Riordan, John biographical material 263 "Delay Curves for Calls Served at Random" 100-19 addendum 1266 Ritchie, Alistair E. biographical material 520-1 "Throwdown Machine for Telephone Traffic Studies" 292-359 "Rose by Any Other Name" (H. S. Os- borne) 773 "Rotational Relaxation in Nickel at High Frequencies" (W. P. Mason) abstract 1262 Rowen, John H. biographical material 1516 "Ferrites in Microwave Applications" 1333-69 Rubber (s) conductive 1261 hard 513 synthetic 513 Rudisill, J. A., Jr. "Simple Phase-Angle Measurement Technique" abstract 257 Ryder, E. J. "Mobility of Holes and Electrons in High Electric Fields" abstract 1511-12 Salt (paramagnetic) magnetic resonance 74 Sawyer, B. "Silicon p-n Junction Alloy Diodes" abstract 512-13 Schimpf, L. G. "High-frequency Transistor Tetrode" abstract 774 INDEX 27 "Junction Transistor Tetrode for High-Frequency Use" abstract 517 Schlaak, N. F. ''Single-Sideband System for Overseas Telephony" abstract 514 "Science and Scientists in Telecommuni- cations" (R. M. Burns) 1259 "Segregation of Two Solutes, with Par- ticular Reference to Semiconduc- tors" (W. G. Pfann) abstract 258 "Selective Voice Control Problems" (W. E. Kock) abstract 771 "Selenium Rectifiers — Factors in Their Application" (J. Gramels) 1469-92 Semiconductor (s) DC field in a "swept intrinsic" 665-94 diode gates 1137-54 silicon, electron spin resonance 1511 solute segregation 258 space-charge limited emission 1512 statistics of the recombination of holes and electrons 259 "swept intrinsic" configuration, DC field distribution in 665-94 "Semiconductor Diode Gates" (Luther W. Hussey) 1137-54 Serial Adder transistor 509 Shift Register transistor 509 Shipley, F. F. "Automatic Toll Switching Systems" 514 Shive, J. N. "Optical Position Encoder and Digit Register" abstract 508-9 "Properties of Germanium Phototran- sistors" abstract 1263 "Properties of M-1740 p-n Junction Photocells" abstract 514 Shockley, William "Anistropic Elasticity with Applica- tions to Dislocation Theory" abstract 1509 "Interpretation of e/m Values for Electrons in Crystals" 515 "Microwave Observation of the Col- lision Frequency of Electrons in Germanium" 1258 "Solid State Physics in Electronics and in Metallurgy" abstract 257 "Space-Charge Limited Emission in Semi-Conductors" abstract 1512 "Statistics of the Recombinations of Holes and Electrons" abstract 259 "Transistor Electronics: Imperfec- tions, Unipolar and Analog Tran- sistors" abstract 515 "Unipolar 'Field-Effect' Transistor" abstract 515 Short Arc characteristics 1493-1506 Silicon carrier density and mobility 507 electrical properties 507 electrons and holes, field dependence of mobility 1511 lithium diffusion 1509 p-n junction alloy diodes 512-13 properties 507 semi-conductor, electron spin reso- nance 1511 temporary traps 1260 varistors 514 "Silicon p-n Junction Alloy Diodes" (Pearson and Sawyer) abstract 512-13 "Similarity Between Vowel Formants and the Formants of Musical Instru- ments" (W. E. Kock) 1261 "Simple Attachment for Low Tempera- ture Use of an X-Ray Diffraction Camera" (E. A. Wood) 1264 "Simple Phase-Angle Measurement Technique" (J. A. Rudisill, Jr.) abstract 257 28 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 "Single-Crystal Germanium" (Teal, Sparks and Buehler) abstract 257 Single-Phase 115-Volt Power conversion to three-phase 230-volt form 256 "Single-Sideband System for Overseas Telephony" (N. F. Schlaak) abstract 514 Slade, F. D. "Mechanized Billing of AMA Toll Messages" 1264 Slepian, D. "On the Number of Symmetry Types of Boolean Functions on n Vari- ables" 1264 Slichter, W. P. "Impurity Effects in the Thermal Con- version of Germanium" abstract 257 "Solute Distribution in Germanium Crystals" 1512 Smith, K. D. "Properties of Junction Transistors" 774 Soil-Block Tests creosotes, residual, characteristics, of weathered blocks 479-87 evaluation 132-60 Solder Joint 525 Solderless Wrapped Connections 523-610 "Solderless Wrapped Connections" Introduction (J. W. McRae) 523-4 Part I — Structure and Tools (R. F. Main n a) 525-55 Part II — • Necessary Conditions for Obtaining a Permanent Connection (Mason and Osmer) 557-90 Part III — Evaluation and Perform- ance Tests (R. H. Van Horn) 591- 610 Solid (s) magnetic resonance 384 " nature of 517 paramagnetic, electron resonance 393 "Solid Lubricants" (W. E. Campbell) 1508 "Solid State Physics in Electronics and in Metallurgy" (W. Shockley) abstract 257 "Solubility and Diffusion Coefficient of Carbon in Nickel : Reaction Rates of Nickel-Carbon Alloys with Barium Oxide" (Lander, Kern and Beach) abstract 771 Solute (s) germanium crystals, distribution 1512 segregation during directional solidifi- cation of an ingot 258 "Solute Distribution in Germanium Crystals" (Slichter and Kolb) 1512 Sound Pressure ear canal 512 Sound Recording dial photomagnetic intermediate studio recording 769 magnetic 509 mechanical 509 photographic 509 processes comparison 509 re-recording 769 three-phase power from single-phase source 256 "Southern Bell Switches to Chemical Brush Control" (S. G. Pugh) abstract 1263 Southern Pine Sapwood stakes, testing 427 "Space-Charge Limited Emission in Semi-Conductors" (Shockley and Prim) abstract - 1512 "Space-Charge Limited Hole Current in Germanium" (G. C. Dacey) abstract 1509 Sparks, M. "Junction Transistor" abstract 516 "Single-Crystal Germanium" abstract 257 Spectrogram (s) speech 771 Speech information-bearing elements 773 phonetic content 771 spectrograms 771 INDEX 29 "Spherical Model of a Ferromagnet" (Lewis and Wannier) abstract 510 Spin-Latticp: Relaxation Sec K(>l;i.\- ation "Standardization of Rigid Coaxial Transmission Lines" (Drvostep and Lebert) 1009 Stang, L. R. "Telephone Answering Services" 516 Stansel, F. R. "Transistor Equations" abstract 1013 "Statistical Fluctuations of Radio Field Strength Far Beyond the Horizon" (S. O. Rice) abstract 773 "Statistical Study of Selective Fading of Super-High Frequency Radio Signals" (R. L. Kaylor) 1187-1202 "Statistics of the Recombinations of Holes and Electrons" (Shockley and Read) abstract 259 Sterilization wood preservation 152 Stewart, J. A. "Traffic Engineering Design of Dial Telephone Exchanges" 516 abstract 257 Stilwell, G. R. "Mean Free Paths of Electrons in Evaporated Metal Films" 514 Stone, H. A., Jr. biographical material 521 "Ferrite Core Inductors" 265-91 Strain (s) solderless wrapped connections 557 Stress (es) solderless wrapped connections 557 "Stress Relaxation in Plastics and Insu- lating Materials" (E. E. Wright) abstract 517-18 Struthers, J. D. "Copper as an Acceptor Element in Germanium" abstract 256 Stubner, F. W. "Acceleration Effects on Electron Tubes" 1203-2<) biographical material 1270 Suhl, Harry biographical material 777 "Theory of Magnetic Effects on the Noise in a Germanium Filament" 647-64 Sulfur Hexafluoride ions, negative, formative of 1007 "Superconducting Properties of Cobalt Disilicide" (Matthias and Hulm) abstract 1011 "Superconductivity in the Cobalt-Sili- con System" (B. T. Matthias) abstract 256 "Surface Properties of Germanium" (Brattain and Bardeen) 1-41 Swedish Creosote Evaluation Tests 489-90 Switches dynamic measurements of reciprocat- ing phenomena 1413 electrolysis, improved 259 Switching Circuits arcing of electrical contacts 1231- 44, 1493-1506 contact erosion 1493 Switching Systems AF-type relay 510 automatic 507, 514 balanced polar mercury contact relay 1393 calls delay 100 cold cathode gas filled tubes 1371 counting, reversible binary 516 crossbar No. 5, traffic study 292-359 crossnet array 406 crosspoints 406, 419 ferroelectric storage elements 506 nation-wide telephone service 507, 1019-36 non-blocking networks 406-24 rapid line insulation testing 627, 631-3 toll 514, 1019-36 transistors 506 voice frequency signals 1371 30 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Tanenbaum, M. "Magnetoresistance Effect in InSb" 1263 Tape-to-Card Conversion in the AM A system 1260 Teal, G. K. * 'Single-Crystal Germanium" abstract 257 Tebo, Julian D. appointed editor opjp 523 Telecommunication (s) science and scientists in 1259 "Telephone Answering Services" (L. R. Stang) 516 Telephone Calls See Call(s) "Telephone Industry in National De- fense" (C. A. Armstrong) abstract 507 Telephone Numbers nation-wide plan 512 Telephone Poles, Posts fiber stresses 1262 preservatives, evaluation 772 testing 425^505 Telephone Set (500 type) circuit, improved 611-26 "Telephone System in National De- fense" (CM. Mapes) abstract 258 Telephone Traffic delay formula 100 throwdown machine 292-359 "Telephony Development in the United States" (A. B. Clark) abstract 768 Television connections, solderless, wrapped 525/ political conventions, Chicago 1263 terminals, L3 coaxial system 915-42 transmission, influence of echoes 1511 "Television Coverage of the National Political Conventions" (Ralston and Wickline) abstract 1263 "Temporary Traps in Silicon and Ger- manium" (Haynes and Hornbeck) 1260 "Tensor Notation and the Sampling Cumulants of /^-Statistics" (E. L. Kaplan) abstract 770 "Ten-Stage Cold-Cathode Stepping Tube" (D. S. Peck) abstract 773 Terminal (s) connections, solderless, wrapped 523- 610 Tetragonal Crystal ND4D2PO4 properties 510 Tetrode (s) transistor, high-frequency 774 transistor, junction, for high-fre- quency use 517 "Theoretical Study of an Antenna-Re- flector Problem" (W. C. Jakes, Jr.) abstract 770 "Theory of Magnetic Effects on the Noise in a Germanium Filament" (Harry Suhl) 647-64 Theuerer, H. C. "Properties of Thermally Produced Acceptors in Germanium" abstract 256 Thomas, D. E. "Low-Drain Transistor Audio Oscil- lator" abstract 516 "Transistor Amplifier — Cutoff Fre- quency" abstract 516 Thomas, U. B. "Lead-Acid Stationary Batteries" abstract 257 "Three-Phase Power From Single-Phase Source" (A. L, Holcomb) abstract 256 "Threshold Law for Single Ionization of Atoms or Ions by Electrons" (G. H. Wannier) abstract 1512-13 "Throwdown Machine for Telephone Traffic Studies" (Frost, Keister and Ritchie) 292-359 Throwdown Tests call delays 360 Timber Preservation See Wood pres- ervatives INDEX 31 Tip and Ring Ground Test 636 Toll Messages billing, mechanized 1264 Toll Switching nationwide 1019 Toll Telephone Plant plans 513 Toluene as a diluent for creosote treating solu- tions 455 Townsend, J. R. "Dynamic Program for Conversion" 1512 ' 'What We Have Learned in 1952" 774 Townsend, Mark A. biographical material 1516 ''Cold Cathode Tubes for Transmis- sion of Audio Frequency Signals" 1371-91 Traffic See Telephone traffic "Traffic Engineering Design of Dial Telephone Exchanges" (J. A. Stew- art) 516 abstract 257 Transducer (s) wide-band coaxial-to-waveguide, opti- mum piston position 772 Transistor (s) amplifier 516 analog 515 audio-oscillator 516 circuit application noise 511 circuit cutoff frequency 516 circuits, negative-resistance 508 civilian economy 510 computer, digital, block diagrams 255, 508, 769 converters, negative-impedance 1510 counter, reversible binary 516 digital computer, block diagrams 255, 769 electronics 515 equations 1013 field effect type 515 filamentary, large current amplifica- tions 774 germanium, single-crystal 257 imperfections 515 junction: design theory 1271-1312 principles and development 516 properties 774 small signal ac transmission duirac- teristics of 1271 space-charge lay(M- widening 507 tetrode for high-freciueru-y use 517 negative-resistance circuits 508 noise behavior 511 oscillator, multifrequency pulsing cur- rent supply 1313-31 phototransistor See Phototransistor Pierce, J. R., article 1511 p-n-p-n, four terminal 508 point-contact: audio oscillator 516 electrode spacing effect on equiva- lent base resistance 516 frequency response control 515 negative-resistance properties 508 stability 515 serial adder 509 shift register 509 tetrode, high-frequency 774 unipolar 515 "Transistor Amplifier — Cutoff Fre- quency" (D. E. Thomas) abstract 516 "Transistor Electronics: Imperfections, Unipolar and Analog Transistors" (W. Shockley) abstract 515 "Transistor Equations" (F. R. Stansel) abstract 1013 "Transistor Negative-Impedance Con- verters" (J. G. Linvill) 1510 "Transistor Noise in Circuit Applica- tions" (H. C. Montgomery) abstract 511-12 "Transistor Oscillator for Use in Multi- frequency Pulsing Current Supply" (F. E. Blount) 1313-31 "Transistor Physics Simplified" (Gentile and Barotta) 1510 "Transistor Reversible Binary Counter" (R. L. Trent) abstract 516 "Transistor Shift Register and Serial Adder" (J. R. Harris) abstract 509 "Transistors" (J. R. Pierce) 1511 32 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 "Transistors in Our Civilian Economy" (J. W. McRae) abstract 510 "Transistors in Switching Circuits" (A. E. Anderson) abstract 506-7 Translator(s) card, for nationwide dialing 1037 Transmission audio frequency signals, cold cathode tubes for 1371-91 Clogston type conductors, laminated 695-713 coaxial system L3 See Carrier: L3 electric waves, around bends 511 intertoll telephone trunks, design 1019-36 radio, beyond horizon in the 40- to 4,000-Mcband 767-8 radio, mobile, comparison at 150, 450, 900 and 3700 Mc 1513 television, influence of echoes 1511 ^waveguide, low-loss 1011 "Transmission Design of Intertoll Tele- phone Trunks" (H. R. Huntley) 101&-36 "Transmission Properties of Laminated Clogston Type Conductors" (E. F. Vaage) 695-713 Transmitter (s) "Nonsynchronous Time Division with Holding and with Random Sampl- ing" (Pierce and Hopper) abstract 258, 513 single-sideband 514 sinusoidal radio wave 773 "Traveling-Wave Tube" (M. E. Hines) 1510 Traveling Wave Tube Power Ca- pacity 1509 Trent, R. L. "Transistor Reversible Binary Coun- ter" abstract 516 TrunkCs) intertoll, transmission design 1019-36 Tuffnell, T. L. biographical material 1018 "L3 Coaxial System — Application of Quality Control Requirements in the Manufacture of Components" 969- 1005 Turner, E. H. "New Non-Reciprocal Waveguide Me- dium Using Ferrites" 1512 "Type-0 Carrier Telephone" (E. K. Van Tassel) abstract 1259 "Typical Block Diagrams for a Trans- istor Digital Computer" (J. H. Felker) abstract 255, 769 U "Ultrasonically Induced Cavitation in Water: A Step-by-Step Process" (G. W. Willard) abstract 1513 ''Unipolar 'Field-Effect' Transistor" (W. Shockley) abstract 515 Vaage, E. F. biographical material 777 "Transmission Properties of Lami- nated Clogston Type Conductors" 695-713 Vacuum Tube See Electron tube "Vacuum Tube Electrometer Amplifier" (Hughes and Lander) 1261 Valdes, L. B. "Effect of Electrode Spacing on the Equivalent Base Resistance of Point-Contact Transistors" abstract 516 "Measurement of Minority Carrier Lifetime in Germanium" abstract 517 Van Horn, R. H. biographical material 777 "Solderless Wrapped Connections — Part III — Evaluation and Perform- ance Tests" 591-610 Van Roosbroeck, W. "Large Current Amplifications in Filamentary Transistors" 774 INDEX 33 "Properties of Thermally Produced Acceptors in Germanium" abstract 256 Van Siclen, H. E. "What We Did to Cut Costs of Finish- ing Telephone Woodwork" 1264 Van Tassel, E. K. "Type-0 Carrier Telephone" abstract 1259 'Vapor Pressure of He^ and He* Mix- tures" (R. A. Nelson) 772 Varistor(s) copper-oxide 514 germanium 514 Varney, R. N. "Drift Velocities of Ions in Krypton and Xenon" abstract 517 "Drift Velocity of Ions in Oxygen, Nitrogen, and Carbon Monoxide" abstract 1013 Viscometer rolling ball, calibration of 1010 Voice Control selective 771 Voice Frequexcy Signals switching 1371 Vowels (s) formants, and the formants of musical instruments 1261 formants, band width 1507-8 synthetic, auditory tests 1011 W Waddell, R. A. biographical material 1018 "L3 Coaxial Sj^stem — Application of Quality Control Requirements in the Manufacture of Components" 969- 1005 Walker, J. G. "Magnetic Crystal Anisotropy and Magnetostriction of Iron-Nickel Alloys" abstract 1008-9 Wallace, R. L., Jr. "High-Frequency Transistor Tetrode" abstract 774 "Junction Transistor Tetrode for High-Frequency Use" abstract 517 Wannier, Gregory H. biographical material 263 "Connection Formulas Between the Solutions of Mathieu's Equation" abstract 1014 "Motion of Gaseous Ions in Strong Electric Fields" 170-254 abstract 259 "Nature of Solids" abstract 517 "Spherical Model of a Ferromagnet" abstract 510 "Threshold Law for Single Ionization of Atoms or Ions by Electrons" abstract 1512-13 Warner, A. W. "High-Frequency Crystal Units for Primary Frequency Standards" abstract 260 Water cavitation, ultrasonically induced 1513 plating room waste disposal 1012 Wave See Electric wave Waveguide (s) directional coupler 510 multimode round, dominant wave transmission 256 non-reciprocal medium, using ferrites 1512 transmission, low-loss 1011 "We Begin a New Institute Year" (D. A. Quarles) abstract 514 Weathering creosote and creosoted wood 457 line insulation resistance 629 wood preservatives 150 Weighted Sample Values approximating the mode 1010 Weiss, M. T. "Magnetic Double Refraction at Microwave Frequencies" 517 Weiss, P. R. 34 THE BELL SYSTEM TECHNICAL JOURNAL, 1953 Weiss, P. R. — continued: "Exchange Narrowing in Paramag- netic Resonance" abstract 1257-8 "Western Electric's Service with Stand- ards" (K. B. Clarke) abstract 507 "What We Did to Cut Costs of Finishing Telephone Woodwork" (H. E. Van Siclen) 1264 "What We Have Learned in 1952" (J. R. Townsend) 774 Wick, R. F. "Hall Effect Modulators and 'Gyra- tors' Employing Magnetic Field Independent Orientations in Ger- manium" abstract 1010 Wickline, B.D. "Television Coverage of the National Political Conventions" abstract 1263 Wiener, F. M. "In Search of the Missing 6 Db" abstract 512 Wilkinson, Roger I. biographical material 521 "Working Curves for Delayed Ex- ponential Calls Served in Random Order" 360-83 Willard, G. W. "Ultrasonically Induced Cavitation in Water: A Step-by-Step Process" abstract 1513 Williams, H. J. "Magnetic Study of Low Temperature Transformation in Magnetite" 1264 Windeler, A. S. biographical material 777, 1270 "Polyethylene Insulated Telephone Cable" 1245-56 WireCs) concealment 770 connections, solderless, wrapped 523- 610 copper-steel, electroformed, develop- ment 1260 drop, electroformed conductor for 1099 polyethylene insulation 1245 Wood, E. A. "Polymorphism of ND4D2PO4" abstract 257 "Simple Attachment for Low Tem- perature Use of an X-Ray Diffrac- tion Camera" 1264 Wood Poles fiber stresses 1262 Wood Preservatives evaluation 120-69 Woodwork Finishing cost cutting 1264 Wooley, M. C. "Polyethylene Terephthalate — Its Use as a Capacitor Dielectric" abstract 1264-5 "Working Curves for Delaj^ed Exponen- tial Calls Served in Random Order" (Roger I. Wilkinson) 360-83 Wrapping connections 543-5, 551 Wright, E. E. "Stress Relaxation in Plastics and Insulating Materials" abstract 517-18 Wright, S. B. "Higher Frequencies for Ground-Air Communications" 1265 X-Ray Diffraction Camera attachment for low temperature use 1264 Xenon ions, drift velocity 517 Yaeger, R. E. "Optical Position Encoder and Digit Register" abstract 508-9 Young, W. R. "Comparison of Mobile Radio Trans- mission at 150, 450, 900 and 3700 Mc" 1513 Zone-Melting Principles 513 Printed in U.S.A. \