llBH.n Cflpr MtmEf 3781 -m ■»- ■ - o 6'- From the collection of the ^ m o PreTnger t a V Uibrary San Francisco, California 2008 ttri^- THE BELL SYSTEM TECHNICAL JOURNAL A JOURNAL DEVOTED TO THE SCIENTIFIC AND ENGINEERING ASPECTS OF ELECTRICAL COMMUNICATION EDITORIAL BOARD F. B. Jewett H. p. Charlesworth W. H. Harrison A. F. Dixon O. E. Buckley . O. B. Blackwell D. Levinger M. J. Kelly H. S. Osborne W. Wilson R. W. King, Editor J. O. Perrine, Associate Editor TABLE OF CONTENTS AND INDEX VOLUME XVII 1938 AMERICAN TELEPHONE AND TELEGRAPH COMPANY NEW YORK 3781 PRINTED IN U. S. A. THE BELL SYSTEM TECHNICAL JOURNAL VOLUME XVII, 1938 Table of Contents January, 1938 New Transmission Measuring Systems for Telephone Circuit Maintenance — F. H. Best 1 The Impedance Concept and its Application to Problems of Reflection, Refraction, Shielding and Power Absorption- — 5. A. Schelkunoff 17 On the Theory of Space Charge Between Parallel Plane Electrodes — C. E. Fay, A. L. Samuel and W. ShocUey 49 A Carrier Telephone System for Toll Cables — C. W. Green and E. I. Green 80 Cable Carrier Telephone Terminals — R. W. Chesnut, L. M. Ilgenfritz and A. Kenner 106 Crystal Channel Filters for the Cable Carrier System — C. E. Lane 125 Crosstalk and Noise Features of Cable Carrier Telephone System — M. A. Weaver, R. S. Tucker and P. S. Darnell 137 A New Single Channel Carrier Telephone System — H. J. Fisher, M. L. Almquist and R. H. Mills 162 April, 1938 Studies of Telephone Line Wire Spacing Problems — , /. A. Carr and F. V. Haskell 195 Variable Equalizers — H. W. Bode 229 An Explanation of the Common Battery Anti-sidetone Sub- scriber Set — C. 0. Gibbon 245 The Occurrence and Effect of Lockout Occasioned by Two Echo Suppressors — Arthur W. Horton, Jr 258 Characteristic Time Intervals in Telephonic Conversation — A. C. Norwine and 0. J. Murphy 281 Radioactivity — Artificial and Natural — Karl K. Darrow 292 3 BELL SYSTEM TECHNICAL JOURNAL July, 1938 Hertz, the Discover of Electric Waves — Julian Blanchard 327 Instruments for the New Telephone Sets — W. C. Jones 338 Transmission Features of the New Telephone Sets — A. H. Inglis 358 Spectrochemical Analysis in Communication Research — Beverly L. Clarke and A. E. Ruehle 381 High Speed Motion Picture Photography— PF. Herriott 393 An Optical Harmonic Analyzer — H. C. Montgomery 406 Magnetic Shielding of Transformers at Audio Frequencies — W. G. Gustafson 416 Coaxial Cable System for Television Transmission — M. E. Strieby 438 Stabilized Feedback Oscillators — G. H. Stevenson 458 The Discovery of Electron Waves — C. J. Davisson 475 October, 1938 Ultra-Short-Wave Transmission and Atmospheric Irregularities — C. R. Englund, A. B. Crawford and W. W. Mumford 489 Amplitude Range Control— 5. B. Wright 520 Devices for Controlling Amplitude Characteristics of Telephonic Signals — A. C. Norwine 539 The Exponential Transmission Line — Chas. R. Burrows 555 The Bridge Stabilized Oscillator — L. A. Meacham 574 Effect of Space Charge and Transit Time on the Shot Noise in Diodes—^. /. Rack 592 Fundamentals of Teletypewriters Used in the Bell System — E. F. Watson 620 The Dielectric Properties of Insulating Materials — E. J. Murphy and S. 0. Morgan 640 The Bell System Technical Journal Vol XVII January, 1938 No. 1 New Transmission Measuring Systems for Telephone Circuit Maintenance By F. H. BEST Transmission measurements on telephone circuits have long been recognized as essential aids in the furnishing of good service. Re- cent development work has produced new testing methods and apparatus which greatly simplify and expedite transmission meas- uring. This paper gives a brief description of the existing and the new arrangements. ABOUT two decades ago pioneering work was being done in the development and introduction of methods and apparatus for the measurement of the transmission characteristics of local and toll telephone circuits, it having been realized that with the increasing complexity of the telephone plant, these measurements were necessary to insure satisfactory transmission performance. Transmission meas- urements now have a well established place in maintenance activity and development work is constantly in progress to improve the testing apparatus so that it can be operated more rapidly and conveniently, will cost less or do more things, and also to improve testing methods. The principle employed in making transmission measurements has not changed. A standard testing power is supplied to one end of a telephone circuit and the power received at the other end is measured. The ratio between these powers, expressed in decibels, is a measure of the transmission loss or gain in the circuit. The magnitude of the testing powers is small, the sending power for measuring being one- thousandth of one watt and the received power ranging from one ten-thousandth to one-millionth of a watt. Sensitive meters must be used, or the power amplified before measuring it. Direct-current meters are more sensitive than alternating-current meters and it is the practice to convert the weak received alternating current to direct current by means of vacuum tube or copper-oxide rectifiers and employ direct-current meters for measuring it. Until recently the developments have been along the line of im- provements in the apparatus itself, there being practically no change 1 2 BELL SYSTEM TECHNICAL JOURNAL in the general methods and arrangements employed. For the large toll offices where much testing is done the measuring apparatus has been installed at one or more points in an office, the installations being known as transmission test boards. From these points trunks radiate to repeaters, test boards, switchboards, etc., and the circuits and equipment to be tested are connected to them by patching cords or switchboard cords. The testing has been done by a trained force of transmission testers, this force often being separate from that of the test board attendants whose work has consisted chiefly in correcting line faults, signaling troubles, etc., which do not require transmission measurements for their location. In the local plant, testing has been done almost entirely with portable transmission measuring apparatus in the hands of a trans- mission testing force who travel from office to office, testing the cord circuits, switching trunks, etc., at intervals of one or two years. The measuring apparatus has been costly and its permanent installa- tion in each office could not be justified. The apparatus used in both local and toll testing has been described in anumber of articles.^- ^'^• During the last few years new methods and apparatus have been developed and radical changes made that greatly improve the situa- tion both from an economic and operating standpoint. The new measuring instruments are much less costly than the types heretofore available and are as stable in operation and as simple to use as am- meters and voltmeters. In the local plant the decrease in cost and complexity and improvement in operating methods has justified per- manent installations in many of the larger offices while the newer forms of portable apparatus are being much more extensively distributed than earlier types. For the toll plant the improvement consists in doing away largely with the centralized transmission testing point. The regular test board force now can make transmission measure- ments at their test boards and the repeater attendants can measure the performance of the repeaters while working on them. The work is so simple and the maintenance forces are so well trained that a special transmission testing force is not required. This change has been brought about by the development of new instrumentalities and improved circuits, chief among which are more sensitive meters, the copper-oxide rectifier and the negative feedback amplifier.^ Until recently, the most sensitive meters which were suit- able for general use would not measure the weak power used in trans- mission measuring so that amplification of this power was required. Using the alloy steels now available, meters of much greater sensitivity and equal ruggedness have been manufactured and transmission losses NEW TRANSMISSION MEASURING SYSTEMS 3 of 20 db can be measured without an amplifier and without using abnormal testing power. Vacuum tubes formerly used to rectify the received alternating current testing power have been largely super- seded by the copper-oxide rectifier which is small, inert and requires no added power for its operation. The combination of this rectifier with sensitive meters greatly simplifies and reduces the cost of the measuring apparatus. Figure 1 shows the simplicity of the circuit of a transmission measuring set which will measure losses up to 20 db. While indicating meters have been used in transmission measuring for many years, it is only recently that they have been calibrated directly in db. This is the preferable way of measuring and its adop- tion has been delayed only because of the limitations of available apparatus and circuits which prevented a stable device from being developed. The earlier copper-oxide rectifiers were too unstable for CALIBRATING RESISTANCE WV DECIBEL METER Fig. 1 — Simplified circuit of transmission measuring set having a range of 20 db without an amplifier. measuring and until the development of the negative feedback ampli- fier, all vacuum tube amplifiers varied in amplification with changes in filament and plate voltage and the aging of tubes. The voltage of available power plants, while sufificiently stable for commercial use, caused the pointer of the meter in a transmission measuring set to fluctuate and over intervals of a few hours the change might be as much as one db. Smaller changes occurred frequently. These changes were slow enough so that measurements could be made, but they required frequent adjustment of amplifier gain to compensate. With such instability, the most rapid and accurate measurements can be made by using the comparison method of testing where a known "standard" power is attenuated by calibrated potentiometers or net- works until it equals the unknown received testing power. A meter in this case serves merely as a means for telling when the two are equal and the result is read from calibrated dials on the attenuator. A majority of the measuring sets now in the plant operate on this 4 BELL SYSTEM TECHNICAL JOURNAL "flip flop" principle. The use of these dials generally requires that the measuring instrument be built as a unit and that measurements be made at its location. The latter requirement often prevents its installation at the most desirable point because space is not available. For toll circuit testing in the larger offices where a considerable number of routine and trouble locating tests are made, the required frequency range and loss range are beyond the scope of a simple copper-oxide rectifier and meter. The ability to make level measure- ments is also a desirable feature. To meet these requirements, an amplifier is provided in the receiving circuit. By applying the nega- tive feedback principle to the amplifier and rectifier of the latest toll transmission measuring system a remarkably stable circuit has been obtained so that the meter can be calibrated directly in db and oper- iNPUT ^eoo^ Fig. 2 — Simplified circuit of reverse feedback amplifier rectifier used for toll circuit maintenance. ated for long periods without adjustment. This circuit, which is shown in simplified form in Fig. 2, consists of a high-impedance input transformer T bridged across a 600-ohm terminating resistance which may be removed when level measurements are made, two pentode tubes 1 and 2, a copper-oxide rectifier R and a meter M. This com- bination has much more amplification than is required, so the excess is used to improve stability by introducing part of the output voltage into the input circuit of the first tube in such phase relation with respect to the applied input voltage that the net input voltage is reduced. This reverse or negative feedback voltage is introduced into the grid circuit of the first tube through resistances A, B and C by connecting one of the rectifier terminals to the movable arm of the potentiometer P. These resistances, and resistance D form the cath- ode drop resistance of the grid circuit and any potential applied across them aff"ects the potential on the grid of the tube. In the position shown, the reverse feedback is a maximum and the net amplification NEW TRANSMISSION MEASURING SYSTEMS 5 of the amplifier is a minimum. Moving the potentiometer arm to the lowest step gives less feedback and the amplifier net gain is greater. The principle of stabilization is as follows: If a constant potential is applied to the input terminals of the amplifier a constant potential will also be applied to the grid of the first tube. As long as the tubes or the rectifier do not change in characteristics the output of the rectifier will likewise be constant. Now if through any cause, such as a change in tube characteristics or in the rectifier, the gain of the amplifier-rectifier should increase, the output will increase and the neutralizing voltage fed back into the input circuit will also increase. This will reduce the voltage on the grid so that the net output voltage of the amplifier-rectifier will not be changed noticeably. Conversely, if a change in the tubes or the rectifier should cause a reduction in gain, the voltage fed back into the input circuit will decrease, there will be less feedback voltage and the output voltage will be substan- tially the same as before. The reverse feedback amplifier-rectifier is so stable that once it is adjusted to have the proper characteristics it will remain constant for long periods. The meters used with this amplifier-rectifier have a range of 15 db. This is less than the required measuring range so that it is necessary to increase it by changing the amplifier gain in steps of 10 db. The reverse feedback amplifier lends itself readily to this as the variation of a single resistance in the feedback circuit is sufficient and no ex- pensive balanced attenuators are required. In Fig. 2 this is done by potentiometer P. In practice this resistance is controlled by relays which form a part of the amplifier, these relays in turn being remotely controlled by keys, jacks or dials at various points in the office. The db meters can be located where desired without reference to the location of the amplifier-rectifier. They may also be placed in lantern slide projectors which throw a greatly enlarged meter scale on a screen so that it can be read from distances up to 50 feet or more. The new arrangements are extremely flexible, one set of equipment supplying measuring facilities for several different points in an oflfice where previously several sets would be necessary. Where the use of the equipment is intermittent more than one meter may be used with a single amplifier-rectifier. The meters from which the results are read may be of the conventional indicating type which can be mounted on keyshelfs or on vertical panels, they may be of the projector type or they may be of the recording type which records on paper the characteristics which are being measured. All meters are interchange- able, being similar electrically. These new instrumentalities have removed many of the limitations 6 BELL SYSTEM TECHNICAL JOURNAL to the design of measuring systems and the recently developed arrange- ments are therefore better coordinated with other facilities than those which they replace. A few examples will be given to illustrate the recent advances in transmission measuring work. Fig. 3 — Transmission measuring set developed about 1920 and associated genera- tor as used at a switchboard in testing cord circuits or interoffice trunks. Both units are required for either sending or receiving. Figure 3 shows the testing power generator and measuring set developed about 1920 for use in the local plant to measure central office equipment and interoffice trunks with testing power of a single frequency, the frequency employed usually being 1000 cycles. When NEW TRANSMISSION MEASURING SYSTEMS 7 measuring circuits between ofifices, it is necessary either to have dupli- cate sets of equipment at the two ends of the circuit or to connect two circuits together at one end, testing them at the other end as one circuit. This latter arrangement, while not wholly satisfactory, has been the one generally employed because of the greater expense of the Fig. 4 — Latest type of transmission measuring set as used at a switchboard. The associated generator shown in Fig. 5 is usually permanently mounted in the office and testing power is obtained through the multiple as shown. For one-way measure- ments on circuits between offices, the generator is connected to one end of the circuit and the transmission measuring set at the other, both units not being required at each end. other method which has involved not only two sets of expensive testing apparatus but two testers, these being necessary because the testing apparatus required frequent adjustment. If circuits between a group of offices in a city are to be tested between circuit terminals rather than by the looping back method, considerable time is con- sumed in traveling between offices when using this apparatus. 8 BELL SYSTEM TECHNICAL JOURNAL The new apparatus shown in Figs. 4 and 5 not only costs but one- tenth as much and weighs one-quarter as much as the older apparatus but also has electrical advantages which enable a new and improved measuring technique to be employed. The generator is a magneto inductor alternator driven by a 50-cycle or 60-cycle induction motor and gives a constant output without attention. The output is ad- justed at the factory or on installation. Some of these machines have Fig. 5 — One thousand-cycle magneto generator used generally for transmission testing. Cover plate removed to show the generator construction. Overall length 7 inches. been running continuously for about a year without showing any appreciable change in output. They can, therefore, be mounted permanently in an office and the output terminals wired to convenient testing points so that the generator need not be carried around. Because of this output stability, it is not only unnecessary to have an experienced tester at the distant end of the circuit but in the larger offices auxiliary switching equipment is arranged so that testing power can be supplied automatically to one end of the circuit by direction of the tester at the other end, who simply calls or dials a designated NEW TRANSMISSION MEASURING SYSTEMS 9 number over the circuit to be testedJ The testing power is cut off at the sending end when the connection is broken at the receiving end. Tests can be made in the same manner from private branch exchanges and subscribers' stations without a charge being registered against the subscriber. Fig. 6 — Small receiving set as used to measure the transmission loss of a sub- scriber's line. The 1000-cycle generator has been connected to the central office end of the line by calling a designated number. When the telephone instrument has been replaced on the mounting, the meter in the measuring set will read the line loss. The receiving set shown in Fig. 4 is based on the electrical circuit of Fig. 1. It has a transmission range of 20 db, and is provided with all the jacks and facilities required for testing cord circuits, trunks and central office equipment. A still smaller and less expensive receiving set of a similar type having a 10 db range is shown in Fig. 6. It is extremely portable and light and is useful for work where the limited 10 BELL SYSTEM TECHNICAL JOURNAL range is sufficient. Jacks and other testing conveniences have been omitted to save bulk and cost. Fig. 7 — Transmission testboard for maintaining long distance circuits. The new instruments for the local plant are not limited to local testing but also have extensive application in the toll plant especially in the smaller offices which do not have many toll circuits. Because NEW TRANSMISSION MEASURING SYSTEMS 11 of its excellent performance the 1000-cycle generator shown in Fig. 5 will be used generally in the toll plant for 1000-cycle testing. Figure 7 shows several positions of the transmission test board now widely used for measuring toll telephone circuits. Introduced in 1926, it preceded the development of the reverse feedback amplifier-rectifier and the transmission measuring set used in this board is therefore of the comparison type in which the results are read fron calibrated dials. Testing power is provided by a variable frequency oscillator from which frequencies in the voice range can be obtained. The size of this equipment precluded its being installed in test boards which contain the circuit terminals so that trunks between the measuring 1 ^ X ^»- ^ Fig. 8 — New amplifier rectifier for general use in toll transmission maintenance. equipment and these terminals are necessary. In large offices a num- ber of these transmission test boards have been provided. Figure 8 shows the new reverse feedback amplifier-rectifier used with the latest toll transmission measuring system. This simple panel, which is about one-fourth the size of the measuring set employed in the test board shown in Fig. 7, contains everything required at the receiving end of a circuit excepting the meter and the keys for changing the measuring range. With its associated meter and keys, it is less than one-half as expensive as corresponding elements in the trans- mission testboards. It can be used with the variable frequency oscillator shown in Fig. 7 or with the 1000-cycle machine shown in Fig. 5. The meters used with this amplifier-rectifier have a specially de- signed magnetic circuit which gives an evenly spaced scale on the 12 BELL SYSTEM TECHNICAL JOURNAL meter.^ A conventional meter would have large db divisions at one end of the scale and small ones at the other. When a direct reading transmission measuring set is used to make measurements over a wide frequency range, it is essential that the response be the same at all frequencies. Variation of response of the new feedback amplifier-rectifier with frequency is so small that it can be calibrated with 1000-cycle current and measurements made over a wide frequency range without recalibration. From the standpoint of efftciency, the best place for making trans- Fig. 9 — Projection meter as used for measuring transmission at a secondary testboard. The meter is being read by the third man from the right. mission measurements on complete toll circuits is at the test board to which all troubles are reported by the operator and where overall signaling and other operating tests are made. Practically all of the space in this board is taken up by jacks on which the circuits terminate so that the older types of measuring set could not be provided at this point. The new arrangement is ideal for application to this type of board since all that is required in the board are the meter and a few jacks or keys for controlling its range. This type of measuring device can be applied equally well to new and existing boards of various types. This feature is of particular value as the maximum efftciency of high- speed measuring systems can be obtained only when all test boards NEW TRANSMISSION MEASURING SYSTEMS 13 are equipped with them. Two arrangements are available for test board use. One employs a conventional type of meter mounted in the keyshelf or on a panel and the other employs the projection type of meter which is illustrated in Fig. 9.^ The method of operation is simple. When a transmission test is to be made the tester listens until he hears the tone caused by testing power coming over the circuit from the distant generator, then connects the circuit to a jack in which the measuring set input terminates. The meter indicates immediately the net loss of the circuit at the testing frequency. With the projec- tion meter arrangement the lamp in the projector is turned on automat- ically when a connection is made to the test jack. The new amplifier-rectifier will measure transmission losses and Fig. 10 — Noise measuring set. gains and also transmission level. For the latter type of measurement the input impedance of the amplifier is raised to a high value so that it is, in effect, a voltmeter. This change in impedance can be made from a remote point, a relay for making the change being a part of the amplifier. In addition to the measurement of transmission losses, gains and levels on telephone circuits, it is also necessary at times to make measurements of noise on the circuit. This noise may be caused by currents induced by power systems or from power plants in the telephone offices or it may be in the form of crosstalk, sometimes un- intelligible, from other telephone circuits. Noise measurements are now made with meter indicating devices, the latest type of self- contained portable noise measuring set being shown in Fig. 10.^ Where enough noise measurements are made to justify a permanent installation, an arrangement similar to that described for transmission 14 BELL SYSTEM TECHNICAL JOURNAL measurements can be used. While a different amplifier-rectifier is required for noise measurements the same meters and methods of control can be employed as for transmission measurements. All of the methods so far described are manually operated in so far as the recording of the results is concerned. There are occasions when a fully automatic recording device is desirable.'* Such cases are the making of transmission versus frequency runs on repeaters or circuits where measurements are desired over a wide range of frequencies. Another class of measurements are those in which the single-frequency transmission loss of a circuit is to be determined over a long period of time to obtain a measure of the stability of the circuit. For this purpose the new method of measuring transmission is well adapted. With a stable amplifier-rectifier having practically no frequency dis- tortion, the indicating meter may be replaced by a recording meter which will record continuously the received power expressed in db. A fully automatic recording system is shown in Fig. 11. With this arrangement an oscillator at one end of the line supplies testing power to the line, the frequency of the power being changed continuously by a synchronous motor. At the receiving end the recording meter, also operated by a synchronous motor, plots the received power. A complete transmission frequency run on a message telephone circuit can be made in less than one minute. When records are to be made of transmission vs. time, the oscillator frequency at the sending end is fixed and the meter plots the transmission loss at that frequency. The above description has been limited to the types of measure- ments commonly made on complete circuits or parts of circuits. In addition to these there are a number of types of measuring apparatus which are used in connection with the installation of new equipment, changes in installations and the detailed running down of trouble. Transmission measurements have proved to be of great value in maintaining satisfactory transmission performance of telephone cir- cuits. Periodic routine tests avoid service impairment by detecting troubles. Troubles which are thus detected or which cause service complaint are located readily. Another large field of use is in the adjustment of repeaters and complete circuits to prescribed trans- mission characteristics. The importance of this work to the Bell System is indicated by the fact that there are nearly 1000 of the portable transmission measuring equipments and 1500 transmission test boards now in use, with which several million measurements are made annually. The new measuring systems enable this work to be done more rapidly than in the past, and the reduced cost of the equipment is resulting in its greater distribution. NEW TRANSMISSION MEASURING SYSTEMS 15 Fig. 11— Automatic recording transmission measunng system. 16 BELL SYSTEM TECHNICAL JOURNAL References 1. "Measuring Methods for Maintaining the Transmission Efficiency of Telephone Circuits," F. H. Best, Elec. Engg., vol. 43, February 1924, pp. 136-144. 2. "Electrical Tests and Their Applications in the Maintenance of Telephone Transmission," W. H. Harden, Bell System Technical Journal, vol. 3, July 1924, pp. 353-392. 3. "Practices in Telephone Transmission Maintenance Work," W. H. Harden, Elec. Engg., vol. 43, December 1924, pp. 1124-1128. 4. "A Recording Transmission Measuring System for Telephone Circuit Testing," F. H. Best, Bell System Technical Journal, vol. 12, January 1933, pp. 22-34. 5. "Stabilized Feedback Amplifiers," H. S. Black, Bell System Technical Journal, vol. 13, January 1934, pp. 1-18. 6. "Projecting Circuit Performance on a Screen," F. E. Fairchild, Bell Laboratories Record, vol. 13, July 1935, pp. 328-331. 7. "Improved Transmission Measuring System," F. H. Best, Bell Laboratories Record, vol. 14, March 1936, pp. 237-239. 8. "Decibel Meters," F. H. Best, Bell Laboratories Record, vol. 15, January 1937, pp. 167-169. 9. "A New Noise Meter," J. M. Barstow, Bell Laboratories Record, vol. 15, April 1937, pp. 252-256. The Impedance Concept and Its Application to Problems of Reflection, Refraction, Shielding and Power Absorption By S. A. SCHELKUNOFF This paper calls attention to the practical value of a more ex- tended use of the impedance concept. It brings out a certain underlying unity in what otherwise appear diverse physical phe- nomena. Although an attempt has been made to trace the history of the concept of "impedance" and many interesting early sug- gestions have been found, reference to these lies beyond the scope of this paper. Apparently, Sir Oliver Lodge was the first to use the word "impedance," but the concept has been developed grad- ually as circumstances demanded through the efforts of countless workers. The main body of the paper is divided into three parts: Part I, dealing with the exposition of the impedance idea as applied to different types of physical phenomena; Part II, in which the general formulae are deduced for reflection and transmission co- efficients; Part III, presenting some special applications illustrating the practical utility of the foregoing manner of thought. THE term "impedance" has had an interesting history, in which one generalization has suggested another with remarkable rapid- ity. Introduced by Oliver Lodge,^ it meant the ratio V/I in the special circuit comprised of a resistance and an inductance, / and V being the amplitudes of an alternating current and the driving force which produced it. This was soon extended to the somewhat more general circuit consisting of a resistance, an inductance coil and a condenser.- The usage did not develop much further until the use of 1 Dr. Oliver Lodge, F.R.S., "On Lightning, Lightning Conductors, and Lightning Protectors," Electrical Review, May 3, 1889, p. 518 2 It is interesting to note that the first impulse was to introduce a new word rather than to extend the meaning of the old term. Thus in 1892, F. Bedell and A. Crehore write as follows: "From the analogy of this equation to Ohm's law, we see that the expression -» /i?2 _|_ (_ j^^ j is of the nature of a resistance, and is the apparent resistance of a circuit containing resistance, self-inductance and capacity. This expression would quite properly be called 'impedance' but the term impedance has for several years been used as a name for the expression Vi?^ + XW, which is the apparent resistance of a circuit containing resistance and self-inductance only. We would suggest, therefore, that the word 'impediment' be adopted as a name for the expression ^/i?2 -\- (— Leo] which is the apparent resistance of a circuit containing resistance, self-induction and capacity, and the term impedance be retained in the more limited meaning it has come to have, that is Vi?^ + LW, the 17 18 BELL SYSTEM TECHNICAL JOURNAL complex quantities, which had begun early in the nineteenth century among mathematicians, was popularized among engineers by Kennelly and Steinmetz. Then the proportionality relation V = ZI, which had previously been true only if V and / were interpreted as amplitudes, acquired a more general significance, for it was found that this relation could express the phase relationship as well, provided Z was given a suitable complex value. An important generalization came when the close similarity of the laws connecting V and / in an electric circuit to those governing force and velocity in mechanical systems suggested that the ratio "force/ velocity" be called a "mechanical impedance." This usage is now well nigh universal. The next step was a short one : it amounted to extending the term to include also the ratio "force per unit area/flow per unit area" ; that is, "pressure/flux." This usage is well known in such fields as acous- tics, but it has not penetrated as far into the electrical field as con- venience seems to warrant. If we read these remarks with a view to appraising the direction in which future growth might be expected, we are immediately impressed by the strong trend toward interpreting the ratio "force/velocity" in an ever widening sense. It is my purpose in the present paper to indicate some further extensions which I have found to be useful. They are founded upon five basic ideas. The first is to recognize and use whenever possible analogies between dynamical fields in which the impedance concept is common and others (heat, for instance) in which it is not. The second is the idea of extending the F// relation from circuits to radiation fields, in much the same way that the "force/velocity" concept has been made to embrace "pressure/flux" in hydrodynamics. The third is, to regard the impedance as an attri- bute of the field as well as of the body or the medium which supports the field, so that the impedance to a plane wave is not the same as the impedance to a cylindrical wave, even when both are propagated in infinite "free space." The fourth basic idea is that of assigning direction to the impedances of fields. This does'not mean, however, that the impedances are vectors; in fact, they are not, since they fail to obey the laws of addition and the laws of transformation peculiar to vectors. And finally the fifth is a generalization of the idea of a one-dimensional transmission line or simply a transmission line. While apparent resistance of a circuit containing resistance and self-induction only." Frederick Bedell and Albert C. Crehore, "Derivation and Discussion of the General Solution of the Current Flowing in a Circuit Containing Resistance, Self-induction and Capacity, With Any Impressed Electromotive Force," Journal A. I. E. E., Vol. IX, 1892, pp. 303-374, see p. 340. IMPEDANCE CONCEPT AND APPLICATION 19 all physical phenomena are essentially three-dimensional, frequently all but one are irrelevant and can be ignored or are relatively unimpor- tant and can be neglected. In the mathematical language, this means that only one coordinate (distance, angle, etc.) is retained ex- plicitly in the equations of transmission. The paper is divided into three parts. Part I discusses broadly the ratios to which the term "impedance" can appropriately be applied in a wide variety of physical fields, ranging from electric circuits and heat conduction to electromagnetic radiation. In this part the con- cept is gradually broadened until at the end it has acquired the prop- erty of direction mentioned above. Parts II and III consider the general laws governing reflection, refraction, shielding and power absorption, and rephrase them as theorems regarding the generalized impedances. To make the illustrations more effective, familiar ex- amples are chosen. PART I THE IMPEDANCE CONCEPT Electric Circuits In an electric circuit comprised of a resistance R and an inductance L, the instantaneous voltage-current relation is described by the fol- lowing differential equation L^ + RIo=Vo, (1) where Vo is the applied electromotive force. If Vo varies harmonically with frequency /, ultimately /o will also vary harmonically with fre- quency/. What happens is that the solution of (1) consists of two parts, the transient part and the steady state part, the former decreasing exponentially with time and the latter being periodic. The steady state solution of (1), or indeed of the most general linear differential equation with constant coefficients, can be found by means of a simple mathematical device based upon the use of complex num- bers. Thus if Fo and /o vary harmonically, they may be regarded as real parts of the corresponding complex expressions Ve'"^ and /e*"', where/ = co/lir is the frequency. The quantities V and / are complex numbers whose moduli represent the amplitudes and whose phases are the initial phases (at the instant / = 0) of the electromotive force and the electric current. The time rate of change of /o is then the real part of the derivative of /e'"', that is, the real part of ioiIe'^K If we form another equation after the pattern of (1), replacing /o and Vq by the imaginary part of /e'"' and Fe™', and add the new 20 BELL SYSTEM TECHNICAL JOURNAL equation to (1), we shall have L ^ J^ ' + i?(/e-0 = Foe'"'. Differentiating and cancelling the time factor e'"', we obtain {R + icoL)/ = F. The ratio Z = V/I = Fe*"'/-^^""' is called the impedance of the elec- tric circuit. In the present instance Z = R + io)L. In general, the impedance Z = R -\- iX has a real and an imaginary part, the former being the resistive component of the impedance and the latter the reactive. Mechanical Circuits Linear oscillations of a mass in a resisting medium are described by equations identical with (1) and (2) except for the customary differ- ence in lettering d(ve'"^) , , . ,^ ^ . , m~j — - + r{ve"^^) = Fe'^K In this equation, v represents the velocity and F the applied force, m the mass and r the resistance coefficient. The mechanical im- pedance is then Z = r -\- icom. «^ Similarly, for torsional vibrations the impedance is defined as the ratio "torque/angular velocity." Electric Waves in Transmission Lines Let X be the distance coordinate specifying a typical section of an electric transmission line. Let the complex quantities F and / be the voltage across and the electric current in the transmission line.^ Then the space rate of change of the voltage is proportional to the current and the space rate of change of the current is proportional to the voltage ^=-Z/. g=-FF. (2) 3 The time factor e'"' is usually implicit. IMPEDANCE CONCEPT AND APPLICATION 21 The coefficients of proportionality Z and Y are known as the dis- tributed series impedance and the distributed shunt admittance of the Hne ; they depend upon the distributed series resistance R, shunt con- ductance G, series inductance L and shunt capacity C in the following manner: Z ^ R + iooL, Y = G + io^C. (3) In a generalized transmission line Z and Y may be functions of x and may depend upon co in a more complicated manner than that suggested in (3). If Z and Y are independent of x, (2) possesses two exponential solutions : where T = a + ip = 4ZY, ^0 = Vf "^ ^ " T ' (4) It is customary to designate by V that value of the square root which is in the first quadrant of the complex plane or on its boundaries ; the other value of the square root is — F. The two "secondary" constants T and Zo are called, respectively, the propagation constant and the characteristic impedance. The real part oc of the propagation constant is the attenuation constant and j3 is the phase constant. Equations (4) represent progressive waves because an observer moving along the line with a certain finite velocity beholds an un- changing phase of V and /. This velocity c is called the phase velocity of the wave. Setting x = ct m the upper pair of (4), we obtain the condition for the stationary phase - /3c -f CO = 0, c = ^ . Hence, F+ and /+ represent a wave traveling in the positive x-direc- tion. Similarly we find that V" and /~ represent a wave traveling in the opposite direction. Consider two points in which the phases of V and / differ by lir when observed at the same instant; the distance X between these points is called the wave-length. By definition ^j8X = 27r, ^ =^- 22 BELL SYSTEM TECHNICAL JOURNAL If the transmission line is non-uniform, that is, if Z and Y are func- tions of X, then the solutions of (2) are usually more complicated. In any case, however, there are two linearly independent solutions I'^{x) and I^{x) in terms of which the most general solution can always be expressed I{x) = AI+{x) + BI-{x). These independent solutions may represent either progressive waves in two opposite directions or certain convenient combinations of such waves. The corresponding F-functions are found by differentiation from (2) ; thus ^^ Y dx ' ^ ^^ Y dx • The impedance of the F+, /+-wave is then ^o-(x) =^^= -——= --- (log I-). Similarly the impedance of the V~, /"-wave is 7 ~( \ V~{x) \ dl- \ d . . The negative sign in (5) is merely a matter of convention: the "posi- tive" and the "negative" directions of the transmission line are so defined that the real parts of Zo+ and Zq~ are positive. In general, Zo+ and Z^r are not equal to each other. Moreover, there is a considerable amount of arbitrariness in our choice of the basic solutions /+ and /~. Thus, we are brought face to face with the fact that we must regard the impedance as an attribute of the wave as well as of the transmission line. This point of view will become even more prominent when we come to deal with the wave transmission in three-dimensional media. There even progressive waves may have different characters (they may be plane, cylindrical, spherical, etc.) and the impedances of the same medium to these waves will be different. And naturally, it goes without saying that the impedances to like waves in different media may also be different. One could, perhaps, take the position that geometrically similar waves in different media are not really alike if the corresponding "force/velocity" ratios are not equal and that under all circumstances the "impedance" is the property of a wave. However, "intrinsic im- IMPEDANCE CONCEPT AND APPLICATION 23 pedance" will be used to designate a constant of the medium without reference to any particular wave. Vibrating Strings In strings under constant tension r, simply periodic waves may be described by the following two equations : dF / , • N dv iw ^ -T— = — [r -\- tcomjv, -T- = r, ax ax T where m is the mass and r the resistance per unit length of the string. The variable F represents the force on a typical point of the string at right angles to the string and v is the velocity at that point. Hence the characteristic impedance and the propagation constant are given by Zo = J- -. , r = y^ (r -]- lo^m) — . In the non-dissipative case we have simply -7 I T • /^ Zo = ymr, 1 = tu) \ — . Heat Waves Transmission of heat waves is also a special case of the generalized transmission line theory. In the one-dimensional case we have dT _ V dv _ _ dT 'dx~ ~K' dx~ ~ ^ 'dt' where : T is the temperature, v the rate of heat flow, K the thermal conductivity, 8 the density and c the specific heat. For simply peri- odic waves, we obtain dT 1 dv . . „ ■ -— =-— y, — =- to}c8 I . dx K dx Thus the characteristic impedance and the propagation constant of heat waves are 1 _ \i(acb Zo = The ratio "the temperature of the source/the rate of heat flow from the source" is the impedance "seen" by the heat source. 24 BELL SYSTEM TECHNICAL JOURNAL Electromagnetic Waves The transmission equations of uniform linearly polarized * plane waves are : dE . ,_ dH , , • M7 d^= - ^"^^' -5^ = - ^^ + '"^^^' where : E is the electric intensity, H the magnetic intensity, and g, e, ju are, respectively, the conductivity, the dielectric constant and per- meability of the medium. These equations are of the same form as (2). Even the physical meanings of E and H are closely related to those of V and /; thus £ is F per unit length and H is I per unit length. The propagation constant and the characteristic impedance of an unbounded medium to linearly polarized plane waves are : a = ^io}fjL{g + iwe), 'J = V iCO/U (7 tCOjl g + io}€ g + ico€ a These constants are so directly related to the fundamental electro- magnetic constants of the medium that they themselves may be re- garded as fundamental constants. On this account, we call a and rj, respectively, the intrinsic propagation constant and the intrinsic im- pedance of the medium. The intrinsic impedance will frequently occur as a multiplier in the expressions for the impedances of various types of waves. The intrinsic impedance of a non-dissipative medium is simply 7/ — 4yiT^', in air, this is equal to 1207r or approximately 377 ohms.* Thus in the uniform linearly polarized plane wave traveling in free space, the relation between E and H is E = UOttH or E = 377H, provided the positive directions of E and // are properly chosen. An electromagnetic field of general character can be described by means of three electric components E^, Ey, E^, and three magnetic components H^, Hy, Hz. We can form the following matrix whose components can be regarded as impedances : ^ In this connection the word "uniform" is used to mean that equiphase planes are also equi-amplitude planes. * See the letter from G. A. Campbell to Dean Harold Pender reproduced at the end of this paper. IMPEDANCE CONCEPT AND APPLICATION 25 E, E. E. 11/ Hy' H, Ey Ey E, H/ Hy' H. E, E, E. H/ H' H. The algebraic signs preceding the ratios of components with different subscripts are assigned as follows. If a right-hand screw is rotated through 90° from the positive axis indicated by the subscript in the numerator toward the positive axis indicated by the subscript in the denominator, it will advance either in the positive or in the negative direction of the remaining axis. In the former case the ratio is given the positive sign and in the latter the negative sign. This convention happens to be particularly convenient in expressions for the Poynting vector. Thus two impedances are associated with any pair of perpendicular directions, the A;-axis and the j-axis, let us say ; these impedances are : H' If these two impedances are equal, then we define the impedance in the direction of the positive z-axis as follows : 7 = r^ = — H. Similar definitions hold for the impedances in other directions. While the impedances as now defined possess an attribute of direc- tion, they are neither vectors nor tensors because they do not add in the proper fashion. However, in practical applications this lack of vectorial properties does not seem to be a drawback. The above definitions can be extended to other systems of coordi- nates. Let r be the distance of a point P {r, 6, Ha (6) provided the two ratios of the field components are equal. The radial impedance looking toward the origin is defined as the negative of (6). Similarly the "meridian" impedance in the direction of increasing d 26 BELL SYSTEM TECHNICAL JOURNAL Fig. 1 — Spherical coordinates. The positive directions of r-, 9-, and (^-components of a vector are, respectively, the directions of increasing r, 6, and t^ \^^) The relative distribution of the amplitude and the phase of the wave are governed by the factor e-^y ^'° '' and this phase-amplitude pattern is propagated in the direction of the z-axis, the propagation constant being a cos ??. The advantages of this point of view are clear. In attempting to find the reaction of the second medium upon the incident wave, it is necessary to satisfy certain boundary conditions at every point of the interface. This can be insured by requiring the reflected and the refracted waves to have the same phase-amplitude patterns at the interface and by adjusting their relative amplitude and phases to secure the fulfilment of the boundary conditions at some one point. In other words, the problem is reduced to that for which the general solution was given in Part II. The impedance to the incident wave in the z-direction is found from (22) : ^ Ex Eo _ Z^ = jj- = — = 77 sec I?. Hy Ho cos ?? This impedance is seen to be a function of the intrinsic impedance of the medium and of the angle of incidence. For the refracted wave in the second medium the transmission equations are similar to (22) : EJ = {Eo'e~'''y ^'" '^)e~'''' ""^ i+io>t^ Hy' = {Ho' cos lA e-''^ ^'" ^)e-'''' '=°' ^+-', Eo' = v'Hq. *^^^'* The "angle of refraction" \J/ is, in general, different from ??. In our equations we may regard xp merely as a parameter. Its value is obtained from the condition that at the x3'-plane the phase-amplitude pattern of the incident and the refracted waves must be the same, and consequently a sin§ — a' sin \p. (24) In dielectrics this relation is known as Snell's law of refraction. By (23), the impedance to the refracted wave in the z-direction is F ' Zt ■'-'X t I z =777 = ^ sec lA- 44 BELL SYSTEM TECHNICAL JOURNAL The reflection and the transmission coefificients are then obtained from (22) in terms of the impedance ratio , r? sec ?? 7} cos ^l/ ,-_. f] sec ^ t] cos ?? Thus, we have H — -■, — ; — r , 1 E — These coefficients refer to the tangential components of the field. In a similar way we can deal with the case in which the magnetic vector of the incident wave is parallel to the boundary. The parts played by E and // are interchanged and the impedance ratio becomes k=^l^. (26) 7] COS yf/ The cosine factors have changed their places. The general case, in which neither E nor II is parallel to the bound- ary, cannot be treated in the above manner. In this case the com- ponents of E and // which are parallel to the boundary are not per- pendicular to each other, the impedances Z^y and Zyx are not equal to each other and the unique impedance Z^ — Z^y = Zyx, upon which the results of Part II are based, does not exist. In accordance with a suggestion made in Part I, the incident wave must be resolved into components possessing unique impedances in the direction normal to the boundary. It is well known that such a decomposition is possible for ordinary plane waves; the latter can always be decomposed into two components, in one of which E is parallel to the boundary and in the other H is so disposed. It is not surprising that reflection of arbitrarily oriented waves cannot be treated directly. The impedance ratios (25) and (26) for two basic orientations are in general different and the polarization of the reflected wave will be changed. An exceptional case arises when the intrinsic propagation constants of the media are equal. In this case \p = ^, as seen from (24), and the impedance ratio is independent of the angle of incidence and of the particular orientation of the wave. Con- sequently, the reflection and the transmission coefficients depend solely upon the ratio of the intrinsic impedances of the media. Frequently the permeabilities of the media are assumed to be the same, in which case the ratio of the intrinsic impedances is equal to IMPEDANCE CONCEPT AND APPLICATION 45 the inverse ratio of the "indices of refraction" of the media. Much could be said, however, in favor of not making such an assumption when formulating the general results since in many applications the permeabilities may be unequal. Images A few additional interesting results can be obtained for the special case of two semi-infinite homogeneous media having equal propaga- tion constants. If the media are separated by a plane boundary, problems of reflection and refraction can be solved by the method of images. This method is frequently used in electrostatics and one or *S t-S Fig. 6 two simple examples from that science will serve as an introduction to the later generalizations. The field of a point charge g above a conducting plane can be found by assuming another point charge (— q). This "image" charge (Fig. 6) is the same distance below the plane as the actual charge is above the plane, both charges lying on the same perpendicular. The field due to the original charge and to the image charge satisfies the boundary conditions at the conducting plane since it makes the latter an equipotential. This combined field gives the correct resultant field on the same side of the plane as the original charge; on the opposite side the field is zero. If the boundary is the interface between two perfect dielectrics (Fig. 7) with dielectric constants respectively equal to €i and €2, the results are almost equally simple. Above the boundary we have 46 BELL SYSTEM TECHNICAL JOURNAL a reflected field in addition to the original field. This reflected field is produced by an image charge g' = (ei — e2)g/(€i + €2) on the sup- position that the dielectric constant is everywhere equal to ei. Be- low the plane the field is such as would be produced by a charge q" = 2eiql(€i + €2) if placed where the original charge is, also on the assurription that the dielectric constant is everywhere ei. The charge producing the correct field below the boundary would be q'" = 2eiql{f:i + ei) if we were to assume 62 as the dielectric constant of the whole space. Inspecting equations (7) for an electric current element, which we assume to be perpendicular to the plane interface of two homogeneous ♦'I ♦ q'^ Fig. 7 6 - e 1 2 media, we see that the method of images can readily be extended to dynamic fields provided the intrinsic propagation constants of the media are equal. In order to make this conclusion more evident, we replace ico/x in the first equation by the equivalent product 770- and then calculate the component of E tangential to the interface E+ = Ee+ cos d -]- Er+ sine = rjll o or solving for 5o So = 1.527 X 10~*-n7r — 7-1/2 (centimeters). (2) Accordingly, whenever the conditions at the first plane are represented V=0 Vi tpV] I SPACE CHARGE \ LIMITED CURRENTJ -So- OSq I FIRST SECOND PLANE PLANE Fig. 1 — Hypothetical conditions to assist in visualizing the unit of distance So. by / and Vi, distance from the first plane may be measured in units of ^o. The distance in centimeters is then 5 = aSo. (3) Similarly a natural unit of potential is Vi and the potential of any plane at a distance S is then given by V=^V,. (4) Potential Distributions All possible potential distributions to the right of the first plane can now be expressed in terms of ^ as a function of a. The mathe- matical derivation is straightforward and is given in the appendix. ^^ This analogy is useful in getting a physical picture of the units but it must not be carried too far as will be evident when the problem of reflected currents is considered. 52 BELL SYSTEM TECHNICAL JOURNAL The results are shown in Figs. 2, 3 and 4. Figures 6 to 11 refer to current voltage relationships discussed in a later section. To find the potential distribution when a second plane at a distance (T away from the first plane is held at a potential > I 3 Vi'^'^A Cr= DISTANCE IN UNITS OF Sq \^So=I53 X 10""^ ^j Fig. 2 — Potential distributions of the A and B types. The solid lines are the potential curves; the broken lines indicate the transit times. 54 BELL SYSTEM TECHNICAL JOURNAL For the second group of curves (Type B) a certain fraction of the current denoted by Z (and so indicated on the curves) will be trans- mitted while the fraction 1 — Z will be reflected toward the first plane. These potential curves have zero slopes at the reflection planes so that they correspond to solutions of Child's equation on both sides of a so-called virtual cathode existing at the reflection plane. For each value of Z from 0 to 1 a possible potential distribution is obtained. It will be observed that the portions of the solid curves to the left of the zero points are drawn lighter than the rest. These portions correspond to potential distributions resulting from conditions with reflected current and so while applying as extensions of the heavy curves to the right they cannot be entered directly with values of

^ Vr+^'/^. (9) dip d(T ^l^ \: n\ \ \ \ N \' \ \\ \\ A \; \\ \, \ ^1 \ \^ \\ \\ \\ \ \ >^ \ N \\ \N \\ \,^ A \ N, \ A x\ n\ ^^ \^ JD \ \ A xN N^ A V> N ,\ \ v^ 9^ i\^ \ \ ^ \ >N V \\ \\ Q7 UJ± 1- UJZ -> Ui 2 OC \ p\ \ \\ a: ^^^ \^ \ \ A \^^ \, x^ \ N^ ^^ \ ,,. D V ^ \l A ^^ A \^ "n A ^^ \ \ <5 \ ~^ \ A \ \ Jl^ \, A \ \ \ <<1 V^. " \: \, ^ A \ 0 ^ A A \^ A^ \ \, \ \ A A \ \ \ \ ^ A \ c^ \^ \ \ M .N s A \i o\ 1 ■<'\ ^/^ ^ CO 0 -3 to 5 m rr OJ UJ n, 1- >, -■ < — UJ -^ W U, Zr= TRANSMITTED CURRENT IN UNITS OF Lq ('1-0 = 2.33 X 10"^ AjT-j J - - O 64 BELL SYSTEM TECHNICAL JOURNAL \ \ \ \ \ \ \ w \ — ^ ^\ •-S \ \ \ \ \\ \ \ \ ^^ \ \ \ \ \ A \; \\ \ ~_ \ s \ \ \ \ \ w \\ I ^ -X \ \ \ \ \ A v\ \ \ X \ s \ \ \ \ \^ \ \\ \ \, ^ \ Sw \ \ \ A \ \\ \^ k \ y \ \ — ^ \ ^- \ \ \ \; \ \ \ V \\ I \ \ N N \ \ \\ V s \\ [A \\ ^ \ \ \ n:^ ^ ^ \ A V \ \ . \ ^ ^ \ \ s^ A V A \ \ \ \, ^ N, \ *x \ v\ \ \\ \ \ V- \ \ \ n\ A \\ c». ^ •^ \ ^-~, ^ X, \, \ \^ A \ \ \ ^ \ O^ \. \ ^^ \ y\ \ X- CM" \ \, \ ^^^ \ \ \\ -A \ <^ ^ ■^ \ \, ^~^ ~oq~~^ X, \ A ^\^ \ V \ \ / ^ o\ \ N^ ^^^ V A w \ 1 (0 \ \ ^ \^ \ \\ \ ^ ^ \ ^, r\ \ \\ \\ w J □X / / \ \, \ \ ^"^ W \\ \ < t \ _^ \ N \N w ■^ \ ^ ujZ—' 3 uj w \ ^ N, \ \\\ \ \ \ / \ \, \^ \^ \ / OJ \ V --^ ■^ \ \ \\ \ / \ \, N A \ ) \ \ / / o ^ A \ " ~^ \, \ \ / y ^ \ ^ A A \ / / (0 ^ X ^ \ / / ^ >" II d J ---- — — - -<->0""00,,0 Zr= TRANSMITTED CURRENT IN UNITS OF Lq [Lo= 2.33 X 10 " -^ ) SPACE CHARGE BETWEEN PARALLEL PLANE ELECTRODES 65 *. 3 ^^ \ V \ \ \ ^ \, ' \ m \ ^. \, \ \, \ \, \ \ \ 00 \\ \ \, \ \ ^ \ \ \ c\ \ ^> \ s, \ \ \ \ s, \ (£ \\ ^ \, \ \, \ \, \ \ \ v \V \, \ \ \ \ \ \ 'i \ \\ \ V S, \ \ \, \ \ 0. \\ v\ \, \ \, \ N, \ \ i \ (\ ^\ \ \^ \ \ ^ ,% v^ c \ \\ \ s \ a K' \ \ \ . \ \ \ \ \ A \ \, u \ \' \ \, \ \, \ \, \ Vmin- jimum potential n units of v| d V ^^ \ \, S \ \^ \ V \ ^> \ \ \ \ \ \ \ \ \ \, \ \ \ \ \^ \\ \ V \^ \ \ \ \^ n\ \, — — — — — — ^ i_i < d \- t \ \__ ) \ / ^) '-> Vmin_ ' minimum potent in units of v; u \\ 5 — < / / // k %\ ' °/l V\4 d^ \^: A ( \ W / 1 / s\\ v\ / /t3 ol 4 \ ^ 1 d „> { k -> < ji D ^ D a a f 1 ■vj _ D 7> CO - I. 0 u 0 ' J < ^ (M C ) :>1> II 9l 3 J3 >-• _1 Q. 1^ •T3 z u R V III a CO >. w bo IZ Y = INJECTED CURRENT IN UNITS OF Lq (^Lo = 2.33 X 10"° —^) 66 BELL SYSTEM TECHNICAL JOURNAL NX \ I \ \, \ \ 1 1 1 / / / \ \, V, \ ^. \ \ 1 1 / / / / \, \ \ \ \, \ \, 1\ \, ^ / / / / \\ \ \ \ \, \ \ X 1^ / / A \ \ >>4 s \ 1 1 i \ \. V \ ^ \ \ \ \ 1 1 ! h 1^\ \0 \ \ \ \ I \ V 1 1 ■^ 1 r ^ ^ \ V \ \ \ \ N a —1 d ^^ ,\ s^ k s. o 01 c 1 V 5^ld / \ K \ \ ool \ \ \ 1 1 1 /c ^ \ -^; \ V 1 1 / ^ II z \\\ \\\ \ N 1 I 1 1 ^ / 1 9 " V \\ V \ \ \ \ 1 1 1 1 Jro d c \\ \, ^ N, 1 V^ w N 1 o A v. J I \ d Vv \ T3 V \\ \ \ \ \ \ d ^\ \ \ \ \ \ V \ V > \ \ ^~~~+ a ^^ \ \ \ \ o \ \ A \ s \ \ \ 1 \ \ \ S N \^ \ \ — ^ = MINIMUM POTENTIAL f IN UNITS OF V2 V|IN = MINIMUM POTENTIAL IN UNITS OF Vi \\ \ d ^ ^^ \v M o N (O V \'^1 o ^ ^^ l\ o o z (1 < iN ^ ( 3^ A9- 00 9.1 1 9- C 5^ d r = INJECTED CURRENT IN UNITS OF Lq (lo=2.33 X 10-6-^^) SPACE CHARGE BETWEEN PARALLEL PLANE ELECTRODES 67 \ \ \ s, \ \ \ \ \ \ \ \ \ \ \ \ \ s, \ \ \ \ \ A \ \ \ \ \ \ \ \ s, \ \ \ \ \\ >.'^- \ \ ^ \ ■>\ \ % \ % V 9- N \ \ \ \ <>\ ^ \ ^N \ \ \ \ \ \ \ \ \ \ \ CO d \^ \ \ \ \ \ I \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ V \ \ \ A \ i\\ \ \ \ \ \ \ \ \ V \ \ L \ \ \ \ \ \ \ A \ \ \ \ \ \ \ d V A \ \ \ \ \ \ \ A \ A \ \^ y \ A \ \ A \ On ^. \\ ^ \ d \ 8^ d ^\ O Oi ID o o) 0) r^ ct. 0 UJ u ■7 > UJ z < 0 . 1 Q. to n 0 z ■K 0 0 0) < _ II w 5>1> i 9- .5f fO t\j — o (■ 6 ^•"'^'l (^L0 = 2J3X10 6-^^ 68 BELL SYSTEM TECHNICAL JOURNAL rises, finally reaching the limiting values indicated on the upper curve of Fig. 11; the limiting conditions there are precisely the same as for curve a of Fig. 10. Further increase of injected current produces a transition to type B solutions. It seems questionable that this overlap type C condition can exist in a practical case. An investigation of this question involving ex- ternal circuit considerations is beyond the scope of this paper. Transition Between Distribution Types The physical choice between the different possible potential dis- tributions which may exist with a given set of boundary conditions is determined by the sequence in which the boundary conditions are established. Extreme values of any parameter are seen from Figs. 7 and 8 to lie in regions for which only one solution is possible. If the boundary conditions are varied slowly and continuously from these values, the indicated type of distribution will persist until the limit of this region is reached at which time a sudden transition must occur to another indicated type of distribution. Inspection of Figs. 7 and 8 will show that at such transitions only one other type of distribution is ever possible. The determination of the correct physical distribu- tion can thus be made without ambiguity. Certain peculiarities are, however, to be noted. A survey of all possible transitions in which 7 and v' are treated as independent variables will indicate that, starting from extreme conditions and changing conditions continuously in the same direction, distributions of the overlap C type shown in Figs. 4 and 1 1 never occur. A second peculiarity has to do with the unstable region of type B solutions shown on Fig, 9. When this region is entered with insufficient resistance in the external circuit, instability results with a sudden transition to a corresponding stable type of distribution. The space model shown in Fig. 12 has been found to be of value in visualizing problems involving transitions. The three coordinates used in its construction are the second electrode potential

2 and limiting curve d ii y < 2. Now since Vp Vp Fi Vp ' Va-\- — 1^ Proof of the relationship 7 = o-^ is contained in the appendix. .72 BELL SYSTEM TECHNICAL JOURNAL and v,iv, = 1-^^-1 1 — > if IX and a are known, the maximum value of Vg/Vp may be calculated. Fig. 7 shows that a value of 7 just greater than 2 will give the lowest minimum value for

0 and must therefore lead to solutions of the C type. Integrating once more and introducing the unit So gives X = ± 5o(^'/2 ^ 2ai/2)^^i/2 _ „i/2 _|. const. (22) Expressing distance from the first plane in units of Sq, we find two possibilities: CD = -\- { 1, and a = - (^1/2 _ 2/31/2) V^i^2+7i72 + (1 - 2)81/2) Vr+^, (35) which applies for ^ < 1. Corresponding transit times are: r = + (^1/2 + /31/2)l/2 - (1 + /31/2)l/2. (37) X = - (^1/2 + /31/2)l/2 -I- (1 + /31/2)'/2, (38) Integration Constant is Positive — Type A The potential distribution curves of the A type are identical in form with those of the D type, where (p < 1, which result from the 78 BELL SYSTEM TECHNICAL JOURNAL positive value of the constant in equation (7). They differ in nu- merical values in that the current / must be replaced by the value 27 to allow for the reflected current. The correct equation is then 0- = [(1 - 2/31/2) (1 _1_ ^1/2)1/2 _ (^1/2 _ 2/31/2) (<^l/2 _|_ ^l/2)l/2-|2-l/2. (39) The corresponding transit times are given by r = [(1 + ,31/2)1/2 - (^1/2 + ,3i/2)i/2]2-i/2. (40) To the right of ^ = 0 the space is free of charge so that the potential gradient is constant and equal to the value at ^ = 0 obtained by taking the derivative of equation (39). This value is d^^ 4V2/31/-1 ^^j^ da 3 Concerning Completeness We may now review our work and see that no possible space charge distributions can have been omitted. Starting from the fundamental equation (6) we obtain equation (7) with an undetermined integration constant. Setting this constant equal to zero we could integrate once more, obtaining a solution formally identical with Child's equation. If we supposed that the cathode plane defined by the Child's solution lay to the right of the initial plane, then the only freedom left in the solution was represented by Z, the fraction of current passing through the plane. All physically sensible values of Z, i.e., 0 to 1, are included in the solutions. If the cathode is assumed to lie to the left of the plane, then Z must equal 1 and the solution which arises is given by a = 0 in equations (23) or (26), or /3 = 0 in equation (35) — that is, a D solution. For a negative value of the constant, a further inte- gration gave only two possibilities. Each of these was investigated for all possible values of the constant. A similar statement is true for positive values of the constant. As was stated in the text, space charge distributions corresponding to injection from both bounding planes can be handled in formally the same way as injection from one plane ; therefore we may conclude that all solutions to the problem given by specifying the boundary conditions on two planes, subject to the assumptions represented by equation (6), have been determined. SPACE CHARGE BETWEEN PARALLEL PLANE ELECTRODES 79 Equations for Boundary Curves For convenience we append a table of equations for the limiting curves occurring in the figures. Symbol Equation numbers V2)3/J 1 + ¥>"* (1 + ,,1/2)3/22-1/2 (1 +^1/2)3 (1 + ^/4)2 (1 + ¥>V2)3/2 (1 +2^1/2)2(1 _ (^it + 2)K«''« 00 . pl/2) - 1) 1 1 2^1/5 (1 +^1/2)3 (1 +^3/4)2 ^1/2/(1 +,pl,2)S (1 +2.f'/2)2(l - (^1/2 + 2)2(¥>l/» ^3/2 (1 + ,pl/2) 1 1 0 ■ d (1 + 2,pl/2)Vl - y = M M 1 1 M M » I I a i ft '^ Q 1 J M } 1 M I 1 84 BELL SYSTEM TECHNICAL JOURNAL A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 85 single sideband method of transmission is employed, with carrier fre- quencies suppressed. The choice of a group comprising 12 channels was influenced not alone by the requirements of the type K system itself but also by those of other broad-band systems. From the earliest stages of the broad-band development it was recognized that there would be considerable advantage from the standpoints of flexi- bility of interconnection, of minimum development effort, and of large scale production of equipment units, if the designs of different broad- band systems could be so coordinated as to enable the same design of channel terminal equipment to be employed for each. A common 12-channel terminal unit developed for this purpose is used in the type K system. 15.0 m 10.0 o (O 5.0 o _l ^ 2.5 tr -2.5 -5.0 1 1 1 1 1 / ,' 1 1 / 1 FIVE LINKS 7 \s > / 1 / "~"'~- SINGLE LINK 500 1000 1500 2000 2500 3000 FREQUENCY IN CYCLES PER SECOND Fig. 3 — Transmission frequency characteristics of overall circuit. The spacing of the channels in broad-band systems is important from the standpoint of the channel selecting circuits and the width of the derived voice circuit. As discussed in a recent article, a uniform 4000-cycle interval has been adopted for the different channels of all broad-band systems.^ The speech band width obtained with this spacing is in keeping with recent improvements in telephone instru- ments and other parts of the telephone plant. Overall transmission- frequency characteristics for a single link and a five-link connection are shown in Fig. 3. Cable Attenuation The type K system is designed to be applied to the No. 19 AWG (0.9 mm.) pairs commonly found in existing cables. (The Morristown 86 BELL SYSTEM TECHNICAL JOURNAL system used 16-gauge pairs.) Because the conductors are small and closely spaced, with paper and air dielectric, the attenuation of a non-loaded 19-gauge pair at the frequencies involved is inherently high, as will be seen from Fig. 4. Because of the high attenuation, the repeaters must be placed much closer together than is necessary for voice-frequency cable circuits. Fortunately this effect is partly offset by the fact that it is possible, as discussed later, to use higher gains in the carrier repeaters. The cable pairs exhibit the rise in attenuation with frequency which is familiar in most transmission circuits. This effect is brought about largely by the increase in conductor resistance, due to skin effect, and 4.5 4.0 3.5 3.0 2.5 9 1.0 0.5 II0°F 0^ ^ ^ ^ ^ - -^ ^ ^ ^^ y ^ ^ /^ ^ ^ 10 15 20 25 30 35 40 45 50 55 60 65 FREQUENCY IN KILOCYCLES PER SECOND Fig. 4 — Attenuation of 19-gauge non-loaded cable pair. the increasing dielectric losses. More important than this, however, is the fact that the resistance of the wires and the other "constants" of the cable pair undergo variations with temperature, which in turn affect the attenuation. The magnitude of the result for a representa- tive non-loaded 19-gauge cable pair is illustrated by the curves of Fig. 4, which show, respectively, the attenuation for an average tempera- ture, assumed to be 55° F., and for 0° F. and 110° F. The latter values, often taken as the extremes of annual variation for an aerial cable, are in fact frequently exceeded. One reason for this is that when the sun is shining directly on an aerial cable, it may assume a temperature from 15° to 25° above that of the ambient air. The A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 87 range of temperature variation (and attenuation variation) for an aerial cable may be half as much in one day as in an entire year. For an underground cable, changes of temperature occur quite gradually and the total annual variation is about one-third of that for an aerial cable. These relations between the attenuation of a cable pair and the frequency and temperature are of fundamental importance in the design of the type K system. First of all, since the attenuation at 60 kilocycles is about 4 db per mile, the total attenuation for a cable circuit of the length used in designing the type K system, i.e., 4000 miles, would be approximately 16,000 db. This must be offset by a corresponding gain. In the next place, differences in the attenuation at the different fre- quencies would, if uncorrected, become so great that signals of the less attenuated channels would overload the repeaters, while those of the more attenuated channels would drop down into the noise region. Hence, each repeater must be given a gain-frequency slope which is complementary to the attenuation slope of the line. Finally, the changes of transmission due to temperature variations and other causes must be compensated so precisely that the net vari- ation in each channel is held within very narrow limits. The method of doing this is explained later. Here it is interesting merely to con- sider the magnitude of the problem. For the top channel, assuming a 4000-mile circuit, the annual variation in attenuation of an aerial cable pair might jbe approximately 2000 db. The systems thus far installed have, of course, been limited to much shorter distances than this. Even if the change of attenuation with temperatures were related to frequency by a simple law, correct compensation over the frequency range would be far from easy. To a casual inspection the differential between any two curves of Fig. 4, for example those for 55° F., and 110° F., will not appear serious. This differential, which becomes very large for a long circuit, is a complicated function of the frequency. The attenuation differential with temperature can be considered as made up of two components, one which is independent of frequency and another which varies with frequency. The former component, which is much the larger, requires a gain adjustment which is uniform or flat over the frequency range of the system. The latter component is frequently referred to as the "twist." For the range from 12 to 60 kilocycles, the maximum change of attenuation with temperature occurs near 28 kilocycles. Hence this frequency has been used as a datum point in determining the twist. The shape of the twist com- 88 BELL SYSTEM TECHNICAL JOURNAL ponent is apparent from Fig. 5, which shows the net loss per mile at temperatures of 0° F., and 110° F., assuming that the attenuation has been equalized so as to obtain a flat characteristic at 55° F., and that the gain is then adjusted so as to hold the transmission constant at 28 kilocycles as the temperature varies. Although the twist is small enough so that it need not be corrected at each repeater, it is too large to be allowed to accumulate over a very long distance. 0.06 0.04 1 ^ 0.02 O I/) QO -^0.04 \ V \ \ V I10°F ■"55^ -— -^ / ^ ~"o^ -^ - / J 0.06 0 5 10 15 20 25 30 35 40 45 50 55 60 65 70 FREQUENCY IN KILOCYCLES PER SECOND Fig. 5 — Twist characteristics of 19-gauge non-loaded cable pair. Crosstalk As noted above, crosstalk between opposite directions of transmission is avoided by using two separate cables (or shielded compartments in the same cable). To prevent crosstalk in offices, special measures are employed. There remains the problem of "far-end " crosstalk between pairs in the same cable which transmit in the same direction. The pairs are packed closely together, and substantial crosstalk occurs between them because of small departures from symmetry and slight imperfections of twisting. However, by abandoning the use of phantoms and by connecting small adjustable mutual inductance coils between each carrier pair and every other carrier pair, sufficient cross- talk reduction is obtained to permit transmission up to 60 kilocycles on a substantial number of pairs.^ The scheme is illustrated in Fig. 6. In the original Morristown cable, the crosstalk was reduced in part by separating the 16-gauge carrier pairs from one another by 19-gauge quads which served as spacers. With existing cables, however, the use of spacers would be impracticable since this would require resplicing the cable at every joint, and therefore reliance must be placed largely on balancing. Since the number of combinations to be balanced increases approximately as the square of the number of pairs employed for carrier, the number of balancing coils required for even a moderate A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 89 complement of carrier pairs becomes quite large. With 40 carrier pairs, for example, the number of coils required is 780. The balancing coils are mounted on panels as shown in Fig. 7 and are connected together in a crisscross arrangement. Each repeater section is balanced separately, the balancing panels being located in the repeater station. REPEAT- ERS INTERMEDIATE LINE SHOWING PATHS OF CROSSTALK r T — TT ^ U- :u: ; \ I — 1- ■' '- J L BALANCING COILS Fig. 6 — Method of balancing out crosstalk. Other measures are necessary to supplement this balancing tech- nique. To reduce the crosstalk coupling, and also to average the transmission characteristics of different pairs, the carrier pairs in dif- ferent quads of an existing cable are respliced about every mile so as to approach random splicing. The crosstalk coupling between the two sides of a quad is reduced by test splicing at the middle of a repeater section and the quads are split at repeater points. In a new cable, of course, the desired splicing arrangements are introduced at the time of installation. The carrier pairs are also transposed from one cable to the other as indicated in Fig. 1. This avoids interaction crosstalk that would take place, through the medium of the voice- frequency pairs, between the high-level carrier outputs on one side of a repeater station and the low-level inputs on the other side. There is of course, a similar effect between carrier pairs at any point in a repeater section. This is much less serious since no level difference is involved, but it does tend to limit the effectiveness of the balancing over a range of frequencies. Reflections resulting from impedance irregularities reverse the direc- tion of propagation and therefore produce far-end crosstalk from near-end crosstalk. This crosstalk cannot readily be balanced out 90 BELL SYSTEM TECHNICAL JOURNAL over a range of frequencies. To avoid it, it is necessary that the im- pedances of successive lengths of cable pair be substantially uniform and also that the impedance of the equipment be closely matched to the characteristic impedance of the cable pair. Fig. 7 — Crosstalk balancing panel. Noise The cable pairs are fairly well protected by the lead sheath from external electrical disturbances. However, high-frequency noise originating in the voice-frequency repeater stations due to relay opera- tion, etc., would, unless prevented, enter the cable over the voice- frequency pairs and thence would be induced in the carrier pairs. Such noise would be excessive at the low-level carrier inputs. It is avoided by connecting, in each "voice-frequency pair in the "low level" cable on each side of a voice-frequency repeater station, a coil which suppresses longitudinal noise currents. Similarly, it is necessary to A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 91 keep high-frequency noise from entering the cable where open-wire pairs tap into it and frequently also where branch cables are con- nected. For this purpose simple noise suppression filters are em- ployed. With these and other measures the noise on the carrier pairs can, at the highest frequencies where the amplification is greatest, be brought within a few db of the basic noise due to thermal agitation of electricity in the conductors themselves. Velocity of Transmission Voice waves travel through loaded cable circuits at from 10,000 to 20,000 miles per second, the higher speed being used for the longer circuits. On very long connections, even if echo suppressors are employed this velocity results in transmission delays which introduce difficulties in conversation.^ The use of non-loaded conductors for the type K systems results in an overall velocity of transmission of about 100,000 miles per second, a speed so high that such difficulties are greatly reduced and satisfactory telephone conversations are possible over the longest distances for which connections may be required. Repeaters Since the noise level in the cable circuits can be made quite low, the carrier currents may be permitted to drop to levels below those used on voice-frequency circuits or on open-wire carrier circuits, and the repeaters may have higher gains. In the cable carrier system, the noise has been so reduced that the level of the top channel at the repeater input may on the average be dropped about 60 db below the voice level at the transmitting switchboard. The amplifier gains at the top frequency range from about 50 to 75 db, and the output level of each of the 12 channels is about 10 db above that at the switchboard. The average repeater spacing is about 17 miles. The tube which was developed for the gain stage of the amplifier is a pentode with indirect heater. The heater requires a potential of 10 volts and a current of 0.32 ampere and the plate 150 volts. The tube in the power stage is similar in type but requires a heater current of 0.64 ampere at 10 volts. With this power tube a feedback of about 40 db has been found to provide a satisfactory reduction of inter- channel modulation.^ Both tubes were designed to have long life with very reliable performance. Description of Amplifier Each repeater comprises two amplifiers of the type illustrated in Fig. 8. A schematic diagram of the amplifier circuit is shown in 92 BELL SYSTEM TECHNICAL JOURNAL Fig. 9. Three stages with impedance coupling are used and the feed- back circuit is connected between the plate circuit of the last tube and the grid circuit of the first. The amount of gain and the slope of the gain -frequency characteristic are controlled by the condensers and line equalizer in the feedback circuit. EQUALIZER IN FEEDBACK TUBE-TESTING JACKS GAIN- REGULATOR CONDENSER GAIN-CONTROL CONDENSER REVERSIBLE SYNCHRONOUS RECEIVER MOTOR DIAL INDICATES "POSITION OF REG- ULATOR CONDENSER Fig. 8 — Line amplifier. RECEIVER MOTOR /ER 1 \ LINE EQUALIZER TO FLAT-GAIN MASTER CONTROLLER - FLAT-GAIN REGULATOR CONDENSER (14 DB RANGE) FLAT-GAIN CONTROL CONDENSER (10 DB RANGE) ! + Fig. 9 — Schematic of line amplifier. The line equalizer has the same attenuation slope as the line. In the introduction of this equalizer in the feedback circuit careful atten- tion to phase shift requirements was required. Four types of equalizers are available, for different repeater spacings, to compensate for the cable distortion which occurs at a temperature of 55° F., additional means being provided to compensate for variations which occur as the cable temperature swings away from this value. The solid curve of Fig. 10 shows a repeater gain characteristic with one of these equalizers in the feedback circuit. A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 93 The correction introduced by the line equalizers is subject to errors which, although small at each repeater point, become important for a moderate length of system. Supplementary equalizers have been designed to correct for these. Two types of deviation are considered, 75 65 m 60 m o UJ Q 55 50 45 40 35 ^^-" ^^'' ^-'' ^ " ^.- '■^ -^ ^'' " ^ ^^^' \ 1 .^ ^ -^ ^•' ,-'' ^•^ ^ ^^^^ -"'' ^^ ^ .-,-' .I-' \ CHANGE IN GAIN OBTAINED FROM FLAT- GAIN REGULATOR CONDENSER y^ / 15 20 25 30 35 40 45 50 FREQUENCY IN KILOCYCLES PER SECOND Fig. 10 — Amplifier gain characteristics. 55 70 that of the cable and that of the amplifier. As the characteristics of cables manufactured at different times show slight departures from one another, two shapes are required to correct their deviations. There is one correction for concave deviations and another for convex. 2 10 15 20 25 30 35 40 45 50 FREQUENCY IN KILOCYCLES PER SECOND Fig. 11 — Characteristics of cable deviation equalizers. 65 The amplifier requires but one type of correction. The characteristics of these equalizers are shown in Figs. 11 and 12. The equalizers for amplifier deviations are used about every 10th repeater and those for the cable deviations, at distances of 300 to 400 miles. At normal temperature (55° F.), the correction applied by these networks will. 94 BELL SYSTEM TECHNICAL JOURNAL for a 500-mile system, result in a frequency characteristic which is flat to within less than 2 db over the range from 12 to 60 kilocycles. For a longer circuit further corrective measures will be provided. 3 2 / ^, / '\ / N \, / \ / \ \, / ) y 0 •s -^ 15 20 25 30 35 40 45 50 FREQUENCY IN KILOCYCLES PER SECOND 55 Fig. 12 — Characteristics of amplifier deviation equalizer. Regulation for Temperature Effects. The method adopted for controlling the repeater gain to compensate for temperature changes is similar to that which has been found satis- factory for voice-frequency cable circuits. This is the pilot wire method in which a pair of cable conductors extending over the section GALVANOMETER POTENTIOMETER AND DIAL GEARED ■ TO MASTER TRANSMITTER MOTOR REBALANCING MECHANISM TRANSLATES GALVANOMETER DEFLECTION INTO ROTATION OF MASTER TRANSMITTER MOTOR MASTER TRANSMITTER MOTOR SETS POSITION OF SYNCHRO- NOUS RECEIVER MOTORS Fig. 13 — Flat gain master controller. to be regulated forms one arm of a Wheatstone bridge. This bridge is designed for automatic self-balancing and the mechanical motion required for establishing the balance has been made to adjust the gain of the amplifiers. The d-c. resistance of the pilot wire gives an A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 95 accurate indication of the temperature of the carrier pairs which determines their attenuation to a close approximation. The motion of the bridge mechanism is communicated to the repeater ampUfiers by means of self-synchronizing motors, a master motor being associated with the bridge and an individual motor with each amplifier. With aerial cable a flat gain correction must be made at every repeater. With underground cable the flat gain correction may be -3 -5 \ \ REGULATOR STEP ^ - \ V 44^ V \ s^ ^ ■^ ^"^ ^^ - 1 \ \^ 33, "^ 22 / -^ — ■ ■~-iL / / ^ "~^ ^^ ^"^ - / / ^>.» ■^.^ / ^ »«. / 0 5 10 15 20 25 30 35 40 45 50 55 60 65 FREQUENCY IN KILOCYCLES PER SECOND Fig. 14 — Characteristics of twist regulator networks. omitted at some repeater points. Figure 13 show^s a master controller with its galvanometer, driving motor, and self-synchronizing motor. The air condenser in the feedback circuit of Fig. 8 makes this cor- rection. The small self-synchronizing motor which may be seen in this figure is geared to the condenser and it moves the air condenser corresponding to the motion of the master motor. The resulting change in repeater gain is virtually the same for all frequencies in the transmitted band. In Fig. 10 the repeater gain is plotted against frequency for three angular positions of the condenser. 96 BELL SYSTEM TECHNICAL JOURNAL As was mentioned earlier, an additional correction for the residual effect or "twist" is required about every six repeaters to supplement the flat gain adjustment. This distortion is a function of frequency and has been found to vary from cable to cable. A network the characteristics of which are shown in Fig. 14 has been developed to meet this condition. Certain fixed resistances in the network are selected to correspond to the length and twist characteristic of the cable section considered. A variable resistance in the network is adjusted automatically using a control similar to the flat gain regu- lator. Figure 15 gives the transmission characteristics of a 150-mile regulator-controlled circuit under two temperature conditions. -2 9°F^ -^ ■^.N. — — ^__ ^^ \ ^ ^ 15 20 25 30 35 40 45 50 FREQUENCY IN KILOCYCLES PER SECOND 55 60 65 Fig. 15 — Overall transmission-frequency characteristic of 150-mile line. Auxiliary Repeater Stations Cable carrier systems are expected to be used largely on existing toll cable routes which now carry voice-frequency circuits. The aver- age spacing of the stations housing the voice-frequency repeaters on these routes is about 50 miles. The same buildings with their power plants will also care for the cable carrier repeaters. Since the maxi- mum spacing for the carrier repeaters is about 19 miles, additional carrier repeaters must be provided at intermediate stations (two is the usual number). The various design features of the equipment to be located in these stations have been made the subject of extensive development work and field tests. These stations are designed to function with a minimum of attention and are visited at intervals for routine testing work or as required by some emergency, but resident maintenance forces are not planned for them. The present equipment is expected to be suitable not only for auxiliary stations on existing cable routes but also for cases where a greater spacing than 50 miles between the attended stations may be desired on new routes. A voice-frequency repeater station for a single cable and a cable carrier auxiliary station are shown to approximately the same scale A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 97 in Figs. 16a and 16b, respectively. Many of the existing voice- frequency stations are even larger than that shown in Fig. 16a. The auxiliary building shown in Fig. 16b has about 600 feet of floor space Fig. 16a — Voice frequency repeater station on single cable route. with a ceiling height sufficient to take care of 11'6" relay racks. This building will house 100 repeaters with necessary auxiliary equipment, thus providing ultimately for a total of 1200 carrier circuits. The interior of a typical auxiliary station is shown in Fig. 17. "2"- Fig. 166 — Auxiliary cable carrier repeater station. The main power plant for the repeaters consists of a 152-volt storage battery, which is continuously floated across a grid-controlled rectifier fed from the 60-cycle power mains. The voltage of the entire 98 BELL SYSTEM TECHNICAL JOURNAL battery supplies the plate voltage for the tubes. Each amplifier re- quires about 22 volts for the tube heaters and this is obtained by dividing the battery into seven sections, each section supplying several amplifiers in parallel. Additional power supplies of 55-volt alternating Fig. 17 — Interior of auxiliary repeater station. current and 140-volt direct current are required for the regulator system. In the station, there are alarm circuits, which signal the nearest attended office if trouble develops. There are alarms for blown fuses, high or low battery voltages, power failures, etc. A telephone order wire to the nearest attended station is provided for the maintenance force. A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 99 In addition to the line amplifiers with their regulating equipment, there are racks mounting the crosstalk balancing coils. There are also sealed terminal units between the outside cable or the balancing units and the office cable. These furnish access to the line or equipment through jacks. Terminals The minimum distance over which a cable carrier system can be operated economically is determined in large measure by the cost of the terminal apparatus. Hence, the field of usefulness of the system is greatly increased by keeping the terminal cost as low as is consistent with satisfactory performance. Numerous developments during the past few years in connection with modulation, filtering and methods of carrier supply have all contributed materially toward this end. At the same time, the standards of performance have not only been maintained, but in many respects substantially improved. Channel and Group Modulation In the design of the terminals for the type K system, a number of circuit arrangements were considered, the final choice being influenced to a considerable extent by the conditions imposed upon the filters. As noted above, the desirability of using the channel terminal equip- ment in other broad-band systems, such as those for open-wire or coaxial cable, was also an important factor. The circuit arrangements selected have a first stage of modulation which raises the voice fre- quencies of the 12 channels up to a range of 60 to 108 kilocycles. This range is favorable to the use of crystal type band filters, '^ w^hich have transmission characteristics superior to the coil and condenser type and seem to be no more costly. For the type K system, a single stage of group modulation shifts the frequencies to the range required on the line, 12 to 60 kilocycles, and a similar stage at the receiving end returns them to the 60 to 108-kc. range. Other carrier systems will also use the 60 to 108-kc. channels and by group modulation shift them to the desired position in the frequency spectrum. The band filter occupies a space on the relay rack equal to 1/8 of that required by the coil and condenser type which was used in the earlier model of this system. Its attenuation characteristic in the transmitting region is flat to within 1 db over a range of about 3100 cycles. Immediately outside of this range the attenuation rises very rapidly, thus permitting very efficient use of the frequency spectrum. Another new device on the terminal is the copper-oxide unit used in the modulating process. These units are expected to show a stability 100 BELL SYSTEM TECHNICAL JOURNAL of the same order as that of coils and condensers, and require practic- ally no maintenance as compared to vacuum tubes. The translation of the channels from the 60 to 108-kc, range to the position required for cable carrier, 12 to 60 kilocycles, is made by a stage of group modulation. A copper-oxide group modulator is used and a carrier frequency of 120 kilocycles. The reverse of this process in a similar group demodulator at the receiving end steps the frequency back to its original range, 60 to 108 kilocycles. These processes of modulation take place at points of low-energy level in the circuit with a comparatively high level of carrier, so that the inter-channel crosstalk which results from unwanted products of modulation is unobjection- able. Low-pass filters are inserted after the group modulator and demodulator, and amplifiers with flat gain characteristics are supplied to raise the levels of the output currents of the group modulators or demodulators. Carrier Supply The carrier frequencies which are required at a terminal are obtained from the harmonics of a base frequency. The carrier supply system is common to as many as 10 systems in one office. This simplification was made possible by the selection of the channel frequencies as mul- tiples of a base frequency, 4 kilocycles being chosen for this system. This base frequency is produced by an oscillator in which the control element is a tuning fork, the whole unit being designed to have the necessary output and frequency stabilities. The output of the oscil- lator is amplified and fed to a circuit which produces the desired harmonics. All of the carrier frequencies which are required for the different channels as well as for group modulation and demodulation are obtained from these harmonics. A small coil with a permalloy core is the important agent in this process.^ Failure of the 4-kc. supply, or failure of the 120-kc. supply used for group modulation, would cause failure in the channels of all systems operating from this supply. Provision is made for such a contingency by an emergency carrier supply which is automatically switched into service when the regular supply fails. This reserve source duplicates all of the parts of the regular supply, 4-kc. fork, amplifier, harmonic producer, and amplifier for the 120-kc. carrier. Assembly The different panel units which make up the terminal of a type K system are assembled on a functional basis with similar panels of other K systems, the channel modulator-demodulator panels in one A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 101 bay, the carrier supply in a second, the group modulator and demodu- lator in a third, etc. The compactness of the equipment makes it possible to mount the modulators and demodulators for 18 channels on one 11 ft. 6 in. bay 19 inches wide. Signaling The same type of ringdown signaling equipment is used with the channels of this system as with the voice-frequency toll circuits. A 1000-cycle tone, interrupted 20 times per second, is impressed on a channel terminal, modulated, and transmitted over the carrier system in the allotted channel band. At the far end, it is demodulated to operate the receiving end of the standard voice-frequency signaling circuit, or to be transmitted along an extended voice-frequency circuit to its terminal. Fig. 18 — Vacuum lube test set. Telegraph and Program Applications Voice-frequency telegraph can be superimposed on any of the carrier channels as is now done on the three-channel open-wire systems. Equipment is being developed to include a program channel on the cable carrier system. This will be done by devoting to the program circuit the frequency space occupied by two of the 4-kc. speech bands. 102 BELL SYSTEM TECHNICAL JOURNAL System Maintenance Arrangements are provided whereby the tubes may be tested on a routine basis as has been done in voice-frequency practice. The ampUfier panels, however, are provided with test jacks which are con- • m • iM Fig. 19 — Testing oscillator. nected to resistances in the plate circuits. A reading of the voltage across the resistance gives a measure of the plate current for the associated tube without disturbing the amplifier performance while the amplifier is in service. A portable tube testing set. Fig. 18, has been designed for this measurement. A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 103 Provision is being made for the removal of an amplifier from an active circuit without interruption of service. A spare amplifier at each repeater station can be substituted for the active one by con- necting it to jacks at the sealed terminal and operating associated relays to make a quick transfer. Apparatus is also furnished which permits the substitution of a new link between attended points for one which develops trouble. A complete high-frequency circuit for each direction of transmission will generally be reserved as a spare. It can be substituted for any working HS^* Fig. 20 — Transmission measuring set. high-frequency circuit without interfering with service by paralleling the transmitting ends of the spare and working circuits and patching the receiving ends through relays. The operation of a key controlling these relays substitutes the spare circuit for the working one with a transient disturbance of but 1 or 2 milliseconds. Three pilot frequencies, 15.9, 27.9 and 55.9 kilocycles, which are produced at the transmitting terminals, may be used to check the levels at the main repeater points and the receiving terminals. This is done by means of a special testing circuit which can be bridged across a pair to detect the level of the pilots without interference to service. 104 BELL SYSTEM TECHNICAL JOURNAL A heterodyne oscillator having a frequency range from 60 cycles to 150 kilocycles has been developed for use in testing this and other carrier systems. Its frequency is calibrated at 60 cycles against the power mains and at 100 kilocycles against a quartz crystal. This oscillator is shown in Fig. 19. A portable test set, developed for measuring transmission gains and losses with high precision, is shown with the cover removed in Fig. 20. Conclusion The type K system makes possible the application of carrier to toll cables of existing type, whether installed underground or aerially. The blocks of 12 circuits each, which it furnishes, seem to be a con- venient size for routes where large numbers of circuits are concen- trated. It is to be expected, of course, that substantial modifications and improvements will be made in this system through further develop- ment effort. In its present form, however, it constitutes an important stage in the history of carrier development. Plans already under way call for the application of large numbers of such systems to meet rapid growth in long distance traffic. This new system forms merely one phase of a concerted development , effort on broad-band carrier transmission systems.^* ^^ There is every indication that, taken collectively, these broad-band systems will have far reaching effects upon the toll telephone plant of the Bell System. A transition is already under way from the time when carrier was used only on open wire, and comprised only a small part of the toll plant, to a time when carrier systems will furnish a major part of the toll circuit mileage of the Bell System. The type K system is clearly destined to play an outstanding part in this evolution of the toirplant along carrier lines. References 1. "Carrier in Cable" by A. B. Clark and B. W. Kendall, Electrical Engineering, Vol. 52, page 477, July 1933; also Bell Sys. Tech. Jour., Vol. XII, p. 251, July 1933. 2. "Carrier Systems on Long Distance Telephone Lines" by H. A. Affel, C. S. Demarest and C. W. Green, Bell Sys. Tech. Jour., Vol. VII, pages 564-629, July 1928; also Electrical Engineering, Vol. 47, pp. 1360-1367, October 1928. 3. "Transmitted Frequency Range for Circuits in Broad Band Systems" by H. A. Affel, Bell Sys. Tech. Jour.,\ol XVI, p. 487, October 1937. 4. A. G. Chapman, U. S. Pat. No. 1863651; M. A. Weaver and O. H. Coolidge, U. S. Pat. No. 2008061; M. A. Weaver, U. S. Pat. No. 2080217. 5. "The Time Factor in Telephone Transmission" by O. B. Blackwell, A.I.E.E. Transactions, Vol. 51, pages 141-147, March, 1932; also Bell Sys. Tech. Jour., Vol. XI, pp. 53-66, January 1932. 6. "Stabilized Feed-Back Amplifiers" by H. S. Black, Electrical Engineering, Vol. 53, p. 114, January 1934. I A CARRIER TELEPHONE SYSTEM FOR TOLL CABLES 105 7. "Electrical Wave Filters Employing Quartz Crystals as Elements" by W. P. Mason, Bell Sys. Tech Jour., Vol. XIII, p. 405, July 1934. 8. "Magnetic Generation of a Group of Harmonics" by E. Peterson, J. M. Manley and L. R. Wrathail, Electrical Engineering, Vol. 56, No. 8, p. 995, August 1937; also Bell Sys. Tech. Jour., October 1937. 9. "Wide-Band Transmission in Sheathed Conductors" by O. B. Blackwell, Bell Telephone Quarterly, Vol. XIV, p. 145, July 1935. 10. "Systems for Wide-Band Transmission Over Coaxial Lines" by L. Espenschied and M. E. Striebv, Electrical Engineering, Vol. 53, pp. 1371-1380, October 1934; also Bell Sys. Tech. Jour., Vol. XIII, pp. 654-679, October 1934. 11. "Modern Systems of Multi-Channel Telephony on Cables" by A. S. Angwin and R. A. Mack, Journal of the Inst, of Electrical Engineers, Vol. 81, No. 941, p. 573, November 1937. Cable Carrier Telephone Terminals * By R. W. CHESNUT, L. M. ILGENFRITZ and A. KENNER This paper describes the circuits, performance and equipment features of the terminals of a new 12-channel carrier system for application to existing toll cables. The 12-channel group of ter- minal apparatus has been designed also to form a basic part of the terminals of other carrier systems now under development, such as the type J system for open wire and the coaxial system. Introduction A BOUT twenty years ago the first commercial carrier telephone ^ ^ system was installed between Baltimore and Pittsburgh. Until] recently, telephone circuits were obtained by carrier methods largeh on open-wire lines. The notable exceptions were on short deep sec submarine cables.^- ^ Ten years ago, experiments were initiatec which have now resulted in the design of a carrier system which can b« applied with substantial economy to existing long distance toll cables on land. Its general features are described in another paper.^ The| present paper describes in detail the circuits and performance of the carrier terminals of this system. General Features The carrier system for existing cables, designated type "K," is de- signed to provide twelve telephone channels in the frequency range between 12 and 60 kilocycles, using one non-loaded 19-gauge papei insulated cable pair in each direction. Previous carrier systems em- ployed for open-wire lines used vacuum tubes for the modulating oi translating circuits and electrical filters composed of coil and condenser| networks for separating the frequency bands associated with the re- spective channels. The terminals of the new type "K" system are simpler and yet provide improved performance by using copper oxide bridges for the modulation function and quartz crystal filters ^ for the| separation of the individual channel bands. The quartz crystal filter is economical only in a comparatively high- frequency range, necessitating the use of high intermediate frequencies. The high intermediate frequencies are reduced by a second stage ofj modulation to the desired range of frequencies for transmission over the] * Presented at Winter Convention of A. I. E. E., Jan. 24-28, 1938. 106 CABLE CARRIER TELEPHONE TERMINALS 107 line. Copper oxide bridge circuits again are used for this group modu- lation stage. In all cases they are connected to suppress the carrier. To provide the various carriers required for modulation and demodula- tion, a carrier supply system has been designed somewhat along the lines of zn office power distribution system using bus bars and pro- tective arrangements for the various carriers. Each carrier supply system is capable of supplying as many as ten carrier terminals, or a total of one hundred and twenty two-way channels. Because of the large number of circuits involved, every effort has been made to provide reliable operation of the carrier supply and common terminal equipment. The terminal and carrier supply equip- ment is designed to permit maintenance tests for checking the per- formance of amplifier tubes and to permit switching between regular and spare equipment without interruption of the large number of circuits involved. The emphasis placed upon ease of maintenance and the necessity for more careful handling of higher-frequency circuits have resulted in new equipment design features. These include new cable terminals, new shielded office cabling, and panels arranged for front wiring and maintenance which are mounted on racks having wiring ducts at both edges of the bays. In the following sections a more detailed descrip- tion is given of the circuits, their performance, equipment and main- tenance features. Circuits The frequency allocation for one direction of transmission and a block schematic of one terminal are shown in Figs. I and 2, which sup- plement each other and need little explanation. The twelve voice bands shown at the left in Fig. 1 are modulated individually in the channel modems.* This forms a 12-channel block lying between 60 and 108 kc. which is then modulated in the group modulator by a 120 kc. carrier to move the block down in the range from 12 to 60 kc. for transmission to the distant terminal. On the receiving side the processes are reversed. One of the channels, as well as the group modem of Fig. 2, is presented in more circuit detail in Fig. 3. This shows the circuit from the point where the voice comes into the carrier system to the point where the twelve carrier sidebands go out onto the cable and vice versa. At the left the four-wire terminating circuit serves, not only as a device to transform from a two-wire to a four-wire circuit, but also as a * The term "modem " has been coined to mean a panel or equipment unit in which there is both a modulator and a demodulator to take care of both the outgoing and the incoming signal. 108 BELL SYSTEM TECHNICAL JOURNAL high-pass filter to eliminate, from the input to the carrier system, noises below about 200 "cycles, such as telegraph harmonics, 20-cycle ringing, 60-cycle power, etc., which may be present on connected voice- frequency circuits. Otherwise these noise frequencies, which are below the voice range, would modulate and pass through the terminal to load unnecessarily the carrier repeaters along the line, as well as to inter- fere with the level indications of the pilot channels. TRANSMITTING END RECEIVING END ^ GROUP CARRIER CHANNEL MODULATOR ^CHANNEL OUTPUT / CARRIERS I \r 108 ^68 1 k64 I D^vi GROUP MODULATOR . OUTPUT PILOT Iv 13 n POSITIONS r«i-^— y*56 „ I Ii-J»52 "-^ 1 iO- J*48 U-°-o U-Q,24 ''■' L3__Q.30 ^,^-0.16 _ VOICE BANDS CHANNEL DEMODULATOR INPUT 108 D -104 l±. I5 D I 8- GROUP DEMODULATOR INPUT 12 .-Q.S2 I i°— Q.48 / J-D.44 I -8-D.40 i-Q.36 ^-D*32 I ---438 I 1-4.0 ^-4,6 I 1-°.I2J 1/ D ' VOICE BANDS Fig. 1 — Frequency allocation. From the terminating equipment the circuit loops through jacks which have paralleled contacts for reliability. The level at this point is — 13 db compared with the transmitting toll switchboard, which level is expected to be generally used in the Bell System for all multi- channel carrier telephone systems. Then comes the channel modula- tor which consists of four copper-oxide discs, each three-sixteenths of an inch in diameter, potted in a small can. This makes a very simple and inexpensive modulator which is much more satisfactory than tubes. CABLE CARRIER TELEPHONE TERMINALS 109 VOICE TERMINATING CHANNEL MODEMS CIRCUITS I , MODULATORS, DE- MODULATORS AND DEMODULATOR |yQI(-g 1, AMPLIFIERS ^ CIRCUITS Tp ' UU ♦r UU "1 64 - MODULATOR AND DEMODULATOR BAND FILTERS Fig, 2 — Block schematic. 110 BELL SYSTEM TECHNICAL JOURNAL >■ Q .ujo3 mjm u lllllllllllllllll J Ki m AA^ V os vr r/ 1- ^a ' •^. ..^■-**' ' 00-5 z o _^ o — 0 2_i UJ I- 3 1 iQCi CABLE CARRIER TELEPHONE TERMINALS 111 ^!< ■ m 2< li- £ iillMi^il MK gtoQz - 5h LU -if-iu if <5o UJ I ILUJ ■ ^ 1 Zl- a LU 5 t- f ^z s s 1 ^$ C ) <►- \ I/) LU CO -1 I UiK bX) tu O U. I—^M^^ 112 BELL SYSTEM TECHNICAL JOURNAL It seems to have an indefinite life. (Some have been on Hfe tests as modulators for about five years.) The carrier power required is about 1/2 milliwatt to modulate satisfactorily a single telephone circuit level of — 13 db. The modulator produces the usual two sidebands and the lower one is selected by the quartz crystal channel filter described in another paper.^ This sideband, joined by eleven others, is stepped down to about the iterative impedance of the shielded office cabling. In the office cabling the twelve channels pass through the high-frequency patching bay to the double balanced group modulator of copper oxide where they are joined by three pilot channel frequencies. The group modulator uses the same copper oxide as that in the channel modulator described above, but the carrier power is about 50 times greater (about 25 milliwatts) in order to keep down unwanted modulation produced between the twelve sidebands. To that same end the level of each sideband is made low (— 46 db), and the double balanced type of circuit is used to balance out some of the undesired products. It also balances out the twelve incoming bands in the range 60 to 108 kc. from the output and so simplifies the following group modulator filter. From a level of — 57 db the twelve channels, now in the range from 12 to 60 kc, are amplified to + 9 db for delivery to the 19-gauge pair in the lead covered toll cable. The amplifier is a three-tube negative feedback type, using pentodes and operating with 154 volts plate bat- tery which is composed of the usual 24-volt filament battery and 130-volt plate battery in series. The last tube is a power tube and does not overload until a single-frequency output of about one watt is reached. On the receiving side in Fig. 3, the twelve incoming channels, in the range from 12 to 60 kc, pass from the amplifying and regulating equipment,^ to the group demodulator. This is identical with the group modulator described above and transfers the twelve channels to the range 60 to 108 kc The channels are then amplified to a — 5 db level by an amplifier of the negative feedback type using two low- power pentodes with 154-voIt plate battery as described above for the transmitting amplifier. From there the twelve channels are separated by the filters which are identical with those on the transmitting side, and are then demodu- lated and amplified to a + 4 db level as shown for one channel in Fig. 3. The demodulator is identical with the modulator but it is poled oppositely on the carrier supply so that the d-c components of modulation in the modulator and demodulator neutralize each other ■1 CABLE CARRIER TELEPHONE TERMINALS 113 and thereby avoid developing an undesirable voltage bias. The poling also reduces somewhat the amount by which stray frequencies have to be suppressed in the carrier supply. The demodulator amplifier has a slide wire gain control rheostat to equalize channel levels, which func- tions by changing both the grid bias on the tube and the amount of negative feedback which is introduced by the rheostat. The sliding contact in the slide wire is made practically free from contact trouble by the space current of the tube flowing through it. As the rheostats are only about 1000 ohms and small in size, they can easily be mounted at a distance from the amplifier in the voice-frequency jack field. The carrier supply for the twelve channels from 64 to 108 kc, and for the group modems of 120 kc, is derived in the circuit shown in Fig. 4. A regular generator is shown at the top in solid lines and an emergency generator at the bottom is shown in dotted lines. Between the two is an automatic transfer circuit (in dotted lines) which trans- fers to the emergency whenever the regular generator fails to supply the proper amount of 120 kc. to the 120 kc. bus. At the upper left-hand corner is shown a 4-kc. tuning fork, of an alloy having a low temperature coefficient driven by the tube to its right to operate as an oscillator of very stable frequency. The next, or control tube, amplifies the 4 kc. to drive the push-pull power stage where a power of about 4 watts is developed. This passes through the 4 kc. filter to the non-linear coil where odd harmonics of 4 kc. are pro- duced. The underlying principles of operation of this coil have been published.^ To derive even harmonics of 4 kc, the copper-oxide bridge is used which rectifies about half the energ\^ of the complex wave of odd harmonics but, by balance, greatly reduces the amount of the odd harmonics present in its output. Odd harmonics are obtained at one point and even at the other. This separation into odd and even har- monics by the balance of the copper-oxide bridge provides effective loss of about 30 to 40 db and reduces the requirements on the carrier supply filters which follow. The two branches pass through hybrid coils to the banks of channel carrier supply filters. These separate the frequencies and feed them to twelve carrier supply bus bars, one for each channel frequency. From these the individual modems are fed through protective resis- tances so that an accidental short circuit on one of the modems will not cut off the carrier supply to the others. The hybrid coils permit the two generators to be connected so that either can feed into the same bank of channel carrier supply filters without being reacted upon by the other. No switching is required when changing from regular to emergency supply. BELL SYSTEM TECHNICAL JOURNAL |L CABLE CARRIER TELEPHONE TERMINALS 115 H The 120-kc. carrier which feeds the group modulators of ten systems ^r a total of one hundred and twenty talking channels must be very- dependable. Therefore separate filters are used for the regular and emergency supply and separate amplifiers for the large power required by group modulators. Regular and emergency distributing buses are provided. Each group modulator and each group demodulator is wired through protective resistances to the regular bus and through another set of protective resistances to the emergency bus. With this arrangement an accidental short circuit even across one of the busses or across one of the output coils of one of the 120-kc. amplifiers will not stop the whole supply of 120 kilocycles. The 4-kc. oscillator of the emergency generator is in constant operation so that when it is needed no time is required to start it, but the grid bias on the second tube is held above its cutoff value by the automatic transfer circuit. This prevents the 4 kc. from going further until called for in an emergency. An emergency is indicated when there is no 120-kc. supply on either the regular or emergency bus. When this happens, the copper-oxide rectifier in the transfer circuit gets no 120 kc. and so loses its rectified voltage. This triggers off one or both of the two gas-filled tubes (multipled for safety) which increases the grid bias on the control tube of the regular generator to stop its 4 kc. supply and at the same instant restores the bias to normal on the control tube of the emergency to let its 4 kc. pass through and put the whole emergency circuit into opera- tion. The keys in the transfer circuit are provided for maintenance purposes, and to return from emergency to regular operation, since the gas tube circuit is arranged to transfer automatically in only one direction. The pilot supply circuit is shown in Fig. 5. The 3.9-kc. tuning fork oscillator at the left supplies that frequency, through the three transformers, to the three copper-oxide modulators the carriers of which are obtained from the regular channel carrier supply bus-bars as shown. The three filters, which are identical with channel carrier supply filters, select the lower sidebands to be used for pilot frequencies at 64.1, 92.1 and 104.1 kc. The three pilot frequencies are distributed to the different systems through protective resistances from a bus-bar as shown. They are set 100 cycles off the carrier frequencies to obtain locations of minimum interference from carrier leak and other sources. Signaling circuits do not form an integral part of the carrier terminal equipment. Signaling equipment of a type already widely used in the Bell System for toll circuits, is connected between the toll switchboard and the four-wire terminating set of the individual channel. 116 BELL SYSTEM TECHNICAL JOURNAL 3.9KC OSCILLATOR I 66 KC I FILlERS J 96 KC I ■A^/v^ 64.1 KC K^^ 92.1 KC kW-' 104.1 KC I — ^^A^- r\AAr ---■-^vAA/— \ SYSTEM 1 (^J\ A A \ OTHER SYSTEMS Fig. 5 — Pilot supply circuit. Transmission Performance In general, the performance requirements set down as objectives in the development of this system were based on the assumption that five carrier links operating in tandem and over a 4000-mile circuit should give satisfactory, high-grade service. \ FIVE CHANNELS IN TANDEM EACH WITH VOICE TERMINATING SETS \ / _y _^ } J ■ ^ 10 9 o z g > z LU < 1 •| 1 J_ ONE CHANNEL WITHOUT VOICE TERMINATING SETS LIMITS WITHIN WHICH ALL CHANNELS FALL \v ' 1 1 1 \ / y 1 \ ---• — -_-- V — ~ .-__ '" 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 3.6 FREQUENCY IN KILOCYCLES PER SECOND Fig. 6 — Channel frequency characteristics. t I CABLE^CARRIER TELEPHONE TERMINALS 117 The channel frequency characteristic which has been attained in the terminals is shown in Fig. 6. The solid curve below shows the fre- quency characteristic of a representative channel, while the dotted curves near it show the limits within which the characteristics of all single channels, so far measured, would fall. Above in the figure is shown the characteristic of five representative channels in tandem, each channel having its two voice terminating circuits included. The delay distortion and time of transmission, contributed by all terminal apparatus at both ends of a system except voice terminating sets, are shown in Fig. 7 for a single channel. I V) 6 Q z O 5 O ^ LU (rt _J S 3 z a. 2 5 ^- I \ \ ONE CHANNEL V \J V ^ \ . ^^ ^ 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 FREQUENCY IN KILOCYCLES PER SECOND 3.2 3.4 3.6 Fig. 7 — Delay characteristic. The channel modulators have been adjusted so that they will cut off the peaks of excessively loud talk to prevent overloading the carrier repeaters or other parts of the circuit, but this cutting is not enough to degrade the quality of speech. The single-frequency load curve of one complete channel is plotted in two ways in Fig. 8. The frequency stability of the oscillating tuning forks is expected to be within ± 1 X lO"*^ parts per degree Fahrenheit on all systems, with negligible variations due to other causes. The amplitude stability of each frequency at its distributing bus is expected to be within ± 1/4 db over a period of months. The impedances of the bus-bars are suffi- ciently low so that crosstalk from one system into another through this path is unimportant. The effectiveness of the protective resistances at the carrier supply bus-bars is such that a short circuit on one modu- lator or demodulator will increase the loss in the remaining modulators and demodulators less than 1/2 db. The speed of switchover to emer- gency carrier supply is such that the disturbance to transmission will be less than 10 milliseconds. The effect on speech is not detectable. 118 BELL SYSTEM TECHNICAL JOURNAL i=! 12 _l2 lij _ 10 / / / / / -^ y / r / / / 2 4 6 8 10 12 14 16 18 20 22 lOOO-CYCLE INPUT (O LEVEL) IN DECIBELS ABOVE 1 MILLIWATT Fig. % — Channel load curve at 1000 cycles. Maintenance Features Since the type "K" system provides more circuits in a group than ever before, it is essential that appropriately better maintenance facilities be furnished. Wherever vacuum tubes are used, jacks have been included to permit testing the condition of the tube by plugging in a new type of test set. The testing of a working tube with this set will not produce an appreciable reaction on performance of the circuits involved. When it has been determined that a tube in the common equipment is nearing the end of its useful life, a special transfer cord circuit is used to remove the circuit involving the tube from service and to substitute a spare circuit temporarily while the defective tube is replaced. This transfer from a regular to a spare and vice versa can be made without effect upon service. CABLE CARRIER TELEPHONE TERMINALS 119 In a type K terminal office transmission tests are made at the four- wire test board shown in Fig. 9, A, where the incoming and outgoing voice-frequency circuits appear, and at the high-frequency test board shown in Fig. 9, B. At the former, four-wire talking, monitoring and testing circuits have been provided for voice-frequency maintenance. Fig. 9 — Test positions: "A" — voice frequency; "B"- — high frequency. Adjustment of the equivalents of the individual channels can be made from this point as previously described. Much can be done from this position by means of monitoring and noise measuring to diagnose troubles. At the high-frequency testboard the circuits are brought through jacks and high-frequency measuring apparatus is provided. Measure- 120 BELL SYSTEM TECHNICAL JOURNAL ments can be made on operating systems to determine the performance of the intermediate repeaters and regulators with respect to level and equalization over the frequency range from 12 to 60 kilocycles. Loss and gain measurements also can be made either between this point and the voice-frequency four- wire test board, through the carrier terminal equipment or through the next adjacent repeater or terminal office at high frequencies. It is possible to test the high-frequency portion of the terminal and to substitute a spare, by patching or rapid transfer, for a defective or potentially defective group modulator, transmitting amplifier, group demodulator or receiving amplifier. Some of the high-frequency testing equipment is shown mounted on the middle bay of Fig. 9, B, of which one of the most important units is the 1 to 150-kc. test oscillator located at the center of this bay. It is a heterodyne type of oscillator which covers the frequency range with a continuous film strip scale about 300 inches long. Its maximum output is about one watt and this varies less than 1 db over the entire range. It is provided with built-in calibrating features and can be set to any frequency with an absolute accuracy of about 25 cycles. It is used as the tuning control of the pilot level measuring circuit. An auxiliary scale on the oscillator permits tuning the measuring circuit directly in terms of frequency. The pilot level measuring circuit is of the double heterodyne type and includes a copper-oxide modulator which is supplied with carrier from the heterodyne oscillator, an intermediate frequency 130-kc. crystal filter of 10 cycles band width, a high-frequency amplifier for this frequency, a copper-oxide demodulator supplied with carrier from a 129-kc. fixed frequency oscillator, a voice-frequency amplifier and calibrating circuit. The input impedance of the measuring circuit is high so that when it is bridged across a line pair at the high-frequency testboard jack fields it does not produce appreciable loss to the line. The circuit permits measuring each of the three pilot frequencies to check levels and equalization of operating systems. The panels com- prising the circuit are shown below the oscillator in Fig. 9, B. Mounted on a shelf just below the oscillator is the transmission measuring set, which contains a highly accurate thermocouple and meter combination with calibrating circuits, wide range repeating coils, a test key circuit, and attenuators, one of which can be set in steps of 1 db up to a total of 90 db. Equipment Features Because of the large number of systems likely to be terminated in an office, the jacks are concentrated in a group of bays located together CABLE CARRIER TELEPHONE TERMINALS 121 for ease in patching and testing. There are in general five major divisions of the terminal equipment consisting of channel modem bays, group modem bays, carrier supply bays, high-frequency testboard, and four-wire voice-frequency patching board and associated voice termin- iiiiniiiifliiiiii fi'iiiiiwiwiiiiiii: iwiiiiiiininiiiii imurm Fig. 10— Cabling of high-frequency jack bay. ating equipment. The general arrangement in an office is such as to simplify the cabling between various groups of equipment The cabling of a high-frequency jack bay, shown in Fig. 10, illustrates the congested wiring condition occurring when a large number of heavy 122 BELL SYSTEM TECHNICAL JOURNAL shielded wires is run to one location. Because of this congestion the jacks in this bay are mounted or removed from the front. The concentration of equipment in the modem unit is made possible by the small size of the copper-oxide bridges and the filters. Fig. 11,^, shows twelve modem units for two systems on adjacent bays with space left at the bottom for the six modems of a third system. Fig. 11 — Carrier equipment bays: "A" — channel; "B" — group; "C"- pilot supply. :arrier and The group modems are about the same size as the chanijel modems and include a modulator, a demodulator, a transmitting amplifier, and auxiliary receiving amplifier with associated filters. Fig. \\, B, shows three units for three systems with space for six additional units at the bottom of the bay. The carrier supply equipment for ten systems is mounted in one bay as shown in Fig. 1 1 , C, which includes the regular and emergency gen- erators, transfer unit, distributing equipment and pilot channel supply CABLE CARRIER TELEPHONE TERMINALS 123 panel. The bays of this type are located near their associated channel equipment because the supply is chiefly for channel modems. One carrier distributing unit provides for the even and another for the odd harmonics. All terminals and bus bars of these units which are com- mon to the ten systems are protected by insulating covers. The four-wire voice-frequency jacks for all the systems in an office will ordinarily be grouped in associated bays, one of which is shown on Fig. 9, A. A bay will accommodate five systems as an average, that is, 60 voice circuits including the necessary pads and telephone set. The high-frequency testboard is an arrangement of sealed test terminals, high-frequency patching jacks and high-frequency testing equipment mounted on bays as shown in Fig. 9, B. Only a few high- frequency patching jacks were required initially and these were there- fore mounted above the sealed terminals. This arrangement of bays with the addition of a high-frequency patching bay at the right of each sealed terminal bay will accommodate 100 systems. The carrier pairs are split off from the main toll cables at splices in the cable vault. The input circuits are carried thence in lead covered cable to the cable crosstalk balancing bays and thence to the input sealed terminal. The output pairs run directly from the output sealed test terminal to the splice in the cable vault. The remaining high-frequency wiring from rack to rack is shielded wire. Conclusion The carrier telephone terminals for the type "K" system which have been described are simpler, occupy less space and provide better trans- mission performance than multi-channel carrier terminals used pre- viously in the Bell System. As part of a general development of broad-band transmission systems, it is very desirable to employ equip- ment which can be used in common with several systems. The 12- channel bay, much of the carrier supply and all of the voice-frequency terminating equipment of this type "K" system terminal will be used to form corresponding parts of the terminals for the 12-channel open- wire system and the coaxial system, both of which are under develop- ment. This not only has simplified the development work, but also will result in greater mass production of these common parts and pro- vide desired uniformity of voice-frequency terminating levels and maintenance arrangements. References 1. "Carrier Current Communication on Submarine Cables," H. W. Hitchcock, Jour, of A. I. E. E., October 1926; Bell Sys. Tech. Jour., October 1926. 124 BELL SYSTEM TECHNICAL JOURNAL 2. "New Key West-Havana Carrier Telephone Cable," H. A. Affel, W. S. Gorton and R. W. Chesnut, Bell Sys. Tech. Jour., April 1932. 3. "Carrier Telephone Systems for Toll Cables," C. W. Green and E. I. Green. Presented at the 1938 Winter Convention of the A. I. E. E. Published in this issue of the Bell System Technical Journal. 4. "Crystal Channel Filters for the Cable Carrier System," C. E. Lane. Presented at the 1938 Winter Convention of the A. I. E. E. Published in this issue of the Bell System Technical Journal. 5. "Magnetic Generation of a Group of Harmonics," E. Peterson, J. M. Manley and L. R. Wrathall, Electrical Engineering, August 1937. Crystal Channel Filters for the Cable Carrier System * f By C. E. LANE SINCE the channel selecting filters used at the terminals of the twelve-channel cable carrier system are the principal filters in the system this paper is concerned primarily with these. Their importance is evident from the fact that they represent over one-third of the cost of the system terminals. Many new features appear in these channel filters. The most outstanding is the use as filter elements, along with inductance coils and condensers, of plates cut from crystalline quartz. It is for this reason they are called "crystal filters." In addition, however, the inductance coils, some of the condensers, and also the filter assemblies have in them new features. Only after a number of years of laboratory experimentation with filters using crystal elements, studying their advantages and limitations, was the cable carrier system planned to use such filters. There are twelve channel filters which transmit the lower side bands derived from the modulation of the speech signals with carrier fre- quencies spaced 4 kilocycles apart from 64 to 108 kilocycles. An insertion loss frequency characteristic which applies for each of the twelve filters is shown in Fig. 1. Regarding a 10 db loss increase as the cut-off as compared with transmission at 1000 cycles, the voice- frequency band for a single-carrier link, largely determined by the characteristics of the channel filters, extends from approximately 150 to 3600 cycles. For five links the band extends from about 200 to 3300 cycles. This is a 600 or 700 cycles wider frequency band than the present three-channel open-wire carrier system. The maximum delay distortion in the transmission band of each of the filters is about 0.4 millisecond. As many as ten of these filters may appear in tandem in the longest talking circuits. The total delay distortion in such cases would then not exceed 4 milliseconds. This is not objectionable since the average listener can not observe the effect of delay distortion unless it exceeds about 10 milliseconds. A representative filter schematic is shown in Fig. 2. The condenser shown by the dotted line at the left is used only in the two lowest frequency filters to obtain * This is a companion paper to other papers covering different parts of the twelve- channel cable carrier system. t Presented at Winter Convention of A.I.E.E., New York, N. Y., Jan. 24-28, 1938. 125 126 BELL SYSTEM TECHNICAL JOURNAL 90 80 70 m 60 50 40 30 20 1 h 1 j \ > ^ ^ \h v V. \ f V ___y -4 -3 -2 -I 0 +1 +2 FREQUENCY FROM CARRIER IN KILOCYCLES + 5 Fig. 1 — When plotted in cycles removed from the carrier frequency, the insertion loss frequency characteristics of each of the twelve crystal channel filters are for all practical purposes identical. o^W> oviik> ^T^j^r^^w^ ^JMLr^^Jm/ -^W^- BALANCED I INDUCTANCE /WV) vQikl^ CRYSTAL ELEMENT MECHANICALLY ONE ELECTRICALLY TWO Fig. 2 — The schematic circuits of each of the twelve filters are the same except for the addition of the condenser shown by the dotted lines which appears only in the two lowest frequency filters. CRYSTAL CHANNEL FILTERS 127 an impedance transformation internal to the filters and thereby permit the use of crystals of practical thicknesses for these filters. However, the equivalent circuit for each of the twelve filters is the same. In the system the filters work in parallel at one end and between terminating impedances of 600 ohms. The two condensers appearing in the series arms at the left end of the filter schematic are used in obtaining satisfactory operation of the filters in parallel and otherwise might be omitted provided the inductance at this end was made smaller at the same time. Figure 2 indicates the separate physical elements and the manner in which these are connected in the filters. In considering the per- formance of the filters the crystal elements are replaced by their equivalents, an inductance and capacitance in series, shunted by a second capacitance as shown in Fig. 3. Also, the condensers in o-YA^WT^ cwWV^15MH AAArO WW5 Fig. 3 — The schematic circuits of the filters are each equivalent to two lattice sections with a resistance pad between them, resistances at each end and at the paralleling end a coil and condenser which resonate at the mid-band frequency. shunt across the filter are shown inside the lattice combined with the direct capacitance of the crystals and the inductances are relocated in series in each lattice arm. In making this conversion, however, the eff"ective resistance of the inductance coils are, for reasons which will appear later, shown remaining outside the lattice. Also the capacitances and the portion of the inductance which are used solely for purpose of paralleling are left outside the lattice. The basis for the conversion from Fig. 2 to Fig. 3 is shown in Fig. 4. Before considering further the filter as a whole, the nature of the crystal elements and the reason for using them will be considered. It is common knowledge to those familiar with the performance of electrical wave filters that the energy loss unavoidably associated 128 BELL SYSTEM TECHNICAL JOURNAL with inductances imposes limitations upon the filter characteristics obtainable. Capacitances may be designed so that the energy dis- sipation is small and negligible as compared to that in the inductances. With ideal reactance elements entirely free from dissipation, filters might be designed for any band width with as little loss in the band as wanted and at the same time frequencies might be rejected outside the band by any amount desired, no matter how near such frequencies c^VW Fig. 4 — It does not alter the transmission properties of a network such as shown in Figure 4A to remove the impedances in shunt and outside the lattice and replace them by impedances of equal magnitude in shunt across each lattice arm nor by removing the impedances in series with the lattice and replacing them by series impedances inside the lattice of twice the magnitude of those removed. were to the edges of the transmitting band. Of course the sharper the filter cut-offs, other requirements being the same, the more complex the filter structure would be even neglecting dissipation. The greater the dissipation in the filter elements, the greater the loss in the trans- mitting range of the filters and the greater the number of cycles required for this loss to rise from the relatively low and uniform loss in the transmitting band to the high loss wanted outside the band. In the design of channel filters for carrier systems, the presence of CRYSTAL CHANNEL FILTERS 129 dissipation in the filter elements is costly in that the channels must be spaced farther apart than would otherwise be necessary, thereby wasting frequency space. At the same time the loss to transmitted frequencies must be made up for by amplification. The amount of dissipation in a reactance element is measured by the ratio of the effective resistance component of its impedance to the reactance com- ponent at any frequency. The reciprocal of this ratio is called the Q of the reactance and hence is a measure of efficiency or freedom from dissipation. In the design of inductances in the form of wire wound coils, it is generally not practical to obtain Q's much in excess of 200 or 300 at any frequency. The quartz crystal element used in the filters as previously stated is equivalent electrically to a two-terminal reactance consisting of an inductance and capacitance in series shunted by a second capacitance. For the Q of the inductance in the equivalent circuit of the crystal element a value of 15,000 or more can readily be obtained. It is for the purpose of utilizing this high Q inductance and obtaining the benefits therefrom that crystal elements are used in these filters. 64 68 72 76 80 84 88 92 96 100 104 108 FREQUENCY IN KILOCYCLES PER SECOND Fig. 5— The length of the crystal elements used in the different filters varies about inversely as the frequency of the filter band location. The filter schematic in Fig. 2 shows crystal elements in each filter section; the two in the lattice arms and the two in the series arms in each case are identical. Electrically there are four crystals in each section but for reasons of economy and for convenience in handling and adjusting the crystals those in corresponding arms are physically one. This is possible since the filter is a balanced structure and the two like crystals vibrate in unison. Figure 5 is a photograph which shows a representative double crystal element taken from each of the twelve filters. The four crystals in the lowest frequency filter range from 40.2 millimeters to 41.8 millimeters in length and those in the highest frequency filter from 23.8 to 24.3 millimeters. The thickness of the crystals in all four of the lowest frequency filters are 0.63 mi- limeters, in the next four filters 0.82 millimeters, and in the highest frequency filters 1.1 millimeters. Uniformity in thickness is main- tained as far as practicable since it contributes to economy in manu- 130 BELL SYSTEM TECHNICAL JOURNAL facture of the crystal. Within the range using the same crystal thickness the impedance and frequency differences, called for by the design of the different filters, can be provided by variations in width and length of the crystals. The ratio of width of the crystals to their length ranges from about 1/2 to 4/5. The major surfaces of the crystals are plated with a thin layer of aluminum deposited by an evaporation process. This plating is divided along the center line lengthwise of the crystals to form the two ELECTRICAL AXIS = X MECHANICAL AXIS=Y OPTICAL AXIS = Z Fig. 6 — The crystal elements are out with their major surfaces perpendicular to the electrical axis of the natural quartz, with their side surfaces making a small angle with the mechanical axis, and with their end surfaces making a small angle with the optical axis. electrically independent crystals. Since the crystals vibrate longi- tudinally with a node across the middle, they are clamped at this node in mounting. Figure 6 shows the orientation of the crystal plates with respect to the natural axes of the quartz from which they are cut. The plates are cut as accurately as is practicable to the dimensions computed making a small allowance in length and then the crystal is finally adjusted in an electric circuit by grinding the end of the crystal until the resonant frequency falls within five or ten cycles of that desired. CRYSTAL CHANNEL FILTERS 131 Considering again the filter schematic as a whole (Fig 3) and neglecting the dissipation in the crystals and condensers, the filter may be regarded as made of two lattice filter sections having ideal reactance elements, that is, elements free from dissipation. The location of the effective resistance of the coils outside the lattice, for purpose of per- formance analysis, shows how at the end of the filter these resistances may be regarded as part of the terminating impedance between which 2500 2000 1500 1000 -1500 -2000 -2500 -3000 1 I 1 / ^'^ / / 1' / /. / y f ! y / /I < / / / V / / / r j 1/ r / / / / Zx ^ / / / / X / / / / / / ..^. . / / / / / ; I / I I II jl 1 1 1 1 -7 -6 -& -4 -3 -2 -I 0 +1 +2 +3 +4 -1-5 FREQUENCY FROM CARRIER IN KILOCYCLES Fig. 7 — In a lattice filter section transmission occurs for frequencies where the impedances for the two arms of the lattice are of opposite sign and attenuation peaks of very high loss occur where the impedances cross. the filter works, and how between the filter sections the resistances may be combined with a shunt resistance to form a resistance pad which matches the image impedance of the two filter sections. The effect, then, of the coil resistances is primarily to provide a flat loss over the entire frequency range and does not affect appreciably the shape of the loss characteristic furnished by the reactance inside the lattice sections. 132 BELL SYSTEM TECHNICAL JOURNAL In considering the performance of lattice type filter sections, it is common practice to sketch together the frequency reactance curve of the two lattice arms Z^ and Zy. This is done for one of the filter sec- tions and is shown in Fig. 7. In the frequency range where the two curves are of opposite sign the filter transmits, and where they are of 1600 1000 10 600 5 O ■Z 400 2 -200 -400 -600 -1000 y / / / « / / / REACTANCE 1 1 1 A ^ \ 1 1 1 \ 1 1 1 / RESISTANCE \ 1 ; / 1 1 1 / / / / / / f / REACTANCE / / / / / -7 -6 -5 -4 -3 -2 -I 0 +1 +2 +3 +4 + FREQUENCY FROM CARRIER IN KILOCYCLES Fig. 8 — The large reflection losses occurring within the transmission band of the filters and near the edges of this band has the effect of narrowing the width of the transmission band. the same sign there is attenuation. At the point where the two curves intersect there are attenuation peaks of very high loss. The reactance curves of Fig. 7 are for the filter section accountable for the pair of attenuation peaks shown in the filter characteristic which are the closer to the edges of the transmitted band. For the other section the cross- CRYSTAL CHANNEL FILTERS 133 over points of the two reactance curves are farther away from the band, since this section is responsible for the outer pair of attenuation peaks. The design of the filters consisted in determining values for the in- ductance coils, condensers, and crystals, such that the reactance curves of the lattice arms of the filter passed through infinity and intersected with each other at the desired frequencies and, at the same time, in determining the impedance level for all of the elements such that the filters would have the right image impedances. The curves of Fig. 7 would seem to indicate a somewhat greater band width for these filters than shown by the insertion loss characteristic of Fig. 1. The reason for this can best be explained by referring to the image impedance of one of the filter sections as shown in Fig. 8. Within the band the image impedance is, of course, a pure resistance which varies with frequency. It is about 800 ohms at mid-band frequency and falls rapidly to zero near the edges of the band. Assum- ing the effective resistance of the coils, which is about 100 ohms, as belonging to the terminating impedances, the filter sections actually work between impedances of about 700 ohms. This means that large reflection losses occur at each end of each filter section near the edges of the transmission band where the image impedance of the filter is very small. It is these reflection losses that are responsible for the actual transmission band being much narrower than it would be with the filter sections terminated in their actual image impedances. The filter sections are designed with 800 ohms image impedance at mid- band frequency instead of 700 ohms to make the band flatter and somewhat wider than it would be otherwise. When a number of band filters are operated in parallel it is generally necessary to connect across the paralleled end a two-terminal network to correct for the distortion that would otherwise be present in the highest- and lowest-frequency filters in the group. A circuit of the network used for this purpose with the channel filters is shown in Fig. 9. The filters employ crystal elements in order to obtain abrupt dis- crimination between wanted and unwanted frequencies and at the same time to secure low and uniform loss in their transmitting bands. This characteristic must not only be obtained at the time the filters are assembled and adjusted but must be maintained throughout the service life of the filters and not appreciably affected by temperature variations. This imposes severe stability requirements upon the elements used in the filters. The crystal elements themselves are very stable when properly designed and once adjusted will retain at a given temperature their frequencies of resonance within one or two cycles seemingly indefinitely. Their temperature coefficient is only 134 BELL SYSTEM TECHNICAL JOURNAL about twenty-five parts per million per degree centigrade, which is not objectionable. The obtaining of inductance coils and condensers that were adequate in stability for use in conjunction with the crystals required consider- able development effort. The inductance coils were required to have not only a high degree of stability with respect to temperature and time but also a high ratio of reactance to effective resistance, low modu- lation, and at the same time be small in size. Air core coils might have been designed for the purpose but they would have been quite large. The coils used are of the toroidal type wound on about one and three- fourths inch rings of molybdenum-permalloy. To reduce eddy current losses the cores are made of very fine powder and then annealed to reduce hysteresis losses. The particles are mixed with insulating 1 o ^m^ CWT\ HH Fig. 9 — A two-terminal reactance network is connected in shunt across the filters at their paralleling end to improve the characteristic of the highest and lowest fre- quency filters. material and formed into rings by extremely high pressure. The inductance of the coils has a temperature coefficient of less than 40 parts per million per degree centigrade. The cores of the coils for the higher-frequency filters are wound with finely stranded wire to help secure good Q's (about 225). Because of the high impedance of the coils called for by the filter design, care is taken to make the capacitance between the windings and the core and between the windings and the case as low as practicable and also to make stable all such small capacitances as must be present. The two extra condensers used at one end of each filter for paralleling purposes are of a high grade mica type. The other condensers are all quite special. The fixed ones, ranging in magnitude from about 7 mmf to 100 mmf, are made by plating short lengths of high grade glass tubing inside and outside with silver. Because of the intimate associ- ation of electrodes with the surfaces of the tubes and the low expansion coefficient of the glass used, a condenser is obtained that has a tem- perature coefiicient comparable with that of the coils and crystals. No aging effect has been observed. It will be noticed that four small CRYSTAL CHANNEL FILTERS 135 adjustable air condensers appear in each filter section. These are used to secure precise initial adjustment of the filter capacitances. To protect the filter elements against moisture, the filters are hermetically sealed in a container made from a rectangular section of seamless brass tubing with closely fitting plates soldered in each end. One end plate carries four metal-glass seal terminals and a nozzle through which dry air is blown after the filter is assembled. The other end plate is provided with a small hole for the escape of the drying air. After the drying operation the hole in the nozzle and the hole in the opposite end of the filter are closed by soldering. The elements that make up the filter are assembled on a chassis which is completely wired and then slid into the container in assembly. Figure 10 is a photograph showing this chassis and the arrangement of the \ Fig. 10 — The filter parts are assembled and wired on a chassis which is slid as a unit inside the filter container. elements. The elements are located in such a way as to use very short wiring connections which reduce the magnitude of any stray admit- tances. The wired filter chassis is carefully adjusted by setting the values of the air condenser such that for each filter section the reson- ance frequencies looking into each end of each section occur where they theoretically should. This compensates for the effect of small capacitances between the filter parts. In the design of the filter parts care is taken to use no material which absorbs moisture readily since such moisture would later be released and raise the relative humidity of the air inside the filter. If a potential much in excess of about twenty volts is applied across crystal elements at frequencies near resonance, the crystals will break from the mechanical strain of their vibration. The maximum safe voltage across the channel filters at the resonant frequencies of the 136 BELL SYSTEM TECHNICAL JOURNAL crystals is considerably greater, however, since at resonance the full voltage does not appear across the crystals. In normal use the voltages across the filters will be very much less than twenty volts. Other filters forming part of the terminal apparatus are the group modulator and group demodulator low-pass filters, the channel and group carrier supply filters, and the pilot supply filters. The group modulator and demodulator filters are of the low-pass type employing only coils and condensers as elements. The group carrier supply filter is the same in schematic and mechanical design as the crystal channel filters described. The pilot supply filters and the channel carrier supply filters are equivalent in schematic to one section of the channel filters; but of course they are only about half the size and are hermetically sealed in the same manner. References For a further discussion of crystal filters the reader is referred to "The Evolution of the Crystal Wave Filter" by O. E. Buckley, Journal of Applied Physics, October 1936, and "Electrical Wave Filters Employing Quartz Crystals as Elements" by W. P. Mason, Bell System Technical Journal, July 1934. Crosstalk and Noise Features of Cable Carrier Telephone System * By M. A. WEAVER, R. S. TUCKER and P. S. DARNELL CROSSTALK and noise are important factors in cable carrier transmission as outlined in the paper "A Carrier Telephone System for Toll Cables" by Messrs. C. W. Green and E. I. Green. Crosstalk and noise limit the number of carrier channels which can be utilized in any one cable, not only by limiting the number of channels which can be placed on a single pair, but by limiting the number of pairs which can be used. Noise also controls the transmission loss which can be permitted between repeaters. Without the crosstalk and noise reduction measures described in this paper, the number of carrier channels per cable would be so few and the spacing between re- peaters so short, that the type K carrier system would be impracticable. Crosstalk To utilize existing toll cables in the Bell System for frequencies up to 60 kilycycles required the solution of many new crosstalk problems because: (1) Crosstalk increases rapidly with the frequency, (2) Non- loaded carrier pairs due to their high speed of propagation are especially suitable for very long distances and hence the crosstalk requirements per unit length are relatively severe, (3) The large gains of the carrier repeaters amplify certain crosstalk currents much more than in the case of voice frequency circuits. Two general effects need to be considered: intelligible crosstalk must be prevented; and, a large number of circuits crosstalking into a par- ticular circuit must not contribute an undue amount of noise. The second effect is called babble, since it consists of a multiplicity of un- related voice sounds which, in the aggregate, are unintelligible. An important feature is the use of different cables for opposite direc- tions of transmission. This makes the major crosstalk problem the reduction of crosstalk between pairs in the same cable used for trans- mission in the same direction. The crosstalk currents due to trans- mission at one end of a disturbing circuit through the distributed couplings with a disturbed circuit tend to arrive at the distant end at the same time since the currents via any of the couplings travel sub- * Presented at Winter Convention of A. I. E. E., Jan. 24-28, 1938. 137 138 BELL SYSTEM TECHNICAL JOURNAL stantially the same distance. This makes it possible to greatly reduce the total effect of these distributed couplings by the use of small ad- justable mutual inductance coils connected between pairs at one point in each repeater section. If nothing more were done, there would still be objectionable cross- talk since currents from the outputs of carrier repeaters could crosstalk into voice frequency circuits and these circuits could then again cross- talk into other carrier frequency circuits at points near their repeater inputs. This effect is minimized by transposing the carrier pairs from one cable to the other at carrier repeater points. At common voice frequency and carrier frequency repeater points there would be an unsatisfactory crosstalk path from a carrier repeater output into all the wires not used for carrier frequencies and from them through coupling between ofilice wiring into similar wires in the other cable and finally into carrier repeater inputs in the second cable. This crosstalk is minimized by the use of carrier frequency suppression coils in the voice frequency circuits. These coils also serve the purpose of preventing carrier frequency noise originating in voice frequency cir- cuits from being transmitted into the cables and inducing noise at points near carrier repeater inputs. Near-End Crosstalk Near-end crosstalk is the result of coupling between circuits trans- mitting in opposite directions, while far-end crosstalk is the result of coupling between circuits transmitting in the same direction. Near- end crosstalk coupling between different carrier circuits of the same frequency must be kept very small, particularly near a repeater point, since crosstalk from the output of a repeater into an opposite directional pair near the input of its repeater will be greatly amplified by this repeater. Crosstalk between carrier circuits within the offices is kept low by careful shielding, segregation, suppression of spurious paths through battery supply, common grounding arrangements, etc. Since the type K system operates on a "four-wire" basis, different electrical paths are used for opposite directions of transmission. Satisfactory near-end coupling in the outside plant is obtained, there- fore, by placing east bound pairs in one cable and west bound pairs in another. When two cables have relatively heavy sheaths as in the larger Bell System cables, their coupling is sufificiently small even with the two cables in close proximity. CROSSTALK AND NOISE FEATURES 139 I I I I > I \^ ) C I I I I I I I I CARRIER PAIRS 140 BELL SYSTEM TECHNICAL JOURNAL Interaction Crosstalk The crosstalk currents from a carrier repeater output into voice frequency circuits in the same cable must be limited, since they cross- talk again into carrier circuits near repeater inputs and, consequently, are amplified by the high gain repeaters. Intermediate circuits most responsible for crosstalk of this type are made up of combinations of pairs and phantoms and the sheath, i.e., longitudinal paths. One case of crosstalk of this kind would occur if the same cable were used for carrier pairs transmitting in the same direction on both sides of a repeater. This is prevented by transposing carrier pairs from one cable to the other at each repeater point, as shown on Fig. 1. A second interaction crosstalk problem is encountered at the com- mon voice and carrier repeater points and involves coupling between cables as well as in the same cable. Here the coupling path is from carrier repeater outputs to intermediate circuits in the same outside cable, back into the common office over these intermediate circuits and then via office coupling to intermediate circuits in a second outside cable and from there to carrier repeater inputs connected to pairs in the second cable. Referring to Fig. 1, a set of noise (and crosstalk) suppression coils is encountered in this path. The high longitudinal circuit impedance of these coils minimizes this interaction crosstalk. Far-End Crosstalk Far-end crosstalk currents are subjected to line attenuation and amplification similarly to the main transmission currents, and do not have extra amplification as in the case of near-end crosstalk. Further- more, far-end crosstalk currents due to couplings at different points along the line tend to arrive at the distant end of the disturbed circuit at the same time. Hence a considerable portion of the far-end cross- talk over the type K frequency range, which occurs between circuits transmitting in the same direction in the same cable, can be neutralized by introducing compensating unbalances at only a comparatively few points, such as one per repeater section. The far-end crosstalk reduc- tion problem is greatly simplified because phantom circuits are not used for carrier operation. Theoretically, for the same precision of match between the im- pedances in the two directions at the balancing point, the crosstalk re- duction would be about the same whether the balancing is done at an intermediate point or at either end of a repeater section. Balancing will be done at repeater inputs rather than at an intermediate point, such as the middle, because it is practicable to obtain repeater im- CROSSTALK AND NOISE FEATURES 141 pedances matching the average Hne impedance sufficiently well so that the effectiveness of balancing is reduced only slightly. Nature of Far-End Crosstalk Coupling The coupling between two cable pairs in a short length may be represented by a mutual admittance and a mutual impedance. The former is due almost entirely to capacitance unbalance, which varies but little with frequency, so that its effect could be practically balanced out by means of a simple condenser. The latter, however, involves a complex mutual inductance of the form Ma + jMb, because of the proximity effect of the wires of a pair and of other cable conductors.^ As shown on Fig. 2, both components vary considerably with fre- -lOMb ^.i r- — "i j- ~^^' ^ Ma ^ > -< k- / J NO. 19 GAUGE QUADDED CABLE, 55-FOOT LENGTH ^ ' Mb )— < 1.0 ; 0.8 (0 U 0.6 I o 0.4 ) i 0.2 < a: 0 -0.2 0 5 10 15 20 25 30 35 40 45 50 55 60 65 FREQUENCY IN KILOCYCLES PER SECOND Fig. 2 — Mutual inductance between cable pairs in terms of value for Ma at 10 kilocycles. quency; Ma on the average decreasing as the frequency increases while Mh in the general case is of negative sign and reaches a maximum value at 56 kilocycles. Type of Balancing To obtain maximum reduction in crosstalk it would be necessary to use a condenser for balancing the mutual admittance and an inductance coil for balancing the mutual impedance or to use some equivalent complex network. Experimental balancing in a particular cable using the coil-condenser method reduced the mean crosstalk over the type K 1 "Cable Crosstalk — Effect of Non-Uniform Current Distribution in the Wires," R. N. Hunter and R. P. Booth, Bell System Technical Journal, April 1935. 142 BELL SYSTEM TECHNICAL JOURNAL range about 20 db, which is close to the maximum reduction possible with a universal type of balancing unit. The reduction is limited by the fact that two pairs having identical crosstalk couplings in each of two short lengths at different points in the cable will not produce two identical elements of crosstalk current at a circuit terminal because: (1) Cable circuits are not perfectly smooth. Reflections, as at junc- tions of reel lengths or at terminals, alter the two crosstalk currents differently, (2) The propagation constants of each circuit vary slightly from reel to reel in a random fashion and therefore the two crosstalk currents are of slightly different phase and magnitude, (3) In any short length the disturbing circuit produces crosstalk currents in inter- mediate circuits, which are propagated along these circuits and cross- talk again into the disturbed circuit at various points, producing an additional crosstalk current at the circuit terminal. At any frequency, this interaction crosstalk current has a random phase and magnitude relation to the crosstalk current for the short length considered by itself, and depends also upon the position in a repeater section of the short length. A 20 db crosstalk reduction is not required, considering the number of K systems anticipated in any one cable. Studies were made, there- fore, to determine whether satisfactory results could be obtained with a less expensive type of balancing, as outlined below. The effects of frequency and circuit impedance on crosstalk coupling are as follows: (1) Crosstalk in a short length due to capacitance and to inductance coupling increases about directly as the frequency in- creases for circuits whose impedance is independent of frequency. (2) Crosstalk due to capacitance coupling varies directly as the im- pedance of the circuits while that due to inductance coupling varies inversely as the impedance. Changing the impedance from about 800 ohms for loaded voice circuits to about 135 ohms for non-loaded carrier circuits and changing the frequency from about 4 kc. to 60 kc. increases the crosstalk due to capacitance coupling by a factor of about 2.5 and that due to inductance coupling by a factor of about 90. Capacitive coupling in existing cables was reduced by design to as great a degree as practicable, particularly for the most closely asso- ciated circuits, because it is of most importance in the loaded voice fre- quency case. These same design measures also reduce inductive coupling but not to the same extent. Capacitive coupling decreases rapidly with separation due to the shielding effect of copper in inter- vening circuits while inductive coupling is not much affected by inter- vening copper wires. To minimize magnetic coupling it is necessary to use different lengths of twist for the pairs. Existing cables have relatively few lengths of pair twists. CROSSTALK AND NOISE FEATURES 143 As the net result, capacitance coupling is no longer all important, inductance coupling at 60 kc. actually predominating by a factor of about 3 to 1 in existing cables. Capacitance balancing should, there- fore, be less effective than balancing designed to reduce the inductance coupling. Tests have shown that capacitance balancing alone gives a crosstalk reduction of about 11 db while inductance balancing alone gives a reduction of about 16 db. Since the latter reduction is suffi- cient, except possibly for small cables or special cases, the type K balancing has been designed on this basis. Far-end crosstalk currents due to the two kinds of coupling have phase relations not differing from zero or 180 degrees by more than about 15 to 40 degrees, depending on whether the upper or lower type K frequencies are considered. There is, therefore, a tendency for either type of balancing unit to annul both kinds of coupling. To obtain as much as 16 db reduction it is necessary that the fre- quency characteristic of the balancing coil simulate that of the cable (Figure 2). This was found practicable, as discussed later, by shunting the primary (or secondary) of the coil by a properly designed impedance. Size of Balancing Coil To meet the crosstalk requirement it is necessary to balance each carrier pair against every other carrier pair. If 50 carrier pairs were used, there would be 49 balancing coils connected to each pair for balancing to all the other pairs, a total of 1225 coils. For convenience, adjustable coils having the same mutual inductance range and the same self-inductance are used. Hence, the insertion loss per coil, resulting from the self-inductance and resistance of the coils, must be kept small. In addition, the self-inductance of the coil presents a problem from the impedance standpoint. To keep the impedance at any point in the balancing panel as nearly like the average cable im- pedance as practicable, the self-inductance of a series of coils must be neutralized by capacitances shunted at suitable intervals. It is very desirable, therefore, to use coils whose self and mutual inductances are no larger than actually essential. Consequently, an attempt has been made to keep the maximum crosstalk before balancing low. Due to special measures, described below, it appeared that the maximum inductance unbalance per repeater section could be kept be- low about 1.3 to 1.5 microhenries, with the possible exception of side- to-side unbalances, and trial balancing coils were designed accordingly. Crosstalk Reduction Before Balancing Changes in the original splicing are made at approximately 6000-foot intervals, i.e., at points where voice frequency loading coils must be 144 BELL SYSTEM TECHNICAL JOURNAL removed from the carrier pairs. In most existing voice frequency toll cables the 19-gauge quads were spliced as three groups, one a two- wire circuit group, one an east bound four-wire circuit group and the third a west bound four-wire circuit group. Ordinarily, the carrier pairs will be selected from the four-wire groups because these groups are usually larger than the two-wire group and since the quads within a group are spliced at random there is less chance of a large value of coupling be- tween pairs of different quads, i.e., two pairs are less apt to be recur- rently in a relation of high coupling. The carrier pairs are divided equally between the two four-wire groups, in order that the least number of four-wire voice circuits will be lost. In cables with large four-wire groups it is satisfactory to maintain the grouping arrangement on the pairs converted to carrier. In such cables, however, one four-wire group is in the center or core of the cable and the other group in the outer periphery. In order that all circuits will have about the same velocity and attenuation and be sub- jected to about the same temperature conditions for both transmission and crosstalk reasons, one (four-wire) carrier group in these cables will be spliced to the other (four-wire) carrier group and vice versa at the 6000 -foot intervals. In cables with relatively small four-wire groups, there is more chance of two pairs being recurrently in a relation of high coupling. To re- duce this chance, a special splicing arrangement has been devised for use at the 6000-foot intervals. With existing splicing the maximum coupling in cables with small groups is about 2.5 times that for cables with large groups. This ratio is appreciably reduced by the special splicing, likewise reducing the maximum mutual inductance that must be supplied by the balancing unit. The foregoing was with particular reference to crosstalk between pairs in different quads. Crosstalk between pairs in the same quad (side-to-side crosstalk) is an additional problem. A quad consists of two twisted pairs of wires which are twisted together to permit the use of voice frequency phantom circuits. Since the two sides of a quad are so closely associated, side-to-side crosstalk is generally much greater that than between pairs of different quads. The electrical size of the balancing unit, therefore, is determined by the side-to-side crosstalk, which is reduced by "poling." To apply poling, the quads are carried through as quads for an entire carrier repeater section. From measurements of side-to side crosstalk in phase and magnitude, quads in one half repeater section are chosen and spliced to quads in the other half in such manner as to partially neutralize the side-to-side crosstalk. In effect, quads in one half-sec- tion serve as balancing units for the other half. CROSSTALK AND NOISE FEATURES 145 In most existing toll cables the side-to-side capacitance coupling was reduced when the cables were installed, by means of test-splicing within the 6000-foot sections. Obviously, for poling to be effective it is necessary to operate mainly on the inductance component. The poling measurements, therefore, are made at about 1 kc. where an approximate measure of the inductance component can be obtained directly since the capacitance and inductance components of the crosstalk are at an angle of almost 90° at this frequency. Figure 3 shows the crosstalk results obtained by means of 1-kc. poling on 14 repeater sections. It has been shown that this 9 db reduction is within 2 to 3 db of the -50 -55 -60 -65 -70 -75 -80 BEFORE POLING __^ p J ^ , r- ^ ^ ^ -— --' / V ^ ^ AFTER POLING ( ^ 15 20 25 30 35 40 45 5C FREQUENCY IN KILOCYCLES PER SECOND 55 60 65 Fig. 3 — R.M.S. side-to-side far-end crosstalk per repeater section from measurements on 14 repeater sections. maximum reduction possible with much more complicated poling involving measurement and consideration of both components at carrier frequencies. After side-to-side poling, coil balancing cannot be expected to give as much as 16 db reduction in crosstalk. This is unimportant, how- ever, as long as the required reduction can be obtained more economic- ally by the combined methods rather than by balancing alone. Crosstalk Balancing Coil Since the voltage which causes the crosstalk current in the disturbed circuit may be in either a clockwise or counter-clockwise direction, the balancing device, for flexibility reasons, should be capable of establish- ing voltages in either direction. A balancing coil was developed, therefore, which in operation may be likened to that of two separate transformers with simultaneously movable cores. The primary wind- 146 BELL SYSTEM TECHNICAL JOURNAL ings are in series, as are the secondary windings, and are connected as shown in Fig. 4, for example. The relative direction of each secon- dary winding is the same, whereas the relative directions of the primary windings are reversed. With the cores in mid-position, the voltages induced in the two secondaries are equal in magnitude but opposite in phase, and the net induced voltage in the disturbed circuit is zero. As the cores are moved toward the left the respective components of the voltage induced in circuit 3-7-8-4 increase in a counter-clockwise direction and decrease in a clockwise direction, the net result being a voltage in a counter-clockwise direction. Such a setting of the balanc- 2 — Is DISTURBING CIRCUIT I ^ ■^W^ Rs AAAr SHUNT L.R ^wy^ A^^- ■VA ^mu- DISTURBED CIRCUIT Iw ' CORES Iw e ■nm^ Fig. 4 — Schematic of a simple balancing coil designed to produce a complex mutual impedance. ing coil would be used to counteract a clockwise crosstalk voltage, the amount of departure of the cores from mid-position being dependent on the magnitude of the crosstalk voltage being counteracted. Move- ment of the cores toward the right produces the opposite effect This device, disregarding any proximity effects therein and the effects of the shunt, acts to set up a net voltage e which is in phase quadrature with the disturbing current /. Hence, e — — jwml, (1) in which m is the net mutual inductance of the device. To obtain the required mutual impedance characteristic, the primary (or secondary) windings of the coil are shunted by an inductive resistance. Let the effective self-inductance and resistance of the line windings (primaries) be denoted by L and R, respectively, and the current through these windings by /«,. Let the effective self-inductance and resistance of the shunt be denoted by Lg and Rs, respectively. At balance, that is. CROSSTALK AND NOISE FEATURES 147 when no crosstalk current flows in 3-7-8-4 due to / (the disturbing current), the current /„. is Rs(R + Rs) + c^~Ls(L + Ls) (R + R,r + o^KL + LsY J ^(RsL - RLs) {R + i?.0- + 0,'iL + Ls)' _ I = La-jby, (2) (3) where a and b are, respectively, the coefficients of the real and imaginary parts of the expression. Hence, with a shunted coil the voltage induced in the disturbed circuit is: e = — joomlw = — jwima — jmh)I. (4) The mutual impedance, Zm, equals jw{ma — jmb), or the effective mutual inductance AI of the balancing coil may be written Af = Ah+jAh, (5) wherein Ala = ma and Alb = — mb. Assuming R, L, Ls and Rs to be constant with respect to frequency of current and position of the cores, it is seen from (2) and (5) that for any core setting, Ala and Afb are functions of frequency only and their ratio at a given frequency is theoretically constant throughout the operating range. To keep inductance L constant irrespective of the mutual inductance settings, the length of the coil windings, the length of the magnetic cores and their spacing with respect to the winding spacing are so related that the change in inductance of one primary (or secondary) 2 3 — 4 -WV-OKJ^ -VWvMiLr- — 5 .--6 .--7 -—8 Fig. 5 — Schematic of winding arrangement of trial balancing coil. winding caused by motion of its associated core is equal and opposite to the change caused by the movement of the core associated with the other primary (or secondary) winding. To keep R low over the type K frequency range, cores of finely powdered molybdenum permalloy pressed into a cylindrical form are used. 148 BELL SYSTEM TECHNICAL JOURNAL Because of other requirements which a balancing coil must satisfy, the winding arrangement actually employed is shown in Fig. 5. The simple device of Fig. 4 is not balanced from the standpoint of longi- tudinal crosstalk for any coil setting except that of zero mutual induc- Fig. 6 — Trial balancing coil construction. Container is 432" in length and 1/^" in diameter. tance. The Fig. 5 arrangement is such that theoretically there is no magnetic coupling between the two circuits for longitudinal currents in either one, regardless of the coil setting. Unless the capacitance be- tween primary and secondary windings can be kept very small, the resultant admittance unbalance produces crosstalk which is not com- { i CROSSTALK AND NOISE FEATURES 149 pletely balanced out when the coil is adjusted. The turns of conductor in the Fig. 5 coil are so located that this side-to-side capacitance un- balance is less than 5 micro-microfarads. The capacitances between wires of either the primary or secondary winding do not afifect the unbalance but contribute a part of the capacitance loading which compensates for the line inductance of the coils. In the actual balancing coil, shown in Fig. 6, the windings are located in channels cut in a fibre tube which is secured to a head carrying the winding terminals and a bushing through which passes the threaded brass rod supporting the two cores. Below the head are small spool Fig. 7 — Arrangement of inner w inding of trial balancing coil. forms on which the shunts are wound. Insulating material such as bakelite is used to obtain proper spacing of the two cores. The rather unusual manner in which the turns are applied is illus- trated in Fig. 7, which is a closeup view of the two wires forming the inner winding. These two wires alternately cross over each other, progressing along the axis in opposite directions of rotation. The outer winding is similarly applied. This type of winding eliminates all splices within the coil, removing hazards incident to interior splices. The complete coil assembly is enclosed by an aluminum container which serves the dual purpose of a shield and a convenient means of holding the coil for mounting purposes as this container fits snugly into k 150 BELL SYSTEM TECHNICAL JOURNAL an aluminum cup riveted to the assembly panel. The windings are dried and impregnated and the space between the coil assembly and container is filled with insulating compound. The mutual inductance of a typical coil varies as shown in Fig. 8 as the cores are moved. The range, with the shunts disconnected, is approximately + 1-6 to — 1.6 microhenries, which is covered in about 16 turns of the screw (a total core travel of 0.5 inch). With the shunts connected, the effective mutual inductance at a given setting becomes less as the frequency rises, the two components. Ma and Mb, varying with frequency as shown in Fig. 9. To determine the proper values of Ls and Rs for the shunt, allowance must be made for the complex mutual inductance inherent in the coil due to proximity effect within the windings. — cr ^ FREQUENCY = 1 KILOCYCLE NO SHUNTS \,, ^ \ ^ "^ . - 4 6 8 ID 12 14 l( NUMBER OF TURNS OF ADJUSTING SCREW Fig. 8 — Mutual inductance of trial balancing coil. 18 20 The series inductance of the balancing coil without shunts varies as shown in Fig. 10. As the cores are moved, the inductances of windings l-A-B-2 and S-C-D-6 (Fig. 5) behave as shown by their respective curves, one increasing as the other decreases. The sum of these two curves is shown by the dotted line, and the measured value of 1-5-6-2 is shown by the solid line. It is seen that the overall self-inductance of 1-5-6-2 is constant to within ±0.1 microhenry. The difference between the curves (about 0.1 microhenry) is caused by the slight mutual inductance existing between winding \-A~B-2 and 5-C-D-6, which is negative owing to reversed winding direction in this side of the balancing coil. The measured inductance around 3-7-8-4 would slightly exceed that obtained by adding the inductances of the two sections owing to positive mutual inductance between the two ends. These end effects could be reduced by greater separation of the two sets of windings, but this refinement is not necessary. CROSSTALK AND NOISE FEATURES 151 1.00 0.99 0.98 •^ < 0.92 0.91 0.89 0.87 -0.01 -0.02 -0.04 \- — --. --— .._ ' \ WITHOUT SHUNTS ""■"■ ---. •*•»,, \ \, \ - \ \ N k. SHUNTS \ s. \ \ \, s \, s . 2 ^_ -0.10 -0.11 -0.12 -0.13 -0.14 ■" — - —J ---L.,. ^^^ SHUNTS *■' "--. ***^ s. \ \ \ \ V \ WITH \ ^ ■^ ■ 10 15 20 25 30 35 40 45 50 55 60 65 FREQUENCY IN KILOCYCLES PER SECOND Fig. 9 — Variation of Ala and Mb components of trial balancing coil with frequency. 152 BELL SYSTEM TECHNICAL JOURNAL When the shunts are connected, the inductance around 1-5-6-2 is lowered slightly, and the effective resistance is increased. To simplify the capacitance loading and in order not to introduce more resistance in one cable pair than another, the balancing coil assembly is so arranged that shunted and non-shunted windings are alternately introduced into a pair. 5.5 5.0 4.5 t 4.0 O 3.5 < 3.0 2.0 1.5 SUM OF l-A-B-2 AND 5-C-D-6 ■^^^^^ rcr:^ ■ 1 "^:rr: ^^^ »^ MEASURED 1-5-6-2 SHUNTS OPEN ___. "*■■»»- 5-C -D-6 "v \ ."' ^.-" ,-'*" "-^. -:.-r B-2 --''" " *v ^^- * — - - 0 2 4 6 8 10 12 14 16 18 20 NUMBER OF TURNS OF ADJUSTING SCREW Fig. 10 — Series inductance of non-shunted trial balancing coil. Balancing Panels In assembling the balancing coils on panels, the same number of coils should be traversed on each of two pairs before reaching the coil that balances these two pairs, in order that the phase shift up to this balancing coil on one pair will be essentially the same as that on the other pair. If these phase shifts differed materially, the coil setting for minimum crosstalk when one pair is the disturbing circuit might be quite different from the best setting when the other pair is the dis- turbing circuit. To obtain this equality objective a "criss-cross" ar- rangement, as shown schematically on Fig. 11, was devised, whereby the number of coils on one pair up to a particular balancing coil never differs by more than one from the number of coils on the other pair up to this same balancing coil. CROSSTALK AND NOISE FEATURES 153 For economic reasons it is undesirable to install a complete panel for the ultimate number of pairs, possibly 100 in some cases, but rather to install sections conforming more closely to the circuit growth. The placing at different times and properly connecting of sections ob- tained from the 100-pair criss-cross panel and at the same time main- taining service on operating circuits appeared rather formidable. This problem was solved by the use of two types of criss-cross panels; an 4-6 .2-4 1-2 1-2 1-3 3-5 5-7 7-9 9-11 11-13 13-15 i5-l7 17-19 BALANCING POSITION FOR PAIR COMBINATION INDICATED Fig. 11 — Schematic of criss-cross wiring for 20-pair balancing panel, designed to maintain phase equality of coils. intra-group panel for balancing within one group of carrier pairs and an inter-group panel for balancing pairs in one group against pairs in a second group of equal size. In the present design, an intra-group panel takes care of 20 pairs (190 combinations) and an inter-group panel of the 400 combinations between two 20-pair groups. To maintain phase equality through a number of panels, it is necessary to install them following a definite pattern. Figure 12 shows a suitable pattern for the 15 panels required for 100 pairs. 154 BELL SYSTEM TECHNICAL JOURNAL In the criss-cross scheme (Fig. 11) the side-to-side combinations, which are those marked 1/2, 3/4, 5/6, etc., appear twice, i.e., along the left and right edges of the panel. Advantage of this is taken by instal- ling balancing coils at both locations. This is done because one side- to-side coil of about 1.3 microhenries may not be large enough in all cases in spite of the fact that the mean side-to-side crosstalk has been reduced 9 db by poling. 15 BALANCING PANELS FOR 5 GROUPS NOS. 1,2,3,4 AND 5 AS INDICATED- (EACH GROUP 20 PAIRS, 100 PAIRS TOTAL) I AGAINST 3 3 AGAINST 5 2 AGAINST 3 3 AGAINST 4 4 AGAINST 5 I AGAINST 5 1 AGAINST 2 2 AGAINST 5 2 AGAINST 4 Fig. 12 — Allocation of balancing panels designed to maintain phase equalization of coils at all stages. Panels with suitable cross-connections between them are in- stalled in following order. For first group — Install 1 Add second group — Add 2, and 1 against 2 Add third group — Add 3, 1 against 3, and 2 against 3 Etc. Balancing Procedure As stated above, the far-end crosstalk in a repeater section can not be balanced out completely over the frequency range with a single bal- ancing unit. To determine the balanceable as distinct from the non- balanceable crosstalk, involves crosstalk measurements in phase and magnitude at a number of frequencies, using each pair of a two-pair combination as a disturbing circuit in turn. The balanceable crosstalk may then be separated from the non-balanceable crosstalk by computa- tion. Balancing by this method would be impracticable because of the time required. As a practical scheme, it has been shown that balancing at a frequency of about 40 kc. will produce satisfactory results over the type K range even though part of the non-balanceable crosstalk may CROSSTALK AND NOISE FEATURES 155 be neutralized at this frequency. This is theoretically undesirable since the crosstalk reduction at other frequencies is impaired. To prevent undue interference into operating carrier circuits when balancing, a frequency falling between the transmitted bands must be used. For this reasorv, the balancing coils are adjusted at a test frequency of 39.85 kc. and a measurement to check the suitability of the adjustment is made at 28.15 kc. Figure 13 shows the crosstalk vs. frequency before and after coil balancing by this method on three repeater sections. Additional Crosstalk Remedial Measures Although poling as well as balancing is done to reduce side-to-side crosstalk, this crosstalk is still considerably greater than the pair-to- pair crosstalk. For this reason, side-to-side crosstalk is diluted among OQ _ 0 5 10 15 20 25 30 35 40 45 50 FREQUENCY IN KILOCYCLES PER SECOND Fig. 13 — R.M.S. far-end crosstalk per repeater section from measurements on 3 repeater sections. the pair-to-pair combinations by a system of quad-splitting at repeater points. The crosstalk after balancing (Fig. 13) is considerably higher at the upper end of the frequency band than at the lower end. , Consequently, if circuits were set up to use the same channel throughout, the crosstalk in the upper-frequency channels would be materially greater than that in the lower-frequency channels. In order that all circuits may be equally satisfactory from the crosstalk standpoint, a system of special channel assignments in successive intervals, say 500 to 1000 miles, can be used. This will tend to equalize both the crosstalk and the noise on all circuits, thus permitting a somewhat cheaper design then if each channel had to meet the crosstalk and noise limits by itself. 156 BELL SYSTEM TECHNICAL JOURNAL Noise Besides babble, many other sources of noise need to be considered in cable carrier design. Figure 14, which shows the approximate mag- nitude of several of these if no means are taken to suppress them, indicates the noise at the end of a single 17-mile repeater section when 50 30 2 20 Cj 15 -5 __ - E ^ ^ ^ ^ ^ ^ / ^ y y D^ ^ ^ ^ > y ^ - y ^ ^ ^ -^ c ^ ^ ^ ^ ^ B — ^ ^ A^ ^ 10 15 20 25 30 35 40 45 FREQUENCY IN KILOCYCLES PER SECOND 50 55 Fig. 14 — Noise, prior to suppression measures, per repeater section at output of repeater whose gain equals line loss. A — Noise from thermal agitation. B — Thermal agitation plus tube noise. C — Noise from voice frequency telephone repeater office. i?^Noise from telephone and telegraph repeater office. E — Noise from heavy static on open-wire tap close to carrier repeater input. amplified by a repeater whose gain equals the hot-weather line loss. Curve A shows the unavoidable lower limit of noise, that produced by thermal agitation of the electrons in the cable conductors and the repeater.^ This amounts to about 2 X 10~^^ watts per telephone " J. B. Johnson, Phys. Rev., 32, 97 (1928); H. Nyquist, Phys. Rev., 32, 110 (1928). CROSSTALK AND NOISE FEATURES 157 channel per repeater section, at the repeater input. If there were no other noise sources, the repeater section length would necessarily be limited by this effect. Curve B shows the sum of thermal noise and noise due to the vacuum tubes in the repeaters, which is little in excess of thermal noise alone. The other three curves show noises of considerably higher magnitude which require suppression in order to arrive at an economical carrier system. Curve E shows the order of magnitude of noise on carrier circuits due to connecting open- wire pairs directly to non-carrier pairs in the outside cable near the carrier repeater input. The source of the noise is heavy atmospheric static of a magnitude experienced several times during the summer. 10 tf)0 _J2 5> zo •15 ^ " ^ ~ ~— 7-^=-- ^ sf^- ^- rrr! — E^ — — - '^ ^ ^ ^^^. B-^ -^ ^^ ■^ A^ ■^ ^ ^ ^^ C-^ ^ "*>«» 10 15 20 25 30 35 40 45 FREQUENCY IN KILOCYCLES PER SECOND 50 60 Fig. 15 — Noise, subsequent to suppression measures, per repeater section at output of repeater whose gain equals line loss. A to £— Same sources as in Fig. 14. F — Noise from heavy static induced directly into outside cable. The other curves show typical magnitudes of noise originating in the existing telegraph and voice frequency telephone plant; this is gen- erated in existing repeater stations and transmitted by the non-carrier pairs to the outside cable where it is induced into the carrier pairs. Curve D represents the situation at a combined telephone and tele- graph repeater station, and Curve C, the situation at a station where there are no telegraph repeaters. Figure 15 indicates the results after suppression measures have been applied. As shown, at the top frequency, which controls the carrier repeater section length, these sources of noise have been reduced to be well below thermal plus tube noise. It is also shown that the noise due 158 BELL SYSTEM TECHNICAL JOURNAL to heavy atmospheric static induced directly into a carrier pair in the outside cable is below thermal plus tube noise at the top frequency. There are additional types of noise, not shown, whose sources lie within the carrier system: e.g., modulation in amplifiers, inter-system cross-induction, battery noise. While control of such noise is an integral part of the fundamental carrier system design, it is not the purpose of this paper to cover this class of noise. Conductors Tapping the Carrier Cable Carrier noise may come from open-wire pairs which connect to conductors in the cable. Its sources may be static; corona on power lines; power line carrier or other carrier frequency voltages on power lines paralleling the open wire; induction from radio telegraph stations; or carrier frequency voltages arising in the office to which the open CARRIER PAIR ^ COUPLING WITHIN CABLE CARRIER REPEATER '-f=^ NON-CARRIER PAIR u OPEN-WIRE PAIR CARRIER- ^^.^^ FREQUENCY I xSLJ' ' NOISE VOLTAGE j TO GROUND ~=- Fig. 16 — Schematic of path followed by induction from open-wire taps. wire is connected, such as voltages generated by d-c. telegraph or telephone signaling systems. The limited experience to date indicates that, in a long cable carrier system, the effect of heavy static will be larger than that of the other sources if telephone and power supply plants are coordinated so as to be satisfactory from voice frequency and low frequency standpoints. Branch cables connected to the carrier cable have a similar but generally smaller effect than that of open-wire taps. Figure 16 illustrates the path followed by this induction. A voltage to ground impressed on the open-wire pairs passes by secondary in- duction over to the carrier pairs in the cable, and, on account of the unbalance to ground of these pairs, produces a metallic voltage on these pairs at the repeater input. The effect may be greatly reduced by interposing a filter at the junction of the open wire and the cable. { CROSSTALK AND NOISE FEATURES 159 It is necessary to filter only the longitudinal circuit at an open-wire tap, because: (1) the voltage to ground on the open wire is larger than the metallic circuit voltage, and (2) the coupling between the longitudinal circuit and the disturbed carrier pair is greater than the coupling between metallic circuits. Figure 17 is a schematic diagram of the longitudinal filter developed for a phantom group. It consists of two longitudinal retardation coils and a set of condensers connected between the line wires and the cable sheath. This filter has relatively high carrier frequency longi- tudinal impedance to minimize effects of impedance in the ground con- r 1 PAIR 1 IN PAIR 2 IN ^W5^ ^wr- ^im^ I I "J I I IGROUND n ^wr- ^im^ L2 ^im^ ^wr- PAIR I OUT PAIR 2 OUT I I Fig. 17 — Schematic of longitudinal filter. nection. The major portion of the carrier frequency impedance of the coils is obtained by designing them to have high core loss at these frequencies. The filter has little effect on voice frequency transmission , precaution having been taken to hold the transmission loss, crosstalk and unbalance to ground to low values. Noise Arising in Existing Repeater Offices The noise caused by carrier frequency voltages generated in existing repeater offices is due to d-c. telegraph, telephone speech and signaling voltages, power supply, etc. Figure 1 shows the path by which they reach the carrier plant and the means used to suppress them. In this 160 BELL SYSTEM TECHNICAL JOURNAL figure, N represents a source of carrier frequency voltage in a repeater office, connected to a voice frequency pair which transmits this voltage into the outside cable where it is induced on the carrier pairs. These voltages are reduced by inserting suppression coils in the longitudinal voice frequency paths at the junction between the office and the outside cable connected to carrier inputs. The design of coils giving the requisite carrier frequency suppression without appreciably affecting voice frequency transmission on the circuits in which they are connected was difficult. One coil is used for each phantom group. Each coil has sixteen windings, four for each line wire. These windings are so paired and disposed about the core O 25 1 / I, / A / 1 y / 1 1 1 SUPPRESSION ^ y ^ 1 1 ^ 1 / / IMPEDANC ---" y — — — ' ** 70,000 2 60,000 ? 50,000 < i,000 Z O 10,000 15 20 25 30 35 40 45 FREQUENCY IN KILOCYCLES PER SECOND Fig. 18 — Longitudinal impedance and suppression of noise suppression coils. as to make possible very small side-to-side and phantom-to-side cross- talk between line windings. They also permit obtaining very small leakage flux in both the sides and the phantoms; hence the coils intro- duce very small transmission loss in their voice frequency circuits. The leakage impedance of the coils plus the impedance of the cable stub used to connect them into the circuit is held down so that the effect on repeater singing and echoes in the voice circuits is very small. The coils are so wound that their longitudinal inductance is in anti- resonance with their distributed longitudinal capacitance at ap- proximately the top cable carrier frequency, resulting in a large increase in their suppression in this critical frequency range. The longitudinal I I CROSSTALK AND NOISE FEATURES 161 impedance of one of these coils, and the approximate suppression which a set of them provides, are shown in Fig. 18. In addition, the carried circuits are carefully separated, electrically and physically, from existing voice frequency circuits in common re- peater stations. To this end the carrier pairs in the outside cable are brought out on the line side of the noise suppression coils into a separate cable connected directly to a sealed terminal. From this terminal they are carried in shielded wire to the units in the carrier office and then to a similar sealed terminal leading to the outside cable in the opposite direction. Filters for filament and plate battery supply are included in the carrier amplifiers and additional filament battery supply filters are provided at the carrier fuse panels. A New Single Channel Carrier Telephone System * By H. J. FISHER, M. L. ALMQUIST and R. H. MILLS The single channel carrier telephone system described in the following paper is designated the Type H. It is characterized by several new features, making it applicable not only to the needs of telephone companies but also to those of railroads, power sys- tems and oil companies. It replaces the Type D single channel carrier system, more than 500 of which are now in operation in the Bell System, and, in addition because of its lower cost is applicable to shorter distances. It therefore marks another step in extending the use of carrier. Reduction in size and provision for operating on a-c supply simplify its installation, and its porta- bility makes it well suited to provide emergency circuits. ON open-wire lines where the growth is not rapid, there is frequently need for adding telephone circuits one at a time. When the Type D single-channel carrier telephone system was developed a few years ago it became possible to meet this need without stringing additional wires. ^ More than 500 of these systems have been placed in service in the Bell System plant. A new single-channel carrier telephone system, known as the Type H, has recently been developed and is now being applied. This new system offers improved per- formance, and also, because of its lower cost, is applicable to pro- viding service over shorter distances than were economical with the earlier system. The Type H system, which is characterized by a number of new features and special developments, is applicable not only to the needs of telephone companies but also to those of railroads, power systems, and oil companies. ^ In the first place it is designed to operate either on alternating current or on direct-current plate and filament supply. A repeater is available to extend the range of operation. Through the use of specially designed but simple filters the system can be employed on circuits which are equipped with bridged telephone * Presented at Winter Convention of A. I. E. E., New York, N. Y., January 24- 28, 1938. Published in Electrical Engineering, January 1938. 1 " Carrier Telephone Svstem for Short Toil Circuits," H. S. Black, M. L. Alniquist and L. M. Ilgenfritz, A. I. E. E. Transactions, Vol. 48, January 1929, pp. 117-140. ''"Carrier Telephone Systems — Application to Railroad Circuits," H. A. Affel, Proceedings of the Association of American Railroads, Telephone and Telegraph Section, October 1936, pp. 654-672. 162 NEW SINGLE CHANNEL CARRIER TELEPHONE SYSTEM 163 TYPE D CARRIER TERMINAL TYPE H CARRIER TERMINAL Fig. 1 — Installation of Type H carrier telephone system at Charlotte, North Carolina. 164 BELL SYSTEM TECHNICAL JOURNAL Stations at intermediate points as is frequently the case in railroad operation. A unique feature is the use of opposite sidebands of the same carrier frequency for opposite directions of transmission. The upper side- band is used in one direction and the lower sideband in the other, the carrier being suppressed. For the modulators and demodulators copper oxide "varistors" are employed in place of vacuum tubes. The amplifiers are single stage, employ pentode tubes and are stabilized in performance by feedback. The filters have been simplified in con- struction by the use of coils with a new type of core material and by improved designs of paper condensers. The size of the new terminal has been so reduced that it occupies less than 40 per cent of the space required for a Type D terminal, as indicated in Fig. 1. The equipment may be mounted on racks as is customary in telephone offices, or a complete terminal or repeater may be mounted in a small cabinet. Single-channel carrier systems have been used in the Bell System principally for short open-wire toll circuits. Thus, the Type D systems are for the most part between 50 and 200 miles in length. The Type H system, since it includes a repeater, can be used for greater distances, and due to its lower cost is economical for shorter distances. General Description of System Basic System The basic system consists of two terminals one of which is referred to as an "east" terminal and the other as a "west" terminal, as indi- cated in Fig. 2. The two terminals differ only in minor respects, the differences being due to the fact that at one terminal the upper side- band is transmitted and the lower sideband is received, while at the other terminal the reverse takes place. In order to simplify coordina- tion between various types of carrier systems operating on the same pole line, the frequencies between 7400 cycles and 10,150 cycles are transmitted in the east to west (or north to south) direction, and the frequencies between 4150 cycles and 6900 cycles in the west to east (or south to north) direction. The frequency allocation of the Type H system and those for the Type D and the three-channel Type CS system are shown in Fig.' 3. All three types may be operated on the same pole line. The circuit arrangement is given in greater detail in Fig. 4, which shows a schematic diagram of one terminal, with the exception of the power supply circuit. Each terminal is made up of a transmitting NEW SINGLE CHANNEL CARRIER TELEPHONE SYSTEM 165 ^5^1 . — di^o: 0

  • -Ro£ 8o LU 1^ — _i UJQ. o:< ^ -' S > CD o O O z o — <7) > (O UJO o \- "-U1 I ' < I I I I I I! LVt: Il/'UJ ^ 11 << O UJ uS . S5 SES E- — =tt=rl^ o-a a "> -;^u.« O t/5 -5 E^ M O — CT t« 3 ca ^ O u^ O o vO IM '-I p-1 ID o lO ■D l/V On 00 -* On On >o >o 1-^ Ov O 00 r^ 00 1^ ID 00 o >o ::; o - O ■■o « o •a ? lO ID >> ■o o •o On ■* fO tr> lO •D ■o ^ 00 tN ^ -Tj «5 m >o E Tf O -* r^ - Ui lO >D ^^ to O ■* >o o O V in ID ID ^ ts O rO vO a 1^ O) lO lO ID "^ o rO >o On ID o c o u. *-» '> — o o O O H 0. ""o ro vO NO C '-I &0 a! H a t/3 .ID C^ >* M S g.2 s CO o r > *j ni > 5-s -^-^ n p m il> p ID ^ S S C C O iS-2 M a.t S S E.2 S ° C O ol o • P O &"-c M 5 ft o '^ O •So Si to>0 O m ::e-- "Cm 2 ^ w V c c° M C J3 bo 0] *~* •5 & « „ tn-5 nJ •5 oofcc ^ 5. So bi) E 0) ca <" o S U, V -, \ \ \ A \ \ \ \ \ \ V \ \ \ \ \ ^ y MAXIMUM \ Y \ AVERAGE \ I \ \ \ \ \ \ V \ \ \ \ \ > \ \ \ \ \ \ \ \ \ \ \ \ I \ \ V \ \ \ \ \ 0 6 ro TO 5 10 II 16 21 26 31 36 41 TO TO TO TO TO TO TO 15 20 25 30 35 40 45 WIND VELOCITY IN M 46 51 56 61 66 71 76 81 86 TO TO TO TO TO TO TO TO TO 50 55 60 65 70 75 80 85 90 LES PER HOUR Pig. 9 — Distribution of the annual maximum and five-minute average wind velocities. < %^ a o S IE H - s Pi o o ^ r»5 -(< o O vo O lO o 1 O lO© 1 •# lOVO o ■no o O '^ (M Ht/3 51? •d • 1 .1: ■3: Q cr E E 'o : 'o : aj a! a a (/; on O 00 55 1 1 00 ^1^1 1 o \d 1 1^ o SS 1 1 \2\\ 'O O O O O , lO lO ID so O O ^O O AAA ' AA O O "1 o o -*' so so O O AAAA O o ^ dll 1 1 |. •3: : : Q o- - - ^ o : ; ; o: : M n) R O. "1^ vD 00 r'^V:^ O 00 "2 fco i 1 1 i 15 51 1 1 0 1 ID-* o ■o 1 i;^ 1 1? o o 6 1 AA IT) ir> IDO T(<1D 1 1 1 oo -f ID 1 o o ? 00 1 1^1 1 ID ^ q lOOlOlDlD rD -t •* "CO AA IDIDO -* -t ID OO'D ID ID ID IDIDO -t -f ID OOlD IDID -t A rDrDrO •f-l'H* rDf^^D ^ III III III I —"csro —'CSrD -^tSfO -- I I !00 !-*t- A o ID oo A O »D ID ID ID »D '^'aa '"aa IDO A ©■DO VDOiO ID ©ID '♦■'O'D -t.iD>0 -J> ID O AA A ©©©O ■!f ID I^ 1^ A Q' rO r*"i r*^ -t« ^ III I -< tS r»5 'H ( 1 ^ -H CS <^ u-i O ''^ O "^ _ • -* ■* lO 1/5 ° - A 9 1 o n 1 - -^ 1 o o o A 1 C t^ 1 "-, 1^ 1 I toil o -^ 1 -i- M Tf f»5 o o ^ 1^ r^ O OOOm s ^ A b « • . J3 a u sL« a; rt H73 0) ^- - ■ •o 0) a Q '5: ; : CT 0) C '-' bc C s a X - -t vO X ■^ c m c I/' ' o o r'i c 0 OC 'l- o ^. C 1 "" o c ~; -^1 ^. 1 re vO f^ ^ q o d \0 1- LT. O ! 1 ir LT ifl O C r-" cs 1 1 fO Tj< lA f^ tn o>« "^ . -> r O t' ^ 1 t r^ q U-. O I O lA o o O fN CS f^ 1 ro '^ Tf ■* ro ro O 1/5 I U-. _ o 1 g ^ M 1 -t f*-. ^ 1 fc m 0
  • 5 = ^ s 1' ' g n ?3 c a ft 2 a a ft ^ -r :5 ^ 'S: : f 2 t § c ^^ ^ -i« ^ D O) re to ^ bO ai'C' bi'vT si"? m'R c c Jj c ?; S 0.5 o 'oc : 'o aj 5 B '0 cs'G rt cd ca ft g ft rt ft cfl ft a ftW .D-X ftx ftcfl c/: cr, wx -^ 73 ^73 ^ rr^ ^c ro ^ '^ •* c32 ^ ^c OJ O OJ *^ u *^ u iU 0) * P h .3 I'i cs ft ; i"— ^ ft -•o._ - 1,0 0- = *" c ^ ° S 8 'S '3) c i 3-Sn S — ft-' n ^ 1 •"" ^ S u a; C oj 3- a '^ -C : ■S-3 -"S-o I =2 fe ftO-g o lo J- i-o aj'- o^ — -■ p o y oj 2 re C -j: OJ fe S oo D c ,, SI g O 5, M Mi^ = ^ o in So c3:f: 1- Ji M — ' r^ r- 4J ^ rt g M K'rt dj T 1; ^ rt > e CO -" =0 ° "> i cj aj ^ 5 il; C o o t!— o o ca .^2 a) ca-o 9..'^-^H ■=: ca. . cai 5T3 CC -Si '52 c" ry^ .- — -: 3 aj 1-, ■ C'o S S- ftfc-^ « o 1^0 ftbi c ^^^vo M ca O '3 c ■" il ») " fto =^-i: g 5 s tz^ I ^ ii c 2 £- c •"•^I^ F ji S 3 i ■" 1- — T3 "- ' ■5"o>yOt Ti > V. -n ^ =^ N ^ ■-• "cop"" m2 -g 3-' — o ca •/. J o > 1) t. ft 2 aj -.^ ■*:;w :a a;—- c "^.t-* — J3 2_ ftfty .5i-G*^- -•J 5 aj M.2 OT CQ I ta t.^ B S I 5 c S"? ft 3 Ut/) -^ 0^ ca en M< jT) X m o«w < ^*" Ji 5 M J D ca M-r £ cs < gj^ .5 5 =3 O-Z 3 .n 13 en " ca tn ■;; ■" o M aj aj-- " ^-c I 60-^ •— I "^ Mw c 3 t.„ ^: "a^-^ -"^ c ca.o u 2Ass=|:;2ai.s^ ■O; •"•" n— a^ S 3 aj '^ ^<; li i o. H o v-a M a P ca c-g c = g g . 2 ■•« o aj * m 212 BELL SYSTEM TECHNICAL JOURNAL were obtained when the sags in the wires were small. Threshold velocity data for the cases of larger sags were obtained by repeating the test in some instances with the wires so tensioned that in cold weather the sags were equivalent to those ordinarily prevailing in warm weather. In considering the threshold velocities given in this article and in Table I the relative frequency of occurrence of these velocities has an important bearing upon the amount of contacting to be expected. In Fig. 9 the curve for maximum (comparable to threshold) wind velocities experienced at Chester, shows that for velocities greater than about 20 miles per hour the frequency of occurrence decreases as the velocity is increased. For example, in terms of five-minute periods, winds in a velocity cell of 36 to 40 miles per hour and those of higher velocity have been found to occur approximately twice as frequently as winds in a cell of 41 to 45 miles per hour and higher. Thus, in the vicinity of Chester, a wire arrangement with a threshold velocity of say 40 miles per hour would be expected to be subjected to winds that would cause contacts during approximately twice as many five-minute intervals in a year as an arrangement with a thresh- old velocity of 45 miles per hour. At higher wind velocities an increase of five miles per hour in the threshold velocity will be attended by a greater per cent reduction in the number of five-minute intervals during which winds of sufficient velocity to cause contacts will occur. Since these results were obtained from tests using short lines which were relatively rigid as compared to long lines it was thought that this feature should be given consideration by supplementing these tests with a few representative tests on a longer line. Accordingly, ad- vantage was taken of a toll line, in the Pocono Mountains of eastern Pennsylvania, which was to be dismantled and an 18-mile section of the line was equipped with four pairs of wires each with a different spacing. The pairs were connected to a recorder which registered a contact of practically the same definition as the recorder at Chester. Data from this line were recorded for approximately two winter seasons. The results were in substantial agreement with those obtained from the tests at Chester. Equilibrium Position of a Span of Wire in Natural Winds With regard to the theory ^ relative to the equilibrium position of a span of wire under the influence of a steady wind, a study was conducted at the Chester site to investigate the applicability of this theory under the varying conditions of natural winds. Owing to the ^ Loc. cit. TELEPHONE LINE WIRE SPACING PROBLEMS 213 gustiness of natural winds it was apparent that the study would have to be conducted on a statistical basis. The equilibrium position of a span of wire in a steady wind can be defined by the angle between the vertical plane through the supports and the plane of the suspended wire. This angle is given by equation (1) in Appendix I. Briefly, the problem was to determine this angle for a large number of cases over the complete range of natural wind velocities experienced and to determine the degree of agreement between these values and the angle given by the theory for the corresponding steady wind velocities. A pair of 0.165-inch diameter hard drawn copper wires with a lateral spacing of 16 inches was installed in a 260-foot span with the supports at the same level. The two wires were maintained at equal sags throughout this study. To prevent movements of the pole supports four guys were used on each pole, three 120° apart attached at about two-thirds the height above the ground and a head guy attached at approximately the top of the pole. The wires were approximately normal to the prevailing northwest winds and were located in close proximity to and at approximately the same height as the graphic recording anemometer and wind vane used in the wire spacing study. A Pathe motion picture camera was modified to take a continuous picture of a point (center of span) on each wire. The camera was mounted rigidly directly under the wires at the center of the span. A fine platinum wire was attached to the camera just above the film to provide a fixed zero reference point on the film and a mechanism was provided for synchronizing the wind velocity chart and the motion picture film. With this equipment the wires were photographed when at rest, with no wind present, and a number of pictures were obtained at various wind velocities ranging from zero to approximately sixty miles per hour. Figure 10 is a photograph of a section of film showing the wire images and the reference line. Examination of the films disclosed that except on rare occasions when the wind velocity was low the wires were continually in motion and the point photographed on each wire was represented by a wavy line. In most cases, during the occasions when the wires were being photographed, the wind was approximately normal to the line. The variation in the distance of each wire image on the film from the zero reference line (with reference to the distance when the wires were at rest) provided a means of determining the actual horizontal displace- ment of the wires for any recorded wind velocity through the use of the ratio of the actual spacing between the wires to the spacing between the wire images on the film, equation (2), Appendix I. The angle of 214 BELL SYSTEM TECHNICAL JOURNAL \ i FIXED REFERENCE LINE WIRES WIND DIRECTION 1 SECOND Fig. 10 — A photograph of a section of motion picture film showing the transverse movement of wires in wind. TELEPHONE LINE WIRE SPACING PROBLEMS 215 H^ W 1^ ffl b < O H in Z O q •o 45.0 41.5 50.0 58.0 59.0 59.0 49.5 40.5 50.0 60.5 51.5 49.0 48.5 55.0 50.0 49.0 52.5 46.5 54.5 49.5 q in qqioioqinqq>oqq'0"i'oio>iiqqinq o6d-'r^'-ooo6-*od>ocsd(siot~-'ao6>oaJ-i 00 « o o o»'^in"^ir)ooo»'50ioqq»oqqqqqq q o o CO 00 •* q Oio>')ir>iOimoO"TOO>«OOOinOOOO O 00 O 1> M Q 1 OOOO^lOOM^lOOlO'-'POfOt-.CNOO^^OOt^ 00 "1 loiriO'oiomqqqioirjw^qu^ioqinqqi^ CN r-.' 0> -< O -■ "0 ^ U-! -< O^ tc' m' vd ro r-1 ui •*' (S ro Tf ~5 d ^ •* lOiOiOiOiOOi'lOOtoOOOOOOOiOOiO ^ (U csio-^vOOoot^iO'^r^nou^'^i'oooocNioaor^i'^ Tf t^ q iooqqq>^qq>^q>';'';'ou^qqq'/iqio t^^odt^oCOCNt^oOOOfCrriOr^OeSO'^O'^ 00 00 o o q oto»riioooto»oo>oto»ou^qLcto»oqqio ■*'*Tt>Tf-*T(iTt»OioiOu^O'00>00»00»0 lO O 00 •a nj ':^0^t^t^0^0^^^O'O^^v0O'O^OfOOOf0t^00 fO ■" 4J o 6 qqo'oqqqu^ioquoi^qioiou^ioqx^x^ 00 3 q o»oqoio»oio»oqqiou^iopqqqq"^_q tStS(Sr-l(SrS(N(SCSC-)(NO)tNfStS(S(S!N(NfO 1-0 S o q ooo»noqioioqio»oioioto»ou^qq»oq t^od>o-<^oIt-^ooc-icNoi-<'avod>oiod>df^-' o T3 3 O to irjooiooi/^iotooqioqioqqqi^w^ioto t-.^^dr^'v6rO^CSCSfcdOOO^^^'Ou^?N ts £ o U 00 q ooou^u^i'ioooooqioqioi^^ioioioq 00 fO 00 a >o lO v6 qqqqioq'oioqioquiqqin. qqi^ioio o6 o q "(5 > •* q qqqioOLoioinqqqinioqqqq'^'oio do^dd ^-*O^dt^0^C^00CS0^00C^^^l^t^O^O fO >d o P-1 q o q q •o lo >o lo lo lo >o q q lo i/> "o >« q q >o q ^.^ddddo»o6t-.^osfsoJdo^o6odddo>o> o d i~o fO o o d oioioqqioioqqqioqtoqqqqqqq do^o6o6o^^^t^O\O^OO^O^a^OOQO-^0^0\a^OO OS 00 d 00 q 00 loiooi^ioooqoioioioioioqqu". qqo '-^cscsodododr^^oooot^oooot^oowr^oot^oo 00 5 J >, 1 > XI « 01 fid v Q i .si "m a < _o a; o .s ?8 _3 "(3 > (LI < ""b C o ■> Q •o to •o a a w 216 BELL SYSTEM TECHNICAL JOURNAL deflection of the suspended wires from a vertical plane in a natural wind could then be computed, equation (4), Appendix I, since the sine of this angle is equal to the horizontal displacement divided by the stretched sag of the wires. If the supports were assumed to be rigid the stretched sag could be calculated directly through the use of equation (3) in Appendix I. However, despite the precautions taken to prevent movements of the pole supports it was found that the poles would bend when the tension in the wires varied. For this reason the deflection of each pole for various wire tensions was measured and this factor was taken into account in determining the stretched sag. The correction applied to the stretched sag as computed from equation (3) was in some cases as great as three inches. Furthermore since the length of wire varies with temperature the particular temperature at which a picture was taken had to be given consideration in computing the stretched sag. Following the procedure described above, 20 values of the angle representing the equilibrium position of the wires were obtained for each two-mile-per-hour cell of transverse or normal wind velocity over a range of 17 to 55 miles per hour. The average experimental angle was computed for each velocity cell and the degree of dispersion of the individual values was determined in the regular manner. Table II gives the values of the angle of deflection of the wires calculated from the experimental data, also the average angle and the best estimate of the true standard deviation for each two-mile-per- hour wind velocity cell. For comparison the theoretical angle of deflection as computed through the use of equation (1) is given for each cell. The maximum and the minimum angles determined experimentally might be plotted versus the theoretical angles, but these data furnish no definite measure of the dispersion since maximum and minimum values depend upon the number of observations made. For this reason the degree of dispersion was determined for single observations and for averages by obtaining an estimate of the true standard deviation which is independent of the number of observations. Figure 11 shows the frequency distribution of the angles determined experimentally and also the " three-sigma " limits for the wind velocity cell of 25.1 to 27.0 miles per hour. In Fig. 12, the average experimental angle for each velocity cell was plotted against the theoretical angle as given by equation (1) and a regression or trend line was determined for these points. For comparison with this line is given a reference line of exact agreement. The "three-sigma" limits for single observations and for averages of 20 observations, were also plotted against the theoretical angle in I TELEPHONE LINE WIRE SPACING PROBLEMS in Fig. 12. The standard deviation shows a tendency to increase as the angle of deflection is increased. However, as might be expected from the use of natural winds for the experiments in place of the steady wind 28 o e AVERAGE EXPERh^ MENTAL ANGLE a' I — 3cr' O- "F= I I 1 a' -3a' WMM ^ 1V20 THEORETICAL ANGLE a= 16.5 NUMBER OF EXPERIMENTAL ANGLES PER CELL = 20 3 SIGMA LIMITS FOR SINGLE OB- SERVATIONS AND FOR AVERAGES OF 20 OBSERVATIONS WERE CAL- CULATED WITH RESPECT TO _ AVERAGE EXPERIMENTAL ANGLE a THESE LIMITS ARE SHOWN AS: a'±3(?' FOR SINGLE OBSER- VATIONS , a'± 4=- FOR AVERAGES OF 20 V20 OBSERVATIONS WIRE —0.165-INCH DIAMETER COPPER a' + 3cr' A I 0 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 EXPERIMENTAL ANGLE (OC') IN DEGREES Fig. 11 — Distribution of experimental angles of deflection of a suspended wire in a transverse wind velocity cell of 25.1 to 27.0 miles per hour. assumed in the theory, there was irregularity in the "three-sigma" limits. For this reason a regression line was determined for all the points in each particular "three-sigma" limiting group and the lines were drawn. This graph shows the agreement between the angle of deflection of the wire as determined by the experimental method and that given by the theoretical equation (1). The area between the two extreme limit lines represents approximately the range within which single values determined experimentally would be expected to fall. Likewise, the area between the two inner limit lines represents approximately the range within which averages of 20 values for a particular wind velocity cell would be expected to fall. In general, the results indicate that the theory for the equilibrium position of a suspended wire under the influence of an assumed steady transverse wind is applicable within reasonable limits to a wire sub- jected to the varying conditions of natural winds. An Accelerated Method of Test Testing the merits of various arrangements in natural winds is a slow procedure in which it is necessary to await the course of nature. 218 BELL SYSTEM TECHNICAL JOURNAL TRANSVERSE WIND VELOCITY IN MILES PER HOUR THEORETICAL ANGLE (a) IN DEGREES TELEPHONE LINE WIRE SPACING PROBLEMS 219 In order to narrow the scope of the test in natural winds to the most representative and promising of the various proposed arrangements, an accelerated method of test was used for characterizing the proposals in a preliminary way. This accelerated method was suggested by the above mentioned theory concerning the equilibrium position of a span of wire in wind. From this theory it was conceived that the relative merits of a type of wire arrangement could be determined by successively deflecting one wire of a pair outward and upward and releasing it to swing towards the other wire through an increasing series of known angles until the two contacted. The theoretical wind velocity corresponding to this minimum angular displacement producing contacts would then be determinable through the use of equation (1). While it was realized that such a procedure does not simulate the contacting of wires in natural winds, it was thought that from a relative point of view the arrangement requiring the greater angular displace- ment and therefore the higher theoretical velocity to produce con- tacting might also require a higher natural wind velocity. Further, it was felt that there was a possibility of being able to determine not only the relative merits of types of arrangements but also their natural wind threshold velocities by correlating the data obtained by the accelerated method with those obtained in the natural wind tests. Accordingly, a number of tests were made using arrangements on which considerable natural wind data were available. The test set-up comprised a pair of 0.165-inch diameter hard drawn copper wires installed in proximity to the ground with suitable means for changing spacings and sags at will. In order that the deflection of the wire might be accurately controlled and the results reproducible, a series of rods used as guides were mounted rigidly on a vertical support at the center of the span in a plane at right angles to and in proximity to the wires. The points of these rods were positioned in the arc of a circle with a radius equal to the sag in the manually deflected wire. The intervals between the points of the rods in terms of theoretical wind velocity were five miles per hour. The arrangement of rods represented a range of 20 to 80 miles per hour. The test apparatus is shown in the accompanying Fig. 13. In the first stages of these tests it was found that a pair of wires would not always contact for the same minimum angle of deflection. Experiments showed, however, that during the absence of any natural wind there was a minimum angle for a given wire arrangement which would produce contacting five times in five consecutive trials. This refinement of the method was then adopted and the theoretical 220 BELL SYSTEM TECHNICAL JOURNAL velocity equivalent to this angle was termed the accelerated method threshold velocity or the velocity at which contacting begins. It is, therefore, not necessarily the minimum angle or velocity at which the wires can be made to contact but the minimum angle or velocity at which the wires will contact for each of five consecutive trials in practically still air. i Fig. 13 — Apparatus for the accelerated test. In order that the contacts recorded by the accelerated method should be comparable in definition to those recorded by the natural wind method, the wires used in the accelerated test were connected to the same recording apparatus as was used in the natural wind tests. From the results obtained through the use of the accelerated method of test an empirical equation was developed for the case of a pair of wires with equal sags. This equation (6) is given in Appendix II. The comparable empirical equation (5) for natural winds, referred to above, is also given in Appendix II. With these two equations it is possible to determine the expected natural wind threshold velocities of a wire arrangement through the use of the accelerated method of test. An equation (7) for this use, which was obtained from the above two equations (5) and (6), is also given in Appendix II. While empirical equations were developed only for the case of a TELEPHONE LINE WIRE SPACING PROBLEMS 221 pair of wires with equal sags, the appHcability of the accelerated method is not limited to this case. It was also used where the sags in the two wires of a pair were unequal and, as referred to later, for determining the most promising design of insulating disc (described below) for use in natural wind tests as a means of mitigating the contacting on pairs with the wires spaced, 3, 4 and 6 inches. Anti-Contacting Insulators As stated above it was found that wires spaced 3 or 4 inches con- tacted in wind velocities rather commonly experienced. Some Fig. 14 — Insulating disc. contacting was also recorded on pairs with wires spaced 6 inches. In giving consideration to means of increasing the natural wind threshold velocities of such circuits to those occurring less frequently, two types of anti-contacting insulators were developed. One type, Fig. 14, was a perforated disc of insulating material. When this type was installed on one wire of a pair it was not in contact with the other wire of the pair except when forced there by the action of the wires in wind. The other type, Fig. 15, was a rod-shaped insulating spacer. This spacer bridged the two wires of a pair in the span. The insulating discs used were 3 and 4 inches in diameter. The arrangements of these discs tested in natural winds comprised one, two or three discs per span per pair of wires. When one disc was used, it was placed at the approximate center of the span on the wire 222 BELL SYSTEM TECHNICAL JOURNAL of the pair to the windward side of the Hne. When two discs were used they were placed on the windward wire at one-third of the distance from each support. The three-disc arrangement comprised the two-disc arrangement with the third disc placed at the center of the span on the other wire of the pair. Fig. 15 — Insulating spacer. In selecting the disc and these sizes, various shapes and sizes of insulators (one of which is shown in Fig. 13) were tested using the accelerated method of test referred to above. The circular type was found to give as good results as any other shape and had the advantage of being simple in design. The insulating spacers were used one per span per pair of wires located at the approximate center of the span. Any insulating spacer bridging the wires of a pair constitutes an additional line leakage path and from this standpoint is undesirable. In this respect the discs have a distinct advantage over the spacers. As stated above, they are not normally in contact with both wires and when such contacts take place they are generally of short duration. Then too, in a long line with wires equipped with discs it is improbable that more than a short section would be affected at one time. The thought was that owing to the additional line leakage provided by the spacers, their use would be confined to the occasional long span and the use of discs to the shorter spans. For this reason the tests involving discs were confined to spans of 160 feet and less and those involving spacers to spans of 160 and 260 feet. The dimensional characteristics TELEPHONE LINE WIRE SPACING PROBLEMS 223 of the spacers important from the standpoint of mitigating contacting were tested. These related to the length of the spacer between wires. It was not known whether this distance should be equal to, less than or greater than the spacing between the wires of a pair at the crossarm. The first tests indicated that when this distance was 3^ inch greater than the wire spacing, there was a tendency for the pair to roll and the wires to twist around each other. Later tests were confined, therefore, to spacers with a distance between the wires the same as the spacing or 3^ inch less. The data obtained in the natural wind tests on wire arrangements where discs and spacers were used are given in Table I. The effect of insulating discs was to increase the threshold velocities over those for similar arrangements without discs by about 5 to 20 miles per hour. The 4-inch diameter disc has some advantage over the 3-inch diameter disc. Three discs per span or even two discs give results somewhat better than those obtained with a single disc but the gain is relatively slight. The spacer which holds the wires in the center of the span the same distance apart as the spacing at the crossarm, in general, increased the threshold velocities about 10 to 30 miles per hour for pairs with wire spacings of 3 and 4 inches at equal sags and suspended in spans of 160 and 260 feet over comparable arrangements of unequipped wires. No information is available for the case where the wires of a pair have unequal sags. The data for the 160-foot span give some comparison between the effectiveness of discs and spacers. In the case of wires spaced 3 inches with one 4-inch diameter disc, contacts occurred at a threshold velocity of 35 miles per hour while the comparable figure for the spacer was 55 miles per hour. Abstract Conclusion In these tests typical toll telephone wires were placed in spans of 100, 130, 160 or 260 feet with sags of 4 to 45 inches (depending upon the span length and temperature) and with horizontal spacings between the wires of a pair of 3, 4, 6, 8 or 12 inches. It was found that during the absence of glaze the wind velocities normal to the line when swinging contacts began to occur increased with the wire spacings and also with span lengths (if wire tensions were increased so as to maintain a given sag) and decreased when the sag was increased. An empirical equation based upon this relation and the data (Table I) has been developed for the case of wires at equal sags. As a brief example of the results, in the case of a 130-foot span, it was found that wires 224 BELL SYSTEM TECHNICAL JOURNAL spaced 12 or 8 inches were practically free from swinging contacts in wind velocities below about 70 miles per hour, wires spaced 6 inches contacted at velocities around 50 miles per hour and wires spaced 4 or 3 inches contacted at the more common velocities of 30 or 40 miles per hour. In the absence of glaze, when the wires of a pair were at unequal sags, about 3 inches difference in the 130-foot spans and about 6 inches in the 260-foot spans, there was in general a somewhat greater tendency toward contacting in the shorter spans and a lesser tendency in the longer ones than when the wires were at equal sags. When glaze was on the wires their action was more erratic and swinging contacts were more general and occurred at lower wind velocities than when glaze was not present. Regarding the theory ^ relating to the equilibrium position of a suspended wire in a steady wind, tests were conducted at Chester, New Jersey, in which the displacements of a copper wire were photo- graphed in natural winds. It was found that there was general agreement between the angle of deflection of a wire as determined by this theory for a given steady wind and the angle obtained experi- mentally in a comparable natural wind. The experimental substantiation of this relationship led to the development of an accelerated method for a quick and economical preliminary classification of various wire arrangements. This method was useful in selecting from among several similar arrangements the most promising ones for test in natural winds. An empirical equation based on the data obtained by the accelerated method for the case of equal sags was developed for expressing the relationship between accelerated method wind velocity, span length, wire spacing and sag. By combining this equation with that developed from the natural wind data a third equation was obtained which was used in determining expected natural wind threshold velocities from the accelerated method results. In regard to the anti-contacting devices included in the study with the wires spaced 6 inches and less, it was found that: 1. A 4-inch diameter insulating disc placed at the approximate center of the span on one wire of the pair increased the normal wind velocities at which contacting began by 5 to 20 miles per hour over those for the same arrangements unequipped. The use of three discs per span or even two gave an improvement over the use of one but the gain was relatively slight. ' Loc. cit. TELEPHONE LINE WIRE SPACING PROBLEMS 225 2. A rod-type insulating spacer used to bridge the wires of a pair in the approximate center of the span was somewhat more effective than one disc, increasing the normal wind velocities at which swinging contacts began by about 10 to 30 miles per hour over those for the same arrangements unequipped. In general, the higher the threshold velocity, the less frequent will be the occurrence of those winds which will cause contacting. There- fore, if the threshold velocity of a wire arrangement is increased 5 or 10 miles per hour or more by the addition of an anti-contacting device there will be a decrease in the amount of contacting occurring de- pendent upon the original threshold velocity and the amount of the increase. For example, in the vicinity of Chester, New Jersey, an increase in the threshold velocity of a wire arrangement from 40 to 45 miles per hour results in a reduction of about 50 per cent in the number of five-minute intervals during which winds of sufficient velocity to cause contacts will occur. At higher wind velocities an increase of five miles per hour in the threshold velocity will produce a greater per cent reduction in the number of five-minute intervals during which winds of sufficient velocity to cause contacts will occur. In regard to the field installations with a spacing of 8 inches between the wires of a pair, referred to at the beginning of this article, no serious difficulties have been encountered except in certain sleet areas where there has been some wrapping and freezing together of the wires. In these locations insulating spacers have been installed on a few pairs and their behavior is being followed. The installations in which a 6-inch spacing has been used have been confined to the warmer sections of the country and no serious trouble has yet been encountered. APPENDIX I The expression for determining the angle of deflection or equilibrium position of a suspended wire in a steady transverse wind as given in the theory ^ is tan a = , (1) mg cos 7 '^ ^ ^ where a = Angle between the plane of the suspended wire and a vertical plane through the supports, V = Steady transverse wind velocity (miles per hour), m = Mass of unit length of wire (slugs), g = Acceleration of gravity (feet per second per second), k = Ratio of wind pressure per unit length of wire to square of velocity, and 7 = Angle of inclination of line through supports to the hori- zontal. 226 BELL SYSTEM TECHNICAL JOURNAL In the study to determine the extent to which this theory would check under the varying conditions of natural winds only the case where 7 = 0° (both supports at the same level) was considered. The camera used in this study was equipped with a Tessar F-3.5 lens with a nominal focal length of 40 mm. The films were analyzed with the aid of a motion picture projector. In this projector the film passed over a glass plate on which was engraved a linear scale graduated in hundredths of an inch. The pictures, together with the graduated scale, were projected on to a screen. This method provided a ready means of determining the horizontal displacement of the wire images on the film. The actual wire displacement was then determined through the use of the following relationship: where Li = Distance from wires to camera lens, L2 = Distance from camera lens to film, Di = Spacing between wires, and D-i — Spacing between wire images on film. The two wires were maintained at equal sags throughout this study. The equation for determining the stretched sag of a wire if the supports are assumed to be rigid is as follows: , , 3L ,_ „^ 3wL* _. .. + -(L-i?)a = g^. (3) where a — Stretched sag, R — Unstressed length of wire, L — Span length, ^ .4 = Cross-sectional area of wire, £ = Modulus of elasticity, and w = Resultant of wind pressure and gravity components. As explained in the text, even though the poles were strongly guyed, the supports moved when the tension in the wires varied and cor- rections were applied to the stretched sag to take account of this movement. After determining the horizontal displacement and the stretched sag of the wires the experimental angle of deflection (angle of equi- librium position) was calculated by means of the equation: sm «' = -7- , (4) a TELEPHONE LINE WIRE SPACING PROBLEMS 227 where Li = Horizontal displacement of wires, a' = Corrected stretched sag of wires, and a = Angle of deflection determined experimentally as distin- guished from the theoretical angle (a). The details of the method of determining the equilibrium position of the wires for a particular natural wind velocity were as follows: The horizontal displacement of the two wire images on the film at each wave crest and trough was determined. The displacement for the two wire images at each crest and at each trough was averaged. Next the mean displacement on the film of the crest and trough was calculated and also the mean velocity * was determined for this particular time interval. The mean dis- placement on the film was then converted to actual wire displace- ment through the use of equation (2) and the experimental angle was determined by equation (4). This average angle was taken as the equilibrium position of the wires for this mean velocity. APPENDIX II Empirical Equations From natural wind tests on arrangements of wires in which both wires of a pair were maintained at equal sags it was found that in the absence of glaze threshold velocities increase with the spacing between the wires and the span length and decrease as the sag increases. An empirical equation obtained from an analysis of the results is as follows: " J O.ICO.3 "12.1 V.. = 22aY^]^ , (5) where V^ = Natural wind threshold velocity (miles per hour), L = Span length (100 to 260 feet), S = Wire spacing (3 to 12 inches), and d = Sag of wires at rest (4 to 45 inches). The data upon which this equation was based comprised approximately fifty cases where swinging contacts actually occurred. Regarding the degree to which this equation represents these data, there were only about five cases which deviated as much as five miles per hour in terms of threshold velocity and of these only one deviated as much as seven miles per hour. The nomogram given in Fig. 8 was constructed for this equation (5). * When the direction of the wind was not normal to the line the normal component of the velocity was determined by multiplying the wind velocity by the cosine of the angle between the actual direction of the wind and the norma! to the line,- 228 BELL SYSTEM TECHNICAL JOURNAL The comparable empirical equation for the accelerated method of test is lOF wh( F^ = F = 1 - 0.692 F' 2^0.06^0.2 (6) hi -^ ^ J ^^-^u. ^ ^ ^ r- ^ ^ ^ "oo^ ' ^ ^ y y — 0.125 ■^^ /^ r" / /\ — — — 0.25 /t f 0.5 / " f 2.5 0 0.5 1.0 1.5 2.0 2.5 3.0 3-5 4.0 OJ UJQ Fig. 1 — Characteristics of a simple variable structure. There still remains the possibility, however, of obtaining a network in which the distortion can be kept within tolerable limits over a given range. The quantities Y^, Fo, and Z, which are, of course, all functions of frequency, allow us to determine the transfer admittance at three values of R. The transfer admittances at other settings will then be fixed. If we suppose for simplicity, that the extreme characteristics, corresponding to Y^ and Fo, are set by the engineering requirements on the structure, the problem reduces to that of so choosing Z in relation to these quantities that the distortion is as small as possible at inter- mediate settings of R. Of the variety of possibilities open in the selection of Z, one in par- ticular commends itself by the simplicity and symmetry of the results to which it leads. It is given by the condition 232 BELL SYSTEM TECHNICAL JOURNAL Z=^^R,, (3) where i?n is an arbitrary constant which represents, physically, a reference value for the variable resistance. With the help of this con- dition (2) becomes -^(1 + n z Y^ VFoF. ^^^^ , (4) which can be rewritten in a slightly different notation as where e~*, e"^", x and e'f stand respectively for the quantities F, V Fn F^, R . Z Ao Ao The significance of the assumption made in equation (3) is apparent from an inspection of equation (5). When R — Ro the total loss 6 of the circuit is equal to ^o- The quantity do can therefore be described as the average or reference loss of the circuit, corresponding to the average or reference value of R. It is represented by the middle curve shown on Fig. 2. Setting R = 0 or R — oo gives the symmetrically located extreme curves do zt

    aRp ^2(a2-,) aRp J, 2(a2-,)| aRp e* = a ep+* -J 1 1 1 Lji 1 1 I I I < I 0.5 1.0 1.5 2.0 2.5 3.0 3.5 Wp Fig. 3 — The simplest type of symmetrical variable equalizer. Fig. 4 — Adjustment of the variable characteristic by the addition of an auxiliarj' network. 236 BELL SYSTEM TECHNICAL JOURNAL The efifect of the added network is easily understood from the pre- ceding equations. It will be noticed that although these equations were written under the assumption that i? is a real quantity, they will still be valid if R is complex. We need therefore merely to replace R by the impedance of the auxiliary network terminated by the variable resistance. If we represent this impedance by Zr, the appropriate expression is ^^-^"l+xtanhV'' ^^^ where i^ is the transfer constant of the added network and x is, as before, the ratio of the variable resistance to Rq. Since reciprocal val- ues of X still correspond to reciprocal values of ^- , all of the preceding Ro conditions of symmetry in the resulting family of characteristics are maintained. The simplest formulation for the new 9? is secured from equation (6). Upon replacing the ".r" of this expression by -^- we Rii readily find that the equation becomes X — 1 0 = 0'

    — e 6 4 ^ --^ — -*— __3jU— — ^^^ — — 1 ^^fe ___^,____^ -<^ 0.53 r- p--<^ ^-^IT" -—o , _0;27 2 ^^ ---^ ^SIT^ - ^~"^^ ^Cl '^ - 0 ) 200 300 400 500 600 700 800 900 FREQUENCY IN KILOCYCLES PER SECOND Fig. 6 — Characteristics of the structure shown by Fig. 5. 238 BELL SYSTEM TECHNICAL JOURNAL satisfy the condition expressed by equation (3). For example, any structure having the general configuration of Fig. 7 will meet this con- dition provided the impedances Zi, Z^ and Z3 are so related that when the network is considered as a 4-terminal structure transmitting from a-a' to b-h' it has a constant resistance image impedance equal to i?o at a-a' . In contrast to the network of Fig. 3 in which both ^0 and

    ^ ^^ ^ ^ / eo y" / X ^ ^' ^ ^ So-* ^ ^ Fig. 8 — Variable equalizers with varying reference characteristics. A third simple network is shown on Fig. 9. Its properties are the converse of those secured from the structures of Fig. 8. The reference loss, ^0, is now a constant, while ^ varies with frequency in a manner 'See, for example, "Distortion Correction in Electrical Circuits," O. J. Zobel, Bell Sys. Tech. Jour., July, 1928. 240 BELL SYSTEM TECHNICAL JOURNAL which depends upon the choice of Z. An additional control of 0 can of course be obtained by the introduction of an auxiliary structure in front of the variable resistance. As in the previous example, the illustrative curves are drawn on the assumption that the characterizing 2a^Ro .* = = K-57fe) Ro^ 2a 2a2-2a+i ^ — ' -— " ©0+* A y^ / r a =0.9 / \l A eo ^ ' V \ \. eo-t 1 0.5 1.0 2.5 3.0 1.5 2.0 OJ LOq Fig. 9 — A variable equalizer requiring only one general impedance branch. impedance Z is a simple inductance. It will be observed that the curves "see-saw," the attenuation at certain frequencies increasing while that at other frequencies is decreased. This phenomenon depends upon the choice of the parameter a. It disappears when a is assigned either extreme value i or 1, and becomes most pronounced at the intermediate value a = 1/V2. A similar effect can also be VARIABLE EQUALIZERS 241 produced in the networks we have already considered since, as equation (8) shows, the variable attenuation will change sign if the phase shift of the auxiliary network is allowed to increase beyond 45°. In the fourth structure, shown by Fig. 10, both ^o and ip are variable. 4^^^ pT 1 y^ 1 W^ z„ Ro ' ' ^21 2 Z2I 1 0 ZmZoj-Ro / / f / / ^ -^ y y / y y eo+<^ / A ^y^ 2Ro a'Zsi 2^21 6-^ = 1 + ^ m :iRo Ro pQo =[-^r ^11 ^21 ~ Ro 26 / " / / 2" = i^Ro / / / ^ / / / (a y — " / '' '^ y ^ eo-* // V /eo /; / / / / / 7 / 00+* / y. / ^^ ^ y 1.5 2.0 UJ Wo 3.5 Fig. 11 — A variable equalizer with zero loss at one frequency for all settings of the controlling element. elaborate and difficult to deal with. It possesses, however, the salient property that when Zn = 0, perfect transmission of power from one terminating impedance to the other is secured at any setting of the variable resistance. This property suggests that the network may be VARIABLE EQUALIZERS 243 useful for systems where it is necessary to introduce variable equali- zation without attenuation of the channels having the lowest signal level. The final structure is shown in Fig. 12. Its chief point of interest I — O— ' 2Z2 iZ| xz, Z2 X iz. n 2Z2 y. 14 / / / / Z| = LujLo Z2= ^u;oLo / / ©0+* / / ^ " / ^ ^ ^ / / eo ^ / ^ / ^ -^ ^ ^ ^ *\ \ "^ ^^ — ■ eo-* 1.5 2.0 Wo 2.5 3.0 3.5 Fig. 12 — A variable equalizer adapted to a general controlling impedance. is the fact that the controlling element, instead of being a variable resistance, is a variable general impedance, which has been labelled xZ\ in the drawing. In practical cases, of course, Z\ will ordinarily be a simple resistance, inductance, or capacity. In addition to the variable branch the network also includes two fixed branches propor- 244 BELL SYSTEM TECHNICAL JOURNAL tional to Zi and three fixed branches proportional to a second general impedance Z2. The change from a variable resistance to a variable impedance makes little difference in the analysis. It is merely neces- sary to replace each R and i?o by a Z or Zo in every equation. So long as equation (3), with the appropriate modification, is satisfied, as it is in this structure, the resulting family of attenuation character- istics will have the same general symmetrical form as those obtained from the resistance controlled devices. A set of curves illustrating this point is shown at the bottom of Fig. 12. They are drawn on the assumption that the Zi and Z2 impedances are respectively inductances and resistances. The structure of Fig. 12 will still function satisfactorily as a variable equalizer if it is turned inside out so that the Zi/2 impedances become the terminations and the central shunt branch becomes the variable. In this event the present variable impedance must be set at its nominal value. The resulting structure is essentially the inverse of the network of Fig. 12. In the same way, of course, each of the other configurations which have been described can be replaced by its inverse. An Explanation of the Common Battery Anti-sidetone Subscriber Set By C. O. GIBBON THE telephone transmitter serves to convert sound waves into their electrical facsimile; but in performing this primary function the transmitter also acts as an amplifier. Under some conditions the electrical power output of a transmitter may be more than a thousand times as great as the acoustic power activating it. Part of this greatly augmented power is dissipated in the circuit of the telephone set; part is impressed upon the telephone line, whence it is propagated on to the distant listener; and part finds its way into the receiver of the same set, where it is reconverted into sound waves. Speech or noise, picked up by the transmitter and reproduced by the receiver of the same set, is called sidetone. Noise picked up and amplified by the transmitter and heard as sidetone tends to obscure incoming speech, thereby impairing re- ception. Similarly, the sound of his own voice, heard more loudly than normal as sidetone because of transmitter amplification, impels the talker involuntarily to lower his voice; thus impairing the reception of his speech at the far end of the connection. The consequent desirability of reducing sidetone has long been recognized, and operator and subscriber sets which accomplish this have been developed. Circuit schematics of the common battery sidetone and anti-sidetone subscriber sets at present standard in the Bell System are shown in Fig. 1. The anti-sidetone set has become increasingly common during the past few years, and because of the improvements in effective transmission which it affords, bids fair ultimately to be well nigh uni- versally employed. It is, therefore, not surprising that numerous requests have arisen for an explanation of this anti-sidetone circuit which may be more easily followed than one based on the methods usually employed in network analysis. The present paper provides an ex- planation by means of diagrams with a minimum of mathematical treatment which it is believed those to whom the mathematical approach does not appeal will find helpful in picturing the behavior of this circuit. The explanation given in this paper is, however, confined to idealized conditions. No concern is given to whether the conditions necessary to exact attainment of the balances described are actually feasible; 245 246 BELL SYSTEM TECHNICAL JOURNAL nor is any attempt made to discuss questions of practical design be- yond pointing out something of the nature of the problems involved. Equations for this anti-sidetone circuit are given and discussed in an appendix, and a vector diagram is shown which illustrates graphically relations among the currents and voltages under the ideal condition of exact balances. Rearrangement of Circuit Patterns Simplified explanations of anti-sidetone sets are most frequently based upon analogues with balanced arrangements resembling the SIDETONE CIRCUIT (S) ANTI-SIDETONE CIRCUIT (a) \ WINDING :^ WINDING \ -N \ NET- \ ^ B WINDING d\ \W0RK ^ ^\ N ^ > > A / >LINE \^(V 6y; ^ > L yx/ X/RECEIVER R TRANSMITTER'-r^M t i-\>j Fig. 1 I-ig. 2 3a -i{^^ = I2a I OJM^ lA I \ B -+• '••3A ^(1) ^3A XS «M] '^ Fig. 9 Group III — Components of transmitting E.M.F.'s and assumed mesh currents 250 BELL SYSTEM TECHNICAL JOURNAL when transmitting (see Fig. 5 of Group III), it will be looked upon as the same passive impedance in tandem with an impedanceless generator whose e.m.f. equals the variations in voltage drop (of the battery sup- ply current) across the transmitter due to the changes in its impedance which occur when the transmitter is agitated by sound. This e.m.f. thus replaces in the circuit the sound engendered variations in the transmitter impedance, thereby permitting this impedance to be treated as a constant. Being impedanceless, this generator may, without effect upon the circuit, be replaced by two impedanceless generators — each having the same e.m.f. as the first — connected in parallel as shown in Fig. 6. The direct connection between points a and h, however, is shunted by the impedanceless path ach, so that the direct connection ah may, without effect, be broken as in Fig. 7. Hence, the two equal e.m.f. 's in Fig. 7, acting simultaneously, are equivalent to the single e.m.f. in Fig. 5; and the mesh currents in the two figures are, therefore, identical. Hereafter, Fig. 7, rather than Fig. 5, will be considered the transmitting condition. This transmitting condition may, however, be broken into two com- ponent conditions. By the fundamental principle known as the Super- position Theorem, the currents in Fig. 7 are equal to the sum of the currents which would result from each of the two e.m.f.'s acting alone. In other words, the transmitting currents in Fig. 7 are equal to the sum of the corresponding currents in Figs. 8 and 9. But by a second fundamental principle called the Reciprocity Theorem, the current at any point Z in a circuit, due to an e.m.f. at any other point Y, is equal to the current which would result at Y from an equal e.m.f. at X. Applying this to Figs. 8 and 9, in which Ei = £2, the mesh currents pointed out by the arrows joining these two schematics are equal, viz.: Ifl = I^s and m = m. (1) Of the above components of the transmitting currents, those in Fig. 9 are due to an e.m.f. acting in mesh 1, i.e., in series with the line impedance. This, however, is also the condition when receiving, as will be seen by comparing Figs. 9 and 4. Neutralizing Balance — ^Receiving Efficiency Consider next the purpose of winding C. It is, of course, desirable that the transmitting and receiving efficiencies be undiminished by the anti-sidetone arrangement. If it is possible so to adjust the couplings among windings A, B and C that the current If^ in Fig. 4 or 9 is zero, the balancing branch can then be disconnected without effect COMMON BATTERY ANTI-SIDETONE SUBSCRIBER SET 251 E| - E2 T(l2)lr(l)+TC2)('BY THE SUPER- '1"; ■ IS IS ]^ POSITION THEOREM U,^, BY THE RECIPROCITY THEOREM AND £[=£2 ,(!)_' (1)1 DUETO^r(l)_, '•IA-|MS|_WINDINGCj ^2A ^ r(l2)i,(l) + r(2) / BYTHESUPER- I lA — 1|A^^IA \POSITION THEOREM FIGURE 8S FIGURE 9S IGURE 9A >!y FIGURE E| = E2 Group IV — Transmitting efificiency — sidetone balance 252 BELL SYSTEM TECHNICAL JOURNAL upon the currents, and the circuit will thereby be reduced to the side- tone circuit. Hence, with this ideal adjustment of the couplings, the receiving efficiency of the anti-sidetone circuit will be the same as that of its sidetone complement. Although equally satisfactory designs could be worked out with other polings of the coil windings, the relative inductive directions among windings A, B and C in the circuit here dealt with are such that, if a current were passed through all three in series, windings A and B would be inductively aiding; and C would be inductively opposed to both A and B. Returning to the above condition for maintaining the receiving efficiency, namely, that I'i\ be made zero, the windings of the coil must be so adjusted that the sum of the two voltages induced in winding C by its inductive couplings with windings A and B is equal and opposite the voltage drop across the receiver. With the windings poled as just stated, this requires that (+ Pp^Z^^c) + (- mZBc) = -{- I'ilZn). (2) This voltage balance expressed by eq. (2) will be referred to as neutral- izing balance: its attainment requires the coil windings be adjusted to meet the relation shown by eq. (6) in the appendix. It is important to note that neutralizing balance, and hence the efficiency relations which depend upon it, are independent of the impressed e.m.f. £i, of the line impedance Zl and the self-impedance of winding A , and of the network impedance Z.v and the self-impedance of winding C. This of course follows from the fact that none of these quantities is involved in eq. (2). Transmitting Efficiency "- It will now be shown that the transmitting efficiency of the anti- sidetone circuit is the same as that of its sidetone complement; and that this equality, like that of the receiving efficiencies, results from the neutralizing balance effected by winding C. This is true if, with equal transmitter e.m.f.'s in the top and bottom diagrams of Group IV, the line currents are equal, viz., if 7(12) _ Ti\2) To prove this relation, refer to Fig. 7 A at the bottom of Group IV and move up step by step to Fig. 9A, observing the relations between mesh currents indicated by the arrows. It will be seen that COMMON BATTERY ANTI-SIDETONE SUBSCRIBER SET 253 But in Figs. 9A and 95, due to neutralizing balance, as was shown in discussing receiving efficiencies, r(i) _ n\) 3n^ ni) _ r(i) Finally, continuing from Fig. 95 upward to Fig. 75, it is seen that J- 1.S I -'2S ~ -'IS • Hence, ni2) _ ni) I /-(i) _ ni) _i_ rd) _ 7'(i2) -^lA — -'lA T^ -'2.4 — -'is T^ -'25 — -* IS • Primary Purposes of Winding C and of Neutralizing Balance The above relations between the efficiencies of the anti-sidetone circuit and those of its sidetone complement are, however, merely incidental to the primary purposes of winding C and of the neutralizing balance which it provides. The major purpose of winding C is that, entirely apart from its neutralizing action, the voltages induced in it through its couplings make it possible to obtain sidetone balance by adjusting Zn] i.e. — referring to Fig. 1A at the bottom of Group IV^ — given any value of Zl, it is theoretically possible so to adjust Zn that P^^ = — I^^^^K The current through the receiver under the trans- mitting condition, i.e., sidetone, will then be zero. Neutralizing balance permits this adjustment of Zn to be made without affecting the circuit efficiencies. Sidetone Balance The following discussion of sidetone balance will proceed on the assumption that the couplings of winding C with windings A and B have already been adjusted for neutralizing balance, since this con- dition is required to maintain the circuit efficiencies. This approach is merely a matter of convenience, however, for it will be indicated that the impedance of TV required to effect sidetone balance is the same whether the neutralizing balance is taken into account or ignored. Although sidetone balance is made possible by the couplings of winding C, and the impedance of N needed to reduce sidetone to zero does de- pend upon the values to which the self and mutual impedances of this winding have been adjusted, neither the attainment of sidetone balance nor the value of Z^ required to provide it depends upon the existence of neutralizing balance. If sidetone is to be zero, the voltage across the receiver under the transmitting condition, i.e., the sum of the voltages across C and N, must be made zero. Expressed in terms of the voltages in Fig. 7 A at the bottom of Group IV, this requires that n^rzAc - i'h'Zbc - nTiZc + z^) = o. (3) 254 BELL SYSTEM TECHNICAL JOURNAL Here, at once, the dependence of sidetone balance upon the presence of the inductively coupled third winding is apparent. Without winding C the impedances Zac, Zbc and Zc in the above expression would all be zero, the terms in which they occur would drop out, and the requirement for elimination of sidetone would reduce to Zn = 0, i.e., a short circuit across the receiver. The remainder of this discussion of sidetone balance can be carried out more conveniently in terms of the mesh currents than in terms of the above voltages. As already noted, the voltage balance just examined is equivalent to requiring that I2A and I'^f '^^ Fig. lA be made equal and opposite. But I2A is the sum of the two components, I2A ii^ Fig- 9^ and I2A in Fig. 8^ ; and, because of neutralizing balance, -^3A^ = If A in Fig. 8^1. Furthermore, by the Reciprocity Theorem, the component I2A always equals /j^^ ; and the latter, like the former, is independent of Zn- The condition for sidetone balance may, there- fore, be expressed in terms of the currents in Fig. 8^ as Ifl + ITa + IfA = 0. (4) The question, then, is whether N can be so adjusted that the sum of the three mesh currents in Fig. 8^ is made zero; and it is fairly evident such an adjustment for any given value of Zl is theoretically possible. Since /j^l is known to be independent of Zn, and because inspection shows the circuit to be symmetrical with respect to L and N, it appears that I^A must be independent of L — an intuitive inference which the Reciprocity Theorem confirms. The value of I2A depends, of course, upon both Zl and Zn. Hence, with the value of /j^l remaining fixed as Zn is varied, and with I'iX independent of Zl but under the direct control of Zn, it may be concluded possible to meet eq. (4) by a suitable choice of Zn for any given value of Zl- The value of Zn required to attain sidetone balance is shown by eq. (7) in the appendix. With N so adjusted that sidetone is zero, it is obvious the receiver impedance may be changed in any way whatever without upsetting the sidetone balance. The same is true of any change in the im- pedance of the transmitter; because this, being equivalent to a com- pensating change in the transmitter e.m.f., would cause all mesh currents to change in the same proportion ; thus leaving the balance expressed by eq. (4) undisturbed. Hence, the impedance of N re- quired to provide sidetone balance is independent of the receiver and of the transmitter. But as has already been seen, the couplings of winding C necessary to provide the neutralizing balance in eq. (2) do depend upon the receiver and transmitter impedances. The sig- nificance of this observation is that although the value of Zn required COMMON BATTERY ANTI-SIDETONE SUBSCRIBER SET 255 to provide sidetone balance does depend upon the values of the coup- lings of C, neither the attainment of the balance in eq. (4) nor the impedance of N required to provide it depends upon the balance in eq. (2) being met. In other words, the neutralizing balance expressed by eq. (2), and the sidetone balance expressed by eq. (4), are mutually independent; either may be attained without the other. Practical Considerations With the simple types of coil and network permitted by economic and space limitations, the balances upon which the above performance of the anti-sidetone circuit depends can be obtained exactly only with a given line and at a single frequency. For practical purposes, how- ever, exact balances are needless. Sound leakage under the receiver cap and conduction through the head structure fix a limit beyond which further reduction in sidetone is not of value. Actual designs, therefore, aim at the best compromise in reducing sidetone over the voice range and the range of line impedances important in practice, as judged by the resulting effective transmission performance obtained with the instruments employed. Under typical plant conditions, designs now in service reduce the volume of sidetone with present instruments to a level averaging around 10 to 12 db below that of the complementary sidetone sets. APPENDIX Algebraic Solution of Circuit Equations Referring to Fig. 10, the following circuit equations may be written: -7* - -AC- .. Zg ^-— Zbc—- +- Fig. 10 — ■* The poling of windings A and B is series aiding. Winding C is poled in series opposition to windings A and B. {Z L-\- Z A-\- Z t) I iA-\- {Zt — Zab)J2a— ZacIza = Ei {Zj'—Zab)IiA-{-{Zt-\-Zb-\-Zr-\-Zs)I2A-\- {Zii-\-ZBc)IiA =E2 — ZacIia-\- {Zr-\-Zbc)I2a-\-{Zr-\-Zc-\-Z.w)I3a = 0 (5) 256 BELL SYSTEM TECHNICAL JOURNAL M fe U '^ O'O oj-c ■; rt >H S2 u u 0) O) 0) ^ >> I COMMON BATTERY ANTI-SIDETONE SUBSCRIBER SET 257 These cover both transmitting and receiving conditions: when trans- mitting, El = E2\ and when receiving, Ei — 0. The relation which the induction coil must meet in order to provide neutralizing balance can be determined by solving eqs. (5) under the receiving condition £2 = 0, and imposing the requirement that /3'J = 0. This gives as the relation to be met. ZaC ZaB — Zt />.n (6) Zbc + Zr Zt + Zb + Zr + Zs In like manner, the value of Zn needed to provide sidetone balance can be determined by solving eqs. (5) under the transmitting condition El = £2, and imposing the requirement that I2A + -^3^^ = 0. This value of Z\, regardless of whether or not eq. (2) is imposed as a further condition in its derivation, is found to be r^ ry V \ Z ,\ c{Z A C 4" ZbC " ZaB " Zb " Z ^) ,_, ZjV — ^BC — Z.C -\ ^ j y 1 7^ • \l ) ^ L -r ^A -\- ^AB Note that Zn is here independent of Zt and Zr, except as these may enter implicitly as factors affecting the impedances at right in de- signing the coil for optimum performance with specified instruments. In other words, the transmitter and receiver may be changed without disturbing the sidetone balance. Such a change would, however, upset the neutralizing balance, thereby altering the efficiencies from those of the sidetone circuit. Vector Diagram Relations among the component mesh currents in an anti-sidetone circuit of this type under ideal conditions of exact neutralizing and sidetone balances, are illustrated by the vector diagram in Fig. 11. As all of the current vectors indicate current per volt impressed, those for the mesh currents under the receiving condition in Fig. AA are identical with those under the component of the transmitting condi- tion in Fig. 9A. Vector sums of the mesh currents show the current through the receiver and that fed into the line when transmitting, and illustrate the sidetone balance. Vectors of the three voltages acting around the third mesh in Figs. 4^ and 9 A are also shown, together with their summation. The latter illustrates the neutralizing balance of eq. (2). The Occurrence and Effect of Lockout Occasioned by Two Echo Suppressors By ARTHUR W. HORTON, Jr. "The Time Factor in Telephone Transmission" by O. B. Blackwell (B. S. T. J. January 1932) deals with a number of problems which arise in connection with telephone circuits having long transmission times. This paper discusses one such effect, the occurrence of lockout caused by the echo suppressors involved in a long telephone connection. The occurrence of lockout is shown to cause an increase in repetition rate, which is ordinarily small for circuits as now used commercially. The increase in repetition rate is approximately proportional to the number of lockouts occurring and to their mean duration, or to the per cent of time locked out. The expected number of lockouts is shown to depend upon the characteristic time intervals of conversational speech, the relay hangovers, the delay of the circuit and location of the echo sup- pressors with respect to the ends of the circuit. Subject to certain restrictions, the expected number of lockouts increases with the delay included between the echo suppressors, and is nearly inde- pendent of the delays between the suppressors and the circuit terminals. The mean duration of lockouts is shown to be proportional to the relay hangovers. Introduction WHEN carrying on a conversation over a telephone circuit of moderate length, the subscribers are ordinarily unaware of any limitations imposed upon the free interchange of information. As the length of the circuit is increased the time factor ^ becomes increasingly important and may become manifest in a number of ways. One result of the time factor is the occurrence of echoes which become apparent when the speech energy reflected from the end of the circuit is delayed in returning to the talking subscriber. When the circuit is equipped with an echo suppressor to render this efifect unnoticeable, or when a long connection of two such circuits is made, the action of the suppres- sors is such as to make the circuit inoperative in the opposite direction to which speech is being transmitted. Consequently the subscribers are no longer able to interchange information with the ease and rapidity that would be enjoyed on a shorter circuit. 1 "The Time Factor in Telephone Transmission," O. B. Blackwell, Bell System Technical Journal, January 1932. 258 THE OCCURRENCE AND EFFECT OF LOCKOUT 259 A circuit equipped with a single echo suppressor is always operative in one direction, and although both subscribers may start to talk at about the same instant, one or the other will always obtain control of the circuit and his speech will be heard by the other subscriber. The principal difficulties encountered on circuits of this type become ap- parent when the hangover times of the relays are large. There is some difficulty in interrupting since the relays do not release during the pauses between words, and a quick response following a pause by the first talker may reach the suppressing relay before it has released, result- ing in a mutilation of the initial part of the response. When two echo suppressors are used, as is the case when two circuits each equipped with an echo suppressor are connected in tandem, similar difficulties may be encountered. In addition, lockout, or blocking of transmission in both directions, may occur and may persist for an ap- preciable time. Since neither subscriber is aware that the other is talking, both may continue talking until one or the other of the relays releases during a pause and enables the circuit in the appropriate direc- tion. Thus neither subscriber will be conscious of the fact that a lockout has occurred unless he realizes from the context that some part of the conversation has been lost. This paper discusses the manner in which lockouts can occur, and presents the results of a series of tests to determine their effect upon conversation as measured by repetition rate.^ These results indicate that the repetition rate increases with the per cent of time during which lockout occurs. It is shown that the locked out time can be approxi- mately calculated in terms of the circuit constants and suitable charac- teristic intervals of conversational speech, and the calculated values can in turn be used to predict the effects of lockout on repetition rate. In terms of the effect upon the talkers, a lockout may be considered to occur when speech currents from one talker are prevented from reaching the other talker by one of the suppressing relays and those same speech currents operate another suppressor in such a way that speech currents from the latter talker are prevented from reaching the former. This description of lockout should not be considered as a precise definition since it does not specify the duration of a lockout. No definition in terms of measurements made upon speech at the circuit terminals would be free from difficulties in practical application, such as that of determining with sufficient precision the instants at which speech is considered to start and stop, and that of determining the direction of transmission. A definition in terms of the operations of ^ "Rating the Transmission Performance of Telephone Circuits," W. H. Martin, Bell System Technical Journal, January 1931. 260 BELL SYSTEM TECHNICAL JOURNAL the suppressors is somewhat simpler to formulate, but may be difficult to apply when the echo suppressors are separated geographically. In the tests to be described there was no such separation involved and consequently the operation of the suppressors could be readily ob- served and easily and accurately measured. Accordingly for the pur- pose of this paper we shall define a lockout as the condition in which the suppressors are operated in such a way that both directions of trans- mission are simultaneously blocked. In general, lockouts may be caused by speech, or noise, or both, but the term will be used here to apply to the case in which operations of the suppressors have been caused by speech from both ends of the circuit. In the course of a conversation the interchange of speech is ordinarily such that the circuit is alternately disabled by the two suppressors in one direction or the other depending upon the direction of transmission. When a pause of sufficient duration occurs, the party not in control of the circuit may reply at such a time that he obtains control of the echo suppressor nearest to his end of the circuit, and a lockout can occur provided that his speech does not reach the distant suppressor until after the party formerly in control of the circuit has resumed talking and has obtained control of that suppressor. The occurrence of lock- out is therefore dependent upon the time intervals in conversational speech and upon the constants of the circuit. The Manner in Which Lockout can Occur The characteristic time intervals of conversational speech upon which the occurrence of lockout depends, are treated in a companion paper by Mr. Norwine and Mr. Murphy.^ It is sufficient here to define two such characteristic intervals based on a simplified concept of a conver- sation. Neglecting grammatical considerations we can consider speech to be composed of a sequence of vocal intervals defined and separated by silent intervals. The lengths of these silent intervals will be called resumption times. Likewise a conversation may be considered to be composed of an alternate succession of speeches, defined and separated by intervals, the lengths of which will be called response times. An ambiguity occurs when both parties talk simultaneously but, for the purpose of this discussion, it will be sufficient to allow for this situation by admitting negative response times. Figure 1 represents a generalized four-wire circuit equipped with two echo suppressors located at different distances from the ends of the circuit. The transmission times of the different parts of the circuit are 3 " Characteristic Time Intervals in Telephonic Conversation," A. C. Norwine and O. J. Murphy, this issue of the Bell System Technical Journal. THE OCCURRENCE AND EFFECT OF LOCKOUT 261 indicated on the figure with appropriate subscripts and the two direc- tions of transmission are differentiated by the primed and unprimed notation. The suppression points are indicated by arrows which repre- sent an opening of the transmission path when the relays, or other sup- /w+'''+Te=T w — — E Tw+r + re=T Fig. 1 — Schematic of generalized four-wire circuit equipped with two echo suppressors. pression devices are operated. The suppressing relays are specified by a notation which refers either to the particular relay or to its hang- over, or releasing time. According to the definition given above a lockout exists during the time that both the relays he and hj are operated.* With the exception of the beginning and end of the conversation the occurrence of lockout can be described in terms of the resumption and response times following a pause by one talker, and the constants of the circuit. Referring to Fig. 1, and considering the sequence of events following a pause by E, we shall see that two types of lockout can occur. The first type, which is the one usually met in practice, can occur when he < hw + r, and hv> releases after he. A response by W and a resumption by E are necessary to produce a lockout. It will persist as long as both E and PF continue to talk and for an additional time equal to the delay from the end of the circuit to the first relay to release after a pause by one talker, plus the hangover time of that relay. A lockout of this type may be termed a lasting lockout. The second type can occur when hw -\- t < he, and A«, releases before he. It is possible for a response by W to arrive at hu, and operate hj before he has released thus causing a lockout wihch will be terminated when he releases. A lockout of this type, which may be termed a releasing lockout, can occur without a resumption by E, or if E's resumption reaches hJ after /?« releases. If a releasing lockout has oc- ^ Also, according to the definition, when both the relays hy, and hJ are operated, a condition of no practical importance. 262 BELL SYSTEM TECHNICAL JOURNAL I THE OCCURRENCE AND EFFECT OF LOCKOUT 263 curred and Ks resumption operates he before W's response can operate hj , a second lockout which will be of the lasting type, will at once occur. Otherwise W's response will operate hJ giving control of the circuit to W. Experimental Conditions and Data To obtain experimental data of the occurrence of lockout in long distance conversations and to determine the resulting effect on repeti- tion rate, added delay and echo suppressors were inserted at the New York end of a circuit to Chicago, Illinois. This circuit is used as a tie line by the Western Electric Company for the transaction of com- pany business between its Hawthorne plant and New York office. The regular echo suppressor usually associated with the circuit at Pittsburgh was removed for these tests. The circuit arrangement em- ployed is shown schematically in Fig. 2, the added equipment being in- cluded between the dotted lines. This equipment was adjusted to have zero insertion loss and the frequency characteristic was equalized to within ± 2 db from 200 to 3000 cycles. The overall net loss from toll board to toll board was 7 db. The suppressors were 44-A echo sup- pressors operating at a sensitivity of 31 db referred to the zero level point of the circuit, except in those cases specifically mentioned. The added delay circuits were of the acoustic type consisting essentially of a suitable length of brass pipe terminated by high quality loud speaking telephones together with the necessary amplifiers and equalizers to give zero loss over the frequency range from 200 to 3000 cycles. These de- lay circuits were available in units of 0.023, 0.05, 0.08, 0.10 and 0.15 second, and various combinations of these delays were used together with the tie line delay of 0.043 second to obtain the circuit conditions which were tested. The details of the recording mechanism are indicated schematically in Fig. 2. A relay was added in series with the shorting relay of each echo suppressor, so that every operation of the echo suppressor relay was accompanied by an operation of the added relay. The simultane- ous operation of these relays energized two other relays, one of which in turn operated a message register to record the number of lockouts, and the other connected a 20-cycle oscillator to a cycle counter to record the locked out time. Service observers at New York monitored both directions of the conversation and recorded repetitions and other pertinent data regard- ing each call. The circuit conditions tested are shown in Table I, the notation of which corresponds to Fig. 1. For each condition the first line re- 264 BELL SYSTEM TECHNICAL JOURNAL fers to transmission from west to east and the second line from east to west. The hangovers are those of the relays which short the indicated transmission path, for example, the figure 0.186 in the first line of Table I refers to the hangover of the relay at the west end of the circuit which shorts the transmission path from west to east. The designation in Table I indicates the grouping of conditions for observation. All con- ditions having the same numeral in the designation were observed con- currently, the procedure being to observe 25 calls on condition a, then 25 calls on condition h and so on. In this way seasonal variations and uncontrolled effects at the terminals or in the transmission line have been minimized for a group of conditions bearing the same numerical desig- TABLE I Condition T Tw hu, T he Te la 0.139 0.139 0.043 0.043 0.186 0.200 0.073 0.073 0.146 0.146 0.023 0.023 \h 0.193 0.193 0.043 0.043 0.186 0.200 0.100 0.100 0.200 0.200 0.050 0.050 \c 0.293 0.293 0.043 0.043 0.186 0.200 0.150 0.150 0.300 0.300 0.100 0.100 la Same as la Ih 0.139 0.139 0.043 0.043 0.186 0.200 0.073 0.073 0.200 0.200 0.023 0.023 2c 0.139 0.139 0.043 0.043 0.186 0.200 0.073 0.073 0.300 0.300 0.023 0.023 3a Same as Ic 3b 0.316 0.316 0.043 0.043 0.186 0.250 0.250 0.146 0.023 0.023 4a Same as Ic, suppressor sensitivities 28 db U Same as \c, suppressor sensitivities 31 db Ac Same as \c, suppressor sensitivities 34 db 5a Same as \c, suppressor sensitivities 31 db 5b Same as Ic, suppressor sensitivities 41 db 6a Same as Ic, suppressor sensitivities 47 db la 0.116 0.116 0.043 0.043 0.136 0.150 0.050 0.050 0.146 0.100 0.023 0.023 n 0.116 0.116 0.043 0.043 0.136 0.170 0.050 0.050 0.210 0.146 0.023 0.023 THE OCCURRENCE AND EFFECT OF LOCKOUT TABLE I (Continued) 265 Condition T Tw K T h. Te 8a 0.116 0.116 0.043 0.043 0.136 0.150 0.050 0.050 0.146 0.146 0.023 0.023 Sb 0.116 0.116 0.043 0.043 0.136 0.170 0.050 0.050 0.210 0.096 0.023 0.023 9a 0.093 0.093 0.043 0.043 0.136 0.050 0.000 0.000 0.036 0.150 0.050 0.050 9b 0.093 0.093 0.043 0.043 0.136 0.150 0.000 0.000 0.136 0.150 0.050 0.050 9c 0.093 0.093 0.043 0.043 0.136 0.250 0.000 0.000 0.236 0.150 0.050 0.050 10a 0.116 0.116 0.043 0.043 0.136 0.100 0.050 0.050 0.100 0.096 0.023 0.023 106 0.193 0.193 0.043 0.043 0.136 0.150 0.100 0.100 0.150 0.150 0.050 0.050 lOr 0.293 0.293 0.043 0.043 0.136 0.150 0.150 0.150 0.250 0.250 0.100 0.100 11a 0.116 0.116 0.043 0.043 0.186 0.186 0.023 0.023 0.186 0.186 0.050 0.050 \\h 0.216 0.216 0.093 0.093 0.286 0.286 0.023 0.023 0.286 0.286 0.100 0.100 \\c 0.296 0.296 0.093 0.093 0.286 0.286 0.123 0.123 0.286 0.286 0.080 0.080 Notes The values of delay in the column headed Tw include the delay of the tie line and the added artificial delay. Circuit ib arranged with relays he and hy, to short the echo suppressor without shorting the transmission path. nation. Observations were also made from time to time on the tie line without added delay and with a single echo suppressor of special design. Since this condition was not subject to lockout, these observations may be used to give an indication of the seasonal effects, and to correct data obtained from the different groups of tests. These data are designated by the letter n. The data recorded consist of the duration of each call, the number of lockouts per call, the total locked out time per call, and the number of repetitions per call. Calls having a duration less than 100 seconds are not included in the data. Table II gives the number of calls observed, the mean duration of the calls in seconds, the number of lockouts per 266 BELL SYSTEM TECHNICAL JOURNAL 100 seconds (L/lOO), the per cent of time locked out, or locked out time in seconds per 100 seconds (LT/lOO), the number of repetitions per 100 seconds (i?/100), and the number of repetitions per 100 seconds cor- TABLE II Condition Number of Calls Mean Duration Seconds L 100 LT 100 R 100 R' 100 \a 275 451 2.61 1.13 0.40 0.40 16 275 437 2.63 1.40 0.44 0.44 U 275 456 3.68 2.34 0.53 0.53 \n 275 401 0.36 la 200 439 2.62 1.03 0.36 0.36 Ih 200 410 2.21 1.19 0.47 0.47 2c 200 393 3.86 1.54 0.45 0.45 ia 300 424 3.61 2.34 0.51 0.51 36 300 397 5.90 1.85 4a 275 457 3.26 1.63 0.53 0.51 46 275 413 3.34 1.76 0.55 0.53 4c 275 432 3.13 2.02 0.55 0.53 4» 75 396 0.38 5a 275 436 3.99 1.99 0.60 0.53 56 275 425 4.03 2.37 0.56 0.49 5n 250 392 0.43 6a 50 391 8.34 4.26 0.72 0.67 6n 125 390 0.41 la 200 410 3.00 0.79 0.52 0.41 lb 275 387 4.26 1.23 0.55 0.44 In 75 395 0.47 8a 150 387 2.51 0.74 0.49 0.36 86 300 401 4.64 1.30 0.53 0.40 8w 100 393 0.49 9a 150 413 0.83 0.02 0.43 0.36 96 150 403 2.59 0.17 0.42 0.35 9c 150 380 5.42 1.00 0.44 0.37 9n 175 399 0.43 10a 300 401 3.84 0.91 0.46 0.44 106 300 405 4.46 1.78 0.44 0.42 10c; 300 420 5.28 2.66 0.54 0.52 lOw 175 439 0.38 11a 300 413 1.64 0.64 0.42 0.37 116' 300 408 0.94 0.56 0.38 0.33 He 300 430 2.99 2.28 0.56 0.51 llw 300 396 0.41 rected for seasonal variations {R' ji^Q). These rates are obtained by dividing the total number of occurrences, or locked out time by the total duration for each test condition. Effect of Lockout on Repetition Rate It would be reasonable to expect that the repetition rate would de- pend not only on the lockout rate, but also on the duration and type of lockout. Considering the data as a whole there does not appear to be THE OCCURRENCE AND EFFECT OF LOCKOUT 267 any definite relation between the lockout rate and repetition rate, al- though in most cases an increase in lockout rate results in an increase in repetition rate. If we exclude from consideration those cases in which the lockouts are of very short duration and in which releasing lockouts occur, the data indicate a somewhat closer dependence of repe- tition rate on lockout rate. This suggests that the increase in repeti- tion rate caused by lockouts may be proportional to the duration of lockouts and to their frequency of occurrence, or to the per cent of time which is locked out. Fig. 3, which shows the repetition rates. 0.7 10 1 0.6 O O lU 10 §0.5 a UJ Q. ,.^-^ ,.-" y ,'-' k' O £f^ -.-^' ^y"' 0 <' J>-^ 0 . z "-^ o \- < l- UJ Si 0.3 0.2 ^<. o ^^ P.-^" y 1.5 2.0 2.5 3.0 3.5 PER CENT OF TIME LOCKED OUT Fig. 3 — Observed variation of corrected repetition rate with per cent of time locked out. corrected for seasonal variations, plotted against the per cent of time locked out, indicates a reasonable agreement with this assumption. The correction is applied by subtracting from the observed repetition rate the difference between the observed repetition rate for the appro- priate reference or n condition and a rate of 0.36, arbitrarily chosen as equal to the lowest repetition rate observed on any of the n conditions. All of the data are included in this figure. The dashed lines are drawn to include all the data and have a slope estimated as average from considering the data in individual groups. The variability in the data may in part be attributed to the variation in the distribution of lockout durations, since if two distributions have the same mean value but different spreads, the lockouts comprising the distribution which in- cludes a greater number of long lockouts might be expected to have a greater effect upon the repetition rate. With due allowance for the variability of the data. Fig. 3 indicates that the repetition rate increases proportionally with the per cent of time locked out except 268 BELL SYSTEM TECHNICAL JOURNAL possibly for values less than 0.6 per cent. The slopes of the boundary lines are such as to show about 0.1 increase in repetition rate with each 1 per cent of locked out time, and this relation appears to hold for re- leasing as well as lasting lockouts, and for lockouts which may be caused by relay operations by noise. Certain qualifications are necessary in considering the significance of this result. The indicated increase in repetition rate may be partly due to other causes than lockout, as for example the effects introduced by the delay of the circuit, or by the relay hangover, during changes in the direction of speech transmission which are not accompanied by lockout. The net effect of these causes increases with circuit changes which in- crease the per cent of time locked out. Consequently, the latter may be taken as a criterion of the total effect, even though the contribu- tion of the former to the repetition rate may be appreciable. No general significance can be attached to the absolute values of the repetition rates observed in these tests since it is well known that repe- tition rates will differ for identical circuit conditions used with different terminal conditions and by different classes of telephone subscribers. These observed rates are significant only for comparing the relative performance of circuits under the particular conditions of use pertain- ing to these tests. The significance of the results obtained depends upon the assumption that a change in lockout which causes an increase in repetition rate is an undesirable change and the transmission performance is thereby degraded. In the case of certain circuit changes which introduce changes in intelligibility the resulting changes in repetition rate can be used to determine effective transmission ratings,^ expressed in db, of the circuits under consideration. A corresponding procedure might be applied to express the observed changes in repetition rate due to lockout in terms of db, but in the absence of data to establish the equivalence of the ratings for different types of degradation, it has not seemed advisable to do so. Locked Out Time in Terms of Circuit Constants Since these tests indicate that the repetition rate is proportional to the per cent of time locked out we can limit our consideration to the latter as a suitable criterion for measuring the relative merit of circuits equipped with two echo suppressors. To determine the per cent of time locked out we can measure it directly, as has been done in these tests, or it can be calculated in terms of the circuit constants by deter- * "Scientific Research Applied to the Telephone Transmitter and Receiver," Edwin H. Colpitts, Bell System Technical Journal, July 1937. THE OCCURRENCE AND EFFECT OF LOCKOUT 269 mining the average number of lockouts per hundred seconds and the average duration of lockouts in terms of the circuit constants and ob- taining the per cent of locked out time as the product of these two quantities. The average, or expected number of lockouts per hundred seconds can be approximately determined from the circuit constants and the distributions of response and resumption times. It is shown in the appendix that, subject to certain assumptions, the probability of lock- out following a pause is given by ^ = \ ] P^^^^ ^2(.v) dx dy, (1) in which pi(x) dx and piiy) dy are the probabilities that, following a pause, the resumption time will be between x and x -\- dx and the re- V) 12 r \ RESPONSE \ TIMES T / r s 1 1 \ \ \ I 1 V \ RESUMPTION \ TIMES \ \ > \ / \. \ >»_ . y / ■^ ■»«.^ "^— 0.5 1.0 TIME IN SECONDS Fig. 4 — Observed distribution of resumption and response times. sponse time will be between y and y + dy. As suitable approximations to these probabilities we may take the observed distributions of re- sumption and response times. Mr. Norwine and Mr. Murphy, in their accompanying paper,^ give distributions of resumption and response times which are shown in Fig. 4. These distributions are expressed in terms of the total number of resumptions, or responses and conse- quently the data are an approximation to the conditional probability that if a resumption or response has occurred, the resumption or re- sponse time will be between / and / + dt. The use of these data will ^ Loc. cit. 270 BELL SYSTEM TECHNICAL JOURNAL therefore result in calculated values of the probability of lockout which are proportional to the desired probability, and if the value of the integral calculated from their data is p, then P = kp, (2) where ^ is a constant of proportionality which depends on the average number of pauses occurring, and which can be determined by comparing observed and calculated results. The observed lockouts per hundred seconds plotted against the cal- culated probability of lockout for each circuit condition are shown in Fig. 5 for lasting lockouts and in Fig. 6 for releasing lockouts. The .-' P ,p y' ^-» ^jr" *''' n ^ ^ X cr"' y 8 o -# y ^fi 9 ^- ' .,'-' ^ 0 0.05 0.10 0.15 0.20 0.25 0.30 0.35 0.40 0.45 0.50 CALCULATED PROBABILITY Fig. 5 — Observed lasting lockouts vs. calculated probability. data are separated in this way, since the factor of proportionality be- tween observed and calculated results is found to be different for the two cases. This is probably due to the fact that the method of deter- mining response and resumption times was such that some of the nega- tive and shorter positive response times could not be detected. An increase in the number of these response times would result in an in- creased probability of releasing lockouts, which would tend to bring the two sets of data into agreement. Greater accuracy might be obtained if the distribution of response times were to be more accurately deter- mined, but the present data are sufficient for approximate calculations. In both figures the data obtained in a single group of tests are con- nected by dotted lines. The solid lines are the best estimates to repre- THE OCCURRENCE AND EFFECT OF LOCKOUT 271 sent the two complete sets of data. In the case of Fig. 5 the solid line was determined by the method of least squares, omitting the data of group 10. This omission appears to be justified since these data are consistent among themselves and yield a factor of proportionality 2 3.5 ."; 2.0 / / P, / // / ^ yj^ / P '>^ / ^y^ / y y/ / P / / y Z/ y y / f y/ y^ / y y y J y y y / / P y ^/z f / / y f<'/ ' if y / y / y / / / / y y y ^ y' ;: / ^ / y y / / / v/ y^ y y / / ^ / y y Xy y / /^ .^ y '^ / y y y < V y If y y y / y y / 0.10 0.15 0.20 CALCULATED PROBABILITY Fig. 6 — Observed releasing lockouts vs. calculated probability. which is consistent with the rest of the data, and since it is known that many uncontrolled factors may influence the results of one particular group of tests. In the case of Fig. 6 the solid line was obtained by averaging the slopes and constant terms of the individual dotted lines. Both sets of data indicate that about 0.9 lockout per hundred seconds occurs when the calculated probability of occurrence is zero. This is undoubtedly due to non-synchronous action of the suppressors, caused by slight variations in sensitivity, changes in effective hangover caused by changes in sensitivity and by occasional relay chatter. This con- 272 BELL SYSTEM TECHNICAL JOURNAL elusion is confirmed by the tests made with no delay between the sup- pressors, in which lockout is obviously impossible with synchronous action, but in which lockouts were actually obtained in amount con- sistent with the rest of the data. Figures 5 and 6 show that the number of lasting and releasing lock- outs can be calculated from the circuit constants and the distributions of response and resumption times. Approximations which are suffi- 9 0.7 <^ 0.6 o ,^ / L / \ . ^ < K ^ i D o /" y r " c v ^ y ^ y^ 0.2 0.1 0 0.1 0.2 0.3, 0.4 0.5 0.6 0.7 tie + I^W IN SECONDS Fig. 7 — Observed length of lockout as a function of relay hangovers. cient for practical purposes for calculating the number of lasting and releasing lockouts are respectively Li Lr 0.9 + 11.0 />, 0.9 + 17.7/?. (3) (4) The duration of a lockout is obviously dependent upon the way in which the subscribers talk and upon the hangovers of relays //,„ and hj . Lockouts of several seconds duration have frequently been observed but most frequently the duration of lockouts appears to be short and de- termined primarily by the relay hangovers. Figure 7 shows the mean duration of lasting lockouts plotted as a function of the sum of the relay hangovers, he + hj . The straight line is the least square representa- tion of the data, which is Di = 0.002 + 1.16 {h, + hj). (5) The constant term in this equation can be neglected for approximate calculations. THE OCCURRENCE AND EFFECT OF LOCKOUT 273 0.50 T = 0.45 < a = 0.1 O.AO ^ 0.35 ^ ^ 0.30, ^^ >>^ ■ 0.25y / / ^^ \ 0.20 J l^ ■-— ^ / ^ 7 ^^ ^ 0.10 / ^ -^ / 0.05> L ■ ■ 0.05 0.10 0.20 0.25 0.30 0.35 0.40 0.45 0.50 T IN SECONDS Fig. 8 — Calculated probability of lockout as a function of total circuit delay. Since the duration of lasting and releasing lockouts is not the same, releasing lockouts being of short duration, and since the two were not separately observed, the data are insufficient to determine the duration of releasing lockouts directly. However, tentative calculations indi- cate that approximate results can be obtained by assuming that the 274 BELL SYSTEM TECHNICAL JOURNAL mean duration of releasing lockouts is about one-quarter that of lasting lockouts or a D,. = 0.29 (//. + hj). (6) 0.6S T=0.50y^ 0.60 0.65 OAo/ / a = 0.1 / 0.50 0.45 / / / 0.30 >/ / / ^ / 0.40 0.35 / / y // ^ l/ / 0.25/^ / / / 0.30 0.25 / /^ :y / / y ^ // / y 0.15^ / y'^ ^^ / V' y ^ 0,10 // V/ /y y ^ ^^ 0.075 / / / / > ^ 1,—*-" //, ^ / ^ ■ 0.050 M i^^^ =^== "^ 0.025 0 0.2 0.3 0.5 T 0.8 0.9 Fig. 9 — Calculated probability of lockout as a function of the ratio of the delay between the suppressors to the total circuit delay. With the above relations between probability of lockout and circuit constants, number of lockouts and probability of lockout, and mean duration of lockouts and relay hangovers, it is possible to determine the i I THE OCCURRENCE AND EFFECT OF LOCKOUT 275 per cent of time locked out from the circuit constants. Since this is proportional to the repetition rate, a measure of relative circuit per- formance is obtained in terms of the circuit constants. As an example of such calculations let us assume that t,„ = tJ = Te = T,.', T — T and T = T\ and /i,„ = h,^ = 2t,„ + a. Then the constants of integration determined in the appendix become, a = a, b = a -\- T - T, c = a + 2" -f T, and the probability of a lasting lockout is proportional to "a+T+T n^ r'y+T+T P Xa r^a+T+T n^ r'y+T+T p2{y)dy I pi(x)dx + I p2iy)dy I pi{x)dx. (7) Values of this probability for a = 0.1 are shown in Fig. 8 as a function of the transmission time T with r, the delay between the suppressors as a parameter and in Fig. 9 as a function tJT with 7" as a parameter. The curves are not extended beyond T — 0.5 since smaller values are thought to cover the range of practical interest. Furthermore, for large values of T there is some evidence that the effect of the transmis- sion time would be noticed by the subscribers with a consequent change in the distributions of resumption and response times. These curves indicate that for a constant value of r, the delay be- tween the echo suppressors, there is little change in the probability of lockout as the total delay of the circuit T is increased, and for a constant value of T the probability of lockout is approximately proportional to t. To continue with a more specific example, let us consider a telephone connection consisting of two four-wire circuits each equipped with an echo suppressor in the center of the circuit as shown in Fig. 10. In the notation of Fig. 10 the relay hangovers are each equal to r + 0.100 and the constants of integration are a = 0.100, b = 0.100 + r, c = 0.100 + 3t. Since the two circuits are assumed equal, only lasting lockouts are theoretically possible and the curves of Fig. 8 may then be used to determine p in terms of r as defined by equation (7), which in turn may be used to determine the expected number of lockouts from equation (3). The mean duration of lockout is obtained by inserting the value 276 BELL SYSTEM TECHNICAL JOURNAL of the relay hangover in equation (5) giving D = 0.234 + 2.32 r. The product of D with the expected number of lockouts per hundred seconds is then equal to the per cent of time locked out, which is shown in Fig. 10 as a function of t. By using the relation shown in Fig. 3 a . 3.5 9 0.55 - SO.50 !}J0.40 - l^ __ J ^ r -^. T -< > y /" k h ^ ^ ^ ^ ECHO / ^ _ SUPPRESSOR f^ ' \ / / y ^ y ^ ^ 2.5Q 2.0 (u 2 0.04 0.06 0.08 0.10 0.12 T IN SECONDS Fig. 10 — Calculated per cent of time locked out, and repetition rate for the indicated circuit conditions. second scale is shown in Fig. 10 to give the relation between the repeti- tion rate and the delay between the suppressors. This curve shows that the repetition rate increases with the delay between the suppres- sors, at a gradually increasing rate up to a delay of about 0.09 seconds, beyond which the impairment increases linearly with the delay. Summary It has been shown that two types of lockout, lasting and releasing lockouts, may occur in telephone connections involving two echo sup- pressors, and the manner of their occurrence has been discussed. The results of an experimental investigation show that the occur- rence of lockouts causes an increase in repetition rate, which is ap- proximately proportional to the per cent of time locked out. There has been presented a theoretical method for calculating the expected number of lockouts in terms of the circuit constants which de- THE OCCURRENCE AND EFFECT OF LOCKOUT 277 pends upon the characteristic time intervals in conversational speech. The values which have been calculated with experimentally determined constants are shown to agree with the observed values. The average duration of lockouts has been found to be proportional to the hangovers of the relays effective in lockout. Since the per cent of time locked out is equal to the product of the average number of lockouts per hundred seconds and the average dura- tion of lockouts, it may be determined in terms of the circuit constants, and used as one of the criteria of the relative performance of the circuits under consideration. Specific examples of such calculations have been used to illustrate the relations between the expected number of lockouts and the circuit con- stants, and between the repetition rate and the constants of a particular circuit configuration. Subject to certain restrictions on the relations between the circuit constants, it appears that the number of lockouts and the resulting in- crease in repetition rate are approximately proportional to the delay included between the echo suppressors. In conclusion I wish to express my appreciation to my associates who have contributed to this study; in particular to Dr. G. R. Stibitz who first developed the theoretical approach to the problem, to Mr. W. R. Bennett and Mr. B. D. Holbrook who have contributed to the extension of this approach, and to Mr. A. C. Norwine and Mr. O. J. Murphy who obtained the distribution functions used in the calculations and con- ducted the experimental work. Appendix To assist in formulating an expression for the probability of lockout a number of simplifying assumptions have been made, as follows: Pauses in speech are sufficiently separated to be considered as independent events, or in other words the sequence of events oc- curring at one pause have no effect upon those occurring at another. Following a pause each speaker can start speaking only once and only one of three events can occur. 1. The original speaker regains control of the circuit. 2. The other speaker obtains control of the circuit. 3. Lockout occurs. Resumption and response times are independent. The distributions of response and resumption times are inde- pendent of the delay of the circuit and the disposition of the echo suppressor. The operate times of the suppressors are sufficiently small to be neglected. 278 BELL SYSTEM TECHNICAL JOURNAL Let pi{t)dt be the probability that the speaker in control of the circuit will resume speaking in the interval t to t -{- dt after pausing and let p2{t)dt be the probability that the speaker not in control of the circuit will start speaking^^in the interval t to t -\- dt after hearing the other speaker pause. In the latter case / may be negative. Then the proba- bility that, following a pause, a resumption will occur in the interval X to X -\- dx and a response will occur in the interval y to y -\- dy \s given by p\{x) pi{y) dx dy, and the probability of lockout following a pause is given by P = j j pi(x) piiy) dx dy, in which the integration is to be performed over the region in the xy plane which contains those values of x and y for which lockout occurs. Assuming that either subscriber is equally likely to have control of the circuit at any instant, the probability of a lockout following a pause by either party is given by the average of the probabilities for the two parties. In determining the limits of integration there are three cases to be considered. Assuming a pause by E I. //,. < //,„ + r, II. //,„ + r < //„ < h,, -i- T -\- t', III. /;,„ + T + r' < h. In case I only lasting lockouts can occur while in cases II and III both lasting and releasing lockouts can occur. Case I will be used to illustrate a method of determining the limits of integration which can also be applied to cases II and III for which the results will be stated without proof. In Fig. 11 which is based on the circuit of Fig. 1, time is represented horizontally and distances vertically, upward from the central line, which represents the east end of the circuit, for transmission from W to E and downward for transmission from E to W. Consider a pause by E occurring at x = 0 at ^. The line ABDG represents the transmission of this pause to W. The point H obtained by projecting G to the top line determines the point y — 0. The points C and F represent the instants at which he and /?„, release. If E resumes in the interval AI, the resumption will arrive at the input of hw before hw has released as determined by the point F, and E retains control of the circuit. If, on the other hand W responds at any time prior to THE OCCURRENCE AND EFFECT OF LOCKOUT 279 / the response will be blocked by /?,,. until the time represented by K and it will then be transmitted to the end of the circuit as shown by the line JKLM. If now E resumes at any time after P the resumption Fig. 11 — Time relations in four-wire circuit for determination of limits of integration. will be blocked by hj since H^will have obtained control of the circuit. A resumption in the interval IP will result in lockout since W will control hiJ and E will control hr. If W responds at some time after /, say at Q, a similar argument can be used to show that E must resume in the interval SR to cause lockout. If we let /// = a, AI = b, AP = c, it can be shown that AS - 3' - a + b, AR = y — a + c, and therefore the region of integration is defined by b < X < c, a -\- b < X < y — -x < y < a, a -\- c, a < V < '^ , in which a is the time interval the speaker not in control must wait after hearing the other speaker pause in order to enable the response to get through the circuit; b is the time interval after the speaker in control pauses, during which he can gain control by resuming, regard- less of what the other speaker does; c is the time interval the speaker in control must pause in order to make it possible for the other speaker to get a response through the circuit. 280 BELL SYSTEM TECHNICAL JOURNAL These constants have the values a = Kc — {tw + r,„'), h = h^, C = hu, + (r + r'). Case II. By the same method the regions of integration are determined to be, for lasting lockouts y < a, b < X < c, y < a, y — a-{-b ",< > ^ ^, w Uj k +\ *< lAi i Ul S S h H ^ U.1 '■, Q 52 r; h- ^ '^ a o *» i7) iii rr Q •>] u-i u i> O m ^"^ X '^y / \ / N ^' ' , % '^ ? J V ! i ' J* *^ V I,' i \ ''ik' **<"■ ^^.^i f z* ■', S" / J , V' \ . Aii r ' 1 r 1 f I n \1"" I.V 1 f \ V. 1 (1 J 1 A i 1 I 1 ' 1 " t ) 1 i u) % )- ; 11 '! 4 1* IT lU C| ■; ^ 1 t t- J ft -J ^ r u 1 O O -Z ^ CO ^ t i I ^ I c •n «. •J 1, 1. i. ! X 1 K Fig. 2 — Typical sections of oscillographic records. a. Circuit reversals. New York completed talkspurt, heard Chicago's reply, and began another talkspurt. b. Reply by New York during pause by Chicago caused lockout. New York gained control of the circuit. c. Short reply by Chicago during pause by New York caused lockout. New York regained control of the circuit. d. Negative response time. Chicago replied before the completion of New York's talkspurt. TIME INTERVALS IN TELEPHONIC CONVERSATION 287 A few samples from the original oscillograms are shown in Fig. 2. The speech energy in each sample is shown on traces 3 and 4 counting from the top down, the upper being from Chicago and the lower from New York. The cyclic waves on traces 2, 5 and 6 indicate respectively lockout, establishment by Chicago and establishment by New York } These waves were obtained from an oscillator which was concurrently used to drive an escapement-type electric clock for measuring the total call duration. The top oscillogram was selected to show the simplest type of con- versational interchange. It will be seen that New York had been talk- ing but had reached the end of his talkspurt as marked on the film. Approximately 0.4 second later Chicago responded, his talkspurt ap- parently consisting of three syllables, whereupon after a further time of about 0.35 second New York responded and continued talking. The second film was selected to show a less simple type of interchange where- in a long pause within a talkspurt prompted the listener to reply. In this instance the times were such that a lockout resulted. Since the remainder of the talkspurt by the original talker, Chicago, was short and the responding party. New York, continued talking, the circuit was established in New York's direction after the lockout. In the third oscillographic strip Chicago attempted to interrupt, and a short pause by New York permitted lockout to occur; Chicago did not gain control of the circuit. This is an example of concurrent talkspurts, both of which were included in the data. The fourth example was chosen to illustrate a negative response time. In this case Chicago began to reply before the end of New York's talkspurt; no lockout occurred, but the first part of the reply was inaudible to New York due to continued establishment of the circuit in the opposite direction. It may be noted in Fig. 1 that speech from Chicago was recorded 0.25 second before it was heard by New York and that speech from New York was recorded 0.193 second before it arrived at Chicago. Likewise the beginning of each response did not occur at the time shown on the oscillograms but at a time previous by the delay from the talker's position to that of the recording means. To obtain the response times as previously defined each apparent response time was given an ap- propriate time correction. Data Obtained The more detailed observations were made on fifty-one calls with a total recorded duration of a little over 13,000 seconds. At the record- ing speed of 20 feet per minute this resulted in about 4400 feet of * An establishment by a talker is said to occur when his speech energy has gained control of all voice operated equipment in his transmission path. 288 BELL SYSTEM TECHNICAL JOURNAL oscillograms. In all cases recording began at the start of the call, but in some instances recording was stopped before the termination of the call due to lack of recording paper in the oscillograph. The oscillo- grams, ranging in length from 29.6 to 660.8 seconds, represented ob- servations on calls whose mean duration was 430.5 seconds. The speed of recording was such that the time intervals under observation could readily be measured with a precision of db 0.005 second. The conversational elements were measured with this precision and listed in their order of occurrence for each call. The records for all calls were 10 CD (D s- MODE 1 ^^-^ 1 / / /r ^ MEDIAN MEAN / 1 / 1 1 \ \ Y 1 SUMMATION / \ \/ / / \ \ y/\ 1 / \ / \ ' 1 2845 1 OBSERVATIONS / / y \ y 1 1 / / ^ "- 1 1 DISTRIBUTION ,- -■ y > ^^"^ 1- ■- ... 0.5 I 2 TIME IN SECONDS Fig. 3 — Lengths of talkspurts. 70 UJ < O -2 30 20 then consolidated and retabulated in terms of the number of instances of each element whose duration could be included within each of a regular progression of time increments. For all three items of data time cells 0.10 second wide were chosen. The data, when thus cellularized, provided the basis for the construction of histograms from which the time-distribution curves were obtained. These distribution curves and their respective summation curves are given in Figs. 3, 4, and 5. Some of the statistically significant quantities ® are tabulated on the opposite page. The values are time intervals in seconds. Since most telephonic speech syllables are shorter than 0.3 second the modal value of 0.25 second for the length of talkspurts makes it clear that monosyllabic replies are by far the most numerous. From ^ The moie is the value which occurs most frequently, i.e., the peak of the dis- tribution curve. The median is that value above and below which equal numbers of observations lie. The mean is the arithmetic average of all the values observed. TIME INTERVALS IN TELEPHONIC CONVERSATION 289 I in 14 oz pg 12 ^ SUMMATION MEAN . MODE 1 MEDIAN 1 y ^ / -f — ^^ -y 1 1 1 / 1 2811 OBSERVATIONS / 1 *J ^^ 1 1 / S V ^ 1 1 / A ^-- ■-»^ DISTRIBUTION / / — 0.6 0.8 1.0 1.2 TIME IN SECONDS 100 CD 80 o 60 Oz 50 a < UJ 5 40 S 3 O 1/1 10- tr Fig. 4 — Lengths of pauses within talkspurts, i.e. resumption times. io 16 I (O 14 oz i=0 12 $5 Oif? 8 UJ z

    - '2 25 A ^ ■ VV2 / / VS.y/2 / J X FROM OBSERVATIONS WITH /^^ DIFFRACTION APPARATUS. / o SAME - PARTICULARLY RELIABLE /Tr. D SAME -GRAZING BEAMS. /xx x'^ jf X ® FROM OBSERVATIONS WITH REFLECTION APPARATUS .05 /v % Fig. 4 — Test of the de Broglie formula X = hip = Ii/mv. Wave-length computed from diffraction data plotted against l/F^'^, ( V, primar\' beam voltage). For precise verification of the formula all points should fall on the line X = 12.25/ Vi/- plotted in the diagram. I would Hke also at this time to express my admiration of the late Dr. H. D. Arnold, then Director of Research in the Bell Telephone Laboratories, and of Dr. W. Wilson, my immediate superior, who were sufficiently far-sighted to see in these researches a contribution to the science of communication. Their vision was, in fact, accurate for today in ours, as in other industrial laboratories, electron diffraction is applied with great power and efficacy for discerning the structures of materials. But neither of this nor of the many beautiful and important re- searches which have been made in electron diffraction in laboratories in all parts of the world since 1927 will I speak today. I will take time only to express my admiration of the beautiful experiments — differing from ours in every respect — by which Thomson in far-away Aberdeen also demonstrated electron diffraction and verified de Broglie's formula at the same time as we in New York. And to mention, as 482 BELL SYSTEM TECHNICAL JOURNAL closely related to the subject of this discourse, the difficult and beauti- fully executed experiments by which Stern and Esterman in 1929 showed that atomic hydrogen also is diffracted in accordance with the de Broglie-Schroedinger theory. Important and timely as was the discovery of electron diffraction in inspiring confidence in the physical reality of material waves, our confidence in this regard would hardly be less today, one imagines, were diffraction yet to be discovered, so great has been the success of the mechanics built upon the conception of such waves in clarifying the phenomena of atomic and subatomic physics. Abstracts of Technical Articles from Bell System Sources Stability of Two-Meter Waves} Charles R. Burrows, A. Decino and LoYD E. Hunt. The continuous records of the field strength received over a 60-kiIometer path on a frequency of 150 megacycles for the year 1936 are analyzed. Preliminary comparison with other paths of the same length indicate that the magnitude of the recorded variations of the signals may be typical of paths of this length. A reduction in the path length by a factor of two reduced the fading range in decibels by a factor of five. The results are found to be in agreement with an earlier formula. Fading reduced the field 7 decibels below the average value 1 per cent of the time. Loudness, Masking and Their Relation to the Hearing Process and the Problem of Noise Measurement} Harvey Fletcher. It is shown in this paper how to define loudness and loudness level in a quantitative way. Definite procedures are given for determining experimentally the loudness level of any sound heard by any person. For a typical observer a true loudness scale is developed. The relation of the scale to the loudness level scale is determined experimentally. The scale has been found to be very useful for calculating loudness from the noise spectrogram, the noise audiogram, or the overtone structure of the sound. The relation between the masking and the loudness produced by a sound has been quantitatively determined and a formula deduced from this relation which has proved useful for calculating the loud- ness. This formula may be applied with equal success to a normal ear and also to a deafened ear. Evidence has been given that the masking expressed in decibels produced upon any pure tone is equal directly to the agitation of 1.1 per cent of the total nerve endings expressed in decibels above the threshold value for such a patch and at the position where such a tone would be sensed. These loudness relations throw light upon some of the important processes involved in hearing. In particular the data from the masking effects of thermal noise were used to calculate the relation between the position of 1 Proc. I. R. E., May 1938. 2 Jour. Acous. Soc. Amer., April 1938. 483 484 BELL SYSTEM TECHNICAL JOURNAL maximum stimulation on the basilar membrane and the frequency of the tone producing the stimulation. Pick-up for Sound Motion Pictures {Including Stereophonic) .'^ J. P. Maxfield, a. W. Colledge and R. T. Friebus. Although the basic principles underlying sound pick-up for motion pictures have been understood for some time, the ability to carry them out completely in the presence of the requirements of artistry, photography, lighting, etc., has constituted a difficult problem. The paper discusses some of these problems, particularly with respect to the acoustics of produc- tion sets and scoring stages. The problems of stereophonic repro- duction are also discussed in some detail. Practical Application of Telephone Repeaters and Carrier Telephone Systems} J. A. Parrott. The paper discusses engineering problems in the application of telephone repeaters and carrier systems with which railroad communication engineers recently have been particu- larly concerned. The first part of the paper deals with crosstalk, noise, balance and overloading considerations in the design of re- peatered circuits, particularly from the standpoint of selecting the locations of repeaters to obtain the most satisfactory results on existing lines. The importance of securing test data on the wire facilities to aid in this design work as well as to serve as a guide in improving circuit conditions is emphasized. The second part of the paper briefly discusses the application of the HI carrier telephone system and provides transmission data for the preliminary design of the layout of such systems. The Type D and KIO carrier transpositions are described and features of particular interest in their possible use on railroad facilities are discussed. Sorption of Water by Rubber} R. L. Taylor and A. R. Kemp, The effect of several variables on the rate of sorption of water by rubber is discussed. Expressions based on short-time immersion tests are derived which permit calculation of the water content after an extended period of immersion under fixed conditions of temperature and vapor pressure. A sorption coefficient by which one material may be compared with another is suggested, and its application to practical problems is considered. 3 Jour. S. M. P. E., June 1938. * Proc. Assoc, of Amer. Railroads, Telegraph and Telephone Sec., October 1937. ^ Indus. & Engg. Chemistry, April 1938. ABSTRACTS OF TECHNICAL ARTICLES 485 Chemical Studies of Wood Preservation — The Wood-Block Method of Toxicity Assay.^ Robert E. Waterman, John Leutritz and Caleb M. Hill. Actual decay resistance of treated wood is used as the basis for a simple laboratory technic in the assay of materials advo- cated for the protection of wood. In its present stage of development the test is a valuable tool in wood preservation studies. ^ Indus. &• Engg. Chem., Anal. Ed., June 15, 1938. Contributors to this Issue Julian Blanchard, A.B., Trinity College (now Duke University), 1905; A.M., Columbia University, 1909; Ph.D., 1917. Professor of Engineering, Trinity College, 1909-1912; Research Assistant in Physics, Columbia University, 1912-1915. Physicist, Research Lab- oratory, Eastman Kodak Company, 1915-1917; Engineering Depart- ment, Western Electric Company, 1917-1925; Bell Telephone Labora- tories, 1925-. Dr. Blanchard's work has been concerned primarily with special studies in connection with the development of vacuum tubes and radio. B. L. Clarke, B.S., George Washington University, 1921; M.A., Columbia University, 1923; Ph.D., Columbia University, 1924. Bell Telephone Laboratories, 192 7-. Dr. Clarke has been in charge of the work in analytical chemistry since 1930. C. J. Davisson, B.Sc, University of Chicago, 1908; Ph.D., Prince- ton University, 1911; Instructor in Physics, Carnegie Institute of Technology, 1911-17. Engineering Department of the Western Elec- tric Company, 1917-25; Bell Telephone Laboratories, 1925-. As Research Physicist, Dr. Davisson is engaged in work relating largely to thermionics and electronic physics. In 1928 the National Academy of Sciences awarded the Comstock Prize to Dr. Davisson "for the most important discovery of or investi- gation in electricity or magnetism or radiant energy" made in this country during the preceding five years, for his work in this field. In 1931 he and Dr. L. H. Germer received the Elliott Cresson Medals from the Franklin Institute, Philadelphia, and in 1935 he received the Hughes Medal of the Royal Society of London. W. G. GusTAFSON, B.S. in Electrical Engineering, Union College, 1927; Columbia University, 1929-36. Bell Telephone Laboratories, 1927-. Mr. Gustafson is engaged in work relating to the development of transformers and repeating coils for communication purposes. W. Herriott was engaged in astronomical research at the Allegheny Observatory from 1914 to 1917. Research in astronomical and aerial photography at the Research Laboratories of the Eastman Kodak Company followed from 1917 to 1920. Between 1920 and 1925 he 486 CONTRIBUTORS TO THIS ISSUE 487 was engaged in the development of military instruments and of optical apparatus for microscopy, photography and motion pictures at the Bausch and Lomb Optical Company. During the following three years he was in charge of the Scientific Department of the Fairchild Aerial Camera Corporation. In 1928 he joined the engineering de- partment of Electrical Research Products, Inc., coming to the Bell Telephone Laboratories in 1929 to work on optical and photographic problems associated with sound picture apparatus development. In October 1936 he transferred to the Materials Group of the Electro- mechanical Division of the Telephone Apparatus Development De- partment. A. H. Inglis, B.A., Yale University, 1914. Western Electric Com- pany, Engineering Department, 1914-17. Signal Corps, A.E.F., 1917-19. American Telephone and Telegraph Company, Department of Development and Research, 1919-34; Bell Telephone Laboratories, 1934-. Mr. Inglis has been concerned with both equipment and transmission matters of station apparatus, latterly as Station Instru- mentalities Engineer. W. C. Jones, B.S. in Electrical Engineering, Colorado College, 1913. Western Electric Company, Engineering Department, 1913- 25 ; Bell Telephone Laboratories, 1925-. As Transmission Instru- ments Director, Mr. Jones is concerned with the development of telephone instruments and similar devices. H. C. Montgomery, A.B., University of Southern California, 1929; M.A., Columbia University, 1933. Bell Telephone Laboratories, 1929-. Engaged at first in studies of hearing acuity and related problems in physiological acoustics, Mr. Montgomery has been occu- pied more recently with the study and analysis of speech. A. E. RuEHLE, B.S., University of Idaho, 1930. Bell Telephone Laboratories, 1930-. Mr. Ruehle's work has been chiefly concerned with applications of the methods of physical chemistry to chemical analysis. G. H. Stevenson, B.Sc. in Engineering, University of Glasgow, Scotland, 1906; Instructor in Electrical Engineering, University of Glasgow, 1906-07. Messrs. Barr and Stroud, Glasgow, 1907-11. Western Electric Company, Engineering Department, 1911-24; Patent Department, 1924-25. Bell Telephone Laboratories, Patent Depart- ment, 1925-. Mr. Stevenson's work has to do with patent matters 488 BELL SYSTEM TECHNICAL JOURNAL in the fields of wave transmission networks and radio transmission systems. M. E. Strieby, A.B., Colorado College, 1914; B.S., Harvard, 1916; B.S. in E.E., Massachusetts Institute of Technology, 1916; New York Telephone Company, Engineering Department, 1916-17; Captain, Signal Corps, U. S. Army, A. E. F., 1917-19. American Telephone and Telegraph Company, Department of Development and Research, 1919-29; Bell Telephone Laboratories, 1929-, Mr. Strieby has been associated with various phases of transmission work, more particularly with the development of long toll circuits. At the present time, in his capacity as High Frequency Transmission Engineer, he directs studies of new and improved methods of carrier frequency transmission over existing or new facilities. The Bell System Technical Journal Vol. XVII October, 1938 No. 4 Ultra-Short-Wave Transmission and Atmospheric Irregularities BY C. R. ENGLUND, A. B. CRAWFORD AND W. W. MUMFORD Results of an ultra-short-wave fading study are here reported. Transmission was carried out in the range of 1.6 to 5.0 meters, over a 70 mile (112.6 kilometer) ocean path, on 106 days during a period of two years. Both horizontal and vertical polarizations were used and during part of the time a 6-megacycle amplitude, 120-cycle, frequency modulated transmission was added, for the cathode-ray tube observation of the frequency characteristics of the radio path. On 45 mornings records were taken, on vertically polarized radia- tions, during the flight period of the Mitchel Field Weather Bureau plane. Fading was found present practically all of the time. Amplitude changes up to 40 db and fading rates up to 5 fades per minute were found. Simultaneous transmission of the same wave in two polar- izations, and of two waves of different wave-length in the same polarization showed that the horizontally polarized component was practically always, and the shorter wave-length one was usually the worse fader of the pair. The greater part of the time there was no correlation between the fading of these radiation pairs; occasionally, however, and for the slow, smooth amplitude, undulating type of fading, coincidence was observed. The fre- quency sweep patterns showed multiple signal components to be present, with various degrees of relative phase retardation. A tentative explanation is proposed for these phenomena. This theory assumes the presence of a refracted-diffracted signal com- ponent, transmitted along the earth's surface and calculable in the manner of Wwedensky, Van der Pol and Gray, and one or more signal components reflected from air mass boundaries. The air- plane results are shown to be in reasonable agreement with the frequency sweep observations. Boundary heights from 5.5 kilometers down to 1.9 kilometers are measured; below 1.9 kilo- meters other boundaries are indicated. The receiver band, flat over two megacycles, sets the low height limit of resolution of reflecting boundaries at 1.9 kilometers. Most of the boundaries are at the lower heights. 489 490 BELL SYSTEM TECHNICAL JOURNAL A discussion is given of some observations of signal fading at various wave-lengths which have been reported by other ob- servers, and which are apparently referable to the same mechanism as is here proposed. Introduction IN an earlier paper ^ experimental data were presented which indi- cated that the transmission of ultra-short-wave signals was de- pendent upon the state of the atmosphere, in particular upon its water vapor content. The present paper contains the results of a continua- tion of this work where a two-year survey of ultra-short-wave trans- mission over a 70-mile (112.6 km.) ocean path was carried out. Trans- mission was had on 106 days during this period. In planning this work, preparation was made for seeking a correla- tion between atmospheric structure and signal intensity; but from the very first transmission fading was found, and this fading was so per- sistent and intense that the work became essentially a fading study. In the following paragraphs there are discussed, in the order named, Antennas and Locations; Apparatus and Operation; General Charac- teristics of Fading, with samples of records taken; Polarization Effect on Fading, with sample records; Wave-length Effect on Fading, also with records; Distance and Antenna Height Effects on Fading; Frequency Sweep Patterns of Fading, with sample records; and the logs taken during the flights of the U. S. Weather Bureau airplane for taking free air data. The presentation of experimental data is then interrupted to present a theory which explains several of the experi- mental observations. This is followed by further experimental results and checks, and concluding remarks. Antennas and Locations Figure 1 shows the layout of the radio circuit. The transmitter was erected at Highlands, New Jersey on the edge of a steep hillside. This edge made an angle of about 45° with the transmitter-receiver direction. Below the edge of the hill lay a strip of land slightly above sea level (seven to eight feet) and beyond was Sandy Hook Bay. The altitude at the antenna foot was 119 feet. The antennas con- sisted of a vertical rhombic terminated in its surge impedance with carbon lamps, a horizontal rhombic with the same termination, an unterminated inverted "Vee" antenna and a half- wave doublet. This doublet was equipped with a flexible transmission line which permitted it to be raised to the top of the antenna supporting mast. These antennas were supported on a central 60-foot (18.3 meter) lattice mast surrounded by four 30-foot (9.15 meter) poles. I UL TRA-SHOR T-WA VE TRA NS MISSION 491 The receiver was located on a plot of land at East Moriches, Long Island, New York. This plot was immediately at the edge of Moriches Bay and was only slightly (approximately four feet) above sea level. The same antenna equipment was supplied here as at the transmitter. Except for the transits across Sandy Hook, Fire Island Beach and Smith Point, the wave path was over sea water. A second receiving site at West Sayville, at the edge of Great South Bay, was briefly occupied, using portable receiving equipment. This site was 52^ miles (85 km.) from Highlands. Fig. 1 — Map of ultra-short-wave transmission path between Highlands, New Jersej', and East Moriches, Long Island. Apparatus and Operation In all, three transmitters were installed at Highlands. The first one, of 100 watts output, covered the wave-length range of 5.0 to 3.5 meters. It was equipped with a motor-driven single-turn short- circuit loop which, coupled with the tank circuit coil, produced a 120- cycle frequency modulation of six megacycles amplitude. For cali- bration purposes there was added a low-gain double-detection receiver which used an intermediate frequency of one megacycle and was con- nected so as to pick up an input from the transmitter. The beating oscillator of the receiver was set for the center of the transmitter fre- quency sweep and the receiver output triggered a gas tube connected 492 BELL SYSTEM TECHNICAL JOURNAL to the transmitter tube grids. The transmitter grids thus received a voltage pulse each time that the transmitter frequency passed through one megacycle above or below the beating oscillator frequency. Each transmitter frequency sweep was thus marked with two pulses spaced two megacycles apart. The second transmitter had Lecher wire tuning elements, covered the wave-length range of 3.5 to 1.2 meters and had a power output of 30 watts at 1.5 meters. It was in operation simultaneously with the first transmitter for six months and then was replaced by transmitter No. 3. The third transmitter was coil tuned, covered the wave-length range of 4.9 to 2.8 meters and had a power output over this range of 55 watts down to 35 watts. It was operated simultaneously with the first transmitter except for the first six months. All three transmitters were arranged for voice modulation through a simple grid input, and the first one was thus used for one-way com- munication during the entire period of operation. Normally, unmodulated waves were transmitted and were observed as rectified direct current in the output of the double detection re- ceivers. These receivers had attenuators, variable in steps of 1 db, in the intermediate frequency amplifier circuits and the attenuators were geared to the pens of manual recorders. The operators kept the output current constant by means of the attenuators just mentioned, and there resulted a record of signal amplitude versus time. Some use was made of the Esterline-Angus type of milliampere recorder for automatic recording but no linear scale recorder of this type could handle the amplitude range of the fading encountered. For the reception of the frequency modulated transmission a tuned radio-frequency receiver, with a three-megacycle band-width centered on 66 megacycles (4.55 meters), was constructed and its rectified output was applied to one pair of plates of a cathode ray oscillograph. A linear sweep voltage, manually synchronized with the transmitter 60-cycle power voltage, was applied to the second pair of plates. The oscillograph pattern thus pictured the frequency-amplitude charac- teristic of the radio circuit in toto. Over the frequency range where the receiver band was flat (two megacycles) the curve gave the ap- parent ether characteristic. With a motion picture camera this characteristic was permanently recorded. Fading Characteristics, General The fading was always slow compared with that observed on short waves. Except for the rapid fluctuations produced by airplane reflec- UL TRA -SHOR T- WA VE TRA NSMISSION 493 tions, a record speed of ^ inch (1.6 cm.) per minute was sufficient. This was our standard speed. Amplitude changes up to 40 db and fading rates up to 5 fades per minute were observed. It is difficult to describe the fading in any other way than by the records. From a transmission standpoint a curve giving the per cent of time during which the signal is above the abscissa value is useful. a ^^ ^^ 4.7 METERS JULY 23, 1934 5=00 A.M. EASTERN STANDARD TIME Fig. 2 — Fading extremes, vertically polarized transmission; inverted "V" antennas. 8:00 A.M. e;30 A.M. EASTERN STANDARD TIME Fig. 3 — Extreme amplitude, normal fading rate, vertically polarized transmission; inverted "V" antennas. 494 BELL SYSTEM TECHNICAL JOURNAL > H < 20 _i UJ a 10 /> ^\fi yr \Aii hiy y 1 "i 1 11-30 A.M. EASTERN STANDARD TIME Fig. 4 — Development of "scintillation" fading on vertically polarized ULTRA-SHORT-WAVE TRANSMISSION 495 il 1 A fU^V ^%(h^ h\\f\ iv^ff\(^l, (- 1 = 00 P.M. 20 10 Afrk f^ V W^ ^i%'Ub| /fiVlii^ 2:30 P.M. '''HuJ|i| iW^ JJlfiywl t/^Wi^ AUG. 29, 1934 3:30 P.M. EASTERN STANDARD TIME transmission, 4.74 meters wave-length; inverted "V" antennas. 496 BELL SYSTEM TECHNICAL JOURNAL ^\^\ \r \i\ r \f L^A/ .r A v\ vu V Wl/ 1 V SEPT. 20,1934 20 10 6 00 PM 7 00PM 830PM ^ A A r\ A A_A^A A /\ / )A kA rv\ fsfrw^^r ..y ^ / y 1 r SEPT. 2 0,1934 8:30PM 900PM 9:30PM lOOOPM 1030PM HOOPM r.^ V\ / — V yV/^ \. ^-w> ^/^ A I" A -^/ \! I /T V vv V./| 'N W vv ■ 1 U SEPT 21, • 1934 llOOPM 12;30AM 1 OOAM /^Aat-A % SEPT. 2 1,1934 1 30AM 2:30AM 3:00AM 330AM 4:00AM — . / ■\_ j^—-^ 'r K ^^ A.,/ A^/ V^' V ^N/\ \l 1 'Vv SEPT. 2 1^1934 40 30 4:00AM 4:30AM 5:00AM 5 30AM EASTERN STANDARD TIME Fig. 5 — Twenty-four hour run, vertically polarized ULTRA-SHORT-WAVE TRANSMISSION 497 ,— . ^/ ' — \ ^'--1 \r\ fV r^ V^ ^ ^^■^ SEPT. 21,1934- 30 20 m 10 ^ 9 00 AM z ? 50 7 00AM 7:30 AM 12 OONOON I OOPM 8 30 AM 900 AM rN.. / ^ \|V\ "M /W ^-r\ /\ 1^ V /^ ^V A I ^y V V \) ! SEPT. 21,1934 IIOOAM nSOAM r--^ rW, /■^ r^ a/^ A, /^ ^^) n^ I /If A^ .r V r n ' \flvy SEPT. 2 1,1934 1:30PM 2:00PM 2:00PM 2 30PM 3 00PM 3 30PM 4:00PM 4:30PM S-' -\/^ ^A r\ ^^. A J K ' IaA A^ y v\> ^V rv^ ^ SEPT. 2 1,1934 40 30 4:30PM 500PM 5:30PM 6:00PM EASTERN STANDARD TIME 6:30PM 7:00PM transmission, 4.74 meters wave-length. 498 BELL SYSTEM TECHNICAL JOURNAL Such a curve can also serve to check on the theoretical explanation of the cause of fading in certain cases. Thus if the fading is due to the combination of two radiation components in assigned random ampli- tude relation and arbitrary or random phase relation, a curve can be calculated from probability considerations and compared with the experimental curve. ^ Such a simple mechanism was inadequate for our fading most of the time. Moreover, the fading changed enor- mously from day to day. It is hoped that the samples given in the figures will give an adequate idea of this phenomenon. Only rarely was fading practically absent for periods of an hour or two. Such a period is illustrated in Curve a, Fig. 2. Two days later the extreme fading of Curve h was recorded. It is significant, as will later appear, that the non-fading situation was the one of higher signal. The amplitude range of curve h is nearly normal; the fading rate is much greater than normal, for vertical polarization. In Fig. 3 the fading rate is normal but the amplitude range is excessive. In Fig. 4 a characteristic type of fading, which we have termed "scintillation," is recorded. In this case the fading, initially erratic and of a fairly wide amplitude range, subsides in a characteristic manner to a steady, fast rate oscillation, or scintillation, of moderate amplitude. In Fig. 5 a 24-hour run is recorded. The rambling erratic character of the fading is well shown here. Characteristic deep short-period minima occur at intervals, occasionally they are twinned, some of them have a fine structure at the bottom. There are several "dropouts" where the signal practically disappeared. No sunrise-sunset variations in fading were noticed, though looked for. Diurnal variations could not be established since automatic recording was not available. A seasonal falling off in average signal was noticed in the winter; the 1.6 meter wave, because of its normally low level, dropped below the noise level in the winter of '34-'35. No effect of ocean waves, clouds, or other visible weather phenomena could be established. It is true, however, that to be certain of the non-effect of such phenomena as clouds, a cloud observer at the mid- way point should have been present. In so far as cloud layers make air mass boundaries visible they may well affect the transmission. Cloud bottoms which represent merely the adiabatic dew point level should apparently not cause much signal reflection at these wave- lengths. Effect of Polarization on Fading After some preliminary experimenting it was found that comparisons of two transmissions were worthless unless made on simultaneous recordings. The recorders were therefore fitted with telechron motors ULTRA-SHORT-WAVE TRANSMISSION 499 operating on a circuit of the Patchogue division of the Long Island 60-cycle power network. The resulting timing was faultless and by transmitting the same radiation on crossed antennas, and receiving the vertical and horizontal components separately, a comparison was obtained. In general the horizontal component showed the worse fading, more fades per minute and greater amplitude range. This was always true when the fading on vertical polarization was bad. There was then no noticeable coincidence between the two. When the fading had a smooth long period fade, or "roller," superposed on a short period oscillation, or "line structure," there was at times coincidence between the roller components. Occasionally, with fine structure absent and I 9:30PM lOOOPM Fig. 6 — Composite bad fading, horizontally polarized transmission. 500 BELL SYSTEM TECHNICAL JOURNAL aa Ni Hi0N3aj.9 ivnois 3aiivi3h ULTRA-SHORT-WAVE TRANSMISSION 501 moderate roller fading, a good coincidence between the two records resulted. This is discussed later. Figure 6 is a sample of fading on horizontal polarization, at its worst. This particular specimen shows the superposition of roller and fine structure fading very well. No vertical-polarization record was taken along with this. Figure 7 shows a typical example of fading simul- taneously observed on vertical and horizontal polarization during bad fading conditions. There is no coincidence. Figure 8, on the other hand, records an unusual condition when a mild roller type of fading shows a good coincidence on two polarizations. 30 20 X 10 ^\/A /X-~s ^^r- -^ ^~\ \ -/ \ N VERTICAL 3.5 METERS \l SEPT 26, 1935 50 > 40 30 20 /-\ \/ 'V ^ ^V\ x/" ^ ..,/^ ^^^ -J HORIZONTAL 3.5 METERS -^v SEPT 26, 19 35 1 = 30 RM. 2:00 RM. EASTERN STANDARD TIME 2:30 RM. Fig . 8 — Comparison of simultaneous mild roller fading on horizontally and vertically polarized transmissions. Effect of Wave-length on Fading The double wave-length records are not as contrasty as the double polarization ones. In general the shorter wave has the worse fading, either as higher fading rate, greater amplitude oscillation or both, and the greater the wave spacing the more certain this is to be true. Ex- ceptions have occurred, however, where the fading was much the same, and one record was obtained where the fading rate on 4.7 meters was noticeably greater than on 4.5 meters. Our first simultaneous records were taken at a wave-length ratio of 3 to 1 (4.7 to 1.58 meters) where the fading on the shorter wave was 502 BELL SYSTEM TECHNICAL JOURNAL always worse. The remaining observations were confined to wave- length ratios of 1.5 to 1 and less. A comprehensive set of records was obtained for moderate to small wave-length spacings, down to 1 per cent difference. These records are all for vertical polarization. The few records taken on horizontal polarization happened to be obtained when the horizontal fading was much worse than the vertical fading and the records are too rough for good comparisons. For these small wave-length-difference records the types of fading are more likely than not to be similar on the two wave-lengths. That is, the fading rate and amplitude excursion will average up much the same. More rarely, there will be a similarity between the two fading tracks which is evident to the eye, sometimes as a "retarded" simi- 6=30 A.M. EASTERN STANDARD TIME 7:00 A.M. Fig. 9— Comparison of simultaneous fading on two well spaced wave-lengths, vertically polarized transmission. larity. Occasionally, and usually on the roller type of fading, there will be a marked coincidence between the two records; this coincidence will be better the milder the fading and the smaller the wave-length spacing. Genuine identity was never recorded on different wave- lengths even down to 1 per cent difiference. With scintillation, coin- cidence was difficult to demonstrate; a similarity on the major swings was all that was shown. Figure 9 shows a very marked difference between 4.7 and 3.0 meter fading. This is one of our most contrasty records. Figure 10 shows very slow fading, on two occasions, with wave-length differences of ULTRA-SHOR T-WA VE TRA NSMISSION 503 approximately 1 and 4 per cent respectively. There is good coinci- dence. Figure 11 shows active fading on short rollers for 4.7 and 4.65 meters, a wave-length difiference of approximately 1.1 per cent. There is agreement in major features. Figure 12 shows a case of scintillation 2:00 RM. 2:30 RM 3:00 RM. < 40 O 30 20 10 ^ ./^^ /4..7 METERS "V ^ V V \ V V 4.5 METERS A \a / AUG. 8, 1935 V 5:00 A.M. 5:30 A.M. EASTERN STANDARD TIME 6:00 A.M. Fig. 10 — Comparison of simultaneous slow fading on two slightly different wave- lengths, vertically polarized transmission. superposed on mild rollers, again for 4.7 and 4.65 meters. The time scale is here magnified three times. An in and out similarity can be seen, especially for the rollers. In the section on theory these simi- larities are further discussed. 504 BELL SYSTEM TECHNICAL JOURNAL 30 20 4.7 METERS 10 ^V\A/VyVM| 4:00 P.M 4=30 P.M. EASTERN STANDARD TIME 5:00 P.M. Fig. 11 — Comparison of simultaneous active fading on two slightly different wave- lengths, vertically polarized transmission. 30 20 10 4.7 METERS f^jV, '^"^^Vv./f ry 0f^ yf"^ ifVv 1 1 1 MAR. 6. 1935 1:30 P.M. 1:35 RM. 1:40 RM. EASTERN STANDARD TIME 1:45 RM. Fig. 12 — -Comparison of simultaneous "scintillation" fading on two slightly different wave-lengths, vertically polarized transmission. The time scale has been expanded. Effect of Distance and Antenna Height on Fading In planning this work a survey for a receiving site was made by means of a portable receiver in a car. Later, simultaneous reception ULTRA-SHORT-WAVE TRANSMISSION 505 was had at East Moriches and West Sayville, on three days. The survey data were not sufficient to establish any proposition beyond the statement that the signal strength fell rapidly with distance, with the intensity of fading coming up as the signal fell. The simultaneous two-distance recording showed random fading between the two records with less fading amplitude at the shorter distance. The fading rate was about the same. Unfortunately the recording took place under scintillation conditions, thus giving very poor records for comparison purposes. By mounting two linear doublets on the 60-foot lattice mast simul- taneous recording at two heights was carried out. For the two dou- blets (horizontal, at 14 and 52 feet respectively), a signal level difference of 12 db was observed, in favor of the higher antenna. The fading on the two records was identical. It may be added that, on calibrating the car receiver at East Moriches before moving to West Sayville, identical fading records were obtained with the two antenna systems 150 feet (45.7 meters) apart and substantially broadside to the radiation. Frequency Sweep Patterns of Fading The frequency sweep patterns were of many types, from slow to fast fading and from shallow to deep fading. Apparent path differences from 600 meters down to a few meters occurred. The patterns were usually complicated, indicating that more than two components were present. There is no reason to believe, however, that they were not all due to wave interference.^* On three days the predominant pattern was simple enough to be referable to two waves with a path difference consistently greater than 75 meters. These will be referred to later. In Figure 13 are given three sample runs illustrating a two-component pattern, a three- component pattern with two of the components forming a small path difference pair, and a multiple component pattern. The receiver characteristic is dotted in on one curve of each set. Logs During Weather Bureau Airplane Flights On forty-five mornings recording was carried out during the period of flight of the Mitchel Field Weather Bureau plane. This plane takes off about dawn every morning, when flying is possible, and by means of a meteorograph obtains records of air pressure, temperature and humidity, up to an altitude of about five kilometers. A record of the fading, on 4.7 meters and vertical polarization, was obtained for each of these mornings. In addition, on twenty-six mornings frequency 506 BELL SYSTEM TECHNICAL JOURNAL sweep patterns were photographed at or shortly after the time of flight. These sweep patterns were all on horizontal polarization. From the meteorograph data, kindly furnished us by the United States Weather Bureau, the dielectric constant of the air has been calculated ^ and plotted as a function of the altitude. On twenty-four days there were, above an altitude of 400 meters, changes in the dielectric constant curves equivalent to discontinuities of Ae ^ 10~^. Heights up to 3200 meters were recorded for these. Typical curves JUNE 27,1936 2 COMPONENT JUNE 28jf936 3 COMPONENT 603:55 64 66 69 MAY 19,1936 MULTI-COMPONENT 64 66 69 FREQUENCY IN MEGACYCLES 64 66 69 Fig. 13 — Three sequences of frequency sweep patterns. Horizontally polarized transmission, 4.55 meters mean wave-length. are given in Fig. 18 on the left-hand side. On four days there were small boundaries with Ae < 10~^; on two days there were possible but not definite boundaries, the experimental points being too widely separated in altitude for precision; on five days there were possible boundaries below 400 meters altitude and on ten days the refractive ULTRA-SHORT-WAVE TRANSMISSION 507 index-height relation was an approximate exponential one without any evident boundaries. These data will be referred to later. Theory The fading phenomenon was explicable in several ways. In our previously cited work ^ we found that variable atmospheric refraction was present, the airplane carried receiver being up where the refracted- diffracted field strength was high and dominant. In general variable refraction would be expected to be a slow phenomenon, operating in hours, or even days, rather than in minutes, and much too slow to explain five-cycle-per-minute oscillations, for example. Another explanation was air-mass boundary reflection (or refrac- tion),^ such a boundary readily explaining the rate of signal variation. No Kennelley-Heaviside layer reflection was in question ; this had been quickly ruled out by the experimental data. When, therefore, we elected to transmit the frequency modulated signal, already described, and the oscillograph revealed a cyclic maximum-minimum frequency characteristic of the other path itself, it was evident that there was no possibility other than wave interference left — interference presumably between a direct-diffracted and one or more boundary-reflected components. These boundaries have apparently not been positively identified at longer wave-lengths and for that reason we have tried to get some further experimental contact with them. Attempts, since the closing down of the Atlantic Highlands-East Moriches circuit, to demon- strate an air-mass boundary, any boundary whatever, by high-angle transmission, have failed. No reflected components have appeared. Of course an illy defined, or diffuse, boundary will operate in this manner since only for near grazing incidence can such a boundary give the appearance of a discontinuity for the incident radiation. If we assume such a boundary a few kilometers up, and assign to it a relatively small discontinuity in index of refraction, compared with that of an earth or sea water boundary, then the four components of Fig. 14 will be the only important boundary reflected ones for a radio circuit such as ours. We now, fortunately, have theoretical formulae *• ^ for computing the diffraction of an ultra-short-wave radiation around the earth and the amplitude of the direct-diffracted component can be calculated at once. That is, it can be calculated at once if the air mass has no refractive bending effect upon the radiation trajectory. Since such a bending effect is certainly present at times, and is equally certainly variable, even if only slowly, it must be taken into account. 508 BELL SYSTEM TECHNICAL JOURNAL If the refractive index of the air varies as a power of the distance to the earth's center, it has been shown ^ that the actual state of affairs can be duplicated by a homogeneous atmosphere over an earth, the radius of curvature of which is greater than that of the actual earth and is calculable from the exponent of the height variation function. With this "effective" earth radius, the formulae already mentioned become usable. If the air refractive index does not vary as a power of the distance to the center of the earth we must take that exponent which gives the best first order fit over the height covering the re- fracted wave front, the alternative being a prohibitive complication of the theory. A plausible physical picture of the fading mechanism can now be set up. If we lump the four boundary reflected components in one, and plot as a function of the distance, we have curves "yl " of Fig. 15. REFLECTING BOUNDARY Fig. 14 — Drawing illustrating the four components of a single reflection at an air boundary. Curves "5" are the Wwedensky ^ * and Gray ^ theories. These are for our Highlands-East Moriches circuit with the average effective earth radius of 8500 kilometers and a 1500-meter boundary height. If we now imagine a receiver moving away from the transmitter we shall first traverse the zone of high "5" amplitude with no fading present. The signal amplitude will, for any given near-by point, and for any given antenna ampere-meters, depend on the height of the antenna above the ground and the ground constants. As the distance to the transmitter increases, the falling "5" curve approaches the rising "^" curve in ordinate and we enter a disturbed region where, for any instability of the boundary, more or less complete interference can result and fading will occur. (One such instability occurs when * There is an error in the formula, as given by Wwedensky. It is corrected here. See appendix II. ULTRA-SHORT-WAVE TRANSAilSSION 509 a boundary with an irregular surface is carried past the reflection zone by the normal motion of the atmosphere.) A further increase in transmitter distance and the "5" or residual curve drops out of the picture leaving only the "A'' curve and, presumably, fairly steady signal amplitude conditions. The location of these zones of undis- turbed and disturbed signal will vary from day to day as: (1) the reflection coeflacient and height of the layer change, (2) the effective radius of the earth changes. The effect of the height of the layer is shown in Fig. 16. OQ 20 10 -10 -20 \ k. \ \ ^ "Bv REFRACTED - ^ DIFFRACTED > <;b'h X s. \, s \, 1 1 1 s \ >. \ 1 1 "a; BOUNDARY REFLECTED Ns s N. \ 1 1 A \/ ^ *^ X, ■ — — Ns "^ -^ :^ r^ VI A / \ 1 1 N ^^ -6 If ^\ 1 'a-„ N 1 1 N s. ^. L ] J 1 \, s \ "^ 1 1 \ \, X V 1 1 \ 1 EAST MORICHES 1 ^ 1 \ S N, - 1 f 1 r \ s...,. 60 80 100 120 140 160 DISTANCE FROM TRANSMITTER IN KILOMETERS 180 Fig. 15 — Calculated curves for air boundary reflected and earth refracted- diffracted radiation components, in both vertical and horizontal polarization. Short doublet antennas, 1 kw. power radiated, wave-length 4.7 meters, o- = 5 X 10~^^ (E.M.U.) and e = 80 for sea water. Height of transmitter, antenna 42 meters, of receiver antenna 5 meters, air boundary height 1500 meters, effective radius of earth 8500 kilometers. Since the major lobe of the polar characteristic of any simple an- tenna, such as ours, is directed forward and away from the earth, the signal intensity at the reflecting boundary surface will be comparatively high and will, in some measure, make up for a small reflection coeffi- cient. For longer waves, such as broadcast waves, the high level of the " 5 " curve will move the disturbed zone so far out that the low residual signal level and the Kennelley-Heaviside layer reflections will conceal 510 BELL SYSTEM TECHNICAL JOURNAL or mask the atmospheric boundary reflections. Several observations which can be ascribed to such boundaries have nevertheless been published.''- ^ Obviously, only boundaries lying considerably higher than those discussed here will give the path differences to produce the same type fading at these longer waves. At the same time the ap- parent diffuseness of a boundary will fall off with increase in wave- length, thus removing the restriction of reflection to near grazing incidence angles only. 60 ■^ SH / / c A6= C10)"5 Y- s^e = (io)-'* r 1 \ r> \\. ^ > A \\ W w \ \ J ) ) ) .1.8 3.2 5.6, A6=C30or^ /) ) \ .18 3.2 5.6, Ae=(Xior^ \ 80 70 60 50 40 30 20 RESULTANT FIELD, DECIBELS BELOW FREE SPACE Fig. 16 — Calculated field strength curves showing the effect of air boundary height and density on the reflected radiation component, for the Highlands East Moriches circuit. Transmission path 112 kilometers, over sea water, wave-length 4.7 meters, polarization both vertical and horizontal. Vertical antennas 42 and 5 meters high, horizontal antennas 45 and 9.5 meters high, respectively. This tentative mechanism also explains several other observed features. Thus, for a given type of boundary instability, the fading rate will increase as the wave-length decreases. Furthermore, since the slope of the "5" curve increases as the wave-length decreases, the ULTRA-SHORT-WAVE TRANSMISSION 511 disturbance zone is effectively moved nearer the transmitter and the probability of increase in fading amplitude is enhanced. The usual increase of fading with decrease in wave-length is thus explained. When the wave-length difference is small, on the other hand, the fading type should be much the same on both wave-lengths, as was generally found. The lack of coincidence would arise from the fact that the path difference being a considerable number of wave-lengths, a small wave- length change can introduce a marked randomness in fading without appreciably affecting the type. As has been mentioned earlier, a multiple of reflecting boundaries is the normal condition, rather than that of a single boundary. This circumstance, without invalidating the explanations already given, makes a further elaboration of the theory possible. The "roller" type or component of fading, in particular, requires explanation. In addition to the smooth signal modulation, from which the name has been derived, this type of fading is characterized by showing more or less frequent deep minima or drop-outs and these are often distinctively twinned. Further, the roller component is that component of fading which shows coincidence, in spite of wave-length or polarization dif- ferences. Such coincidence indicates small path difference and this is what we have when a double boundary or stratum exists. Such a stratum would give two "A" components and, if of variable thickness, would, as it was carried along by the prevailing air currents, give the steep, deep, minima at phase opposition thickness. Further, if the stratum contour were that of a hump, thick enough to carry the second "A" component past phase opposition to the first one, the twinned minima would result as the hump entered and left the reflection zone. Occasionally the two "A " components would add properly, with the residual "B" component, to give complete extinction, a result less likely from the phase addition of a single "^" and the "5" component. This explanation of "roller" fading assumes, tacitly, that the "B" component is, at the time, relatively subdued, that is, the disturbance zone has moved inwards due to an increase in the reflection coefficient of the layers or to a decreased "effective" earth radius. The fine structure often appearing at the bottom of a prolonged roller minimum corroborates this, the mutual cancellation of the two "A" components having uncovered, so to speak, the weaker "B" component with its much shorter traversed path. With the "roller" condition characteristic of high "A" component signal amplitude, the "scintillation" condition would be characteristic of low "A" component signal amplitude, the relatively steady "B" 512 BELL SYSTEM TECHNICAL JOURNAL component having superposed on it a small amplitude, variable phase, "yl" component. A relatively low mean amplitude value and the coincidence of scintillation conditions with conditions of convective instability of the atmosphere would thus be explained. All the scintillation records came on days of relatively high wind and convec- tive instability. A turbulent atmospheric condition would dissipate or attenuate any boundaries, especially the lower ones. The rapid flutter about the mean amplitude value is the normal expectation from a high, turbulent, low reflection coefficient boundary. Our two polarization results are qualitatively explicable on the mechanism proposed. As can be seen in Fig. 15, the change from vertical to horizontal polarization results in a relative lowering of the "5" curve without much change in the "^" curve, which should result in increased fading. For our circuit and a boundary at 1500 meters the relative "5" vs. "yl " drop is 13 db. As Fig. 16 shows, the variations of the "^ " components with height are markedly different for the two polarizations. The "Ay" compo- nent falls steadily with height up to 4700 meters; the "Ah'' component has a deep and sharp minimum at 3000 meters after which it rises again. Since most of our observations concerned boundaries at 2000 meters or less, this high altitude disparity between "Ah" and "Ay" does not affect our explanation. The disparity between vertical and horizontal fading should be much more marked for high boundaries than for low boundaries. Further Experimental Curves and Checks The curves given have illustrated the variability in the fading, a variability which no short period of recording can encompass. The tentative explanations proposed have been shown to be in accord with several of the features characteristic of this fading. Certain other experimental results will now be adduced which offer further verifica- tion along somewhat different lines. For the forty-five mornings on which simultaneous recording was carried out during the United States Weather Bureau plane flight, we have calculated, from the airplane data, the values of the "A" and "B" components. As stated earlier, there were twenty-four days when boundaries above 400 meters altitude, and of sufficient distinctness to be fairly accurately estimated (Ae = 10~^) were shown by the meteoro- graph records. For these the "A" components have been computed. By taking the dielectric constant gradient for the first half kilometer, the effective earth's radius was determined and inserted in the Wwe- densky formula to give the "5 " component. These calculated values I ULTRA-SHORT-WAVE TRANSMISSION 513 ("yl " component as triangles, "5" component as circles) are plotted on Fig. 17 together with the maximum and median * observed values. These latter are joined by lines. For the 10 mornings on which no boundaries were evident the calculated "5" component appears to be some 8 db higher than the observed values. With this correction the agreement between observed and total calculated fields is fairly good. A partial explanation of this 8 db discrepancy may lie in the fact that the ocean water trajectory assumed in the calculation differs from the actual one by the land terminals and the three tongues of land intervening. 1 - BOUNDARY EVIDENT AC* tO-S 2 -NO BOUNDARY 3 - SMALL BOUNDARY AetlO"^ 4 - INDEFINITE BOUNDARY 6- POSSIBLE BOUNDARY BELOW 400 METERS -M*-3- a 50 -c^i o °o no4>°0 $ tr^ a — O-Q- _J I I I I I L - rf) -^ to fO , CVJftl „(\1 I I I I I I I I I I I 4- CALC. "A" COMPONENT r^-OBSERVED MAXIMUM 0-CALC."B" COMPONENT L— OBSERVED MEDIAN •LAKEHURST DATA "A" Fig. 17 — Comparison of "^" and "S" radiation components, calculated from the U. S. Weather Bureau free air data, with measured maximum and median signal strengths. Vertically polarized transmission. On the twenty-six morning frequency sweep runs there were only three on which the predominant sweep pattern was simple enough to be interpreted as due to two components with path difference greater than 75 meters. For those days a series of measurements of the film patterns was made by determining the frequency spacing between a maximum and a minimum and calculating the resulting path dif- ference and boundary height. The dielectric constant-height function was also calculated from the Weather Bureau data. These curves are * The signal is half of the time greater and half of the time less than its median value. For random phase with "S" component equal to "A" component the resultant median value signal is V2 X ^ or 3 db up; it falls from this value to "A " as "B" decreases to zero. 514 BELL SYSTEM TECHNICAL JOURNAL plotted in Fig. 18 with the calculated boundary heights set down at the right hand, spread out in time of observation. The boundary height coincidence is pretty definitely located in this manner. Many of the more complicated frequency sweep patterns carried a fine struc- JUNE 23, 1936 \ \ \ I \ \ 1 \ 1 s \ V V« \(e-i)0o)6 > \ \ \ \ \ \, ) 1 S >. • 1 • • • t •V • ••«• *.*' • Li*r ..\*:r s.> •. • i,*4 • •> FREQUENCY SWEEP PATTERNS JUNE 27. 1936 \ 1 I \ \ \ \ \ \ V Vw \ce-i)Oo)' \ \ \ \ k \ 1 \ ) \ \ s \ 1 V V > m ^•.- • T'% m • JUNE 29, 1936 \ \ \ v \ \ \ \ \ { V \ \ \ V V \(e-i)(io)^ ^ I \ \ \ \ \, ) > "> ) • • • • • • ri '/! -1 A * %ii r tt ^— -10 0 10 20 600 600 700 530 TEMP IN ° C DIELECTRIC CONSTANT 2 6 10 14 18 VAPOR PRESSURE MILLIBARS P^ 600 6 30 700 AM EASTERN STANDARD TIME Fig. 18 — Comparison of boundary heights shown by the U. S. Weather Bureau free air data, with boundary heights measured from frequency sweep patterns. Horizontally polarized transmission. ULTRA-SHORT-WAVE TRANSMISSION 515 ture which indicated weak boundaries at higher altitudes up to, roughly, 5.5 km.; most of the patterns, however, were characteristic of layers below two kilometers. The path difference corresponding to two kilo- meters is 85 meters. The theoretical limit of resolution of the amplifier band for a maximum to minimum frequency spacing is Al — 2(C/A/) where Al = path difference, C = velocity of light and A/ = frequency band. For A/ = 2 X 10^ cycles this gives 75 meters, and hence boundary heights at and below 1900 meters are unresolvable by our receiver. It is a remarkable result that the bulk of the disturbing boundaries should lie so low. It was mentioned earlier that several observations referable to air- mass boundaries have been published. In addition there have been reports, for three consecutive years, of long distance ultra-short-wave reception by American amateurs ^ during the month of May. We have copies of the U. S. Weather Bureau atmosphere cross-sections for several of these days and have been curious enough to examine them. On May 9, 1936, during the long distance amateur reception, there was an extensive boundary at 4 km. between an upper Superior air mass and a lower Tropical Gulf air mass. On May 15, 1937, a similar boundary at 4-5 km. had a Superior air mass above a wedge of Transitional Polar to Tropical Atlantic air. Below this at 3-4 km. lay a Transitional Polar Continental air mass. On June 11, 1936, when Colwell and Friend^ report an extra strong 0-2 km. " C" reflection, a subsiding Transitional Polar Pacific air mass lay above a Transitional Polar Continental air mass with the boundary at about 1.5 km. On June 29, 1936, when they reported a very strong 3.5 km. "C" reflection, there existed four wedge-shaped air masses with a Superior air mass over a Transitional Polar air mass at 3-4 kilometers. The wave-lengths used were 186, 125 and 86 meters approximately. These coincidences may or may not be significant but it is very questionable that any boundaries at such altitudes are due to either electron or gas ion distributions. The characteristic properties of North American air masses have been published," as average summer and winter values, and show some marked seasonal differences. The greater dielectric constants for summer conditions are due chiefly to greater water content. For a single air-mass distribution, horizontal stratifications are at a minimum and the radio transmission is via the "B" component. This component can be calculated from the corresponding effective earth radius. The table below gives this radius for three important air mass types. 516 BELL SYSTEM TECHNICAL JOURNAL Effective Earth. Radius Air Mass Type Summer Winter Tropical Gulf- — Tg Polar Continental — Pc- ■ . Superior — 5 . . 1.53 X R .. 1.31 X R .. 1.25 X R 1.43 X R 1.25 X R 1.25 X R " R" = actual earth radius The boundaries between different air-mass types furnish discon- tinuities adequate for radio reflections. The greater the stabiUty of the boundary, the more abrupt it is Ukely to be. In general, when "5" air overlays either "T^" or ''Pc'' air, the resulting boundary is stable. Possible discontinuities, for the three types discussed, may be summarized in the following table. Here the positive sign means that the radiation originates in the more refractive medium. For stability the lower medium is the denser though not necessarily the more refractive. Ae X 106 Altitude Summer Winter SITg SIPc TglPc SITg SIPc TglPc 1.0 Km 100 50 30 20 10 10 -80 -40 -20 55 50 35 25 15 10 -30 2.0 Km 3.0 Km -35 -25 Concluding Remarks ^ The characteristics of this seventy -mile circuit indicate that for ultra-short-wave transmission it rates as a long distance one. If we assume that the air refraction is on the average such that the effective earth's radius is 4/3 the actual one, then the receiving station lay 1400 feet below the line of sight from the transmitter. This is equivalent to 0.57° below the horizon. The reception, using high efifective-height antennas, was good; there was, however, very little lee-way left, above set noise, for reception with simple doublet antennas. Any longer circuit will require to be terminated on elevations such as to keep the intermediate horizon height down. The fading was too slow to be noticeable on amplitude modulated speech unless a deep minimum or drop-out occurred. The circuit was probably unusable for television, most of the time. A system adhering to the R.M.A. standard ^- of 441 lines on an inter- ULTRA-SHORT-WAVE TRANSMISSION 517 laced 60-cycle scanning will have a unit time element of 0.17 micro- seconds. This corresponds to a path difference of 51 meters and only a fraction of this is necessary to produce a ghost. A rough estimate of the boundary height range involved in our fading is one-half to five-and- one-half kilometers. The corresponding path difference range is 8 to 580 meters. As the fading records show, no matter whether the ''A" or the "B" component predominated, the other component was usually present in amplitude only second to the other. It may be pointed out that where a standing wave system exists, ^"^ reflected components with much larger path differences than those recorded here are almost certain to be found. Appendix I In the Wwedensky * paper the author applies his theory to one of the experimental curves from a previous paper of ours. He uses the nor- mal earth radius " R," however, without any correction for air re- fraction. If we assume, as a more probable effective earth radius, the value 4/3R,^ the agreement with our curve is markedly improved. Appendix II In the first Wwedensky paper, Tech. Phys. U. S. S. R. Vol. 2, p. 632, 1935 eq. (7, 1) the sign of the term Ir^p sin 2dm should be minus. Appendix III The fading produced by moving bodies such as airplanes has been referred to in one of our earlier papers.^" It happened one day, during the present investigation, that fading of this type appeared when mechanical recorders were being used and, by speeding up the paper, a record in two polarizations was obtained. The airplane itself (or other cause) was not visible. The results are given in Fig. 19. Again the horizontal component was the worse one. At first the two fadings, both fine and coarse components, were in step; later they passed en- tirely out of step where the fading was so rapid as to smear the paper. These "airplane" fadings were observed, off and on, at other times but were not recorded. References 1. Englund, Crawford and Mumford, Bell System Technical Journal, Vol. 14, p. 369, 1935. 2. Brown and Leitch, Proc. I. R. E., Vol. 25, p. 583, 1937; Norton, Proc. I. R. E., Vol. 26, p. 115, 1938. 3. Ross Hull, Q.S.T., Vol. 21, p. 16, 1937, May. 4. B. Wwedensky, Tech. Phys. U. S. S. R., Vol. 2, p. 624, 1935; Vol. 3, p. 915, 1936; Vol. 4, p. 579, 1937. 518 BELL SYSTEM TECHNICAL JOURNAL 2 J3 ^~ LJ 4-» 'o ? rt biO H c Q -a ID 0^ 03 o < ■^ - Q -o . z (U 0) < 0> rt r, 1- (C -is ;t^ c/1 ^ t^ z .Sf-S cc J= UJ J_l 't •- c o ",' aj 0^ aaNiHioN3ais ivnois 3Aiivi3a ULTRA-SHORT-WAVE TRANSMISSION 519 B. van der Pol and Bremmer, Phil. Mag., Vol. 24, p. 141, 1937; Vol. 24, p. 825, 1937. 5. Miss M. C. Gray, paper to be published.* 6. Schelleng, Burrows and Ferrell, Froc. I. R. E., Vol. 21, p. 427, 1933. 7. Colwell and Friend, Nature, Vol. 137, p. 782, 1936; Phys. Rev., Vol. 50, p. 632, 1936; Colwell, Friend, Hall and Hill, Nature, Vol. 138, p. 245, 1936; Friend and Colwell, Proc. I. R. E., Vol. 25, p. 1531, 1937. 8. Watson Watt, Wilkins and Bowen, Proc. Roy. Soc, A, Vol. 161, p. 181, 1937. 9. Q.S.T., Vol. 21, p. 27, 1937, July. 10. Englund, Crawford and Mumford, Proc. I. R. E., Vol. 21, p. 464, 1933. 11. H. C. Willett, Bull. Amer. Meteor. Soc, Vol. 17, p. 213, 1936. 12. Beal, Television, Vol. 2, p. 15, 1937, R.C.A. Inst's. Press. 13. Englund, Crawford and Mumford, Nature, Vol. 137, p. 743, 1936. * The case of vertical polarization is treated by references 4, that of horizontal polarization by reference 5. Amplitude Range Control By S. B. WRIGHT The art of controlling the amplitude range of telephone signals involves recognition of certain characteristics in addition to those used to specify the performance of ordinary transducers. Funda- mentally, three kinds of characteristics are necessary to distinguish different range control devices. They are (1) the steady-state input-output characteristics, (2) the time actions, and (3) the range over which they function. In some cases, several secondary char- acteristics may be of interest, but they need not be considered in determining to which class a particular device belongs. This paper discusses and classifies these characteristics. Introduction TN a "non-linear" transducer, the output power is not proportional -■- to the input power. Consequently, the ratio of maximum to minimum power at the output differs from that at the input. But the ratio of maximum to minimum power is an expression of amplitude range. A device designed to alter this ratio may be called a range controller. In telephony the term range controller includes many devices ^ having specific names, such as limiters, volume control devices, range reducers, compressors, vogads, expandors, etc. These devices have many prop- erties in common with telephone repeaters, and a repeater may be considered as a special case in which any non-linearity which may exist between the output and input is unintentional. The purpose of one type of range controller is to reduce the range of significant intensities of signals applied to a telephone circuit so as to ease the requirements of the transmission medium with respect to overloading and noise interference. Such a device is placed at the transmitting end of the circuit. When the range is compressed at the sending end of the circuit it may sometimes be desirable to expand it at the receiving end to the original range. This is done with a device having, in general, the same dynamic characteristic as the compressing device, but a range change which is complementary. The purpose of the expandor is to reduce the noise heard by the listener as well as to compensate for whatever characteristic signal modification occurred in ^ For numbered references, see end of text. 520 AMPLITUDE RANGE CONTROL 521 the process of compressing the original wave. Sometimes an expandor is used at the receiving end to reduce the gain in the intervals between the main signals even when no compressor is employed. This is an example of using a range controller to correct defects in the medium. As is well known, the performance of a repeater is specified by such characteristics as impedance, amplification, frequency band, noise, and output carrying capacity. The performance of a non-linear device involves some additional characteristics. The primary ones are (1) the slope of the input-output curve, which tells how the range is changed, (2) the dynamic operation, which tells the manner in which the output varies with time following a given change in input, and (3) the range, which tells the region over which the device exercises control. It may be helpful to imagine a range controller as an amplifier in tandem with an adjustable attenuator, the loss of which may be changed either instantly or slowly to follow in some predetermined fashion changes in the signal. For simultaneous operation, this device could put out a wave which is a simple function of the input, but if the operation were delayed by a definite interval the device would be required to respond in a complex fashion in accordance with a re- collection of what had occurred in the signal during the delay period. Such delayed adjustment would be very crude for intervals comparable with the periods of fundamental speech frequencies. To obtain practical regulation of the delayed type it is necessary to increase the delay beyond this range and base the control upon the amplitudes of the syllables. When the delay is increased to a point where it is comparable with the syllabic periods its usefulness is again reduced. Part 1 — Control Ratio Fundamental Characteristics Figure 1 shows how waves may be altered by a device having a given output-input characteristic, assuming the operation is instan- taneous. As this figure is plotted on a db scale, only the stronger portions of positive values of the wave are shown. A similar diagram could be drawn for negative values. The output-input characteristic, although a straight line in this kind of diagram, would of course be parabola-like if plotted on a current or voltage basis. By selecting points, such as A (or B), on the input wave and determining the rela- tive outputs A" (or B"), the corresponding resultant wave is obtained. In this case, the resultant has a flatter top than the original sine wave, and this illustrates the capabilities of the device in increasing weak signals with respect to the strong ones and also suggests that distortion 522 BELL SYSTEM TECHNICAL JOURNAL may accompany the transformation. Such effects depend upon the slope of the output-input characteristic. The control ratio of a range controller might be defined as the output range in db divided by the input range in db within the non-linear region of interest. The ratio is obtained in such a way as to eliminate transient effects, i.e., using steady-state sine waves. Typical Control Ratios Figure 2 shows some typical output-input characteristics for various transducers having control ratios between zero and infinity. While INPUT IN DECIBELS BELOW ARBITRARY REFERENCE 60 50 40 30 20 10 0, COMPRESSED (OUTPUT) WAVE Fig. 1 — ^The signal modification caused by a non-linear transducer depends upon the slope of the output-input characteristic. these typical characteristics are straight lines there is nothing to pre- vent a range controller having a control ratio which varies with input. However, when complementary action is required at the receiving end it is more readily obtained when the control ratio is constant. Also, some physical elements used in the design of range controllers are most readily adapted to a straight line characteristic. Compressors (that is, devices having control ratios less than 1) may be divided into two classes: (1) Complete * and (2) Incomplete. In a complete compressor (control ratio = 0) the output is held constant within the range of the device. This control ratio gives a maximum * This is not usually of practical interest but is useful as an ideal limit of operation. AMPLITUDE RANGE CONTROL 523 of possible noise improvement and also a maximum of signal modifica- tion. There is, however, no information in the compressed signal INVERSE COMPLETE EXPANDOR IDEAL TRANSDUCER, COMPLETE REPEATER OR COMPRESSOR ATTENUATOR COMPLETE EXPANDOR -RANGE INVERTERS - ♦-COMPRESSORS- -EXPANDORS- -45 0 45 ANGLE OF SLOPE IN DEGREES (FROM HORIZONTAL) (b) Fig. 2 — If transducers are classified with respect to the slope of the input-output characteristics, several fields of action with definite demarcations result. which would serve to indicate how much compression occurred. Consequently, if it were desired to restore the original range, it would be necessary to transmit this information in addition to the compressed 524 BELL SYSTEM TECHNICAL JOURNAL signal. The gain of the restoring device would be guided by this auxiUary information. Hence, the device used to pass the information along is called a "pilot channel." Various types of pilot channels are listed in Part 4 as secondary characteristics of the control. When the control ratio is between 0 and 1 the compression is in- complete. A wave compressed in this manner has the property of being able to cause re-expansion at the receiving end since the output amplitude bears a definite relation to the original, assuming constant transmission over the intermediate circuit. In the field of expandors having a control ratio between one and infinity the signal modification is opposite to that of compressors. Thus a convenient method is available for restoring the original wave shape by using an expandor having a control ratio which is the re- ciprocal of that of the compressor at the sending end. Effects of Control Ratio The control ratio is useful in determining the effectiveness of a device in improving transmission in the presence of noise in the med- ium. When noise alone is acting on the device, the noise determines the action in a manner similar to speech. When both noise and speech are present, the action is determined by the sum of the two. Thus, room noise applied with the speech will be compressed or expanded exactly as if it were part of the speech. In the case of a compressor used at the sending end of a noisy circuit, an input range of say 60 db might be compressed to 20 db, by using a control ratio of 1/3 over the entire input range. At a point where the strongest signals are un- changed, the weaker signals would then be 40 db stronger than when the compressor was omitted. The improvement of the signal and applied noise with respect to noise in the medium thus depends on the difference in ranges at the input and output which depends on the control ratio. A large part of the usefulness of an expandor is in changing the apparent ratio of speech to the noise heard in the absence of speech, since the noise is generally weaker than speech and is made even less compared to speech by expansion. This is in spite of the fact that at any instant the signal-to-noise ratio is the same at the output as at the input. When the noise is comparable with the speech in amplitude, or when the noise is so weak as to be negligible without a controller, there can be no improvement in the noise conditions in using these devices. Between these two limits, the noise improvement rises to a maximum value also determined by the control ratio, and the time actions and range to be discussed. I AMPLITUDE RANGE CONTROL 525 A receiving range controller also changes variations in the trans- mission medium in proportion to the control ratio. Part 2 — Time Actions Instantaneous Control A device having a given control ratio might have its gain changed simultaneously with the applied e.m.f. The signal modification would become greater as the control ratio departed farther from unity and the modified signals would approach rectangular wave shapes at the limiting control ratios. Unless instantaneous compression is limited to a very small part of the signal range, an incomplete in- stantaneous expandor (inverse rooter) is required at the distant end which does the reverse of what is done at the transmitting end to restore the signal to substantially its original form. Due to the characteristics of the compressed signals, however, a transmission bandwidth without appreciable amplitude or phase distortion of two to three times the normal is necessary for high quality transmission. Rectified Control To avoid the necessity of transmitting such a wide band of fre- quencies, as well as to permit the use of a single device without restor- ing, in which case the distortion is limited to a value which is permis- sible from the standpoint of a listener, practical devices do not operate instantaneously. Instead, the gain is controlled by the charge on a condenser, which is controlled by rectified waves. The action of such an arrangement will now be discussed. Consider a wave formed by subtracting two sine waves equal in amplitude, one having a frequency 10 per cent less than the other.* A portion of such a wave is shown in Fig. 3a. This wave is equivalent to a cosine wave of frequency one-half the sum of the two frequencies, as shown by the instantaneous voltages of Fig. ?>a, multiplied by a secondary wave (envelope) of frequency one-half the difference of the two original frequencies. The instantaneous voltages of the wave of Fig. 3a vary from a positive maximum through zero to a negative maximum. Curve a of Fig. 4 is a summation of most of the instantaneous e.m.f. 's of Fig. 3a with respect to their occurrence. About 99 per cent of the instantane- ous voltages are in the ranges shown, the remainder being in the range between the upper and lower halves of Fig. 4. * This illustration is not directly comparable with speech, but it contains some of the' attributes which are comparable in this analysis, besides being readily repro- ducible and relatively simple. 526 BELL SYSTEM TECHNICAL JOURNAL Figure 3b indicates values for the same wave in which the negative ordinates have their signs reversed by means of an ideal full-wave rectifier. The resulting wave contains frequencies which were not present in the original, prominent among them being second and higher harmonics of the original. The range of instantaneous values shown 5 0 RECTIFIED ENVELOPE-^ RECTIFIED TIME Fig. 3 — A wave's amplitude varies from positive maximum to negative maximum. If symmetrical, the amplitude may be expressed as varying in only one direction from zero to maximum by rectification. on Curve b of Fig. 4 is only half that of the instantaneous voltages. About 99 per cent of the values lie in a 60 db range. The instantaneous values of the envelope of the rectified wave follow curves 3c and 4c. In speech the envelope is composed of many rather low frequencies which are determined by the rates of enunciation of syllables. For this reason they are sometimes called the syllabic frequencies. If it were possible to make the control vary as a function of the envelope, the result of using a control ratio of 1/2 on the wave of Fig. 3a would be as shown in Fig. 5c. This was I AMPLITUDE RANGE CONTROL 527 obtained by multiplying the original wave by a factor which is inversely proportional to the rectified envelope. For comparison, the original wave is shown in Fig. 5a, and the result of instantaneous compression I ? 20 5 40 O ^20 C RECTIFIED ENVELOPE — ^ -- ^ X ^ / y/^ b RECTIFIED _[_ 1/ / a INSTANTANEOUS 1 I j /a ^ Fig. 4- 10 20 30 40 50 60 70 80 90 100 PER CENT OF TIME EMF IS ETQUAL TO OR LESS THAN EMF SHOWN -The amplitude range of the wave of Fig. 3 is infinite on a db scale but most of the values are bunched in a much smaller range. by the same control ratio in Fig. 56. It is assumed that the arbitrary reference voltage which is not changed by compression corresponds to the maximum value of the input wave, although any other value might be used instead. It is evident from Fig. 5 that envelope com- 528 BELL SYSTEM TECHNICAL JOURNAL ■£ c j: ol 03 a> s< 3 (Tl O u 0) ex m ■M O CXI OT3 .2 c >- O S 6 o » u a rt w (/) c 0) (fl J2 ^ CO J= a+- ,n is a M UJi' ITl > :^ •^•^ JZ U. a5 + EMF AMPLITUDE RANGE CONTROL 529 pression would result in less distortion than instantaneous compression. The extra frequencies formed that were not present in the original wave are the envelope frequencies, so that the additional band required to transmit this wave faithfully is negligible. Dynamic Operation The measurements ^ and adjustments of speech amplitudes in com- mon use are made with devices that integrate the effects of the wave over certain time intervals. They do this in a rather complicated manner, however, so that it is difihcult to express the resulting quanti- ties in terms that are generally understood. In the measuring instruments the rectified voltages are impressed on a condenser before being sent through a meter. The readings of the meter are, therefore, proportional to the voltage on the condenser modified by the damping of the meter. The voltage is made up of the sum of the elTects of all the instantaneous voltages that have been applied to the condenser from the beginning of time to the instant under consideration. These effects die out so rapidly, however, that the instantaneous voltage on the condenser is practically determined by the voltages received in the immediate past. The condenser may be said to have a memory but a short one. In range control devices, the condenser forms the voltage which determines the amplification of the device. To distinguish this voltage on the condenser from the applied voltage at any instant, we may call the former an "impression" of the original wave. If the time constant RC is small we get strong impressions similar to the rectified applied wave and its envelope, and if it is large we get weak impressions quite different from the applied wave but something like the rectified envelope. Figure 6 shows the impressions of the wave of Fig. 3a, using four different values of time constant RC as compared to P, the period * of the envelope. Figure 7 shows smoothed summation curves of the impressions of Fig. 6 formed during the time P/2. Comparing this with Fig. 4, it is evident that the "bunching" effect for the distribution of impressions is largely between those for the rectified instantaneous and envelope curves. For the longer time constants, i.e., weak im- pressions, this is not the case for the weaker e.m.f.'s. * This is twice the duration of Fig. 3, since only half a cycle is illustrated. It is assumed that C is completely discharged at the time this wave is applied. In prac- tice, the rectifier impedance varies with the applied e.m.f. so that the results are not as simple as in this illustration. In general, the time actions are different de- pending on whether the applied e.m.f. is increasing or decreasing. 530 BELL SYSTEM TECHNICAL JOURNAL / °- J ^i 6( <<^^ o o i-ilrr:=^ 6 ciTf II _J3^lr==:- — 1 — ^ \ \ » ) \ \ \ \ 1 1 cr ~~~-^^y— , ""'' ■ '^c^r \ c^N^'*. 1 IMPRESSION AMPLITUDE RANGE CONTROL 531 Referring again to Fig. 6, it will be seen that for the two smaller values of RCjP the impression curves are composed of (1) the envelope frequency, (2) double the fundamental frequency, and (3) a small delay which can often be neglected. An approximation to envelope com- pression is therefore possible by choosing RCjP to be in the proper range, i.e., .0025 to .025, and making the output vary as a root or power ot the impressions thus formed. Figure 8 shows the result of compressing the wave of Fig. Za by using the impressions of Fig. 6 to determine the amplification. It was 0 10 20 30 40 50 60 70 80 90 100 PER CENT OF TIME IMPRESSION IS EQUAL TO OR LESS THAN IMPRESSION SHOWN Fig. 7 — The amplitude ranges of the impressions shown in Fig. 6 are bunched differently, depending on the time constant. A "volume" measurement means that a given impression is exceeded a small percentage of the time. In speech the peaks are relatively higher than in the wave illustrated. assumed that the amplification varies in inverse proportion to the square root of the impression. The resulting waves for RCjP = .0025 and .025 (medium impressions) are recognizable as something like the original wave. However, for the larger values of RCjP (weak im- pressions), the distortion at the beginning of the wave is quite large. This is because the impressions are formed so slowly that a longer time is required to drive the gain down to the desired value. In order to compare impression compression with instantaneous compression, the ordinates of Fig. 5 and 8 were plotted in Fig. 9. This shows that the greatest possibilities of bunching the waves into a narrow range result from the use of instantaneous compression {b), since the ratio between any value and the maximum is modified by the 532 BELL SYSTEM TECHNICAL JOURNAL c o o o AMPLITUDE RANGE CONTROL 533 a INPUT, UNCOMPRESSED OUTPUT, COMPRESSION BY IMPRESSIONS • OUTPUT, ENVELjOPE COMPRESSION >0 60 70 60 90 PER CENT OF TIME EMF IS EQUAL TO OR LESS THAN EMF SHOWN (POSITIVE VALUES) Fig. 9 — The amplitude ranges of the compressed waves of Fig. 8 are shown, together with those of Fig. 5. The amount of signal modification (and noise im- provement) for any amplitude below maximum may be correlated readily with the time constant. 534 BELL SYSTEM TECHNICAL JOURNAL control ratio. In the case of envelope compression (c), the lag causes a reduction in the amount of compression of the instantaneous voltages and the result is seen to be about half way between curves a and h of Fig. 9. The remaining curves of Fig. 9 are representative of the device ^ used on the long-wave transatlantic radiotelephone circuit. Volume Control To avoid both a large range and also the necessit}^ for a continuous record, practical speech amplitude measuring instruments are not directly concerned with either instantaneous, envelope, or impression voltages. Instead a value is determined, corresponding to an im- pression which is exceeded only a small percentage of the time. This is the principle underlying speech measurements with "volume in- dicator" type of instruments. In the case of speech, which is much more complex than the simple wave we have discussed, curves like Fig. 9 are steeper, i.e., there are relatively more peaks and a larger range to complicate the problem. A particular device capable of compressing according to the re- quirement that the dynamic "volume" range should be reduced, is attained by a combination of several separate range controllers. One is provided to reduce the gain very rapidly when the output volume is too high. A second increases the gain, at a much slower rate, when the impressions formed on the condenser are consistently too low. A third disconnects the condenser from the input when the applied voltage is very small, so that the distortions inherent in change of gain by weak impressions will occur only at times of large and sudden decreases of volume. In the device ^ employed to control volumes applied to a radio transmitter at Norfolk, Virginia, a fourth control provides for rapid partial compression of high peaks, thus improving the modulation. It is unnecessary to re-expand for the purpose of restoring the intelligibility, since the distortion is virtually limited to a change in loudness. Part 3 — Range Range controllers, like repeaters and attenuators, are limited as to the input range they can accept and the output range they can provide. These limits may be due to thermal noise at the low end and output carrying capacity at the high end. Heretofore, in this paper, the terms "input" and "output" have been purposely left somewhat vague so as to be as general as possible. However, the limits of input and output of a range controller take on particular significance when it is considered that the signal input range may difTer both from the AMPLITUDE RANGE CONTROL 535 input range of the device, and also from the range over which control is exercised. This control range may be defined as the difference between the maximum and minimum values of an applied wave over which a device is designed to function in a specific non-linear manner. It is usually expressed in db, and may apply to any measure of the applied signal, such as instantaneous voltage, rms steady-state sine waves, or a dynamic measure such as volume. The values dividing the con- trolled range from the uncontrolled ranges may be referred to as the "control points." Certain advantages in some cases have been found from restricting the control range. This is accomplished by placing one or both of the control points inside the useful amplitude range. The position of the control point may be moved arbitrarily over a wide range by putting an ordinary repeater (or attenuator) in tandem with the range con- troller. A given amount of compression at the high amplitude end of the range gives a real signal-to-noise advantage for a much greater proportion of applied e.m.f.'s than the same amount of compression at the low end of the range. In either case the distortion would be less than that of a full range compressor. When expandors with limited range are used, they are subject to the limitation that variations in the medium are increased, but to a lesser extent than full range expandors. Part 4 — Classification of Range Controllers — Secondary Characteristics Table I, page 536, suggests how the conceptions of control ratio, time actions and range already discussed might be employed to dis- tinguish a variety of devices. In cases where more than one device is covered by a given control ratio and time action the distinction is that the ranges are different. The names of devices used in this table are those which have been used in the past to distinguish the devices one from another. Nomenclature Using the above conceptions of the three primary characteristics, it has been found possible to devise a notation to distinguish all the known devices in this field. As an example of how this proposed system of nomenclature would be applied. Table II, page 537, gives three columns. Column 1 sets forth the arbitrary names that have been used in the past to distinguish certain devices which have come into use. Column 2 gives descriptive names which specify the three fundamental characteristics. In column 3 each device is named by three symbols defining the three fundamental characteristics, and a 536 BELL SYSTEM TECHNICAL JOURNAL TABLE I Classification of Range Controllers Typical Compressors (r < 1) Expandors (r > 1) 1 ime Actions Full Range Limited Range Full Range Limited Range Instantaneous Rooter Peak Chopper, Voltage Limiter Inverse Rooter, (Squarer) Voice Operated Relays, Cross- talk Sup- pressor Syllabic Compressor Limited Range Compressor, Peak Limiter Expandor Noise Reducer Volume Vogad, Range Reducer Volume Lim- iter, Half Vogad Range Restorer classification which tells what the device is designed to do. In this system the numbers preceding the letters specify the input control range in decibels, and the position of a horizontal bar indicates the position of the main signal range with respect to the control range. The letters specify the time actions and in the case of vogads, where several time actions may be combined, an arbitrary combination of letters would be used. The final numbers specify the control ratio, and in the case of vogads, where this might be different depending on whether the input was increasing or decreasing, both values are given, the former first. In this system, definitions of time actions are prerequisite and by way of illustration, the following symbols have been used: / represents instantaneous, meaning very fast adjustment of device 5' represents syllabic, meaning moderate speed adjustment of device V represents volume, meaning a combination of controllers which produces adjustment of device in response to dynamic speech so that the output volume is approximately determined by the input volume. Secondary Characteristics In addition to their three primary characteristics, range controllers may have a number of secondary features which are sometimes im- portant. The outstanding ones are: 1. Bias A neutral range controller is one which holds its setting during the quiet periods between words and sentences and which changes its gain AMPLITUDE RANGE CONTROL 537 TABLE II Comparison of Nomenxlature for Range Controllers Col. (1) Arbitrary ^ 1. Vogad 2. Vogad Combined with Syllabic Compressor 3. Volume Limiter 4. Compandor 5. Noise Reducer 6. Limited Range Com- pressor 7. Peak Limiter 8. Peak Chopper 9. Crosstalk Suppressor 10. Rooter and Inverse Rooter n. Vodas (Singing Sup- pressor Relay) 12. SvUabic Vodas Col. (2)_ Systematic Full Range 45 db Volume Compressor Full Range 45 db Volume Compressor High Range 15 db 1 : 5 Volume Compressor Full Range 60 db 2 : 1 Syllabic Compandor Low Range 20 db 2 : 1 Syllabic Expander High Range 10 db 1 : 2 Instantaneous Compressor High Range 12 db 1 : 5 Syllabic Compressor Hi^h Range 6 db 1 : 100 Instantaneous Compressor Low Range 10 db 10 : 1 Instantaneous Expandor Full Range 70 db 2 : 1 Instantaneous Compandor Col. (3) Symbolic 45 F5/ 23-18 Compressor 45 F55/ 23-18 Compressor 15 VS Compressor 6052 Compandor 2052 Expandor 1072 Compressor 1255 Compressor 6/100 Compressor 10710 Expandor 7072 Compandor 07 so Expandor 05=0 Expandor only when the waves acting on it differ from those just received. This condition sets a new requirement on the range controller which can usually be met by a combination of control circuits. A biased controller is one which returns to a setting corresponding to some fixed or biased intensity when speech is not passing and adjusts itself each time speech begins. A simple compressor is biased since with no input it generally takes a setting of maximum gain so as to be right for the weakest waves that might be applied in its working range or below the working range. It is also possible to bias a range control- ler so as to have minimum gain, or any other intermediate value when no waves are applied. An important secondary characteristic is the rate at which the device returns to the desired "bias" point. Any of the devices listed in the tables might be neutral or biased in either direction, thus increasing the number of possible arrangements. 2. Behavior Outside of Range For inputs outside the working range of a range controller it is important to provide that the amplification of these waves does not cause them to be modified so as to be out of proportion to output signals in the main range. In some cases this is met by choosing a device which follows the same law all the way to zero current. In others, the device may act as a linear transducer, i.e., with range 538 BELL SYSTEM TECHNICAL JOURNAL factor of one outside the working range. Various other combinations of control ratios can, of course, be employed. 3. Pilot Channel In all complete compressors some form of pilot channel is necessary to control the re-expansion if this is required. If the gain changes are slow, the pilot channel may include an operator who changes the gain of the receiving device in a manner complementary to that of the sending device based on aural or visual signals. If the gain changes are too rapid for the operator to follow, the receiving gain may be changed automatically. The pilot channel itself may be a direct or alternating current of variable amplitude or frequency, or in case of carrier or radio, it may be the carrier frequency. In sound reproduction a pilot channel might be a pilot track on the record. Summary In an amplitude range control system, the following characteristics must be specified in addition to the usual repeater characteristics, to determine its design and performance: 1. The steady-state control ratio, which determines how much control is obtained and whether restoration can be made automatically or not. 2. The manner in which the output varies with time, following a given change in input. 3. The range over which it is to function. In specific cases the following should also be considered: 4. The action of the device for inputs outside the working range. 5. If the device is a complete compressor, the type of pilot channel for restoring. 6. The action of the device when signals are removed. Acknowledgment The computations used to obtain Figs. 3 to 9, inclusive, were made by Miss Marian Darville. References 1. " Devices for Controlling Amplitude Characteristics of Telephonic Signals," A. C. Norwine. Presented at A. I. E. E. Pacific Coast Convention, Aug. 9-12, 1938. 2. "Speech Power and Its Measurement," L. J. Sivian, B. S. T. J., Vol. VIII, pp. 646-661. 3. "The Compandor — An Aid Against Static in Radio Telephony," R. C. Mathes and S. B. Wright, B. S. T. J., Vol. XIII, pp. 315-332. 4. "A Vogad for Radio Telephone Circuits," S. B. Wright, S. Doha, A. C. Dickieson. Presented at I. R. E. Convention, June 1938. Devices for Controlling Amplitude Characteristics of Telephonic Signals* By A. C. NORWINE This paper describes a family of devices which automatically respond to signals and control the circuit amplification in such a way as to improve transmission. Their general characteristics are outlined, their differences explained, and some of their applications are listed. Introduction THE transmission of speech energy over electrical circuits is attended by the interesting and sometimes difficult problem of preserving the original signal in spite of limitations in the transmission medium. These limitations include load carrying capacity, inter- ference with other service, noise, change in attenuation with time and many others. Because of special limitations it is sometimes desirable to alter the amplitude characteristics of the speech or other signal energy without, of course, materially lowering its intelligibility. In high quality systems the peak voltage from some speech sounds of a given talker may be over 30 db (some 30 times) higher than from his weakest sounds when there is very little inflection in the speech. Loudness changes for emphasis will increase this range of intensities. Ordinary message systems do not have to contend with quite so wide a range of instantaneous voltages from a single talker, but different talkers under extreme terminal conditions produce about a 45 db range of average voltage, which is additive to that for a single talker. Conse- quently, a voltage range of about 70 db (over 3000 to 1) must be considered for message circuits. In order to accommodate such ranges of intensity to certain trans- mission media such as radio links a new family of automatic devices has been developed. In general all of these contain amplifiers or attenu- ating networks whose loss or gain is changed according to some function of the applied input and which may have a variety of time sequences in their control circuits. It is hoped that by the classi- fication and description of some of these devices their distinguishing characteristics and fields of usefulness will be made somewhat clearer. * Presented at the Pacific Coast Convention of A. I. E. E. and I. R. E. in Port- land, Oregon, August 9-12, 1938. 539 540 BELL SYSTEM TECHNICAL JOURNAL We are to be concerned here principally with those elements allied to the telephonic art, although some applications are to be found in other fields. It is not intended to include those voice operated functions which are essentially switching operations although the distinction in some cases becomes exceedingly fine. Names of volume controlled devices * which have been used in published papers^include vogad,t' ^' ^' * compandor, |' *• ^ and volume limiter.^ Without direct comparison it may not be obvious how these and similar devices differ. First the apparent similarity of several of these devices will be shown in simple diagrams. Next the more important characteristics of a number of devices will be presented in tabular form, followed by descriptions of the different types. These will then be discussed with particular emphasis on their distinctive qualities, with notes on their variants which have some apparent value. General Characteristics of Volume Controlled Devices In Figs. 1 to 10 are shown simplified diagrams of some of these devices. While detailed descriptions of them will be deferred till later it may be pointed out that all those shown contain vario-lossers, and all have paths from the main transmission path to control circuits which affect the vario-lossers. A vario-losser usually consists of a balanced pair of vacuum tubes whose gain is changed by varying the grid bias, or of a network of non-linear elements such as copper oxide or silicon carbide whose loss is changed by varying a current through them. In some special cases it may be a mechanically adjusted variable network. The word vario-losser is thus a generic term relating to a circuit whose loss or gain is controllable. A control circuit ordinarily consists of an amplifier and rectifier whose direct current or alternating current output bears a chosen relation to its input. Thus some control circuits are marginal; they produce no control voltage till the input exceeds some critical value, then produce large control voltages for small additional increments of input. These are used, for example, when it is desired to limit the output of a vario-losser to a definite amount. Another type of control circuit produces a current or voltage which is linear with input expressed in decibels. In combination with a vario-losser whose gain is a linear function of control current or voltage one can produce a device whose gain is a linear function in decibels of the input to the control circuit. * See the footnote on page 543. t " Folume Operated Gain Adjusting Device." X A combination of the names "Compressor" and "Expatidor." AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 541 BIASED CONTROL CIRCUIT VARIO- LOSSER GAIN INCREASER GAIN DECREASER BIASED CONTROL CIRCUIT GAIN INCREASE DIS" ABLER (BLOCKS INCREASER ACTION) VOGAD BIASED CONTROL CIRCUIT VARIO- LOSSER M GAIN INCREASER A GAIN DE-< CREASER GAIN INCREASE DISABLER VARIO- LOSSER BIASED CONTROL CIRCUIT TIME FUNCTIONS SAME AS VOGAD GAIN CHANGES HALF AS GREAT HALF VOGAD IN VARIO- OUT , BIASED 1 CIRC :uiT VOLUME LIMITER FIGURE 3 REGULATING THRESHOLD IN VARIO- OUT LOSSER ' CONTROL CIRCUIT V OLTAGE = K- INPUT VULIAfct COMPRESSOR / \ N VARIO- 01. LOSSER . CONTROL CIRC ;uiT K- VOLTAC INPUT \ ;e = /OLTAGE EXPANDOR JZI 542 BELL SYSTEM TECHNICAL JOURNAL It will be recognized that if the application or removal of the control energy is retarded, the action of the control circuit may be made quite different on transient inputs than on steady state inputs. It will appear later that this is the important distinction between some of the devices to be discussed and that fundamental differences in their functioning are thus brought about. Referring to the figures once more it will be noted that some control devices are connected to the transmission path at the input to the vario-losser. These are known as "forward acting" control circuits. Other controls, connected at the vario-losser outputs, are known as "backward acting" control circuits. This is simply convenient terminology to indicate whether the control energy is progressing in the same direction as the main transmission or is progressing in a backward direction after traversing the main path, usually through a vario- losser. Some backward acting controls function to measure the output of the devices containing them and to make whatever adjust- ments are required. Others are placed in that position to take advantage of the vario-lossers in the transmission paths, i.e., such controls could be replaced by combinations of forward acting controls and extra vario-lossers. In Table I, nine of the volume controlled devices * which have been developed for various commercial and experimental uses are listed with the functions of voltage, time, and frequency which are employed to obtain their respective performances. There is, of course, some latitude in the choice of these functions for any one device. Pending more complete description of the different types in the following paragraphs this table should be viewed as illustrating the general character of the different circuits and also the range of the variables which already have been employed. For example, it will be seen that instantaneous voltage of the signal wave, its short time average value, peak power, syllabic variations, and long time average power have all been used as criteria of gain settings in different circuits. Some devices change their adjustments only when critical values or ranges are exceeded, while others vary somewhat with every syllable if speech, for example, is being transmitted. Some are linear transducers to all but low or high amplitudes while others reduce or increase the output range from that at the input. It will be seen that proper choices of times for gain increase and gain decrease in combination * The names employed do not follow an entirely logical classification, but they are given here because they have had considerable usage. For the same reason the term volume controlled devices is used, although to be strictly correct it might better be sound energy controlled devices, for example, for not all the devices operate in accordance with volume as measured by the well-known class of visual reading meters called volume indicators. AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 543 VARIO- LOSSER VOLTAGE = K- INPUT VOLTAGE CONTROL CIRCUIT LIMITED RANGE EXPANDOR RADIO NOISE REDUCER FIGURE 6 AMPLITUDE A AMPLITUDE IN VARIO- OUT LOSSER ; CONTROL 1 VOLTAGE = ^ K-INPUT VOLTAGE CIRC :uii IN VARIO- OUT LOSSER . BIASED 1 CIRC UIT N VARIO- 0 LOSSER HIGHLY- BIASED CONTROL CIRCUIT AMPLI- TUDE 8 PEAK CHOPPER FIGURE 9 AMPLITUDE BIASED CONTROL CIRCUIT VARIO- LOSSER TWO VALUES OF LOSS CROSS-TALK SUPPRESSOR FIGURE 10 AMPLITUDE 544 BELL SYSTEM TECHNICAL JOURNAL S^ o M U w 1^ J n w .J < o H K^ . 1 t) r- "^ wi >..S £ Part of Input Range Causing Operation •a .a-o "5 0,'c S, "2 3 cil mT3.M;2 3 .2f 3 j3.t: _, •^ oJ 0 _ =3 rt 0 aria! low tude Fixe to h amp < I! < < > >. igh frequency only or multiband Frequenc Range Causing Control •d c "3 •0 c 3 •0 a OS 3 T3 a 3 fc d. fa K fc >. 13 igh frequency only or multiband Sc2 X) c "3 •§ .§ 3 •d c .Q 3 to u, fe X fa 0) D Volume Range Con- trolled* CD 1 ■0 41 0 1) •d 0 a! ►J T3 0 -d . 01 a ^ °^2 p 3 a 3 0 pressor put andor, ut itto pandoi trol ov arate nnel 3 a o o C *J g^«c •0 X c a ™ c Ph O — 3 i> 0 vS "^ •u 0 0 0 41. r; i_ u m u < < U 0 < io of tput ige to put nget 0 i ■^ >- C3 0 10 0 0 0 •0 low pli- es. high lol^^ a .^&a^ (S LO < ^ 0 — o4 -.. tH \-> a 0 0 0 0) (P 60 OJ 1- :3 c 0 3.2 0 •0 bic iations r part ut ran o cd n 03 iH 11 Q, O ^ =3 « >•> ;3 m > S ^> 0.5 < «: (/) C/} I- "C" 0 0 T3 •d a a S u rt a V a L E 0 0 X •d™ 1) Q to 0 u U 1> 0; 'e 0 E a J T3 0 X ft °^ OJ COW OJ a _• • W z-S •d E o.ti a] &0 _3 ^ ^ ^g ;S > 0 u f25 - cs «5 ■* AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 545 O I _. 1 0) u i> tof put nge sing ation 3 3 -J •2 J" > CD (1/ S - ao _ ffi K K < ■" < >. Frequenc Range Causing Control •d •d •0 •0 •d c B 3 3 c rt rt ta ca ca X) /2 J2 J3 J3 3 "3 3 3 3 fa fa fa fa fa y ni "J •0 T3 T3 •d 13 3 ca J2 fa U c J3 C J3 C ta 3 "3 fa 3 fa "3 fa 3 fa 3 fa 4) 'olume Range Con- rolled* E "is &E 0) m 11 m 2 •S 0 " - w Cfl > Y^ j3 1- — « c ^ 0 ^ "ca ' •5^2 == 3 3 a 3 3 •B ° c ag. 3 0 a a I. 0 o o c ir c 3 ii ta Ph O — 3 '"■ •-■ *j 0 < •< < . :a M si ca K tn 3 P< a fa C M 0 c 0 ^^ 3 ° 3 3 I^C ^ S-d 0 s 3 ca c is C ca 3 H 0 0 OD Ss ca to ca nJ il 0 3 ca c fa < fa OJ 3 >.g 3 0 i'.S'^ 0 3 3 V 3 0 U •o ."^ii m 01 3 BC 0 ca il.si' (u a 0 g 3 0 0 rt £ 3.2 u 3 0 13 So B > K 3 ■^ bo 3 ca cail o O cfl I- ai a s rt > e iS ca ca to i! ">. 3 a o.„ >._ C > CAi w > j_ u i 0 (U 0 K u Pi s i 0 u ij 01 (U a > '> so e 0) R a 3 3 0 ^ S 3 a 0 u 3 •d 3 ca ^ ^ e RJ to 0 0 0^ Of Q 3 flH eu 0 ti ui VC t^ 00 ol s > •a S a.t: a 3 c/) • — So > o a; O) -w '^ en t/5 C C U C/l — T) Si OJ 3 0 III CD -r) biO C TD 01 i_ lU m ^ tn 01 • -- j: 0 "^ 03 0 -C e a> en 0 biO c 0 C n cfl 11 4J - d) c •* cS 0) bA n u n! a 0 -n h > 3 "^ , 0 >^ u 546 BELL SYSTEM TECHNICAL JOURNAL with certain gain control criteria make possible a wide variety of signal altering means to meet different requirements. Description of Devices in Table I With this introduction to the combinations of characteristics which are possible it should be less difficult to distinguish between the specific devices discussed in the following paragraphs, which, in addition to describing the devices, contain some comments which should assist in visualizing their forms and their operation. 1. The vogad (Fig. 1) is a device which will maintain at its output speech volume ^ which, over a certain range of input, is relatively independent of the speech volume applied to its input and which, in the ideal case, will not change its gain during periods of no speech input. It makes little or no alteration in the ratios of maximum and minimum instantaneous to average voltages of the speech. 2. The volume limiter (Fig. 3) is a device which is a linear transducer for all speech volumes up to a critical value, beyond which all input volumes produce essentially the same output volume. It is essentially different from the vogad in that its gain approaches the maximum value when input is removed. 3. The compandor (Figs. 4 and 5) is composed of a compressor and an expandor. A compressor is a device whose input-output characteristic on a decibel scale has a slope less than unity * and whose gain or loss is variable under control of the input energy at a time rate which will permit it to follow the syllabic rate of change of speech energy. Simi- larly, an expandor is a device whose input-output curve has a slope greater than unity and whose gain is variable at a syllabic rate under control of the input energy. Thus very shortly after all input is removed the gain of a compressor is maximum and the loss of an expandor is maximum. The reciprocal of the compressor characteristic slope is spoken of as the compression ratio, and the slope of the ex- pandor characteristic is spoken of as the expansion ratio. 4. The radio noise reducer *• ^ (Fig. 6) combines the functions of an expandor which operates in the range of amplitudes where noise and weaker speech sounds lie and a linear transducer which comes into play for all amplitudes exceeding a critical value, which can be set to best suit the atmospheric noise conditions. In other words, the radio noise reducer is a limited range expandor. Inputs which are below the expandor range are subject to transmission at the minimum gain. 5. The limited range compressor (Fig. 7) is a device whose operating * That is, if the input increases by x db the output increases by less than x db. AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 547 range includes a region within which compression at a syllabic rate can take place; at other inputs the device is a linear transducer. Its connecting diagram and time functions are the same as those shown in Fig. 5 except that the control circuit contains a limiting device, so that compression takes place in only a portion of its input range, analogous to the action of the limited range expandor of Fig. 6. As a special case the limited range compressor may have no linear range above its compression range, thus becoming one type of peak limiter. 6. The peak limiter (Fig. 8) is a device whose gain will be quickly reduced and slowly restored when the instantaneous peak power of the input exceeds a predetermined value. The amount of gain reduction is a function of the peak amplitude, and in practice is usually intended to be small to prevent material reduction of the range of intensity of the signal. 7. The peak chopper (Fig. 9) is a device which prevents transmission of peak amplitudes exceeding a critical amount, an essential charac- teristic being that the loss it inserts is completely determined by the instantaneous voltage of the signal. That is, its operating and releasing times are substantially equal to zero. 8. The crosstalk suppressor (Fig. 10) is a device which normally presents a prescribed loss to transmission, which loss is removed rapidly when the input amplitude exceeds a certain threshold and is reinserted at a definite time after the input is removed. It reduces low amplitude unwanted currents such as crosstalk but does not affect amplitudes in the useful signal voltage range. This device differs from the limited range expandor in that the time during which the low loss condition is maintained is considerably greater, so the transition from one gain to the other occurs less frequently. 9. A rooter is an instantaneous compressor. Such a circuit can be made to produce an output whose instantaneous voltage is, for example, the square root or some similar function of the instantaneous voltage applied to the input. An inverse rooter is an instantaneous expandor whose characteristic is complementary to that of the rooter. A combination of rooter and inverse rooter will reduce the load require- ments on a transmission system between the two units but requires that it transmit a wider band of frequencies than that for the original signal, and that it be essentially free from phase distortion. This does not seem to be an attractive arrangement from a commercial viewpoint and is included here simply as an illustration of one of the possible modifications of signal energy. It is not shown" in the group of diagrams. 548 BELL SYSTEM TECHNICAL JOURNAL Variants to the Devices Described In addition to these there are various devices which are essentially modifications of those described. For example, a half-vogad, Fig. 2, may have the same time functions as a vogad, Fig. 1, but the gain changes in the transmission circuit are half as great for the same range of input volumes. Thus in a vogad the range of gain changes in the transmission circuit is equal to the range of input volumes, so that the output volume is the same for all input volumes. In the case of the half vogad the range of gain changes in the transmission path is one- half the range of input volumes, so the output volume range is one-half that of the input. It is also possible to construct a vogad whose output volume range is any desired fraction of the input range. As another example of modification of the devices described, for special appli- cations it may be desirable to incorporate a certain amount of syllabic compression in a vogad. Communication circuits which have separate paths for oppositely directed transmission between the two terminals are usually operated at such an overall loss that with ordinary terminations there will be little tendency for circulating currents to build up to a "singing" condition. Sometimes there may not be a great deal of margin, however, so that volume controlled devices added to such circuits must add loss at some point to counterbalance whatever gain is put in at some other point. Thus a vogad inserted at the transmitting side of one terminal of such a circuit to amplify speech energy from weak talkers must be supplemented by a "reverse vogad" in the receiving side of the circuit. The reverse vogad is simply another vario-losser which is operated upon by the vogad control circuit in such a way that it always has a loss numerically equal to the gain of the vogad. Any vogad gain will be compensated by the reverse vogad loss, so no greater tendency to sing will be effected by the addition of the combi- nation to the circuit. In like manner half vogads must be used with compensating reverse half vogads. Combinations of some of the devices also have interesting charac- teristics. For example, a combined radio noise reducer and peak limiter at the receiving end of a circuit would suppress noise and would also reduce the amplitude of excessively high amplitude signals. Likewise, a vogad, compressor, and peak chopper in tandem in the order named could be made to reduce the range of input signals by a very large amount for transmission over a medium having only a small range between noise and maximum permissible signal. In this case it would be practically impossible to recover the original signal range at AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 549 the receiving terminal of the medium, but the intelligibility of speech over such a system has been shown in the laboratory to be good. Special compandors for high quality service may require compression and expansion which vary with frequency. The exact characteristics will depend upon band width, program material and transmission medium. For transmission media in which the noise reproduced at the receiving end is principally at the higher frequencies an unusual effect is obtained if the usual variety of compandor is used. Low frequencies unaccompanied by high frequencies will cause a gain change in com- pressor and expandor, thus changing the background of high-frequency noise which is not masked by the low-frequency signal energy. The resulting swishing noise has been given the onomatopoeic name of "hush-hush effect." To avoid this, recourse may be had to split band compandors in which the compression and expansion is done only at high frequencies or separately for low and high frequencies. The successful application of the latter method is, however, more difficult than it appears from its simple description. Distinguishing Characteristics It is important to distinguish between the half vogad. Fig. 2, and the compressor. Fig. 4. As shown in Table I the latter operates on syllabic variations and the former on the average volume of the input. Thus the half vogad reduces the range of output volumes to one-half that at the input while the compressor reduces the range of syllabic power at its output to one-half that at the input. In other words, the compressor reduces the ratio of peak to average power on constant volume speech, while the half- vogad simply adjusts for that volume and does not alter the peak ratio. There is, of course, the additional important difference that the half-vogad retains its gain setting during silent periods while the compressor, by virtue of having followed the syllabic power, has its maximum gain during silent periods. Volume limiters. Fig. 3, may be mistaken for vogads. Fig. 1, because during speech input above a certain value the two may produce the same output volume. They both employ something like a measure- ment of average power over periods longer than a syllable to determine their gain settings. The important difference is that a vogad retains its gain setting when speech currents are not present, while a volume limiter approaches its maximum gain during such periods. In terms of the output resulting from a range of input volumes there is another important difference if the volume limiter operates over only part of the input range: the vogad reduces the width of the distribution curve of volumes to a very small value, while the volume limiter moves all the 550 BELL SYSTEM TECHNICAL JOURNAL area under the distribution curve above a certain point to the region near that point, which is its Hmiting volume. This is illustrated in Fig. 11, in which the calculated modifications of a volume distribution by a vogad and by a volume limiter are shown. In the cases "without volume control " and "with a vogad " the distributions are normal, and the standard deviation, a, has its usual statistical significance. With a volume limiter, only volumes above the limiting volume are affected, WITH vogad/ (cr = I DB) 1 1 c / / 1 / // // // WITHOUT // VOLUME CONTROL // ) 1 \ 1 \ 1 \ \ \ \ \ \ \\ +4 V (Cr=l DB FOR VOLUMES ABOVE 0,-1-4, AND +7 DB) \ +7 (0-/ L)B2__^,— -*^ — 7 ■ 1 r y n -15 -10 -5 0 5 10 15 VOLUME IN DECIBELS FROM MEAN VOLUME Fig. 11 — Modification of volume distribution by use of a vogad or a volume limiter. and these higher volumes are redistributed according to a normal law whose standard deviation is 1 decibel, as stated in the figure. It is also important to distinguish between a peak limiter and a peak chopper. Figs. 8 and 9. Naturally they resemble one another since they are intended to permit transmission of signals at higher average amplitudes without excessive loading of transmission circuits. How- ever, they are intended for different classes of service and hence are not interchangeable except in some borderline cases. For the highest grade of transmission harmonic production must be negligible and the reduction in amplitude range of signals small and infrequent. Gain changes must be smooth, though rapid enough to compensate for practically any input wave to be expected. These characteristics are found in the peak limiter now being furnished for use on program networks and radio transmitters."- ^^ For services in which it is desirable to maintain the signal energy at a high value to over-ride noise and in which harmonic distortion must be kept low a peak limiter with somewhat smaller time constants may be used. A high ratio limited range compressor might be suitable in this instance. This device would lower its gain a little more quickly on excessive AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 551 inputs, and it would also reinsert its gain much more quickly; it would affect the naturalness of the sound of the signal more than the slower peak limiter but it would also cause the signal to over-ride noise some- what better. In a third variety of service the harmonic distortion introduced by a limiter is a secondary matter, the prime consideration being that the peak amplitude of the signal shall not exceed a specified value. This may be because higher amplitude signals would produce a tremendous increase in distortion or crosstalk into other channels or would damage expensive equipment farther along in the circuit. For these cases we may use the fastest possible type of limiter, the peak chopper, which simply cuts off any peak exceeding a certain value. The crosstalk suppressor, Fig. 10, is a splendid example of the fine distinction between volume controlled and voice operated switching devices. This device has been described, but in the present state of the art its time functions have not been definitely fixed. If the characteristic of loss versus input is made steep enough and the speed of operation fast enough it will sound like a switching circuit and may in fact be replaced by a relay-switched attenuating network. If made somewhat slower and given a smaller slope of loss versus input it approaches the limited range expandor or noise reducer. Applications and Expected Advantages It may be of interest to give some approximate figures on the magnitudes of the advantages to be obtained by the use of some of these devices. It will be understood that the values to be given are simply illustrative, some having been obtained from field service on particular models and some from tests on laboratory equipment under special conditions. Vogads appear to be most useful in such circuits as transoceanic radio connections, where it is important to properly operate the terminal switching equipment and to transmit over the radio circuit speech energy from loud and weak talkers equally well. It is essential in such cases that noise should not be increased in amplitude during speech pauses, hence the gain retaining feature of the vogad. On such a circuit a vogad will reduce a 45 db volume range to about 2 to 4 db. This is equivalent to expert manual volume control. Volume limiters are in use at the present time to prevent peaks of speech energy in carrier circuits from "splashing" into telegraph channels. '^ Some 5 to 10 db limiting is allowed on loudest talkers, which causes little degradation of the speech channels but makes possible the use of telegraph on the same carrier system. There is no 552 BELL SYSTEM TECHNICAL JOURNAL wide-spread use of volume limiters in point-to-point radio service so far, but in cases in which there is no disadvantage in raising noise in silent periods in speech, such as in push-to-talk installations, proper transmitter loading can be obtained with volume limiters fairly cheaply. One commercial model peak limiter, used as part of a program amplifier ^^' " is capable of introducing a considerable amount of compression without overloading on peaks, but for the preservation of adequate program volume range it is being recommended that only 3 db peak limiting be allowed. This, of course, reduces the range of intensity of the program, but from the standpoint of the listeners it is equivalent to doubling the transmitted power or obtaining the same signal-to-noise ratio with half the transmitted power. Limited range compressors might be used either on land lines to insure full loading or on radio links whose fading is too severe to permit the use of normal compandors. There is no commercial application of either sort at the present time. Peak choppers are, however, used on some high power radio transmitters which might otherwise be tempo- rarily disabled by high peaks in the signal being transmitted. The chief usefulness of compandors is on radio links in which the transmission of a compressed signal with subsequent expansion permits operation through higher noise or with lower transmitter power. On a long-wave transatlantic radio telephone circuit a compandor with 40 db range has been shown to allow an increase in noise of some 5 db before reaching the commercial limit.^ With smaller amounts of noise the noise advantage of the compandor approaches half its range in decibels. This benefit is sometimes applied to a reduction of trans- mitter power. Radio noise reducers have been used to advantage in connection with short-wave ship-to-shore and transoceanic radio telephone service. In the former, routine transmission rating is given on a judgment basis using a merit scale from 1 to 5, 5 being practically perfect transmission and 1 so poor that intelligibility is very close to zero. It will then be seen that the observed improvement of J^ to 1 point in transmission rating due to the noise reducer is of considerable importance. Perhaps more graphic figures are those for transoceanic service, where the reduction of noise in the receiving path not only reduces the noise heard by the listener but also improves the voice operated switching with the indirect result that at times .receiving volume increases of 5 to 15 db are realized.^ As has been noted, the radio noise reducer is a special use of an expandor alone. There are also two interesting applications for a AMPLITUDE CHARACTERISTICS OF TELEPHONIC SIGNALS 553 compressor alone. The first, which uses a fairly high ratio of com- pression, has been mentioned as one type of peak limiting device. The second, using a moderate ratio of compression, is in connection with announcing systems for use in very noisy locations. Its effect is to amplify weak sounds more than strong sounds, which considerably improves the intelligibility through high noise. For quiet locations it is of less value, since the speech sounds lose some of their naturalness in this process. Conclusion In the course of developing various types of the volume controlled devices which have been described means have been worked out for providing almost any combination of time constants, range of control, and other characteristics which may be required. Some devices for which there were specific commercial applications or useful functional characteristics for experimental work have been constructed, with resulting advantages which have been briefly mentioned. There remain many possible ways to alter the characteristics of signal energy such as speech to which these methods are applicable and which await the special needs of new transmission problems. Bibliography 1. C. C. I. F. White Book, 1 bis, pp. 77, 343. 2. C. C. I. F. White Book, 1 bis, pp. 251-3. 3. "A Vogad for Radio Telephone Control Terminals," S. Doba, Jr., Bell Labora- tories Record, Oct. 1938, Vol. 17, No. 2, pp. 49-52. 4. "A Vogad for Radio Telephone Circuits," S. B. Wright, S. Doba, Jr., and A. C. Dickieson, Presented at /. R. E. Convention in New York, June 18, 1938; to be published in Proc. I. R. E. 5 "The 'Compandor' — An Aid Against Static in Radio Telephony," R. C. Mathes and S. B. Wright, Elec. Engg., 1934, Vol. 53, No. 6, pp. 860-6; Bell Sys. Tech. Jour., July 1934, Vol. 13, No. 3, pp. 315-32. 6. "The Voice Operated Compandor," N. C. Norman, Com. and Br. Engg., Nov. 1934, Vol. 1, No. 1, pp. 7-9; Bell Lab. Record, Dec. 1934, Vol. 13, No. 4, pp. 98-103. 7. "Volume Limiter Circuits," G. W. Cowley, Bell Lab. Record, June 1937, Vol. 15, No. 10, pp. 311-15. 8. "A Noise Reducer for Radio Telephone Circuits," N. C. Norman, Bell Lab. Record, May 1937, Vol. 15, No. 9, pp. 702-7. 9. "Radio Telephone Noise Reduction by Voice Control at Receiver," C. C. Taylor, Elec. Engg., Aug. 1937, Vol. 56, No. 8, pp. 971-4, 1011; Bell Sys. Tech. Jour., Oct. 1937, Vol. 16, No. 4, pp. 475-86. 10. "Higher Volumes Without Overloading," S. Doba, Jr., Bell Lab. Record, Jan. 1938, Vol. 16, No. 5, pp. 174-8. 11. "A Volume Limiting Amplifier," O. M. Hovgaard, Bell Lab. Record, Jan. 1938, Vol. 16, No. 5, pp. 179-84. For the sake of completeness the following references are included, although no allusion has been made to them under the specific device-names used in this paper. 12. "tJber automatische Amplitudenbegrenzer," H. F. Mayer, E. N. T., 1928, Vol. 5, No. 11, pp. 468-72. 554 BELL SYSTEM TECHNICAL JOURNAL 13. "High Quality Radio Broadcast Transmission and Reception," Stuart Ballan- tine, Proc. I. R. E., May 1934, Vol. 22, No. 5, pp. 564-629. 14. "Expanding the Music," A. L. M. Sowerby, Wireless World, Aug. 24, 1934, Vol. 35, No. 8, pp. 150-2. 15. "Extending Volume Range," Radio Engg., Nov. 1934, Vol. 14, No. 11, pp. 7-9, 13. 16. "Amplitudenabhangige Verstarker," W. Nestel, E. T. Z., 1934, Vol. 55, No. 36, pp. 882-4. 17. "An Automatic Volume Expandor," W. N. Weeden, Electronics, June 1935, Vol. 8, No. 6, pp. 184, 5. 18. "Die Arbeitsweise der selbsttatigen Regelapparaturen," H. Bartels and W. G. Ulbricht, E. N. T., 1935, Vol. 12, No. 11, pp. 368-79. 19. "Practical Volume Expansion," C. M. Sinnett, Electronics, Nov. 1935, Vol. 8, No. 11, pp. 428-30, 446. 20. "Light-bulb Volume Expandor," Electronics, Mar. 1936, Vol. 9, No. 3, p. 9. 21. "Simplified Volume Expansion," W. N. Weeden, Wireless World, Apr. 24, 1936, Vol. 38, No. 17, pp. 407-8. 22. "Practical Volume Compression," L. B. Hallman, Jr., Electronics, June 1936, Vol. 9, No. 6, pp. 15-17, 42. 23. "Notes on Contrast Expansion," Gerald Sayers, Wireless World, Sept. 18, 1936, Vol. 39, No. 12, p. 313. 24. "Contrast Amplification: A New Development," W. N. Weeden, Wireless World, Dec. 18, 1936, Vol. 39, No. 25, pp. 636-38. 25. "Overmodulation Control and Volume Compression with Variable-mu Speech Amplifier," W. B. Plummer, Q. S. T., Oct. 1937, Vol. 21, No. 10, pp. 31-33. 26. "Limiting Amplifiers," John P. Taylor, Communications, Dec. 1937, Vol. 17, No. 12, pp. 7-10, 39-40. 27. "Low Distortion Volume Expansion Using Negative Feedback," B. J. Stevens, Wireless Engr., Mar. 1938, Vol. 15, No. 174, pp. 143-9. 28. "Distortion Limiter for Radio Receivers," M. L. Levy, Electronics, Mar. 1938, Vol. 11, No. 3, p. 26. 29. "Automatic Modulation Control," L. C. Waller, Radio, Mar. 1938, No. 227, pp. 21-6, 72, 74. 30. "An AVE Noise Silencer Unit," McMurdo Silver, Radio News, May 1938, Vol. 20, No. 11, pp. 46, 55. The Exponential Transmission Line * By CHAS. R. BURROWS The theory of the exponential transmission line is developed. It is found to be a high pass, impedance transforming filter. The cutoff frequency depends upon the rate of taper. The deviation of the exponential line from an ideal impedance transformer may be decreased by an order of magnitude by shunt- ing the low impedance end with an inductance and inserting a capacitance in series with the high impedance end. The magni- tudes of these reactances are equal to the impedance level at their respective ends of the line at the cutoff frequency. For a two-to-one impedance transformer the line is 0.0551 wave- lengths long at the cutoff frequency. For a four-to-one impedance transformer the line is 0.1102 wave-lengths long at the cutoff frequency, etc. The results have been verified experimentally. Practical lines 50 meters and 15 meters long have been constructed which trans- form from 600 to 300 ohms over the frequency range from 4 to 30 mc. with deviations from the ideal that are small compared with the deviations from the ideal of commercial transmission lines, either two-wire or concentric. When an exponential line is used as a dissipative load of known impedance instead of a uniform line it is possible to approach more nearly the ideal of constant heat dissipation per unit length. This makes it possible to use a shorter line. THE exponential line may be defined as an ordinary transmission line in which the spacing between the conductors (or conductor size) is not constant but varies in such a way that the distributed inductance and capacitance vary exponentially with the distance along the line. That is, the impedance ratio for two points a fixed distance apart is independent of the position of these two points along the line. A disturbance is propagated down an exponential transmission line in the same manner as it would be down a uniform line with the addi- tional effect that the voltage is increased by the square root of the change in impedance level and the current is decreased by the reciprocal of this quantity. The exponential line has the properties of a high pass impedance transforming filter. The cutoff frequency depends upon the rate of * Presented before joint meeting of U. R. S. I., and I. R. E., Washington, D. C, April 1938. Published in Communications , October 1938. 555 556 BELL SYSTEM TECHNICAL JOURNAL taper. As the frequency is increased the transfer constant * ap- proaches the propagation constant of the equivalent uniform Hne. At sufficiently low frequencies the only effect of the line is to connect the input to the load. Above cutoff the magnitudes of the characteristic impedances at any point are approximately equal to the nominal characteristic imped- ance * at that point but their phase angles (in radians) differ by an amount which at the higher frequencies is equal to the cutoff frequency divided by the frequency in question. The ratio of input impedance to the input impedance level * of an exponential line terminated in a resistance equal to the impedance level at the output always remains within the range from 1 — fi/fto 1/(1 — fi/f) for frequencies,/, greater than the cutoff frequency, /i. For a 2 : 1 transformation this means that the input impedance remains within ± 6 per cent of the desired value for all frequencies above that for which the line is a wave-length long. For a 4 : 1 transformation under the same conditions the irregularities are twice as great. A transforming network having deviations from the ideal of the order of ± (fi/fY may be made by connecting an inductance in parallel with the low impedance terminal and a capacitance in series with the high impedance terminal. The magnitudes of these reactances are such that their impedances are equal to the impedance levels of the line at their respective ends at the cutoff frequency. Or expressed in another way the capacitance is equal to 2/(k — 1) times the electrostatic capacitance of the line and the inductance is the same factor times the total loop inductance of the line where k is the impedance transforma- tion ratio of the line. Figure 1 shows the theoretical input impedance-frequency charac- teristics for 2 to 1 step-up and step-down exponential lines. Curve 1 is for the line with a resistance termination. At low frequencies the input impedance is equal to the load impedance while at high fre- quencies the line approaches an ideal transformer. Curve 2 is the input impedance of the line terminated with the appropriate resistance- reactance combination. The improvement in the input impedance characteristic for frequencies above the cutoff frequency is evident. At the lower frequencies the input impedance does not approach the terminal reactance but approaches the reactance of the capacitance of the line in parallel with the series terminal capacitance for the step-up line and the reactance of the inductance of the line in series with the shunt terminal inductance for the step-down line. The improvement is not as great as apparent from the figures because the phase angle is * See appendix for definition of terms. EXPONENTIAL TRANSMISSION LINE 557 not improved proportionally. This is easily remedied by completing the impedance transforming network with the appropriate reactance at the input. The resulting input impedance is shown in curve 3. In the "pass" frequency range the maximum reactive component is of 0.2 as 1.0 RELATIVE FREQUENCY f/f 2.0 5.0 • 10 20 50 N. ' -T T -T-r T 1 ' 1 1 111 ' 1 1 ' ' ' • • ' ' ' ~x — 1 III II i. X CUT OFF frequ:ncy VX f,= 0.0551 fn 5.0 \ s. 1 Kline IS l/8 A LONG ''-LINE IS ONE WAVELENGTH LONG AT THIS FREQUENCY ^ N AT THIS FREQUENCY ^ ^v S s \ ^ V "V, -- g -TERMINAL IMPEDANCE CASE 2 8. CASE 3 - 1.0 ■^ } s ^ . Verm INAL IMP EDA N( :e c ,A< >E 1 - / v. — ^ ' y S r \ / \ // N // / >s . /^ 1 ■ o.a / : 0.1 0.5 ti 2.0 i I 3.0 : 001 0-05 0.1 0-5 1.0 4" RELATIVE FREQUENCY f/fQ Fig. 1 — Input impedance characteristics of 1 : 2 exponential lines. Left ordinate scale refers to step-up line. Right ordinate scale refers to step-down line. Curve 1— Resistance termination. Curve 2 — With capacity equal to twice the electrostatic capacity of the line in series with the same resistance, Zj = Zi.{\ — jfilf), for step-up line, or with an in- ductance equal to twice the total inductance of the line in shunt with the same resistance, Zi = Zi/(1 — jfi/f), for step-down line. Curve 3 — Termination as for curve 2 with inductance equal to twice the total induc- tance of the line in parallel with input to the line, Zn = Zi/{1 — jfi/f), for step-up line, or termination as for curve 2 with capacity equal to twice the static capacity of line in series with input to the line, Za = Zi{\ — jfi/f). Curve 4 — Asymptotic value of impedance of capacity of line in parallel with termina- tion for case 2 for step-up line, or asymptotic value of impedance of inductance in series with termination for case 2 for step-down line. Curve 5 — Impedance of shunt inductance added at input for case 3 for step-up line, or impedance of capacity added in series at input for case 3 for step-down line. the same order of magnitude as the deviation of the impedance from the ideal. Besides its application as an impedance transforming network, the exponential line may be used as a "resistance" load of constant known impedance that has a high capability for dissipating power. As such it is capable of dissipating more power in the same length of line than 558 BELL SYSTEM TECHNICAL JOURNAL the uniform line. If x is the maximum attenuation in nepers that can be obtained with a uniform line without overheating, the same length of exponential line will have an attenuation of (e^^ — l)/2 nepers. Exponential lines of the proper length have properties similar to half-wave and quarter-wave uniform lines. The input impedance of an exponential line an even number of quarter wave-lengths long is equal to the load impedance times the impedance transformation ratio of the line. When the length of the line differs from an odd multiple of a quarter wave-length by an amount that depends upon the fre- quency and load impedance, the input impedance is equal to the product of the terminal impedance levels divided by the load impedance. Mathematical Formulation The telegraph equations for the exponential line may be solved by the methods employed in the problem of a uniform line. The resulting equations for the voltage and current at any point along the line are and tx Vx = Ae 5] -(-|)^a.R+(^+l>-..-(^-|) + Be 1 + I e^^^ -4^ 2-(r+|> ^^ + 2;-i-(r-|> Zo 7 Za 7 (1) where Zq 7 1 - B^ + 2 r - (2) ^ ^ log^o ^ log^ ^ log^o .^ ^^^ ^^^^ ^f ^^p^^^ Zj; = Vz/y — Z^e^"" is the surge or nominal characteristic impedance of the exponential line at the point x which is equal to the characteristic impedance of the uniform line that has the same distributed constants as this line has at the point x, Y = "^zy = Vzo3'o is the propagation constant of any uniform line that has the same distributed constants as this line at any point. It is independent of the point along the line to which it is referred, and r = V7^ + 5-/4 = a + j(8 is the transfer constant of the exponential line. EXPONENTIAL TRANSMISSION LINE 559 + 7 and + r refer to the values of the indicated roots that are in the first quadrant. If these equations are compared with those for a uniform trans- mission Hne it is found that the propagation constant is F — 5/2 for voltage waves traveling in the positive x direction and F + 5/2 for voltage waves traveling in the negative x direction. For current waves the corresponding propagation constants are F + 5/2 and F — 5/2. In the terminology of wave filters, F is the transfer constant and 5 is the impedance transformation constant. 5/2 is the voltage transformation constant and — 5/2 is the current transformation constant. The real and imaginary parts of F, a and /3 are the attenuation and phase constants respectively. An important parameter is . 5 which for a non-dissipative line is the ratio of the cutoff frequency to the frequency, as can be seen if we write the transfer constant as r = 7 Vi - v"^, where the indicated root is in the fourth quadrant. For a non- dissipative line V is real and the transfer constant is real or imaginary depending on whether v^ is greater than or less than unity. Hence the exponential line is a high pass filter whose cutoff frequency, /i, is that frequency for which j^ = ± 1. The transfer constant is then less than that for a uniform line by the factor Vl — i'^ so that both phase velocity and wave-length are larger for the exponential line than for the uniform line by the reciprocal of this factor. If we terminate this line dit x — I with an impedance Zi = vi/ii, the ratio of the reflected to direct voltage wave is found to be A 1 + {z,izdNi - p' -jv) ' where the coefficient of the exponential is the voltage reflection coefficient. There will be no reflection if Zi = ZiK^TT^^- + jp) = z,+, (4) which becomes Zie~'^^^~^ " above the cutoff frequency for non- dissipative lines. This is the magnitude of the forward-looking characteristic impedance at x = I as can be seen by dividing the first term of (1) by the first term of (2). Curve 1 of Fig. 2 gives the charac- 560 BELL SYSTEM TECHNICAL JOURNAL teristic impedance of a non-dissipative exponential line looking toward the high impedance end as a function of frequency. At infinite fre- quency the characteristic impedance is a resistance equal to the nominal characteristic impedance but as the frequency is decreased the phase angle of the characteristic impedance changes so that its locus +J V- ® 0.9 -\ ® ■^ ^ 0.8 / ^N »-i^ 0^-° 0.7 0.5 f T > 0.2 0 ( J© Sd ^^L 2.0/— 1.3 -J 1 ____-- % -^ ^ i ^^ 0.9 1 V^i iV^ Fig. 2 — Impedance diagram comparing the forward looking characteristic imped- ance with various terminal impedances. The numbers give the frequency relative to cutoff. The arrows are the vectors Zj — Zj"*" which are a measure of the magnitude of the reflection. A. Step-up line. Curve 1 — Forward looking characteristic impedance, Zi+ = Zie-^^'''~'^^^'f\ f>fi, Zi^ = Zil- i(/i//)(l 4- y!\-Plim, /i > /; Curve 2 — Resistance termination, Zi = Zi; Curve 3 — Capacity resistance termination, Zi = Zi{\ — jf\lf); Curve 4 — Capacity, resistance and inductance termination adjusted for no reflection at twice the cutoff frequency and at infinite frequency; B. Step-down line. Curve 5 — Forward-looking characteristic impedance, Zi+ = Z;[+i(/>//)(l - Vl -/V/i^)], /i > /; Curve 6 — Resistance termination Z/ = Zj; Curve 7 — Inductance resistance termination Z; = Z;/(l — jfijf); Curve 8 — Inductance, resistance and capacity termination adjusted for no reflection at twice the cutoff frequency and at infinite frequency. EXPONENTIAL TRANSMISSION LINE 561 is the circular arc. At and below cutoff it is a pure reactance. If the load is a resistance equal to the nominal characteristic impedance at the terminal as indicated at 2 of Fig. 2, there will be no reflection at infinite frequency, but as the frequency is lowered there will be an in- creasing impedance mismatch with its accompanying reflected wave. This reflection may be materially reduced by inserting a condenser in series with the resistance load as shown by curve 3. Further im- provement results from more complicated networks. Curve 4 shows the effect of adding an inductance in shunt with the resistance load of the resistance-capacitance combination. The arrows indicate the resulting impedance mismatch which is a measure of the reflected wave. The characteristic impedance looking toward the low impedance end is the inverse of that looking in the other direction as shown by curve 5. Shunting the resistance load with an mductance gives the impedance curve 7. Adding a capacitance element gives curve 8. Division of (1) by (2) and substitution of the result of (3) gives the following ratio for the impedance looking into the line at the point x to the impedance level at that point, Z^ _ K{^\ - y-" -jv) -f 1 -f- \K{^\ - t'^ -f-jV) - l]g-^r(;-.) Z- K + jv + Vl - v' - [K - Vr^T^ + jV>-2r('--) where K = ZijZi is the ratio of the load impedance to the impedance level at the terminal. Here as before the indicated root is in the fourth quadrant. Network Characteristics Three parameters are required to specify the characteristics of an exponential line of negligible loss: (1) the cutoff frequency, /i, (2) the length of the line which is perhaps best specified as the frequency, /o = velocity of light/length of line, for which the line is one wave- length long, and (3) the impedance level at some point along the line. We will designate the impedance levels at the low and high impedance ends of the line by Zi and Zi respectively, and their ratio Z2/Z1 by k. When the line is terminated in a resistance equal to the impedance level at the output (5) reduces to Zi , „. o , 1 + i tan f k-"""^ 2r Zi 1 - 7 tan f k"''^ -f ' ^ ^ ^1 ,_„, 1 +tan^/^^ + i-'') ^' H-tan^/-'^-2-'') 562 BELL SYSTEM TECHNICAL JOURNAL for frequencies below and above cutoff respectively. Here r? = — j2Yl is twice the electrical length of the line in radians, sin 2f = Ijv, sin 2^ = V and cos 2^ is ratio of the electrical length of the line to that of a uniform line of the same physical length. For the step-down line the corresponding ratios are the reciprocal of the above expressions. These ratios are plotted in Fig. 1. When /— ^ 0, £i = kZi = Zi — Zi and the only effect of the line is to connect the load to the input. Above cutoff the magnitude of the input impedance oscillates about the nominal characteristic impedance and the phase angle oscillates about the value — 2^(« — /i// for />J> /i) which goes from — 7r/2 to 0 as the frequency increases indefi- 0.1 0.5 1.0 FREQUENCY IN MEGACYCLES Fig. 3 — Input impedance characteristics. Curve 1 — 150 : 600 ohm line, 100 meters long. Curve 2—300 : 600 ohm line, 200 meter? long. Both lines have the same rate of taper. nitely from cutoff. The variation of the input impedance with fre- quency is shown for two lines of different length but the same rate of taper in Fig. 3. The magnitude of the oscillations depends only on the rate of taper and decreases with increase in frequency. The impedance varies between (1 + /i//) and 1/(1 + /i//). The positions of the maxima and minima, however, are determined by the length of the line. They occur respectively at those frequencies for which the line is approximately 1/8 of a wave-length more than an even or an odd number of quarter wave-lengths long. The phase angle is usually negative but has a small positive value when the line is approximately a half wave-length long. The locations of these maxima and minima are the same as would result from terminating a uniform line in an impedance whose magni- tude is the same as the characteristic impedance but has a small reactive component. This suggests adding a compensating reactance EXPONENTIAL TRANSMISSION LINE 563 to the resistance load. From (3) the best single reactive element is found to be a condenser whose impedance is equal to the impedance level at the cutoff frequency. This gives a value oi K — 1 — jv which when substituted in (5) shows that the input impedance is to a first approximation a constant times the terminal impedance. To correct for the reactive component of the input impedance an inductance having an impedance jZi/v which is equal to the input impedance level at cutoff is shunted across the input. The resulting impedance trans- forming network consists of an exponential line with a series capaci- tance at the high impedance end and a shunt inductance at the low impedance end. When terminated in a resistance load at either end equal to the impedance level at that end the input impedance, to a first approximation, is a resistance equal to the impedance level at the input end. In fact the deviations of the input impedance from the ideal for transmission in one direction are just the reciprocal of those for transmission in the other direction. The magnitudes of the series capacitance and shunt inductance that give the improved network may be expressed in terms of the electro- static capacitance and loop inductance of the line. Simple calculation shows that the required series capacitance is equal to 2/(^ — 1) times the electrostatic capacitance of the line and the required shunt in- ductance is equal to the same factor times the total inductance of the line. There is an interesting relationship between these terminations and a simple high-pass filter. The LC product of the shunt and series arms of the filter resonates at /i. If an ideal transformer with transforma- tion ratio k is inserted between the shunt inductance and the series capacitance, the capacitance becomes Cjk and the new LC resonates at /iV^. This is the same frequency at which the series capacitance and shunt inductance that are added to the terminations of the ex- ponential line resonate. Furthermore the reactance of the shunt inductance is equal to the impedance level at the cutoff frequency and the reactance of the series capacitance is equal to the impedance level at the cutoff frequency exactly as in the case of the high-pass filter. By using the exponential line it is possible to construct a network with properties that no network with lumped circuit elements possesses, namely, a high-pass impedance transforming filter. Critical Lengths Besides the characteristics of the exponential line that are sub- stantially independent of the length of the line, it has properties that 564 BELL SYSTEM TECHNICAL JOURNAL depend on the length of the line that are analogous to those of a uniform line a half wave-length or quarter wave-length long. For non- dissipative lines above the cutoff frequency (5) becomes i^cos(|-2^)-fisin^ Z, jY T-^ Z. (8) cos(^^-H2M+jXsin| When the line is an integral number of half wave-lengths long (27? = tt) this reduces to £i = KZ, = kZ2, (9) which says that the input impedance is equal to the impedance trans- formation ratio times the load impedance. The length of exponential line that corresponds to a quarter-wave uniform line differs from an odd multiple of a quarter wave-length by an amount such that / rj - (2« + 1)t\ K^ - I tan i ^ 2 ) " K'+ 1 ^' ^ ^^ for which (8) becomes Z,=^^ (11) Similar expressions exist for the step-down line, but l/iC must be substituted for K in (10) for the length corresponding to the quarter- wave uniform line. With Dissipation ^ An exponential line is an improvement over the uniform "iron wire" line as a resistance load that will dissipate a large amount of power. Provided the attenuation is not too large the current and voltage distribution will be the same as for a non-dissipative line except for the additional power loss so that we may use the equations for an exponential line even though the distributed series resistance and shunt leakage do not vary exponentially with distance. Suppose that the conductor size and resistance that will just dissi- pate the desired input power result in an attenuation constant ao for a uniform transmission line. To a first approximation the conductors can carry the same current irrespective of the impedance level. The current wave will be given by the first term of equation (2) which becomes EXPONENTIAL TRANSMISSION LINE 565 except for a phase factor. In order that the current will not increase, 5 = — 2a. The actual attenuation "constant," will increase with distance down the line so that the current will decrease but not as rapidly as with a uniform line. The total attenua- tion in nepers is approximately 1 /'\ r' o„., /i , 1 P\/e'"oi - 1 1 +-,U a^'-'-dx = 1 +-i- 1 ^— — ^ . (14) At the point where the attenuation of the uniform line is 6 db the tapered line has an additional attenuation of 7 db above the uniform Une or a total attenuation of more than twice. The current has been reduced to less than half. Here an improvement may be made by increasing the dissipation by either changing the wire size or resistivity of the conductor. A greater improvement would result from changing the resistivity because then the capacity for heat dissipation would be the same. Suppose, however, that one conductor material is to be used throughout and the dissipation capacity is proportional to the wire surface; then at this point the wire size could be reduced to 1/2, doubling the attenuation factor. It is already 4 times that for the uniform line, so this increases it to 8 times. The resulting total attenua- tion is 30 db in a length that would have less than 7 db if the line were uniform. If this attenuation were required the length of line could be reduced by a factor of about 4.4. Of course the spacing is very close at the end of this line, but the line could be shorted at the end. This would approximately double the current at the end, but here again the current carrying capacity of the line is more than double the current traveling down the line. With the line shorted the reflected current would be 60 db down, which would not affect the input impedance appreciably. For the first 13 db of attenuation the impedance of the line would be relatively free from changes due to changes in spacing resulting from wind, etc. When the spacing is small enough to be affected by wind, vibration, etc., the attenuation will be great enough to suppress these small irregularities. Experiment In order to verify the foregoing theoretical development, measure- ments have been made on several experimental lines. Figure 4 shows the results of measurements on two such lines. These lines were 566 BELL SYSTEM TECHNICAL JOURNAL constructed of No. 12 tinned copper. ' At the^low impedance end the strain was taken by a victron insulator which also served as a line spreader and terminal mounting. At the high impedance end the strain was taken by 1/4" manila rope without other insulation. The line spacing was adjusted by "lock stitch" tension insulators spaced 1 meter and 1/2 meter apart on the low and high impedance end re- spectively of the 9-meter line. The 3-meter line was supported at the 1/4, 1/2, 2/3, 3/4 and 7/8th points. The impedance was measured by the substitution method. To facilitate the substitution of the reactive component of the line it was bridged by an antiresonant circuit. Pencil leads calibrated on direct current] were used as the resistance standards. Type BW IRC 1/2 ^ 1 >* * * • • i • y * — , • ' -"^ 0^ n° A *• ■A 0.3 f/fo 0.8 Fig. 4 — Input impedance characteristic. Comparison of theoretical curve with experimental points for 600 : 300 ohm lines. Solid circles — 9-meter line. Open circles — 3 -meter line. watt resistances were used for terminations. The solid circles of Fig. 4 represent measurements on the 9-meter line. The agreement with theory is as good as is usually found for actual "uniform lines." In order to check the theory further toward the lower frequency end — beyond the range of the measuring equipment — -measurements were made on a 3-meter line. These measurements are shown by the open circles. The agreement with theory is not so good, but here the lengths of the connecting leads are an appreciable fraction of the length of the line. Preliminary tests on a full size model of exponential line impedance transformer showed deviations from the theoretical that might be at- tributed to improper termination, irregularities along the line, irregu- larities introduced at the change in conductor size or capacitance of the spacing insulators. Since it was impossible to determine which of these was the predominant cause of the deviations from the ideal, it EXPONENTIAL TRANSMISSION LINE 567 was decided to introduce each of these factors one at a time. This test was made on a 600 : 300 ohm Hne constructed of No. 6 copper wire with lockstitch insulators except at the terminals. The correct termination was obtained by tests on a uniform 300 ohm line with the same physical structure at the termination. Of necessity the tying of the wire to the strain insulators at the end introduced a shunt capacitance which augmented the inherent additional capacitance due the "end effect." This additional capacitance is equal to that of a short length of line. Fig. 5 — Photographs of terminations of 300 ohm line, upper right for curve 1 1 lower for curve 2 ^ of Fig. 6. upper left for curve 3 J If the correct amount of inductance is inserted in series with the resistance load the combined effect of the additional capacitance and inductance becomes the same as the addition of a small length of line for all frequencies up to those for which this length of line is an appre- ciable fraction of a wave-length. Accordingly a small amount of induc- tance was inserted in series with the resistance as shown in the right picture of Fig. 5. The input impedance of the uniform line with this termination is given by "Experimental Curve 1 " at the bottom of Fig. 6. 568 BELL SYSTEM TECHNICAL JOURNAL A three-inch length of No. 18 wire was inserted as shown in the lower picture of Fig. 5 and " Experimental Curve 2 " resulted. This reduced the irregularities in the input impedance to about half, so another three 264 10 12 14 16 18 20 22 24 26 28 30 FREQUENCY IN MEGACYCLES Fig. 6 — Lower. Experimental input impedance characteristics of 300 ohm line with terminations shown in Fig. 5. Upper. Input impedance characteristics of 50-meter 600 : 300 ohm line of No. 6 conductors. inches were inserted, resulting in "Experimental Curve 3." Here the maxima and minima are displaced, indicating that the effect of the stray capacitance has been reduced to the same order of magnitude as that due to the deviation of the resistance from the desired value. This EXPONENTIAL TRANSMISSION LINE 569 termination was accordingly removed to the exponential line, resulting in the "Experimental Curve" at the top of Fig. 6. It agrees within experimental error with the "Theoretical Curve." The slight vertical displacement of the experimental curve at the higher frequencies is attributed to deviations in the impedance of the pencil lead, which was Fig. 7 — Photograph of one of the changes in conductor size. used as a resistance standard, from a pure resistance equal to its direct current value. To increase the power carrying capacity of the exponential line, one was built with larger wire size at the lower impedance end. This increased the breakdown voltage by increasing the spacing and con- ductor diameter and at the same time increased the current carrying capacity by decreasing the resistance and increasing the heat dissipat- 570 BELL SYSTEM TECHNICAL JOURNAL ing capacity of the conductors. This was a 600 : 300 ohm Hne con- structed of 20 meters No. 6 wire, 10 meters 1/4" tubing and 20 meters 3/8" tubing. Here again the correct termination was determined by measurements on a 300 ohm uniform line of 3/8" tubing. The total length of terminating loop that gave the best termination was 6)^" in this case compared with lOj^" for the 300 ohm line of No. 6 wire. Since no attempt was made to reduce the variations in input impedance to less than ± 1 per cent these lengths may be as much as an inch off. These measurements indicated that the exponential line would per- form satisfactorily as an impedance transformer if it could be con- structed to have the desired mechanical features without impairing its electrical properties. The greatest difficulty appeared to reside in the 1.06 FREQUENCY IN MEGACYCLES P'ig. 8 — Input impedance characteristics of 50-meter 600 : 300 ohm line of 3/8", 1/4" and No. 6 conductors. insulators. Special isolantite insulators were designed that would be satisfactory commercially and still keep the additional capacity to a reasonable value. Figure 7 shows the construction of the line at the supporting poles where the conductor size changes. The results of measurements on this line are shown in Fig. 8. The solid curve was calculated from the equations developed earlier. The two broken curves are the results of measurements on the line, one without insulators and one with insulators. While the insulators affect the line somewhat they do not increase the deviation from the ideal appreciably. [The improvement in the agreement between experiment and theory in this set of curves over that in Fig. 4 is presumably due to the fact that the comparison resistance for Fig. 8 consisted of 3-IRC EXPONENTIAL TRANSMISSION LINE 571 resistances instead of the pencil lead. With the fixed IRC resistance it was, of course, impossible to adjust the standard to exactly the same value as the unknown. In this case the small difference was determined by using the slope of the rectifier voltmeter calibration.] This line has a maximum deviation from the desired input impedance of ± 6 percent for all frequencies above 4.2 mc. (Measurements were made up to 28 mc.) The phase angle of the input impedance was found to be zero 10 12 14 16 18 20 22 24 26 28 FREQUENCY IN MEGACYCLES Fig. 9 — Input impedance characteristics of 15-meter 600 : 300 ohm line of No. 6 conductors within the accuracy of measurement. From theory the phase angle would be expected to vary between — 0° and + 3°. The curves of Fig. 9 refer to a 600 : 300 ohm line of No. 6 wire 15 meters long. With resistance termination this line has rather large variations in the input impedance but with the addition of the proper reactances the input impedance is flatter than the longer line with resistance termination. At the lower frequencies where the variations in the input impedance were large without the reactive networks, their addition gives approximately the expected improvement. At the higher frequencies the inductance was approximately anti-resonated 572 BELL SYSTEM TECHNICAL JOURNAL by its distributed capacity and the input impedance approaches that for the resistance termination. Conclusion Theory indicates that the exponential Hne may be used as an imped- ance transformer over a wide frequency range. The results of experi- ment show that the desired characteristic'can be realized in practice. Among the applications of the exponential line may be mentioned its use in transforming the impedance level back to its original value after the paralleling of two transmission lines feeding two antennas. It could be used to transform the input impedance of a rhombic antenna down to the usual 600-ohm level of open wire transmission lines. If twin coaxial lines are used inside the transmitter building to eliminate undesired feedback, coupling, etc., the exponential line could be used to transform from the highest practical impedance level of such lines to a practical level of the more economical open wire lines for use outside the building. Appendix The exponential line is a non-uniform line so that the terms "charac- teristic impedance" and "surge impedance" of an exponential line are not synonymous. The terms "surge impedance" ' and "nominal characteristic impedance" ^ may be used synonymously for the charac- teristic impedance of the uniform line that has the same distributed constants as the non-uniform line at the point in question. Expressed 'as functions of the distributed "constants" of the line they are the square root of the ratio of the distributed series impedance to the distributed shunt admittance at the point along the line in question. It will be expedient to refer to the nominal characteristic impedance as the impedance level at the point in question. Schelkunoff ^ has defined the characteristic impedances as the ratio of voltage to current at the point in question for each of the two traveling waves of which iThe term "surge impedance" is defined by A. E. Kennelly on page 73 of "The Applications of Hyperbolic Functions to Electrical Engineering Problems " (McGraw- Hill 1916) as follows: "The surge impedance of the line is not only the natural imped- ance which it offers everywhere to surges of the frequency considered, but it is also the initial impedance of the line at the sending end." Hence the "surge impedance" should be independent of the configuration of the line except at the point in question and in particular it should be equal to that for a uniform line constructed so as to have the same dimensions everywhere as the non-uniform line has at the point in question. 2 The word nominal as used here has the same meaning as in "nominal iterative impedance" as used by K. S. Johnson in "transmission circuits for telephone com- munication" (Van Nostrand 1925). , 3 S. A. Schelkunoff, "The Impedance Concept and its Application to Problems of Reflection, Refraction, Shielding and Power Absorption," Bell System Technical Journal, 17, 17-48, January, 1938. EXPONENTIAL TRANSMISSION LINE 573 the steady state condition is composed. At each point an exponential Hne has two characteristic impedances which are different and depend upon the frequency as well as the position along the line. Because of the change of impedance level, the propagation constants for the voltage and current differ, so that it is convenient to consider the transfer constant ^ which may be defined as half the sum of the voltage and current propagation constants. * Compare with the definition of "image transfer constant" as given by K. S. Johnson in "Transmission Circuits for Telephone Communication." The Bridge Stabilized Oscillator* By L. A. MEACHAM A new type of constant frequency oscillator of very high stability is presented. The frequency controlling resonant element is used as one arm of a Wheatstone resistance bridge. Kept in balance automatically by a thermally controlled arm, this bridge provides constancy of output ampli- tude, purity of wave form, and stabilization against fluctuations in power supply or changes in circuit elements. A simple one-tube circuit has operated consistently with no short-time frequency variations greater than ± 2 parts in 10^. Convenient means are provided for making precision adjustments over a narrow range of frequencies to compensate for long-time aging effects. Description of the circuit is followed by a brief linear analysis and an account of experimental results. Operating records are given for a 100 kc. oscillator. Introduction THE problem of improving the stability of constant frequency oscillators may be divided conveniently into two parts, one relating to the frequency controlling resonant element or circuit, and the other to the means for supplying energy to sustain oscillations. The ideal control element would be a high-(3 electrical resonant circuit, or a mechanical resonator such as a tuning fork or crystal, whose properties were exactly constant, unaffected by atmospheric conditions, jar, amplitude of oscillation, age, or any other possible parameter. The ideal driving circuit would take full advantage of the resonator's constancy by causing it to oscillate at a stable amplitude and at a frequency determined completely by the resonator itself, regardless of power supply variations, aging of vacuum tubes or other circuit ele- ments, or the changing of any other operating condition. This paper, concerning itself principally with the second part of the problem, describes an oscillator circuit which attains a very close approximation to the latter objective. The "Bridge Stabilized Oscil- lator" provides both frequency and amplitude stabilization, and as it operates with no tube overloading, it has the added merit of delivering a very pure sinusoidal output. Oscillator Circuit The bridge stabilized oscillator circuit, shown schematically in Fig. 1, consists of an amplifier and a Wheatstone bridge. The amplifier out- * Presented at Thirteenth Annual Convention of Institute of Radio Engineers, New York City, June 16, 1938. Published in Proc. I. R. E., October 1938. 574 THE BRIDGE STABILIZED OSCILLATOR 575 put is impressed across one of the diagonals of the bridge, and the unbalance potential, appearing across the conjugate diagonal, is applied to the amplifier input terminals. One of the four bridge arms, Ri, is a thermally controlled resistance; two others, Ri and Rz, are fixed re- sistances, and the fourth, Zi = Ri -\- jXi, is the frequency-controlling resonant element. In this discussion Zi is assumed to represent a crystal suitable for operation at its low-impedance or series resonance. A coil and con- denser in series could be substituted, and even a parallel-resonant control element might be used by exchanging its position in the bridge VOLTAGE ATTENUATION VOLTAGE AMPLIFICATION Fig. 1 — Schematic circuit diagram of bridge stabilized oscillator. with i?2 or R3. Operating a crystal at series resonance has the advan- tage of minimizing effects of stray capacitance. The bridge is kept as nearly in exact balance as possible. Assuming that Ri, Ri and Rz are pure resistances, we may write for exact reactive balance, Xi = 0, and for exact resistive balance, -^1 _ -^3 R2 Ri In order that the circuit may oscillate, a slight unbalance is required. Accordingly i?i must be given a value slightly smaller than {RzRzj/Ri, 576 BELL SYSTEM TECHNICAL JOURNAL so that the attenuation through the bridge is just equal to the gain of the ampUfier. It is evident that if all the bridge arms had fixed values of resistance, the attenuation of the bridge would be very critical with slight changes in any arm. This would obviously be undesirable, for the circuit would either fail to oscillate, or else build up in amplitude until tube overloading occurred. The thermally controlled resistance Ri elimi- nates this difficulty. This arm has a large positive temperature coeffi- cient of resistance, and is so designed that the portion of the amplifi.er output which reaches it in the bridge circuit is great enough to raise its temperature and increase its resistance materially. A small tungsten- filament lamp of low wattage rating has been found suitable. It functions as follows: When battery is first applied to the amplifier, the lamp Ri is cold and its resistance is considerably smaller than the balance value. Thus the attenuation of the bridge is relatively small, and oscillation builds up rapidly. As the lamp filament warms, its resistance approaches the value for which the loss through the bridge equals the gain of the amplifier. If for some reason Ri acquires too large a resistance, the unbalance potential e becomes too small or possibly even inverted in phase, so that the amplitude decreases until the proper equilibrium is reached. This automatic adjustment stabilizes the amplitude, for the amount of power needed to give Ri a value closely approaching {RiRzjjRi is always very nearly the same. A change in the amplifier gain would cause a readjustment of the bridge balance, but the resulting variation in R\ or in the amplifier output would be extremely small. The operating temperature of the lamp filament is made high enough so that variations in the ambient temperature do not affect the adjust- ment appreciably. No overloading occurs in the amplifier, which operates on a strictly Class A basis, nor is any non-linearity necessary in the system other than the thermal non-linearity of Ri. As the lamp resistance does not vary appreciably during a high-frequency cycle, it is not a source of harmonics (or of their intermodulation, which Llewellyn ^ has shown to be one of the factors contributing to frequency instability). In contrast to the lamp, an ordinary non-linear resistance, of copper oxide for example, would not be suitable for Ri. A resistance of the thermally-controlled type having a negative temperature coefficient 1" Constant Frequency Oscillators," F. B. Llewellyn, Proc. I. R. E., December 1931. THE BRIDGE STABILIZED OSCILLATOR 577 could be used by merely exchanging its position in the bridge with i?2 or Rz. The frequency control exerted by the crystal depends upon the fact that the phase shift of the amplifier must be equal and opposite to that through the bridge. In the notation of Black,^ applied to the circuit of Fig. 1, E , ,,„ ^=7 = IMI The condition for oscillation is M/^ = 1 l_0, which implies that | /i|3 1 = 1 and d = — \]/. The vector diagrams of Fig. 2 illustrate the frequency-stabilizing action of the bridge by showing the voltage relations therein for two values of amplifier phase shift, 6. When d is zero, as in diagram A, the unbalance vector e is in phase with the generated voltage E applied to the bridge input, and thus all the vectors shown are parallel. They are displaced vertically from each other merely to clarify the drawing. The crystal is here constrained to operate at exact resonance. In diagram B, the amplifier is assumed to have changed its phase for some reason by an amount far in excess of what would be anticipated in practice, 6 hene having a value of + 45 degrees. The important point to be observed is that the ratio of d to the resulting change in the phase angle <^ of the crystal impedance Z4 is very large. That is, the crystal is still operating close to resonance in spite of the exaggerated change in the driving circuit. If the gain of the amplifier were greater, the action of the thermally controlled resistance would keep the amplifier output vector E practically the same in length, making the unbalance vector e correspondingly shorter. The angle 4> would therefore have to be more acute for the same value of 6, and it follows that with increased gain the crystal is held closer to true resonance and the stability is improved. When 6 equals zero, changes in | /u | do not affect the crystal operating phase, but for any other small value of d, gain variations cause slight readjustments of the angles between vectors. The amplifier should accordingly be designed for zero phase shift, and also, of course, should have as much phase stability as possible. ^ "Stabilized Feedback Amplifiers," H. S. Black, Bell System Technical Journal, January 1934. 578 BELL SYSTEM TECHNICAL JOURNAL In this discussion the input and output impedances of the amplifier, i?5 and Rs, are assumed to be constant pure resistances. Actually, changes in the tube parameters or in certain circuit elements are likely to affect both the magnitude and the phase of these impedances. It may be shown, however, that such changes have the same effect upon the bridge and upon the frequency as do changes of about the same I4Z4 ^ I2R2 I3R3 , I1R1 ^ |J-1.^5_1_ e = o \ uz. \ A- -f^ ^ E' 11 LOCUS OF TAIL OF VECTOR e FOR VARYING FREQUENCY •F — O Fig. 2 — ^Vector diagrams illustrating operation of bridge oscillator, with simplify- ing assumptions that R5 is large and that E and E' are strictly in phase. A — At resonance Z4 = Ri+jO d = 0 i?l < i?2 = i?3 = R4 B — Above resonance Z4 = i?4 + jXi X4 Inductive 0 = + 45° i?l < i?2 = i?3 = -R4« R& percentage in [mI or 0; therefore all variations in the driving circuit external to the bridge may be assumed for convenience to be repre- sented by variations in its gain and phase. This leniency with regard to R5 and Re does not apply to the other bridge resistances, however. Ri, R2 and R3 are directly responsible for the crystal's operating phase and amplitude; they should be made as stable and as free from stray reactance as possible. THE BRIDGE STABILIZED OSCILLATOR 579 The effect of the bridge upon harmonics of the oscillation frequency is of interest. Harmonics, being far from the resonant frequency of the crystal, are passed through the bridge with little attenuation but with a phase reversal approximating 180 degrees, as illustrated by the dotted locus in Fig. 2. Thus if the amplifier were designed to cover a band broad enough to include one or more harmonics and if care were taken to avoid singing at some unwanted frequency, a considerable amount of negative feedback could be applied to the suppression of the har- monics in question. Circuit Analysis In the following section, expressions are derived for the frequency of oscillation in terms of the gain and phase shift of the amplifier, the Q of the crystal, and values of the bridge resistances. It is assumed that the latter are constant and non-reactive, and therefore, as ex- plained previously, that all sources of frequency fluctuations apart from changes in the crystal itself appear as variations in | ;ti | or ^. Because the bridge oscillator does not rely upon non-linearity in the ordinary sense to limit its amplitude, the analysis can be based reasonably on simple linear theory. In the near vicinity of series resonance the crystal may be repre- sented accurately by a resistance Ra, inductance L and capacitance C, connected in series. The reactive component of the crystal's im- pedance is accordingly Solving for the frequency, 'LC - 1 (1) 2L^ 1 IL ^ LC \X, LC[ 2 C ^Lc[ 1 + L X, + \1 + m- C 1/X_4 L~^2\ 2 C L C\2 _1 1/X_4 jC 2' 4\ 2 \L + (2) Near series resonance, (X4/2) ^{CIL) < < 1. We therefore disregard powers higher than the first in the series expansion above and obtain the close approximation. 580 BELL SYSTEM TECHNICAL JOURNAL The frequency deviation from resonance, expressed as a fraction of the resonant frequency /o, is thus / — /o CO — COo , A'4 \C /o Wo (4) and in the region of interest, where coL and 1/wC are approximately equal, f — fo . Xi -^4 /o 2coL 2QRi (5) Considering now the bridge circuit, and applying well-known equa- tions,^ we obtain hRr, _ AR, - jBX, E ~ MRi + jNXi ' (6) ^hich and A - R.iR^Rz - RiRa), B = RiRiRc,, M = {R, -f R2)(R,Ra + RM + (Rs + Ra){RiR2 + RM + (i?5 + R6){RlRi + i?2i?3) + i?5(i?1^3 + R2Ri) + 7?6(i?li^2 + RsRi), N = R,{Ri + R, + i?5)(i?2 + i^e) + RiRi{Rs + i^s). (7) The condition for oscillation, as mentioned previously, is jjl^ = 1|0. Putting /x = Ml + iM2, we may write (mi + JM2) ARi - jBXi AIRi + JWX4 which gives the pair of equations fiiARi + ^2^X4 - MRi = 0 HiARi - {fiiB + N)Xi = 0. 1, and (8) (9) (10) For the special case in which the amplifier phase shift is zero (m = 0), these become Ml = ^ = IMI (ii: and X4 = 0. (12) 3 "Transmission Circuits for Telephonic Communication," K. S. Johnson, pp. 284-5. D. Van Nostrand Company. THE BRIDGE STABILIZED OSCILLATOR 581 The latter equation indicates that the frequency is then independent of changes in any of the circuit parameters except the crystal, which must operate exactly at resonance. If the phase of ix differs only slightly from zero, so that jU2 is very small, then it may be inferred from continuity considerations that the frequency is still very nearly independent of all circuit parameters, except of course variations in d, the phase of y.. When Q is limited to values for which ^2^X4 < < iiiARi, (11) still applies closely. Substi- tution into (10) gives X4 = MR A B,xr + N MRS B M + iV (13) and finally from (5) and (13) we obtain the frequency deviation in the form / - /o . Md /o 2Q{B\tx\ +N) (14) As noted above, this expression applies accurately only when 6 is small, as it should be in a well designed bridge oscillator. The effect of variations in the amplifier may be examined by dif- ferentiating (14). For changes in d only, and for those of | /x | , 4/1 M /oj , 1 e~ 2Q{B\y.\ +iV)'^ IX|, dn iMi BMd /oJ 2(2(5 ImI +Ny dd, d\ (15) (16) Equations (15) and (16) have been found to be closely in accord with experiment, although the differentiation is not rigorously allowable (B, M and N being only approximately constant). In the special case where all the fixed bridge resistances (R2 to Re Inclusive) are equal, and |m| is large enough so that Ri has nearly the same value, (14), (15) and (16) reduce to the following: /-/o /o df] /oJ df] /oJ iMl <2(ImI +8) 8 e <2(|m|+8) dd, <2(|mI +8)^ dU\ (17) (18) (19) 582 BELL SYSTEM TECHNICAL JOURNAL These expressions show, as did the vector diagrams, that for optimum stability the amplifier phase shift should be made approximately zero, the crystal should have as large a value of Q (as low a decrement) as possible, and the amplifier should have high gain. Linearity in the amplifier is also desirable, to minimize the modulation effects described by Llewellyn.' When present, these effects appear as variations in ImI and d. One of the significant differences between the bridge oscillator and other oscillator circuits is the fact that its frequency stability is roughly proportional to \ix\. This relationship holds at least for amounts of gain that can be dealt with conveniently. Although increased gain is generally accompanied by larger variations in phase, the two are not necessarily proportional. For example, if greater stability were re- quired for some precision application than could be achieved with a single-tube bridge oscillator, and if the constancy of the crystal itself warranted further circuit stabilization, it could be obtained by adding another stage. The phase fluctuations in the amplifier might possibly be doubled, but the value of ] ju I would be multiplied by the amplifica- tion of the added tube, giving an overall improvement. To illustrate the high order of stability provided by a bridge oscil- lator, let us consider a model composed of a single-tube amplifier and a bridge in which all the fixed resistances are made equal to that of the crystal. We will assume the crystal to have a reasonably high ^ Q of 100,000. The amplifier phase, let us say, is normally zero, but may possibly vary ± 0.1 radian (±6 degrees), and the value of |/x|, nominally 400, maychange ±10 per cent. From (18) and (19) we find A/ /o _ (8)(0-l) - ± 2 17 X 10- r ^ (100,000) (360 + 8) " ^ ^-'^ ^ '^ and (when Q has its maximum value of 0.1 radian) A/ /o (8) (0.1) (40) _ ^ 2 36 X 10-« ^ (100,000) (360 -f 8)2 " ^ ^-^^ ^ ^^ • This example represents the degree of stabilization against circuit fluctuations that can be obtained with a simple form of the oscillator operating under poorly controlled conditions. By stabilizing the power supply and other factors affecting | ^ | and d, and by increasing the gain, the frequency variations arising in the driving circuit may be made negligible compared to the variations found at present in the properties even of the best mounted crystals. * For crystals in vacuum, values of Q as great as 300,000 have been obtained. THE BRIDGE STABILIZED OSCILLATOR 583 Experiment The circuit diagram of an experimental bridge stabilized oscillator is shown in Fig. 3, and its photograph in Fig. 4. The amplifier unit consists of a single high-mu tube Fi with tuned input and output transformers T\ and Ti and the usual power supply and biasing arrangements. The crystal, mounted in the cylindrical container at the left end of the panel, is one having a very low temperature coeffi- cient of frequency at ordinary ambient temperatures. In Fig. 4 it is shown without provisions for temperature control. A high Q is obtained by clamping the crystal firmly at the center of its aluminum- TO 8 < LOAD Fig. 3 — Circuit of experimental bridge oscillator. coated major faces between small metal electrodes ground to fit, and by evacuating the container. Some of the circuit parameters are listed below : Ri = Tungsten-filament lamp, Ri = 100 ohms, i?3 = 150 ohms, Zi = 100 kc. crystal. Characteristics at resonance : Ri = 114 ohms, Xl = Xc = 11,900,000 ohms, Q = 104,000, R5 = Re — 150 ohms (approx.), Ri = 500 ohms, Rs = 200 ohms, \n\ = 422 (52.5 db voltage gain from e to E). 584 BELL SYSTEM TECHNICAL JOURNAL Fig. 4- -Experimental bridge stabilized oscillator without provision for temperature control. Figure 5 shows the resistance of the lamp Ri plotted against the power dissipated in its filament. The large rise in resistance for small amounts of power is due to the efifective thermal insulation provided by the vacuum surrounding the filament and to low heat loss by radia- tion. The lamp operates at temperatures below its glow point, assur- ing an extremely long life for the filament. 23 4-56 789 POWER INTO LAMP IN MILLIWATTS Fig. 5 — Characteristic of lamp used for Ri. THE BRIDGE STABILIZED OSCILLATOR 585 The particular value assumed by Ri in the circuit of Fig. 3 is ap- proximately (R2Rz)/Ri = [(100)(150)]/114 = 131.6 ohms, and hence from Fig. 5 it follows that the power supplied to the lamp is about 3.7 milliwatts. The r.m.s. voltage across the lamp is computed to be 0.70 volt, and across the entire bridge, 1.23 volt. The power supplied to a load of 150 ohms through the pad composed of R-; and Rs is accord- ingly 0.22 milliwatt, or 6.6 db below 1 milliwatt, which is in agreement with measurements shown in Figs. 8 and 9, described below. These quantities are given to illustrate the fact that currents and voltages in K^b -— » a , 1 NORMAL t- OPERATING POINT '' i 1 n 1 » -• — < c,^ 60 80 100 120 140 160 180 200 220 240 260 PLATE BATTERY POTENTIAL IN VOLTS Fig. 6 — Oscillator frequency vs. plate battery potential. a — Ci and C2 tuned for maximum amplifier gain. b — Ci and C2 decreased 5%. c — Ci and Co increased 5%. this type of oscillator may be calculated readily from the values of the circuit elements, and without reference to the power supply voltages or the tube characteristics except to assume that they give the amplifier sufficient gain to operate the bridge near balance, and that tube over- loading does not occur at the operating level. Experimental performance curves for the circuit of Fig. 3 are pre- sented in Figs. 6 to 11 inclusive. Figure 6 shows frequency deviation plotted against plate battery voltage for several settings of the grid and plate tuning condensers. For curve a the amplifier was adjusted at maximum gain, corresponding approximately to zero phase shift as 586 BELL SYSTEM TECHNICAL JOURNAL well. Here the frequency varied not more than one part in one hun- dred million for a voltage range from 120 to 240 volts. Curve h was taken with the two tuning capacitances C\ and Ci decreased 5 per cent from their optimum settings, and curve c with both capacitances increased 5 per cent. These detunings introduced phase shifts of about ± 40° (± 0.70 radian), decreased \)x\ by 0.8 dh and changed the frequency, as shown in Fig. 6, approximately ± 2 parts in ten million. Although the analysis should not be expected to apply 0.1 O - 0.2 b r 1 ( 1 , a , ( , ^— . 1 -• — NORMAL OPERATING POINT 7 8. 9 10 11 FILAMENT BATTERY POTENTIAL IN VOLTS Fig. 7 — Oscillator frequency vs. filament battery potential. a — Ci and d tuned for maximum amplifier gain. b — Ci and d decreased 5 %. c — Ci and Ci increased 5%. accurately for such large phase shifts, calculation of the frequency deviations by means of (18) gives ±1.4 parts in ten million — in fair agreement with the experimental results. As might be expected, curves h and c show somewhat less stability with battery voltage changes than does curve a. Figure 7 presents a similar set of curves for variations of filament voltage. Here, for the "maximum-gain" tuning adjustment, a drop from 10 volts, the normal value, to 8 volts caused less than one part in one hundred million change of frequency, as shown in curve a. THE BRIDGE STABILIZED OSCILLATOR 587 In Fig. 8, the gain of the amplifier and the output level of the oscil- lator are plotted against plate battery voltage, while in Fig. 9 the same quantities are related to filament potential. These curves show that although power supply variations change the amplifier gain, they have but slight effect upon the amplitude of oscillation. This stabilization is produced, as explained heretofore, by the action of the lamp. The oscillator was designed to work into a load of 150 ohms, its output impedance approximately matching this value. It might be expected that variations in the magnitude or phase angle of the load (D 53 O > 51 OSCILLATOR OUTPUT LEVEL / INTO 150-OHM LOAD 1 ,^ _^. 1 H— 1 ' 1 1 1 » 1 NOR OPER> PC VIAL 1 ,1 ^ NT 1 > ^ 1 ^/^ < 1 A A ^VOLTAGE AMPLIFICATION OF VACUUM TUBE CIRCUIT = 20 LOG,o|^x| / 1 r * ■7^, -8 S 60 80 100 120 140 160 180 200 220 PLATE BATTERY POTENTIAL IN VOLTS 240 260 -10 Fig. 8 — Amplifier gain and oscillator output level vs. plate battery potential. would affect the frequency materially even though a certain amount of isolation is provided by i?7 and R^. However, measurements made with (1) a series of load impedances having a constant absolute magni- tude of 150 ohms but with phase angles varying between — 90° and + 90° and (2) a series of resistive loads varying between 30 ohms and open circuit, showed less than one part in a hundred million frequency variation. Graphs of these results have not been included, since they practically coincide with the axis of zero frequency deviation. The tuned transformers Ti and T^ in this experimental model pre- cluded the suppression of harmonics by negative feedback, |/x| being small at the harmonic frequencies. The tuning itself provided sup- pression, however, so that the measured levels of the second and third 588 BELL SYSTEM TECHNICAL JOURNAL harmonics in the output current were respectively 67 db and 80 db below that of the fundamental. This purity of wave form is of course largely dependent upon the absence of overloading. To correct any small initial frequency error of the crystal and to allow for subsequent aging, a small reactance connected directly in series with the crystal provides a convenient means of adjusting the frequency as precisely as it is known. This added reactance may be considered as modifying either of the reactances in the equivalent series resonant circuit of the crystal. Figure 10 shows that for a small 53 > 51 OSCILLATOR OUTPUT LEVEL , INTO 150-OHM LOAD ^ jL., \ r-^^' y \ VOLTAGE AMPLIFICATION OF VACUUM TUBE CIRCUIT = 20LOG,o ||JL| ^ / A \ ■•— NORMAL OPERATING POINT 8 9 10 FILAMENT BATTERY POTENTIAL IN VOLTS ■l^. -6 \ \ s. -10 \ \, \ \, -20 s \, \ \ ■600 -400 -200 0 200 400 REACTANCE IN SERIES WITH CRYSTAL IN OHMS 600 Fig. 10 — Frequency of oscillator vs. adjusting reactance. were obtained over a period of several months. Figure 11 is a photo- graph of a section of this record. It shows the short-time variations of both oscillators plus a small amount of scattering caused by the meas- uring equipment itself. The crystals were temperature-controlled in separate ovens, and the power was supplied from separate sets of laboratory batteries controlled to about ± 2 per cent in voltage. Shielding was ample to avoid any tendency to lock in step. In addition to these small short-time variations, the oscillators exhibited a very slow upward drift in frequency, attributed to aging 590 BELL SYSTEM TECHNICAL JOURNAL of the mounted crystals. This aging decreased in a regular manner with time, the mean drift of one of the crystals being less than one part in ten million per month after three months of continuous operation, « • • 6 J' m m m m m •m m * • 4 ,3 • • « • 2 1 PM - m m ® # 12 11 « • 10 9 8 6 0 10 20 30 40 50 60 70 80 90 100 PARTS IN 108 Fig. 11 — Record of frequency comparison between two bridge stabilized oscil- lators. Full scale one part in a million. Variations less than ± 2 parts in one hundred million. and about a third of this amount after seven months. In most appli- cations, gradual frequency drift is not objectionable even though the required accuracy is very high, for readjustment is merely a matter of setting a calibrated dial. THE BRIDGE STABILIZED OSCILLATOR 591 Application The bridge stabilized oscillator promises to become a useful tool in many commercial fields as well as in certain purely scientific problems such as time determination and physical and astronomical measure- ment. It may be used either to increase the frequency precision in applications where operating conditions are accurately controlled, or else to make such control unnecessary, affording high stability in spite of unfavorable conditions. An interesting application in the field of geophysics has already been made in the form of a "Crystal Chronometer." This chronometer consists of a single-tube bridge oscillator, a frequency dividing circuit, and a synchronous timing motor. It was recently loaned by Bell Telephone Laboratories to the American Geophysical Union and was used with the Meinesz gravity-measuring equipment on a submarine gravity-survey expedition in the West Indies. Although operating under somewhat adverse conditions of power supply, temperature, and vibration, it was reported ^ to be more stable than any timing device previously available, errors in the gravity measurements introduced by the chronometer being negligibly small. ® "Gravity Measurement on the U. S. S. Barracuda," M. Ewing, and "Crystal Chronometer Time in Gravity Surveys," A. J. Hoskinson; pp. 66 and 77 rasp.. Transactions of the American Geophysical Union, Part I, 1937. Effect of Space Charge and Transit Time on the Shot Noise in Diodes By A. J. RACK The theoretical analysis of the effect of space charge upon the "shot noise" in a planar diode shows that for practically all operating conditions, the tube noise is equivalent to the thermal resistance noise of the plate resistance at 0.644 times the cathode temperature. Noise in diodes of other than planar shapes is discussed and it is concluded that the same relation holds. It is shown that transit time produces the same high frequency modi- fication for both the thermal and shot tube noise, and that the tube noise is decreased by transit time. IN the study of noise in vacuum tubes, the effect of the space charge upon the shot noise has been a subject of considerable interest and practical importance. Several papers have been written in which it is shown that the shot noise is decreased by the space charge, and that the tube noise in a diode with space charge is equivalent to the thermal resistance noise of the plate impedance at a temperature slightly greater than half of that of the cathode.^' ^^ ^' * The most compre- hensive analysis was made by Schottky and Spenke. These authors, employing a different method from the one here presented, have obtained the same general conclusions given in this paper, although they prefer to express the result in the form of a modified shot-noise equation, whereas for reasons developed below, the writer prefers the thermal form. The theoretical analysis and discussion presented here was undertaken to show in more detail the extent of the range of the operating condition for which the thermal resistance equivalent of tube noise is valid and to study the effect of transit time upon both the shot and thermal tube noise. For convenience, the paper is divided into three parts. In the first section is given an exact mathematical treatment of the tube noise at low frequencies in a parallel plane diode for any degree of space charge. A discussion of the final tube noise equation obtained through this analysis, and the extension of these results for the planar diode to any other shape diode is given in Part II, where the presentation is such that the section may be read independently of the theoretical analysis in Part I. Through several approximations. Part III treats the effect of transit time upon tube noise in the planar diode. 592 SHOT NOISE IN DIODES 593 Part I — ^General Low Frequency Analysis In the development of the general equations for the direct current in vacuum tubes with space charge, account has been taken of the fact that the electrons are emitted from the cathode with Maxwellian velocity distribution. This fact has been verified experimentally by Germer,^ and the resulting equations for the relation between current and voltage have been derived and investigated by Fry,^ Langmuir/ and others. In the extension of this analysis to tube noise, it is only necessary to assume that the number of electrons emitted with any velocity does not remain constant, but fluctuates with time according to the well-known laws of probability. In the analysis on this basis, the frequencies involved will be considered to be sufficiently low so that any transit time effect is negligible. Below is given a list of the definitions of various symbols to be used in the tube noise study of a parallel plane diode. The practical system of units is employed throughout. n(Uc)dUc = instantaneous rate of emission per unit area of the cathode of electrons with initial velocities between Uc and Uc + duc in the rjc-direction, regardless of the velocity components in the other directions, = no(uc)dUc + 8{Uc)dUc, no(Uc)dUc = average rate of emission of electrons with A:-directed velocities between Uc and Uc + dUc, b{iic)duc = instantaneous deviation from average rate of emission, / = instantaneous anode current per unit area, V — instantaneous potential with respect to cathode of a plane at a distance x from the cathode, V — instantaneous potential with respect to cathode of the potential minimum, u — instantaneous velocity at ric-plane of electrons which had an initial x-directed velocity of Uc at the cathode, x' = instantaneous position of potential minimum, e = charge on electron = — 1.59 X 10~"^^ coulombs, m = mass of electron — 9.01 X 10~^^ grams, h = ratio of dyne cms. to joules = 10~^, e = permittivity of a vacuum in practical units = 8.85 X 10~^* farads/cm., k = Boltzmann's gas constant = 1.372 X 10~^* watts/degree Kelvin, A^ — average total number of electrons emitted per second per unit area from the cathode, T — absolute temperature of the cathode. 594 BELL SYSTEM TECHNICAL JOURNAL In the following analysis, it is assumed that the electrodes of the planar diode are infinite in extent, and that the electron emission is random, so that the equipotential surfaces are parallel planes perpen- dicular to the A:-axis. The potential distribution in such a planar diode operating with space charge is shown in Fig. 1. The origin of coordinates is taken at the cathode, and the potential minimum formed by space charge occurs at a distance x = x' from the cathode. The subscript a will be used to denote the space between cathode and potential minimum while ^ applies similarly to the space between minimum and anode. Of all Fig. 1- — Potential distribution in planar diode. the electrons emitted from the cathode only those whose x-velocity exceeds the value uj corresponding to the potential minimum can penetrate the barrier and proceed to the anode. Electrons with lesser values of initial velocity will come to rest at a point in the a-region where the potential corresponds to their initial velocity and will then return to the cathode. The anode current density is thus given by ■f. I = e i n(u,)duc, (1) while the relation between velocity u and potential V at a given value of X is 2e u^ = u? — hm V. (2) SHOT NOISE IN DIODES 595 A third fundamental relation is Poisson's equation which becomes in the parallel plane case under consideration In the a-region the total charge density is made up of three classes of electrons, namely 1. Those destined to pass the potential minimum and arrive at the anode. 2. Those moving away from the cathode but which will not travel as far as the minimum point. 3. Those returning to the cathode. Corresponding to each class of electrons, there is an associated current, pu, so that each of the three densities pi, P2 or p3, may be expressed by a relation of the form, - = ^- w When it is remembered that the potential and velocity at a given value of X are uniquely related through (2), then it is easy to see that the total density for a given plane in the a-region is given by / fi(u ) I ' fi(u ) Pa = e \ ■ ~ duc + 2e I ' duc, (5) where the first term represents the contribution of electrons in class 1 above, while the second term represents the contribution of electrons in classes 2 and 3. The contribution of class 3 is equal to that of class 2. The lower integration limit v of the second term of (5) represents the initial velocity of an electron which would just arrive at the value of x under consideration before coming to rest and starting back toward the cathode and the limit u/ in both terms represents the initial velocity of an electron which comes to rest just at the potential minimum. Thus, from (2) J^ V and u/ = J~ r. (6) In the j8-region there is only one class of electrons, so that the density is more simply expressed. Thus, ■ Pe = e ] -^:^duc. . (7) The value of p in (5) and (7) may each be expressed in terms of d^V/dx^ by the use of (3), and the integration of these two Poisson's relations for the common boundary "condition that the electric force is 596 BELL SYSTEM TECHNICAL JOURNAL zero at the potential minimum has the following result: , ■ ° — I (m — u')n{uc)duc -\ • I un{uc)dur, (8) {dx) e Jj,^, e Jp {dx) e I (m — u')n{uc)duc, (9) where u' is the electronic velocity at the potential minimum, i.e., {u'Y = Ue- - {2e/hm) V . At this point the analysis departs for the first time from the classic analyses of Fry ^ and Langmuir,^ through the introduction of the concept that the instantaneous rate of emission may be expressed as the sum of an average rate of emission plus an instantaneous deviation. That is, «(Mc) = Wo(m<,-) + 5(Mc), (10) transforms (8) and (9) into the following equations: — I (m — ic )n(i{u,)dUc f ''uc' {dx) where and where 4hm C""' H uno{u,)duo + a{d), (11) e Jo "Zhttt I AhtH I a(8) = I (w — u')8{uc)dUc H I u8{Uc)dUc {dV^ ^2hm r ^^_ ^/)„^(^^)^^^ _^ ^(5)^ (12) {dx) e J„/ ^(5) ^ ^ r (^ _ u')5{Ur)dUr. e Juc' Since the average rate of emission may be expressed by the Max- wellian relation, «o(«.) = laNUce-""'', where _ hm " ~2kf' the indicated integrations in (11) and (12) have as a result, {kry {drra)^ ^ Nhm It _^, {e) {dx) € \a X [e"- 1 +e''P(Vr7) - 2^^ + «(5) (13) SHOT NOISE IN DIODES 597 and {kTY {drjffY- ^ Nhm {e) {dx) € \ a X e" - 1 - e'-PCVr?) + 2 + /3(5), (14) where .=^(F'-F), 2 /"^ ■rfx. The fact that both a{b) and ^{b) are very small greatly simplifies the solution for the distance coordinate x in (13) and (14). The process is to invert the two equations, respectively, extract the square root, and then expand the right-hand side in powers of a(6) and /3(5), respectively. This results in expressions for dx/dr] which can be integrated term by term. However, the small values of a(8) and /3(5) allow powers higher than the first to be disregarded, and hence. F(v') 1 kT' and «^ r Nhm f(v) 3/2 I "i' a(8)dr] (15) L ^-'Y 1. f r Nhm lir , 1 (16) where for convenience r — Jo r el - 1 C?T7 /(>?) Jo [ + e^P(Vr,)-2^^j^'"' Te"- 1 -6''P(V^) + 2^^r' (2) ^o==^(F'- V,), B = /('Jo) = C Fir),') +/(77o) , dFir),') , ^/(r^o) + ^lo' dr]o' dx + ^TJc eIodf(r]c kT dr]o e^ - 1 + (Jil-iJJ'" .[ if 1 - e-P(Vx) + 2 V^]" "' [Vy" + X — y] $(x) (ix. Vtt [Vy- + X — y']d. X 1 - e-P(^'x) + 2 3/2 $(x) = t^ - 1 + e^P(4x) - 2 P(x) =A r%-2^x. Vtt Jo Part II — General Discussion The analysis in Part I shows that as soon as a potential minimum exists, the tube noise in a planar diode is equivalent to the thermal (45) dxdydz dxdydz \ , (46) B (47) 606 BELL SYSTEM TECHNICAL JOURNAL noise of the plate resistance at an effective temperature which is a function of that of the cathode. In general, the effective value of the diode plate resistance tempera- ture for any operating condition is very difficult to obtain because of the complexity of the final noise equations (45) and (46). However, the limiting value of the ratio of the effective plate resistance tempera- ture to that of the cathode, denoted by "X" in (45), may be evaluated very readily for certain limiting conditions. One encounters the first of these conditions when the plate potential and cathode emission are such that the potential minimum has moved just up to the cathode, and is in fact on the point of disappearing. This condition is secured by decreasing the space charge to values less than are required for the formation of a potential minimum away from the cathode. In the equations, it is represented by letting the quantity 770' approach zero, where 770' is the natural logarithm of the ratio of the saturation current to the anode current. For this set of operating conditions, all the electrons emitted from the cathode will go to the anode, and hence the condition is appropriate to the study of pure shot noise. A second condition is obtained when the plate potential is equal in value to the potential of the minimum. Physically, this condition means that the minimum has moved just up to the anode, and requires a negative value for the plate potential. Mathematically, it is repre- sented by a zero value for the quantity 170, where 770 is equal to the difference between 770' and (e/kT)Vp. For negative plate voltages greater in magnitude than that of the potential minimum, all electrons having an initial kinetic energy greater than eVp will reach the anode regardless of the presence of the space charge existing between the two electrodes. For these conditions, the diode becomes a temperature limited current device. A third limiting condition occurs when the plate potential is large in magnitude compared with that of the potential minimum referred to the cathode. In this condition a potential minimum still exists. It is represented in the mathematics by letting the quantity 770 become large. This condition represents the normal operating condition for the diode. As the space charge is decreased, making 770' very small, from (47), the diode plate impedance becomes very large through the action of dF(r]o')ldr]o' which becomes infinite as 770' approaches zero. As all other quantities involved remain finite, the mean square noise current for a very small space charge is SHOT NOISE IN DIODES 607 dfM ell} dr]o Thus, as the potential minimum voltage is reduced to zero, the tube noise as given by (45) reduces to the well known shot effect equation. For some space charge at the cathode, the value of X in (45) has definite limiting values for both very low and for very large plate voltages. For a very small value of 170, that is for negative plate voltage, the value of B defined in (47) is very large because rf/(»7o)M'7o becomes infinite as rjo is decreased to zero. Thus as 170 —> 0 52 r y^-y-'dy = ^ ' (49) Hence, for any value of space charge, the effective plate resistance temperature for negative plate voltages is one-half of the cathode temperature, under the restriction that no potential minimum exists between the cathode and anode. Since the diode is usually operated with a positive plate voltage, the value of the effective plate resistance temperature for a large value of 770 is of more interest. For vo' not equal to zero, and a large value of plate voltage, it can be readily shown that the values of /(ijo) and of D are much larger than any other quantities involved in the equation for X. After a bit of mathematical operation, it may be shown that the limiting values for/(r7o) and D are ^ =W' [t""'" + ^^^'?«''' + . . . - {4yvo"' +•••)] 7^1/4 r 4 ,- m2 L "^ From these relations, the limiting value of X for a large plate voltage is given by X = 3 J% r V23; - ^^ \~y'dy = 3(1-^) = 0.644. (50) Thus, for any value of space charge, as long as a potential minimum exists, a sufficiently large value of plate voltage may always be found for which the effective plate resistance temperature is 0.644 times the cathode temperature. 608 BELL SYSTEM TECHNICAL JOURNAL It is possible to obtain a good approximation for the effective diode temperature for any operating condition by the following method. The values of "C" and "Z)" in (47) may be found without too much difficulty by graphical integration for several different values of y, rjo and rjo'. From the tabulated values for F(rjo') and /(rjo) given by Langmuir, and from the values found for C and D, the integral, S = I ylB - C - Dje~^'dy, Jo (5i: may be evaluated by mechanical means for several values of t^o and rio'. This gives the first integral in (46). It was found practically impossible to calculate directly the contri- bution to X from the last two integrals in (46). However, a rough approximation to them may be found indirectly by the following method : If the sum of the two integrals is denoted by Q, then (46) may be written ^ dfivo) ' drio or Q — B\(df(rio)/dr]o) — S = function of rjo' only. For a fixed value of rjo', the solution of the above equation for several values of r]o should give a constant value for Q. Unfortunately, only the limiting values of X are known. However, if the limiting value of 0.644 is substituted for X in this equation, the calculated value of Q, for a fixed value of r/o', should approach a constant value as t/o is increased since X does assume the 0.644 value for t/o sufficiently large. The limiting value of Q calculated in this manner is the desired contribution to X from the last two integrals in (46). This method of evaluating Q cannot be very accurate since it involves the difference of two quantities of the same magnitude. However, since Q is small compared to the contribution from the first integral in (46), a large error in Q will introduce a much smaller error in the value of X. The values of the effective diode plate resistance temperature calculated in this manner for several different operating conditions are shown in Fig. 2. These curves indicate that the effective diode temperature is 0.644 times the cathode temperature for all practical operating conditions. The values of 170' and 770 may be determined from the following relations: Tjo' = logey^, (52) SHOT NOISE IN DIODES 609 where I^ is the saturation current and I p is the anode current, and (53) T?o ^ '^^' ~ yp ^^p ^ '^^' + "V" ^ 10"* ^p. where Vp is the anode potential, and T is the absolute cathode temper- ature. For T = 900° K, .70 = vo' + n.9Vp. (54) 1.0 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 — 0.2 ^0 3 — 0.5 "^ wl/^' ^^S' ^ _ - "^ 1 ^= =- 1 1.0 30 2.0 4 6 8 10 12 14 16 18 20 22 24 26 28 30 Fig. 2 — Effective noise generator voltage of planar diode. £/- = ^krp{\T)df, rjo' = logey-', 570 = Va' kT Vr, For T = 900° K., rjo = 770' + 12.9 Fp Even for the small space charge condition for which the plate current is eight-tenths of the saturation current (tjo' = 0.2), the value of 770 need be greater than about 25 only before X assumes its limiting value. For a temperature of 900° K., as for oxide coated cathodes, this would require a plate voltage of only two volts. If the plate current were less than eight-tenths of the saturation current for very high plate voltages, then as the plate voltage is reduced, r/o' would increase. For this operating condition X maintains its limiting value of 0.644 for all except negative values of plate voltages. The transition between the various effective planar diode plate resistance temperatures is more clearly shown in Fig. 3. In this 610 BELL SYSTEM TECHNICAL JOURNAL figure, the natural logarithm of the ratio of saturation current to the plate current is plotted as a function of the plate voltage for several constant values of the coef^cient X. These curves show for a fixed positive value of plate voltage that as the space charge is decreased toward zero, by a reduction in the ratio of the saturation current to the plate current, the value of X for moderately large values of space charge increases but little from 0.644, and then for a very low space charge, increases very rapidly to its limiting value given by the shot noise % \ ■ • ■ 1 I ^o.6 1 > . ' ■ • ^ I 1-0.644 I 1 11 m 1 >= 0.644 1 1 \ •.'•';•■•'• •••Vi ■■'ill 1 \ 1 \ V ^. >v=0.5 0.675 V s 'M{ 1 / --^ ^^^ ••'•!•*: 'M f,^'' .— 0.8 -~: -^. .v. •■"•':•• ':iP\ 0.9 — — .— --- •~~i ~^c -==z ^~~ -4 -2 20 22 24 26 28 Fig. 3- — Modification of effective plate resistance temperature produced by space charge. £/2 = ^krp{\T)dJ. which is represented by the axis of abscissae. Thus, the value of X digresses markedly from 0.644 only for the narrow region of operating conditions for which the saturation current is less than 1.25 times the plate current and the plate voltage is less than 2Se/kT volts. For an oxide coated cathode for which T is 900° K., the effective plate re- sistance temperature is 0.6447" for any operating condition for which plate current is less than eight-tenths of the saturation current and the plate voltage is greater than two volts. For a cylindrical diode, the general method of analysis used in the parallel plane case results in equations which are practically impossible SHOT NOISE IN DIODES 611 to solve. The difificulty in these equations arises from the fact that tangential as well as radial initial velocities must be considered in obtaining the total anode current. Since it was shown for the planar diode that the effective temperature of the plate resistance is 0.644 times the cathode temperature for practically all operating conditions, all that is really desired in the cylindrical diode solution is the limiting value of the effective tube temperature. This may be found rather easily from a comparison of the cylindrical diode with the planar tube in the following manner. For a very large space charge, and a high plate potential the radius of an equipotential surface near the potential minimum will be very nearly equal to that of the cathode. Hence, for these operating conditions, the planar diode equations may be applied to this region of the cylindrical diode. In the planar tube, it was shown that for 770' > 3, 770 had to be of the order of unity to obtain the limiting value of 0.644 for X. If the space charge and plate potential are sufficiently large in the cylindrical diode, the radius of the equipotential surface for which 170 is greater than unity will practically be equal to that of the cathode. The cylindrical diode may then be divided into two parts, a planar diode between the cathode and the equipotential surface for which rjo > 1, and a cylindrical diode formed from the remainder of the tube. In any diode, the only source of noise energy is the cathode from which the noise power is transferred to the anode and external circuit through the mechanism of the initial electronic velocities. Furthermore, the same total noise power must be trans- ferred across any equipotential surface between the cathode and anode. In the planar portion of the cylindrical diode as described above, the total noise power crossing any equipotential surface was shown to be 2.576kTdf. This same noise power must be transferred across any other equipotential surface in the cylindrical diode. Hence, the effective plate resistance temperature for the cylindrical electrode tube must also be 0.644 times its cathode temperature. From this line of reasoning, it may be shown that the limiting value of the effective temperature for any shape diode is the same as that for the planar tube with the same cathode temperature. From the experimental data given in his paper, Pearson definitely recognized that the limiting value of the diode plate resistance temper- ature should be between 0.59 and 0.65 of that of the cathode.^ The writer understands that North and Thompson of the R.C.A. in an unpublished paper have obtained the same general result for the effect of space charge upon shot noise in diodes. 612 BELL SYSTEM TECHNICAL JOURNAL In a diode, the tube noise may be expressed equally well and with equal correctness either as a modified shot noise or as a thermal resistance noise. In this paper, the thermal resistance viewpoint was taken for two reasons. First, the coefiicient "X," used in the thermal resistance noise equation £? = Akr,{\T)df, is practically always a constant equal to 0.644, whereas, the factor, "7^," used by Schottky and Spenke in their modified shot noise equation 772 = 2eF'Iodf is always a function of the operating condition. That is, for the operating conditions for which X is a constant, F has the following value: 1.39 [^''^Tr'^^T The second reason for the selection of the thermal resistance noise relation is that power from the motion of the atoms in the cathode is actually transferred to the plate electrode and external circuit through the mechanism of the initial electron velocities. Hence, the tube noise in a diode with space charge is very similar to a thermal resistance noise. Part III — Effect of Transit Time The analysis, in Part I, while giving the correct results for all operating conditions in the ordinary frequency range, is extremely long and cumbersome. It shows, however, that only the limiting values of the effective temperature of the plate resistance are required for most practical cases, and therefore it points the way to make simplifying assumptions which result in a much shorter analysis, and moreover, which allow the analysis to be extended to frequencies so high that electron transit time phenomena become of importance. Thus the final noise equation in Part I shows that for moderately high anode potentials and for the usual excess of cathode emission, a very good approximation may be had by a consideration of the current-voltage relations existing in the jS-region between potential minimum and anode without the necessity of encumbering the analysis by including the a-region between potential minimum and cathode. Moreover, for a large anode potential, the terminal velocities of the SHOT NOISE IN DIODES 613 electrons at the plate are very large in comparison with their initial velocities for practically all of the electrons. This means that the transit time for the various electrons is practically the same for all of them which leave the cathode within a particular very short time interval, even though the initial velocities of the various electrons are statistically distributed among them. It results that the various individual velocities of the electrons in the /3-region may be replaced by an average value, which at the potential minimum may be defined as follows : u'n{ii,)duc -a--^ (55) n{uc)duc f Physically, the meaning of this expression is the average velocity of these electrons which cross a plane in the )3-region close to the po- tential minimum in a unit of time. Inasmuch as the unit of time may be taken to be very small, it follows that (55) expresses the effective instantaneous value of the initial velocity which may, and does, fluctuate as time goes on. On the basis of an equation of the form the planar diode has been extensively investigated by a number of workers and it has been shown ^ that the relation between current and voltage is completely specified as soon as two boundary conditions are given. These may be the initial velocity and acceleration, or they may equally well be the initial velocity and conduction current pu. However, the analysis based on (56) applies strictly to the case where all of the charge moves with the same velocity and hence contains a certain approximation when electrons are considered whose velocities have a certain dispersion around some mean value. The error will be small until frequencies are considered which are so high that a large proportion of the electrons which left the cathode in a time interval which is very short compared with the period of the high frequency arrive at the anode in a time interval which is not small compared with the high frequency period. Normally this means that the error is small even for frequencies so high that the majority of the electrons require several cycles to make their transit from potential minimum to anode. 614 BELL SYSTEM TECHNICAL JOURNAL It is convenient to write the resulting equations in terms of d-c. and first order a-c. values where the initial values of d-c. velocity and acceleration are given, but initial values of a-c. velocity and conduction current are employed. The first order a-c. relation derived by Llewellyn may be written in the form hm time nme (57) where qa and fXa are the initial values of fluctuation conduction current and velocity, respectively, while A, B and C are defined by: A =W - io^^x 4- -^ (2 ■- 2e-'" id 11 B C = - -r^ [_aa{iee-^^ + e-'" - 1) + Uaio^ie-^' - 1)] hmtisP' (58) in which ?? is the transit angle, wr, the transit time being r, and /o is the d-c. current. In the application of these relations to noise analysis, the initial values of velocity, acceleration, and conduction current must be taken at a point in the ^S-region beyond the potential minimum, but just as close to it as possible without encountering conditions where electrons may be moving toward the cathode, for the equations apply only to cases where the electrons are moving in one direction only. The initial point is, however, located so near to the potential minimum that the d-c. acceleration in (58) may be taken as zero. When this is the case, it may be shown that the initial conduction current is equal to the total current. In other words, the initial value of dis- placement current is zero. Under such conditions (57) and (58) reduce to the following expression for the a-c. anode potential in terms of the a-c. component of current and initial velocity : V = /i nme \ 6 + (^""Uaiie + e~^^ - 1) -f '=^ [ide-'' + e-'o - 1]. (59) co'e The term multiplying the a-c. current h in the above equation is the internal high-frequency impedance z of the planar diode. The last^term may therefore be identified with an internal emf. When the initial velocity /x, is expressed in terms of the fluctuation of electron SHOT NOISE IN DIODES 615 velocity, the term gives the equivalent noise generator, E. Thus E=f^{ide-^9 j^e-'9 - 1) (60) and the mean-square value of the noise emf. (at a frequency w) is given by : £2 = ^^" \ie-^6 ^ Q-i0 _ u\ (61) The problem is now reduced to finding the mean square value of initial velocity fluctuation, txa^, which corresponds to electrons cross- ing the potential minimum. This may be done by going to (55) which gives the effective value of the instantaneous initial velocity and separating all quantities, including the lower integration limits into d-c. and a-c. components. Thus n{uc) = no{tic) + 8(uc) uj = Uc + 8Uc u' = u^ -{- bu' U = Ua + IJLa (62) The result may be expanded in series form and products of the 5's may be disregarded inasmuch as the a-c. components are small in comparison with the d-c. The indicated operations have as a result and e f" Ma = -^ 1 (U' — Ua)8 {Uc)dUc (63) (64) The Fourier analysis may be applied to this in the way outlined in connection with (37) and (41) in Part I and gives the mean-square value of velocity fluctuation corresponding to a frequency interval df as follows : — 2e2 p 4ekT,,/. 7r\ AT = T^«/ (" - Ua)-Uc{Uc)dUc = -f^^-dfi 1—1 1 0" ./_ I onm \ 4 / (65) This may be substituted in (62) giving for the effective noise emf. in the frequency range df E/ = 4kTdf\ el, OT- L hmt~ 1 Xi[^- + 2 - 2 (cos d -\- 9 sin 6)']. (66) 616 BELL SYSTEM TECHNICAL JOURNAL The initial average velocity is small so that the low-frequency plate impedance may be written ru = eL or" 12 //me'-' (67) Thus for any transit angle, the mean square noise generator voltage is given by £/ = 12 (^ 1 - -\kr,Tdf\j^[_2 + 0"- - 2(cos 6 + d sin ^)] = ASk{OM^T)rpdf S = 2 + e- - 2(cos d ^ 0 sin (68) For low transit angles, this expression reduces to £.2 = 12 1 kr„Tdf, (69) which is precisely the limiting value obtained by the much longer, but more rigorous analysis. It must be understood that (68) is an approximation since the transit time effect in the region between cathode and potential minimum was entirely neglected, and because the validity of the average velocity concept does fail at the very high frequencies. Some knowledge of the extent of the operating conditions for which the above equations are good approximations may be obtained from the d-c. current-voltage relation. For the boundary conditions assumed, the low frequency current equation derived from the general solution given by Llewellyn reduces to / = - 2.33(F 10'^(x — X |^'[l+2.66^^-i^^]. (70) This equation was shown by Langmuir to be a very good approxi- mation for the plate current for most operating conditions and fails only for very low values of plate voltages. Thus, it may be concluded that (68) is a good approximation for all operating conditions except for very low plate voltages and a small space charge. The plot of (68) given in Fig. 4 shows that the magnitude of the mean square noise generator voltage decreases by five per cent only for transit angles as large as one radian. SHOT NOISE IN DIODES 617 The effect of transit time on the pure shot noise for a low space charge density and a high plate voltage may be obtained quite readily from (57) and (58). Since for a very small space charge, Jo and Ua are small, and a„ large, the equations then reduce to the following ex- pression : Vi = ^ — +-7^a„r- 1 - iiee~''> + e- 1) (71) 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 ^ \ \ \ \ \ \ \ \ s. \ s^ ^ n 0 1 23456789 10 II TRANSIT ANGLE IN RADIANS Fig. 4 — Effect of transit time on both thermal and shot tube noise. E/ = 4SkiOM4:T)rpdf, 7/ = leSIodf. For these operating conditions, the transit time in terms of the d-c. acceleration and the electrode spacing is given by X = tta T- In terms of the external circuit impedance Z/ (72) 618 BELL SYSTEM TECHNICAL JOURNAL SO that (71) and (72) combine to give h qa 1 + Zfioie + e-'" - 1; (73: The mean square shot noise current is thus given by Ir = 1 + Zfioie I (i^ -\-e- (74) The value of the mean square a-c. conduction current at the cathode to be substituted in the above equation may be derived as follows : The total current emitted from the filament was defined as J 100 /»00 /^OO n(Uc)dUc = e I no(u,)diir + ^ I j d{Uc)duc. (75) 0 Jo Jo Hence 5a Jo 8(uc)dUc (76) From (37) and (41), the contribution to the mean square of this conduction current from the frequencies between / and f -\- df is ga~ ..00 2e-dJ I «o(w, Jo )duc = lelodf. (77) With this result, the effect of transit time on the shot noise current is given by lehdf 1 + Z /t(joe -[2 + 6/2 - 2(cos ^ + ^ si in0)]|. (78) where Z/ is the impedance of the external circuit at the frequency/ and x/icoe is the capacitive reactance of the diode at the same frequency. Thus the shot noise current is modified by transit time in precisely the same manner as the noise generator voltage for the thermal tube noise. The effect of transit time upon the shot noise, as indicated in (78), is identical with that obtained by Spenke for the same operating condition of low space charge and high anode potential.* Spenke derives this result through a clever application of a Fourier Series in which account was taken of the effect of transit time upon the wave shape of the current induced in the anode by the electron moving from SHOT NOISE IN DIODES 619 cathode to the plate. The advantage of the method of average veloci- ties used in this paper is that the effect of transit time in both the thermal tube noise and the shot noise may be found. It is noteworthy that Ballantine in 1928 derived an expression for the effect of transit time upon the pure shot noise which is identical to that obtained in this paper.^ In conclusion, the writer wishes to express his appreciation to F. B. Llewellyn whose supervision and numerous suggestions made possible this paper. References 1. F. B. Llewellyn, "A Study of Noise in Vacuum Tubes and Attached Circuits," Proc. /.i?.£., vol. 18, pp. 243-265 (1930). 2. G. L. Pearson, "Shot Effect and Thermal Agitation in a Space Charge Limited Current," Physics, vol. 6, pp. 6-9 (1935). 3. W. Schottky and E. Spenke, "Die Raumladungsschwachung des Schroteffektes," WissetischaflHche Veroff tntlichungen aus den Siemens-Werken, vol. 16, pp. 1-41, Aug. 6, 1937. 4. E. Spenke, "Die Frequenzabhangigkiet der Schroteffektes," Wissenschaftliche Veroffentlichungen aus den Siemens-Werken, vol. 16, pp. 127-136, Oct. 8, 1937. 5. L. H. Germer, "Distribution of Initial Velocities Among Thermonic Electrons," Phys. Rev., vol. 25, 795 (1925). 6. T. C. Fry, "Thermonic Current Between Parallel Plane Electrodes," Phys. Rev., vol. 17, 441 (1921). 7. L Langmuir, "Effect of Space Charge and Initial Velocities," Phys. Rev., vol. 21, 419 (1923). 8. F. B. Llewellyn, "Operation of Ultra High Frequency Vacuum Tubes," B.S.T.J., Oct. 1935. 9. S. Ballantine, " Schrot-Effect in High Frequency Circuits," J.F.I., vol. 206, 159 (Aug. 1928). 10. T. C. Fry, "The Theory of the Schroteffekt," J.F.I., vol. 199, No. 2 (Feb. 1925). Fundamentals of Teletypewriters Used in the Bell System By E. F. WATSON During the past few years the use of teletypewriters has become quite extensive in the Bell System. Simpler and cheaper machines have recently been made available for meeting the simpler service requirements and attachments have been designed to provide additional features for meeting more complex service requirements. This article discusses the fundamental principles and various features of the teletypewriter machines now in common use and explains the more important factors which have been controlling in their development. WITH the growth of Teletypewriter Exchange Service and the general increase in the use of teletypewriters in private line services of various types, questions frequently asked are: How do teletypewriters operate? What is the "start-stop" system? Why is it used? What is a regenerative repeater? This article will attempt to answer some of these questions and explain also the fundamental principles and features of teletype- writers and their auxiliary arrangements as now employed in the Bell System. These have been developed to meet the needs of customers for a typed or similar record form of communication and at the same time be suitable for operation in connection with the Bell System plant. Code For economical transmission over long distances it is fundamental that only a single wire or transmission channel be required to carry the signals. Furthermore, long experience with manual telegraphy on land lines has proved that reliable and efficient operation is secured by using not more than two conditions on the line, such as current and no current or positive impulses and negative impulses, as contrasted with the use of three or more conditions, or current values. The entire telegraph plant of the Bell System as well as practically all other land line telegraph systems have been built on this two-condition basis. The familiar Morse code uses sequences of dots and dashes to represent the different characters of the alphabet and meet the above conditions. This code is not well adapted for teletypewriter control, however, since the signals for different characters vary widely in the time they require, from a single dot for the letter E to a combination of 620 FUNDAMENTALS OF TELETYPEWRITERS 621 several dots and dashes for some of the less frequently used letters or numerals. For machine operation it has thus far appeared desirable in order to obtain simplest mechanisms and to obtain maximum operating speeds with low line signaling frequencies, to have the signals for the different characters of uniform length, that is, each contain the same number of time units. This condition is met by the five-unit code where each character is identified by the impulses in five units of time, and this is the code normally employed in Bell System teletypewriters. Each of the five units of this code may be either positive or negative, current or no current, or either of two values of current, and the permutations provided are 2'^ or 32. These are sufficient for the 26 letters of the alphabet, a space, carriage return and paper feed signals as well as case shifting signals to bring another set of characters into action so as to include numerals and punctuation marks. A chart of this code as used in Teletypewriter P^xchange Service (TWX), is shown below. FIGURES LETTERS A 5 8 B 1 e C $ D 3 E 1 4 F & G Q. o 1- ul H e 1 J 1 2 K 3 4 L M 7 8 N 9 O 0 1 Q 4 R _] UJ CD S 5 T 7 U 3 8 V 2 W / X 6 Y Z Q UJ UJ u. UJ z -J UJ u UJ a: cr < tr UJ UJ _i 10 UJ ce u. z < PULSE 1 • • • • • • • • • • •••• •• PULSE 2 • • ••••• • • • ••• • •• PULSE 3 • ••••••• • ••••• •• PULSE 4 • • • • • • PULSE S • • • • • • • • • ••••• •• Fig. 1 — Chart of five-unit TWX code. The keyboard used for sending this code is shown below. Fig. 2 — Chart of TWX keyboard. It will be noted that this keyboard is similar to the ordinary type- writer keyboard except that there are only three rows of keys instead of four as in the typewriter. In the typewriter keyboard, the lower three rows of keys are used ordinarily for small letters but when a shift 622 BELL SYSTEM TECHNICAL JOURNAL key is also operated they type the corresponding capital letters. The fourth or top row of keys carries the numerals and certain punctuation marks. The teletypewriter types capital letters but not small letters so that by using a shift or Figures key the upper case position of the letter keys is available for the usual punctuation marks and numerals. Thus only three rows of keys are required on the teletypewriter key- board. The operation of the Figures key sends a signal causing the receiving machine to shift to upper case so that numerals and punctua- tion marks will be printed until a Letters or Space signal is sent which restores the machine to lower case. Start-Stop System For transmitting the signals of the five-unit code over a telegraph line, it is necessary to have some system of timing so that each of the five impulses may be properly received, identified and interpreted at each receiving station. The start-stop system is used for this purpose. One arrangement of this system using segmented distributors with revolving brushes, is illustrated in Fig. 3. In this system both sending and receiving brush arms are normally at rest but are maintained under constant torque, tending to rotate them in the direction of the arrows, by constantly running motors driving them through friction clutches. Normally the line circuit is closed and carries current. When a key of the keyboard is operated to send a signal, the start magnet of the sending distributor is ener- gised releasing the sending brush arm and allowing it to rotate. As this brush passes from the stop to the start segment, the line circuit is opened and this open signal transmitted to the receiving station where it causes energization of a start magnet which releases the receiving brush and allows it to rotate. Both sending and receiving brush arms rotate at approximately the same speeds since they are driven from motors running at approxi- mately the same speeds. These motors are either small synchronous motors driven from constant frequency commercial 60-cycle 110-volt power supply or by commutator type motors, equipped with centrifugal governors to hold them at approximately a constant speed, for use on other commercial a-c. or d-c. supplies. Now as the sending brush arm sweeps over the sending face, the impulses of the five-unit code, as set up by the particular key de- pressed, will be transmitted over the line as shown in Fig. 4 for the letter A, and through the action of the rotating receiving brush, Nos. 1 and 2 current impulses will cause the energization of Nos. 1 and 2 selecting FUNDAMENTALS OF TELETYPEWRITERS 623 0^\V:\V^-v^: \ 5.9 "/a SLOWER THAN ) RECEIVING DISTRIBUTOR Fig. 5 — Effect of distorted signals on reception. experienced in service without failure to receive and properly identify each pulse as a No. 1, 2, 3, 4 or 5 pulse. The other traces illustrate certain types of distortions which may be experienced and the condi- tions existing for their proper reception with this adjustment. These traces will now be explained on a purely theoretical basis assuming an ideal receiving machine without mechanical or other imperfections. Trace (b) shows the conditions in case of 25 per cent "marking bias" in the received signals; that is, each marking pulse has been lengthened FUNDAMENTALS OF TELETYPEWRITERS 627 by 25 per cent of one pulse length. It will be noted that under these conditions each pulse will still be properly received and identified. Trace (c) illustrates 50 per cent marking bias in the received signal, and at this point it will be noted that the No, 3 and stop pulses have been so elongated that they are just on the verge of being erroneously received and identified also as No. 2 and No. 5 pulses. This then is a theoretical limit for proper operation with marking bias without a readjustment of the receiving distributor. Similarly traces (d) and (e) illustrate respectively the conditions when the received signals have 25 per cent and 50 per cent "spacing bias," that is each marking impulse has been shortened by this per- centage of one pulse length. It will be noted that 25 per cent spacing bias can be easily tolerated but that with 50 per cent spacing bias the No. 1 and No. 3 pulses are on the verge of failure to be recorded. This then is a theoretical limit for spacing bias in the signals under this adjustment. Traces (f) and (g) show the effect of 25 per cent marking and spacing distortions respectively on the rear end of the stop or front end of the start impulse, all other pulses remaining undistorted. Trace (h) shows the effect of distortions on the selecting impulses alone. By combining this trace with traces (f) and (g) it will be seen that with 25 per cent distortion of the start pulse, 25 per cent distortion of the same sign is the limit for distortion on the front end of marking impulses, or 25 per cent distortion of opposite sign on the rear end of marking impulses. Thus for distortions other than bias, which are apt to affect both start and selecting impulses in the same signal, ± 25 per cent is the theoretical limit of allowable distortion. Traces (i) and (j) show the effects of speed inaccuracies. From these it will be seen that theoretically the sending distributor could be about 8.9 per cent faster or 9.5 per cent slower than the receiving distributor before errors would be experienced. In practical machines there are, of course, inaccuracies due to tolerances of manufacture, and other departures from the ideal so that the above mentioned theoretical limits are not reached. However, all machines used in the Bell System are required to tolerate a lengthen- ing or shortening of the front end of any current impulse of at least 40 per cent of its length and with the same adjustment a lengthening or shortening of the rear portion of any current impulse of at least 35 per cent with the start pulse undistorted. Since bias is nearly always present to some degree in the received signals, and since as interpreted by the receiving distributor it affects only the front end of the current impulses as illustrated in traces (b), (c), (d) and (e) of Fig. 5, the 628 BELL SYSTEM TECHNICAL JOURNAL distributors are usually adjusted for maximum tolerance of front end distortions and then have ample tolerance for such distortions of the rear ends of the current impulses as are experienced under service conditions. Regenerative Repeaters As circuits become longer and more complex, eventually a point is reached beyond which signal distortion becomes so great that the signals cannot be reliably received without error. To overcome this Fig. 6 — Regenerator unit. limitation a device known as a regenerative repeater may be inserted in the line at this point. It has a receiving mechanism similar in principle to the receiving distributor of a teletypewriter and will accurately receive and interpret any signals which a teletypewriter would accurately record. This receiving mechanism is interconnected with a retransmitting mechanism, or sending distributor, which re- transmits the signals reshaped and reformed so as to be substantially free from distortion. In its latest form a one-way regenerative re- peater consists of a receiving magnet and a set of transmitting contacts interconnected by some relatively simple mechanical parts driven from FUNDAMENTALS OF TELETYPEWRITERS 629 a motor. A photograph of such a recently developed regenerative repeater unit is reproduced in Fig. 6. By means of these regenerative repeaters reliable teletypewriter service may be extended to any desired distances so long as the signals in any one regenerative repeater section are not too badly distorted to permit reliable operation of that section alone. Several regenerative repeaters may be operated in tandem on a single very long circuit if required and in fact a number of very difficult long circuits are operat- ing satisfactorily under these conditions at the present time. A point worthy of note and which has not previously been mentioned is that the stop impulse in the code adopted for Bell System apparatus is slightly longer than the other impulses, which facilitates the use of regenerative repeaters in tandem without requiring complex speed control arrangements. General Features of Teletypewriters Teletypewriters are widely used for high speed written communi- cations. Generally speaking, written communications are desired for purposes of accuracy. Therefore, high speed, accuracy and reliability are basic requirements for teletypewriter service. In choosing an operating speed at which distributors of teletype- writers are to be set, several factors must be considered. These are the capabilities of the mechanisms of the machines, the average capabilities of operators for continuous sending at high speeds, the commercial need for high speeds, and the capabilities of the line circuits for transmitting the signals reliably over long periods without excessive distortions or excessive attention for maintenance and adjustment. A satisfactory compromise among these different factors seems at the present time to be about 60 words per minute, or 368 machine opera- tions per minute, which is the speed usually employed in the Bell System. The machines themselves may be arranged and adjusted to be capable of higher speeds up to about 75 words per minute, and it may be that, in the future, service at these higher speeds will be justified under certain circumstances. Accuracy and freedom from breakdown troubles are necessarily inter-related and both required to a very high degree for machines handling important written communications over long distances. To give some idea of the severity of these requirements, we have found from long experience that to produce a good machine we can not be satisfied in our laboratory tests unless the machine is capable of typing at least 1,000,000 consecutive words (6,000,000 operations) without 630 BELL SYSTEM TECHNICAL JOURNAL error or trouble of any kind, and without requiring service attention of any kind other than normal replacement of paper and inking ribbons. For rendering service economically with teletypewriters on sub- scribers' premises, an important requirement in order that the expense of maintenance be not prohibitive is that the machine should not require maintenance attention except at very infrequent intervals. Bell System machines are designed to require routine maintenance attention not oftener than once in two months where the machine is used continuously over periods of eight hours each day. To accomplish this the problem of lubrication has required very careful attention. It has necessitated the provision of oil reservoirs in certain places and the careful selection and specification of oils and greases. Another feature making for economical ma.n c : nance is interchangeable parts. In other words, if a part breaks or wears, it is replaceable by another part of the same type without requiring fitting and usually without readjustment. At times customers wish to use teletypewriters on tables especially designed and arranged to suit the convenience of their offices. For this reason teletypewriters are designed as far as feasible to be self- contained units which can be mounted on any desk or table. All present Bell System teletypewriters employ the start-stop system of synchronizing and are well adapted for the connection of any number of machines to one circuit with facilities for rapid to and fro intercommunication among the various stations. To permit optimum control of intercommunication and interruption of the sending station when desired, a device known as the "break lock" is incorporated in many machines. This device, together with a "break" key located on each machine, provides facilities whereby any station may interrupt a station which is sending, take control of the circuit and send. The operation of the "break" key opens the line transmitting a signal which causes the "break lock" device to function at the station which is sending and automatically stop any further sending from that station until the device is manually restored. This device is very important in the case of transmission from a perforated tape, which is described later. Motor control devices are of importance for stations which are not in continuous use but which may wish to receive messages from time to time from distant stations without requiring an attendant to turn on the machine. Such devices are used both in private line and in TWX services. In the case of a private line it is often desired to have the machine normally idle with the motor stopped but so arranged that. FUNDAMENTALS OF TELETYPEWRITERS 631 when a distant station wishes to send a message, a signal may be sent which will automatically start the motor and condition the receiving machine so that it will properly record the message and then have its motor automatically stopped again at the end of the communication. Various devices are available for this purpose, some operating over the regular signaling circuit and others requiring a separate circuit. Similarly, in the case of TWX service, stations may, if desired, be equipped for unattended service so that, if the station is called and no attendant is present, the teletypewriter motor may be started remotely by the switchboard operator and the station conditioned to record the incoming message at the termination of which the motor can again be stopped by the switchboard operator. Signal bells are usually provided on the machines so that, if it is desired to call an attendant to a working machine or to call attention to a specially important message being received, the bell can be rung by signals sent over the circuit. A general feature incorporated in the design of all modern machines, and one which is not often appreciated, is the so-called "overlap." This feature makes high speed possible by overlapping the selecting and printing parts of the receiving operation. In other words it provides for the typing of one character to take place simultaneously with the reception of the selecting impulses for the next character. Features of Page Teletypewriters Page teletypewriters have been built in several different forms, notably with a moving paper carriage or a stationary paper carriage and with a typewheel or with type bars for printing. An early design employed a moving paper carriage and a typewheel, with an ink roller for inking the characters on the wheel. With this design it was im- practical to make satisfactory carbon copies, the printed record was unevenly inked, and much trouble was experienced due to side printing, that is, unwanted printing of portions of letters adjacent to the desired letter on the typewheel. Furthermore, considerable trouble was had in properly feeding paper from a paper roll through the moving paper carriage. To eliminate these limitations and troubles it was decided that for general service in the Bell System a new machine should be designed to be capable of making as many carbon copies as a typewriter and that it should use type bars and have a stationary paper carriage. This sort of machine was new in the art and required extensive development work to produce a satisfactory commercial design because of the in- 632 BELL SYSTEM TECHNICAL JOURNAL herent difficulties of, moving an automatically operated basket of typebars back and forth in front of the stationary paper. The present standard No. 15 teletypewriter was the ultimate result of this work and has proved very satisfactory in general service over a number of years. It employs a typewriter ribbon for inking, has the paper roll inside the machine cover and makes very satisfactory carbon copies with various types of paper supply without being subject to the paper feed, inking and side print troubles previously experienced. This machine has also lent itself to meeting later demands from business houses for typing either single or duplicate copies on special printed forms as commonly used in modern business practice. By equipping the platen with sprocket teeth and having feeding perfora- tions along the edges of the forms, all copies of these forms are auto- matically held in perfect registration during typing at all stations connected to the circuit. In connection with the rapid handling of these forms a further requirement for automatic tabulation has been met by providing a tabulating device which on the transmission of a certain signal causes all carriages to move over rapidly to any pre- determined position on the form and stop there for the typing of letters or figures in columns perfectly aligned. This device greatly facilitates the rapid transmission and reproduction of orders and the like on organized printed forms. With the advent of TWX service a new situation arose in which many of the machines were only infrequently used and then for very short periods to make a single copy only. To render this service economi- cally it seemed desirable to have a less expensive machine and since narrower capabilities were required this seemed entirely feasible. Accordingly a new machine known as the No. 26 teletypewriter has been developed primarily to print a single satisfactory copy although one carbon copy can be made if desired. To obtain low first cost this machine has a moving paper carriage and to secure a satisfactory printed record it employs ribbon inking and a typewheel arrangement which is a sort of cross between conventional typebar and typewheel designs. This typewheel is an assembly employing a small individual type pallet for each separate character. In the process of printing a character, a striking arm somewhat like the shank of a typebar comes forward and forces the individual type pallet against the ribbon to make an impression on the paper. The typewheel is rotated to different positions to select the different characters to be typed. In this way satisfactory inking and a clear cut impression without side print is obtained, which compares favorably with the record obtained FUNDAMENTALS OF TELETYPEWRITERS 633 on a typebar machine or typewriter. The entire machine costs appreciably less than the more comprehensive No. 15 machine. The No. 26 machine is illustrated in Fig. 7. Fig. 7 — No. 26 teletypewriter. Features of Tape Teletypewriters In the case of tape teletypewriters it is also necessary to have a clean printed record and occasionally there is a need for carbon copies. Accordingly, the tape machine standard for the Bell System is a type- bar machine using an inking ribbon and known as the No. 14 teletype- writer. It is illustrated in Fig. 8. 634 BELL SYSTEM TECHNICAL JOURNAL A feature worthy of note is that with this machine typing always occurs at the same point introducing a problem in connection with platen wear. If the platen were fed by the usual ratchet in, say, 36 steps per revolution, there would be heavy wear concentrated at these 36 points and the platen would require frequent replacement to pre- Fig. 8— No. 14 teletypewriter. serve good printing. To avoid this, the platen is fed through differ- ential gearing so that on a second revolution the typing comes in a different spot from that of the first revolution; thus the wear is uni- formly distributed over the entire circumference. One carbon copy can be made by leading tapes through the machine from two rolls of record paper and one roll of carbon paper. Two carbon copies can be made in a similar way if desired. FUNDAMENTALS OF TELETYPEWRITERS 635 The tapes employed may be either gummed on the back for con- venient pasting on blanks for filing or may be plain paper tapes if the records are of temporary interest only. Also cellophane or similar transparent tape may be used if it is desired to project the record on a screen. A tape out signal is provided on the machine so that when a roll of tape becomes nearly exhausted a bell will ring continuously to give warning of this fact. Where a bell is not desired, the last few feet of tape on the roll are painted red to give similar warning. If desired, this tape printing machine may be used on the same circuit with page printing machines such as the No. 15 teletypewriter, and when so employed is usually equipped with an "end of line indicator" to warn the operator of the approach of the end of the line in the page machine, so that suitable signals may be sent for starting a new line. Features of TWX Switchboard Operators' Teletypewriters Such machines must be small in size to permit their use in a switch- board position, quiet in operation to permit their use in the same room with a telephone switchboard and must be capable of working with any machine employed in the TWX system. To meet these requirements the standard No. 14 tape teletypewriter has been modified in several important respects as follows: 1. It has been provided with a specially designed enclosing cover which reduces the machine noise radiated by at least 5 db more than standard covers. 2. The machine is tilted so as to raise the keyboard and permit the operator to assume a more elevated position nearer the switchboard jack field. 3. It is equippent with an end of line indicator mechanism and lamp to warn of the approach of the end of a line when sending to a page teletypewriter station so that the proper signals may be sent to start a new line. 4. The usual tape feeding mechanism which pulls the tape past the typing point and obscures some of the typed message is replaced by a so-called "push feed" mechanism which acts ahead of the typing point and makes the typed message more fully visible. 5. Many of the operators' machines are provided with specially arranged power supply and governing circuits so that their motors normally run from 115 volt a-c. commercial supply but in case of a power failure can be quickly switched to run from the 130 volt d-c. telegraph battery. 636 BELL SYSTEM TECHNICAL JOURNAL Features of Monitoring Teletypewriters In connection with private wire teletypewriter service it has been found very desirable to have so-called monitoring teletypewriters in the repeater offices to facilitate testing between offices and with the subscriber stations. These machines must be adaptable to work with any subscriber's machine and to be usable for making test measure- ments on circuits. The No. 14 tape teletypewriter has also been adapted to this service. It may be equipped with an end of line indicator to facilitate communi- cation with a page teletypewriter. Also since commercial service is given at speeds of 40 and 60 words per minute, many of the monitoring machines are equipped with two-speed governors and a switch to provide for changing from one speed to the other. These machines are also usually arranged for normal operation from commercial power supply but emergency operation from the 130- volt telegraph battery. For making test measurements over circuits a special orientation scale is provided together with a small crank extending through the cover for quickly shifting the orientation setting to any desired point. With the machine carefully adjusted to be practically free of harmful distorting effects on the signals, it may then be used for measuring distortions in received teletypewriter signals, the scale being arranged to read the total distortion directly in percent of a pulse length. Tape Storage Transmission A heavy volume of traffic may be transmitted rapidly and con- veniently by the use of perforated tape. In this method a machine known as a perforator and having a keyboard like that of the teletype- writer is used for punching the code signals for the message in a strip of paper tape. This may be done with simultaneous typing of the message on the teletypewriter in which case the speed of perforating is limited to the speed for which the teletypewriter is set. If a typed record is not made simultaneously with the perforating, punched tape may be prepared at practically any speed within the capabilities of the operator. This punched tape may then be fed through a device known as a tape transmitter which automatically transmits the message signals from the tape at the maximum speed for which the teletype- writers connected to the circuit are set, which is usually 60 words per minute. The method of transmitting from perforated tape has the distinct advantage of using the line at maximum efficiency at all times as com- pared with direct keyboard sending where pauses in operating the keys FUNDAMENTALS OF TELETYPEWRITERS 637 and interruptions to the operator result in direct losses of circuit time and effectively slower transmission. Another important advantage of the perforated tape method is that errors may be corrected in the tape before transmission with the result that only errorless copy is transmitted on the circuit. This is done in the following manner and is illustrated in the section of perforated tape shown below. If the operator in attempting to write the word THE should strike the keys T and J (in error), realizing her error she back spaces the tape one division, strikes the "letters" key and then the correct keys H and E. The transmission of the "letters" signal will cause no operation in the recording teletypewriters since they are already in the "letters" case, and the word will be recorded correctly as though no error had been made. Similarly, entire words or groups of characters may be erased from the tape if desired. LETTERS FEED SPACE — ^ I ^ SPACE HOLES • • • • • • • •••• • •• •• • •••• • •• •• ••• • • FROM T HE LARGE Fig. 9 — Sample of strip of perforated tape. For TWX service a further advantage of the perforated tape method is that the entire message may be punched in tape and checked by printing, if desired, before a call is placed and a connection established. Then, when the connection is established, the message can be auto- matically transmitted at maximum speed requiring a minimum time for the connection and giving a minimum charge for the call. It is true, of course, that in this method there is some delay between perforation and transmission. For this reason short to and fro mes- sages, as required in setting up a connection, may be better handled by direct keyboard. To facilitate such working, the perforator key- board is normally arranged so that by throwing a switch this same keyboard may be used for direct keyboard sending without perforating. This switch also has an intermediate position in which the keyboard is connected for simultaneous direct sending and perforating. This provides for meeting the needs of certain TWX subscribers who wish to simultaneously type and punch the message so that the typed copy may be checked as it is perforated. The complete page printing set arranged for tape transmission is 638 BELL SYSTEM TECHNICAL JOURNAL known as the No. 19 teletypewriter set and is illustrated in Fig. 10. It employs a No. 15 teletypewriter as the page printing unit. Fig. 10 — No. 19 teletypewriter set. Automatic Retransmission Using Reperforators At times it is desirable to retransmit messages received from one circuit to some other machine or machines on a separate circuit. A unit known as a "reperforator" is often used to facilitate such re- transmission. The reperforator now standard for the Bell System is a start-stop receiving device using the 5-unit permutation code. It is somewhat similar to the receiving-only tape teletypewriter except that the record produced consists of code perforations in a tape rather FUNDAMENTALS OF TELETYPEWRITERS 639 than typing on a tape. This perforated tape is the same as tape pro- duced by a keyboard perforator as previously described, and may be used in an automatic transmitter for retransmitting the message on a separate circuit. The reperforator is usually associated with a re- ceiving teletypewriter on a circuit and may be cut in or out manually or automatically from signals transmitted along with the message signals, so that it will automatically reproduce a code tape for use in automatically retransmitting such messages as desired on some new connection. Conclusion The fundamentals of teletypewriters, as described above, now seem to be fairly well established. The future should bring simpler and cheaper machines, especially where the more difficult requirements do not have to be met, and probably additional attachments and auxiliary features to extend the applications and convenience of operation. The Dielectric Properties of Insulating Materials By E. J. MURPHY and S. O. MORGAN This article discusses the variation of dielectric constant and dielectric loss in the radio and power frequency range with the object of giving a simple picture of the type of mechanism which is able to produce anomalous dispersion in this range of frequen- cies. Some of the general characteristics of anomalous dispersion can be demonstrated as well on a simple and arbitrary model of the structure of dielectrics as on the more complex ones which corre- spond more closely to the actual structure of dielectrics. Such a derivation is given here in order to indicate the significance of the different factors which occur in the formulee which have been proposed to account for the variation of dielectric constant and dielectric loss with frequency. This enables a distinction to be made conveniently between the general characteristics which are shared by several types of dielectric polarization and the special characteristics which are peculiar to a restricted class of polariza- tions or to a particular kind of polarization. II. Dielectric Polarizability and Anomalous Dispersion IN a previous paper ^ the general features of the dependence of dielectric constant on frequency were indicated schematically for the entire range extending from the frequencies used in power trans- mission to those of ultra-violet light. In the range of frequencies below the infra-red (that is, in the electrical range of frequencies) anomalous dispersion is the rule, normal dispersion not having been observed as yet, except for piezo-electric materials, whereas at high optical frequencies normal dispersion is the predominant feature. In the intermediate infra-red region it is not surprising to find a behavior which shows anomalous and normal dispersion in more nearly equal degrees of prominence. It will be recalled that anomalous dispersion is the type of frequency- variation in which the dielectric constant decreases with increasing frequency, while normal dispersion is the reverse of this, the dielectric constant or refractive index increasing as the frequency increases. The use of the term anomalous dispersion to describe the dependence of dielectric constant on frequency in the radio and power frequency range is now widespread, and seems quite appropriate, for it brings out the point that the variation of dielectric constant with frequency in » Murphy and Morgan, B. S. T. /., 16, 493 (1937). 640 DIELECTRIC PROPERTIES OF INSULATING MATERIALS 641 the radio and power range is in certain respects the same type of phenomenon as optical anomalous dispersion. Anomalous dispersion plays a very important part in the behavior of dielectrics in the electrical range of frequencies. It is seldom possi- ble to interpret a set of measurements of dielectric constant or other dielectric properties without encountering some manifestation of anomalous dispersion or of the other characteristic types of behavior which follow as corollaries of it. The two catagories, polarizability and dispersion, include a great deal of the dielectric behavior of insulating materials. This paper will deal primarily with anomalous dispersion, but the theory of anom- alous dispersion is not entirely separable from that of the polarizations of which it is an attribute, so it will be necessary to discuss at least briefly the nature of dielectric polarization. The Relation between Polarizability and Dielectric Constant For our purposes a dielectric may be thought of as an assemblage of bound charges, where this term is intended to include the electrons and positive cores in atoms and molecules, the ions held at lattice points in ionic crystals and, in general, any assemblage of charged particles which are so bound together that they are not able to drift from one electrode to the other under the action of an applied electric field of uniform intensity. Actual dielectrics, of course, also contain some conduction electrons or ions which are free to drift through the material and dis- charge at the electrodes, producing a direct current conductivity. This conductivity is small at ordinary temperatures in materials classified as dielectrics. The positions of these charged particles may be considered to be determined by an equilibrium of forces. When an electric field is applied this equilibrium is disturbed and the bound charges are dis- placed to new positions of equilibrium; then when the applied field is removed they revert to their initial positions. In the equilibrium positions which the charges occupy when a constant electric field has been impressed on the dielectric they have a larger potential energy than in their initial positions. Moreover, they do not revert instantly to their initial positions, and when the retardation is due to friction some of the potential energy of the bound charges is dissipated as heat in the dielectric. When an alternating voltage is applied to the dielectric, we may think of the bound charges as moving back and forth with certain amplitudes, a different amplitude for each different type of bound charge. When the applied electric field is of unit intensity, the sum of the product of 642 BELL SYSTEM TECHNICAL JOURNAL amplitude and charge extended over all of the bound charges in a unit volume of the material determines the dielectric constant of the mate- rial. The energy dissipated as heat by the motions of these bound charges in the applied electric field represents the dielectric loss per second, a quantity which is proportional to the a.-c. conductivity after the d.-c. conductivity has been subtracted from it. The ima- ginary part of the complex dielectric constant is proportional to the dielectric loss per cycle. While the physical meaning of the dielectric constant and dielectric loss can be conveniently described, as above, in terms of the amplitudes and energy relationships of bound charges in their motions in an applied electric field, a more useful basis for the discussion is that provided by the concept of polarizability. In the present application the polarizability is equivalent to the product of charge and amplitude, but it has the advantage of being a quantity which is defined and dis- cussed in the general theory of electricity as well as in that of dielec- trics. The dielectric constant is then found to be related closely to the polarizabilities of the assemblages of charged particles which the dielectric contains. The polarization of an assemblage of charges is a quantity defined in electrostatic theory as the vector sum P = HCiSi, (1) where Si is the distance of the i^^ charge, d, from a point chosen as origin, and the summation is extended over all of the charges in the assemblage, for which Ci is a typical charge. (If the assemblage has no net charge (X^i = 0)- the origin may be arbitrarily located without affecting the value of p.) The polarization is a vector quantity. It can be written as the product of a scalar quantity p, which represents the magnitude or electric moment of the polarization and a unit vector pi which gives the direction of the polarization ; thus p = ^pi. As it will not be neces- sary to distinguish between the properties of isotropic and anisotropic materials in this article the direction of the polarization need not be emphasized. The notation will therefore be simplified, in general, by using the magnitude or scalar part of such vector quantities as the polarization, the electric field intensity and the displacement of charged particles. To illustrate the application of equation (1) let us consider a very simple configuration consisting of two charges + e and — e (see Fig. 1), The vector polarization of this configuration is p = e(si — S2) = ^pi, where p is the magnitude or electric moment of the polarization and DIELECTRIC PROPERTIES OF INSULATING MATERIALS 643 pi is a unit vector in the direction of the vector (si — S2). If now one of these charges is an electron {e — 4.77 X 10"^° e.s.u.) and the other a unit positive charge and they are separated by a distance of the order of magnitude of atomic distances (10~^ cm.), p has the value 4.77 X 10~^^ e.s.u., or 4.77 Debye units. The permanent electric moments of molecules seldom exceed a few Debye units. Let us now apply the definition contained in equation (1) to a dielectric material. In the first place it indicates that if we know the effective positions of the electrons and other charged particles which Fig. 1 — The calculation of the polarization vector by the general method for a verj' simple configuration. contribute to the structure of the material we can always, in principle, calculate the polarization of the body as a whole or any part of it. Actually the calculation of the polarization of a body as a whole or that of unit volume in it is in general a complicated matter involving statistical considerations, but there are special cases in which the result is rather obvious. For example, in a gas or liquid if all orientations of the molecules are equally probable in the absence of an applied field, the value obtained by taking the time-average of the summation indi- cated by (1) is zero. Equation (1) would also give the value zero when applied to all of the ions in a c.c. of a solution because any arbitrarily chosen small volume in the liquid would be as likely to contain a posi- tive ion as a negative ion. 644 BELL SYSTEM TECHNICAL JOURNAL In some crystalline materials equation (1) gives the value zero be- cause there is a suitable symmetry in the configuration of charged particles in the unit cell ; for other solids equation (1) gives a finite value for the unit cell, but zero when applied to a volume of the material large enough to contain a great many crystallites with random orienta- tions; however, there are some macroscopic crystals which have per- manent polarizations. A solid material consisting of polar crystallites with random orientations is analogous, as far as equation (1) is con- cerned, to a liquid or gas containing polar molecules having random orientations; the polarization of the material as a whole is zero in either case. CONDENSER- PLATES • POSITIVE CHARGE O NEGATIVE CHARGE Fig. 2— A dielectric in a condenser. The circles joined by a bar represent "bound charges" of various kinds, including atoms and molecules. Let us now consider a dielectric of any kind occupying the space between two plane, parallel condenser plates of great enough area and small enough separation that the electric field between the plates when they are charged may be considered to be directed normally to them (cf. Fig. 2). Consider the space between the plates of the condenser to be divided into small" cubes of the same size, the purpose of this imaginary division of the dielectric being merely to obtain a representa- tive specimen of the dielectric material. If the cube size is too small the instantaneous value of p obtained by applying equation (1) to all of the particles in a cube will vary appreciably from one cube to another ; DIELECTRIC PROPERTIES OF INSULATING MATERIALS 645 but we can then increase the size of the cubes until p is the same for each cube to a close enough approximation. The polarization in each cube is then representative of that of the dielectric as a whole, ^ and by dividing X! ^i Si for a typical cube by the volume of the cube we obtain the polarization per unit volume, which for the present will be designated as P. This quantity is a statistical mean value involving a summation over a large number of particles; its value depends not only on the structure of the material but upon the effect of thermal motions on the mean positions and orientations of the molecules or other elementary particles in the material. One of the most interesting points in dielec- tric theory is the consideration — pointed out by Debye and at the basis of his theory of polar molecules — that for some types of structure the mean positions of the particles from which P is calculated are unaffected by changes in the amplitude of thermal motions while for another type of structure (consisting of polar molecules free to assume many or at least several orientations) an increase of temperature de- creases P, because the randomness of the orientations of the polar molecules is increased. For many materials P is zero when no electric field is applied, and assumes a finite value only when an electric field is applied, though as has been indicated, some crystalline materials have a finite value of P even in the absence of an applied electric field. In either case, how- ever, the application of an electric field causes the bound charges within the dielectric to be shifted in general to new equilibrium posi- tions, corresponding to the slight change in the system of forces acting upon them, and if the material did not have a polarization before the application of the field, it assumes one; if it did, it assumes a different value of P. The value of P when an electric field E is applied will be designated as Pe, and that when no field is applied by Pq. Then Pe — Po is the polarization per unit volume induced by an applied field £. As the dielectric constant of a material depends upon the magnitude of the polarization induced in it by an applied field, and we are concerned here with dielectric constants, it will be desirable to simplify the notation by setting Pe — Po = P- This gives P a slightly different meaning than it had in the earlier part of the dis- cussion, where it represented the total polarization per unit volume whatever its origin. 2 A detailed consideration of the method of dividing a dielectric up into ele- mentary volumes in order to compute the mean polarization encounters complications which need not be discussed here. A critical analysis of the method of computing the volume density of polarization of a dielectric is given by Mason and Weaver, "The Electromagnetic Field," Chicago (1929); Chapter III. 646 BELL SYSTEM TECHNICAL JOURNAL The relation between the applied electric field, E, and the polariza- tion induced by it per unit volume is given by P = i^ E (2) 47r ^ ^ for isotropic materials. The constant (e — l)/47r is the susceptibility of the dielectric in e.s.u., and e is the dielectric constant, which is defined as C/Co, where C is the capacitance of the measuring condenser while it contains the dielectric and Co is its capacitance when empty. For some purposes there are advantages in considering the actual polarization, which is produced by a discontinuous distribution of charged particles, to be replaced by a vector point function which gives equivalent external effects. Then a vector P may be considered to be associated with every point in the space occupied by the dielectric and the dielectric may be considered to have a continuous volume density of polarization,^ P. In non-isotropic bodies the polarization vector P induced by an applied field E is not always in the same direction as E, but is assumed to be a linear vector function * of E (involving, in the general case, six independent constants), where both E and P are vector point functions. In deriving the relationship between the dielectric constant and the molecular structure of a material it cannot be assumed in general that the local field which is impressed upon the elementary particles in the dielectric is simply the field E which can be computed by dividing the applied voltage V by the distance between the plates of the condenser, the intensity of the field being assumed to be uniform. For there is an interaction between the molecules of the dielectric such that each mole- cule exerts a force on every other molecule. In the absence of an applied electric field these forces combine with other influences to create a distribution for which the polarization per unit volume has the value Po (frequently zero, as has been mentioned). Then when a field is applied each element of volume in the dielectric is put into a polarized condition and in general the forces which it exerts upon the particles in other volume elements changes, because the charges in each volume element have been displaced to new positions. Conse- quently, the value assumed by P in a given cube of Fig. 2 will depend not only upon the direct action of the charges on the plates of the condenser — which determines the strength of the field E — but also * Cf. Mason and Weaver, loc. cit. Chap. III. * Cf. P. Debye, "Polar Molecules," Chemical Catalogue Co., New York (1929), pp. 32-35. DIELECTRIC PROPERTIES OF INSULATING MATERIALS 647 upon their indirect action through the polarization which they create in other elements of volume. The contribution which the polarization of the dielectric makes to the force upon a charged particle in it has been calculated by Lorentz to be (47r/3)P, where P is the polarization per unit volume induced by the applied field. This calculation applies to an array of particles with cubic symmetry and to isotropic materials.^ The internal or local field F is then given by F = E + ^P. (3) E may be thought of as the force which has its origin in the direct interaction between the charges on the plates of the condenser and the charges in the polarizable complex on which attention has been fixed (such as one of the cubes of Fig. 2), while the term (47r/3)P may be regarded as an indirect force coming from the other parts of the dielec- tric by virtue of their polarized state. It is assumed in the theory of dielectrics that the structure of mate- rials is such that P is a linear function of F (or a linear vector function in the case of anisotropic materials) ; then P = kF, (4) where k is the polarizability per unit volume. It can be seen that E 1 - Ak (4a) where A = Air 13, and consequently that the relation between the polarizability k and the susceptibility (e — l)/4x (= K) is whenever (3) is a valid expression for the internal field. The susceptibility can be calculated without presupposing the validity of equation (3) for the internal field, while the value of k depends upon whether (i) or some other expression gives the strength of the internal field in the dielectric. If L is the number of molecules per cubic centimeter, kjL{= a) is the polarizability per molecule. This molecular constant a is called the polarizability of the molecule. By multiplying a. by Avogadro's number iV, we obtain the polarizability per mole of the dielectric: 6H. A. Lorentz, "The Theory of Electrons," p. 138, and Notes 54 and 55. 648 BELL SYSTEM TECHNICAL JOURNAL Na = Nk/L. And if m is the mass of a molecule, Nm = M, where M is the molecular weight, and Lm = p, where p is the density; so that L p and the polarizability per mol may be written as Mkjp. From equations (3) and (4) (or 46) it can be shown that the polariz- ability is related to the dielectric constant by the familiar relation * = .- 47r ^A (5) which however is only valid when (3) is valid — and for some materials (3) is apparently not valid. For gases the term (47r/3)P in (3) is so small as compared with E that F is approximately equal to E and i=ir = 4^. (6) The polarizability and susceptibility are then equal. The physical reason for this is that the ratio of intermolecular space to the space occupied by molecules is much larger in a gas than in a solid or liquid and the direct force exerted by the charges on the condenser plates on a charged particle in the dielectric is then much greater than the in- direct force which they exert through the polarization induced in other molecules. It is customary to call the quantity (47r/3)^ the volume polarization, and it is often denoted by the letter p. The volume polarization may be thought of as 4r/3 times the polarization induced in the dielectric per unit volume per unit applied field. The convenience of using (4irl3)k instead of k comes from the occurrence of the factor 47r/3 in the relation (5) between dielectric constant and polarizability. On dividing equation (5) by the density we obtain a quantity which is called the mass polarization, as it is 47r/3 times the polarizability per gram : 1 li + 2 *i-- (7) 6 p And on multiplying (7) by the molecular weight of the material we obtain M p - 1 47r M , 47r kN 47r,, ,„, DIELECTRIC PROPERTIES OF INSULATING MATERIALS 649 The quantity {4:T/3)Na is the molar polarization, Na being the polariz- abiUty per mole.^ Equation (8), and also (7), expresses the Clausius-Mosotti relation when a is considered to be a constant characteristic of the individual molecule and independent of density. The function of e on the left- hand side of (8) is independent of density whenever a is independent of density. The following relation, analogous to that of Clausius and Mosotti but expressed in terms of the refractive index n, was derived by Lorentz and by Lorenz : p n^ -\- 2 S The left-hand member of this equation is called the molar refraction. Equations (8) and (8a) are equivalent because of the general relation between refractive index and dielectric constant (n^ = e), but owing to the fact that refractive indices are measured at optical frequencies the molar refraction contains only the electronic part of the total molar polarization of the material. Subtracting the molar refraction from the total molar polarization, is one of the methods of determining the amount of polarization contributed by non-electronic polarizations. It has been found that the Clausius-Mosotti relation is not equally satisfactory for all kinds of dielectric polarization. It gives good results when applied to electronic and atomic polarizations. For example, in an interesting paper on materials of high dielectric con- stant, Frank ^ has recently shown that the Clausius-Mosotti-Lorentz- Lorenz relationship aids materially in explaining the behavior of the dielectric constants of crystalline materials of high dielectric constant where the dielectric constant depends upon electronic polarizations. Where the polarizability of a molecule is the sum of the polarizabilities of the atoms of which it is composed it is to be expected that if the relation (5), or (8) or (8a) is valid the sum of the atomic polarizations would be equal to the molar polarization. Experimental agreement ^ The polarizabilities of non-polar molecules and atoms are usually of the order of magnitude of lO"^'* c.c, and the molar polarizations of such substances, conse- quently, are of the order of magnitude of a few c.c, since the molar polarization is (47r/3) X 6.06 X 10^3 times the polarizability of the individual molecule. The polarizability of a conducting sphere is equal to the cube of its radius. And, as atomic dimensions are of the order of magnitude of 10~* cm., it is evident that the polarizabilities of atoms tend to be of a similar order of magnitude to the polarizabil- ities which would be expected if they behaved as conducting spheres, though there are large differences in the ratio of polarizability to volume for different atoms. The molar polarizations of polar molecules are in general larger than those of similar non- polar molecules and may be a few hundred c.c. (Cf. P. Debye, "Polar Molecules," pp. 12-19.) 7 F. C. Frank, Trans. Faraday Society, 23, (4), 513 (1937). 650 BELL SYSTEM TECHNICAL JOURNAL with this requirement has been found in optics where the refractive indices of molecules can be calculated approximately from the molar refraction (eq. 8a) obtained by adding the atomic refractions. This additive property of electronic polarizations has been employed by Frank ^ to interpret the tendency of crystalline materials hav- ing high dielectric constants to be characterized by a high polariz- ability/volume ratio for the atoms or ions of which they are composed. This condition would tend to allow the largest number of highly polarizable particles to be concentrated in a given space, giving, on the additivity rule, a high molar polarization and a high dielectric constant. On the other hand Wyman ^ has pointed out that the Clausius- Mosotti relation is not satisfactory when applied to highly polar liquids, such as water, and has found that for these substances it appears to be more satisfactory to consider that the polarization is related to the dielectric constant by the empirical relation ^+^ ^""k. (9) 8.5 3 The calculation of the internal field by Lorentz, which provides the theoretical basis for equation (8), was made before the theory of polar molecules had been developed, but equation (8) has since been applied tentatively to polar molecules."* The problem of obtaining an im- proved relationship between polarizability and dielectric constant for materials having molecules with permanent electric moments has been studied in recent years by several investigators. ^^ The calculation of the internal field usually involves the assumption that the efifect of the molecules included in a small sphere surrounding the central molecule on which the force is being calculated is negligible on the average because of the random motions due to thermal agitation. On the supposition that such an assumption is not justified in a polar material because of the interactions of adjacent polar molecules, Onsager ^^ has obtained a relation between polarizability and dielectric constant which for high dielectric constants is nearly the same as Wyman's empirical relation, equation (9). A comprehen.sive study of the efifects of interaction between the dipoles of polar molecules has 8 Lqp Clt ^Cf. Wyman, Jour. Amcr. Chein. Soc, 56, 539 (1934); 58, 1482 (1936). i» Cf. Debye, loc. cit., p. 13. " Cf. Onsager, Jour. Anier. Chem. Soc, 58, 1486 (1936); Van Arkel and Snoek, Trans. Faraday Soc, 30, 707 (1934); Wyman, Jour. Amer. Chem. Soc, 58, 1482 (1936); Van Vleck, Jour. Chem. Physics, 5, 320 (1937) and 5, 556 (1937). '^ Loc. cit. DIELECTRIC PROPERTIES OF INSULATING MATERIALS 651 been made by Van Vleck by the methods of statistical mechanics. He obtains an expression which agrees to a second approximation with that obtained by Onsager. Thus it seems that for highly polar liquids the relations between polarization and dielectric constant developed by Onsager, Wyman and Van Vleck may be more satisfactory than the Clausius-Mosotti relationship, though for many other materials the Clausius-Mosotti relationship is apparently valid or approximately valid. In deriving expressions for the dependence of dielectric constant on frequency later in this article the formulae obtained will naturally depend upon which of the equations, (5), (6) or (9), is taken as the relationship between polarizability and dielectric constant. The alternative expressions will be listed. Derivation of a Dispersion Formula The above-described relations between polarization and dielectric constant provide the means of obtaining expressions for the variation of dielectric constant with frequency when we have determined the dependence of polarizability on frequency. As our object is to exhibit the general features of anomalous dispersion shared by several par- ticular types of polarization, it will be sufificient to derive dispersion formulae containing constants the values of which are not specified, but which have a sufficiently obvious physical significance. The derivation given will parallel that of Lorentz in deriving a formula for optical dispersion, ^^ and in fact is simply a special case of it in which certain terms are considered to be negligible by comparison with others. An analogous procedure was used in one of the earliest attempts to explain anomalous dispersion in the electrical frequency range, the theory proposed by Drude '* in 1898. This theory was based upon the hypothesis that anomalous dispersion in the electrical frequency range depends upon a mechanism similar to that to which optical dispersion was attributed, the difference being that the particles which produce anomalous dispersion in the electrical frequency range are so large that some of the terms in the optical dispersion formula can be neglected. The formula which Drude derived for electrical anomalous dispersion yield the same form of variation of dielectric constant with frequency as do the generally accepted theories of the present time, such as the Debye theory; the differences lie in the expressions given for the con- "H. A. Lorentz, "The Theory of Electrons," Chapter IV. See also Korff and Breit, Reviews of Modern Physics, 4, 471 (1932), where a review of the classical theory of optical dispersion is given. " P. Drude, Ann. d. Physik, 64, 131 (1898), " Zur Theorie der anomalien elek- trischen Dispersion." 652 BELL SYSTEM TECHNICAL JOURNAL stants in the formulae in terms of properties of the material. Another adaptation of optical dispersion theory to the explanation of dispersion in the electrical frequency range was proposed by Decombe ^^ in 1912. He employed the Lorentz electron theory for the dispersion of light as a basis for the consideration that if the environment of some of the electrons in dielectrics is suitable their motions in an applied field could produce anomalous dispersion and dielectric loss in the electric fre- quency range. A similar simple and arbitrary assumption regarding the structure of dielectrics will also be employed here. However, it is not proposed as a theory of dielectric behavior but merely employed as a comparatively simple means of deriving and discussing relationships which can be demonstrated as well on a simple and arbitrary model as on the more complex ones which correspond more closely to the actual structure of dielectrics. The relation of the constants in the dispersion formulae which will be derived here to the actual structure of dielectrics will only be indicated in a general qualitative way for the purpose of illustrating the physical nature of the processes involved; no attempt will be made to provide expressions for the dispersion constants in terms of other observable properties of the material. In Fig. 2, let the applied potential be V, where V may vary in general in any way with the time, though in the present discussion it will be considered to vary sinusoidally with the time; the impressed field strength is then given by E — V/d. As in the more general discussion which preceded this, it will be assumed that the imaginary cells pictured in Fig. 2 contain large numbers of polarizable complexes consisting of positive and negative charges in equal numbers held in position by constitutive forces^ — the origin of which need not be specified for our present purposes — such that if they are displaced a distance 5 from their initial positions they will experience a force fs, where /is a constant, tending to restore them to their initial positions; and that while these charges are in motion as a result of the action of the impressed field they experience a frictional force rs, where r is a constant and s is the velocity in the direction of the impressed field: and, finally that their motion is also retarded by an inertia reaction ms, proportional to the mass m and the acceleration s of the particles. The equation of motion for any typical charge e in a polarizable complex having the above-described specifications is ms -f rs + /s = eF, (10) where F is given by equation (3) in materials to which the Lorentz calculation of the internal field applies, by F = E in the case of gases 1^ L. Decombe, Journal de Physique, (5), 3, 315 (1912). DIELECTRIC PROPERTIES OF INSULATING MATERIALS 653 and by other expressions — which in some cases may approximate either to F = E or to F = E + (47r/3)P — for still other materials. The quantities F and 5, are vectors, but for isotropic materials 5 is in the same direction as F. If, following the method employed by Lorentz, we write an equation of the form (10) for each charged particle in a physically small volume 8 (such as the cubes of Fig. 2), multiply each equation by e, add the equations for all of the particles in 8, and divide by the volume 8, we obtain mP -{- rP +fP = ne^F, (11) where P = (l/5)X!e5 and n is the number of charged particles charac- terized by the constants m, r and / per unit volume. The volume 8 may be considered to be that of one of the cubes in Fig. 2. As indi- cated earlier it should contain a sufftcient number of molecules to give a good mean value for P, the polarization per unit volume, but at the same time it should be small enough not to mask significant spatial variations in P. \\'hen the impressed field E is varying sinusoidally with the time at the frequency aj/27r, the local or internal field F tending to displace each charged particle in the dielectric will also vary sinusoidally with the time, though in general out of phase with E, if F is given by equation (3), and can be considered to be given by the real part of Foe'"'. Under these conditions P = kFoe"^' solution of equation (10) for the steady state provided that k^j-. ""'' ,.,,■ (12) k is the polarizability per unit volume and is a complex quantity, since the term zVco in the denominator is an imaginary {i = V — 1). Equations (10), (11) and (12) apply to a dielectric having a single type of polarization characterized by the constants /, r, m, n and e. But in general an applied field induces several types of polarization simultaneously in a dielectric, and if we assume that it induces w types which are independent of each other, the total polarization per unit volume is given by P = kiF + k.F -\- • • • KF. (13) The total polarizability is then the sum of the individual polarizabili- ties, or w k = Z k,. (14) 3=- 1 is a 654 BELL SYSTEM TECHNICAL JOURNAL In this discussion it will be sufficient to consider that the different types of polarization designated by ki, k^ - • • kw differ from one another only in having different sets of values for the constants of equation (12), designated by the subscripts 1,2,3 • • • w; for example, the character of the polarizability k-i is specified by the set of constants mi, ri,/i and «i. In the first place it is evident that when the frequency of alternation of the voltage applied to the dielectric lies in the radio and power range it is possible to select any number of sets of values of m, r,/ which will make the terms ww^ and rw negligible in comparison with / in the denominator of (12). Let mi, ri, /i be an example of such a set of constants and let there be «i particles per unit volume to which these constants apply. Then for this type of polarization equation (12) reduces to ^i-¥- (15) This type of polarization is independent of frequency and will be referred to as an instantaneous polarization or an optical polarization. The main representatives of the instantaneous or optical polarizations are the electronic and atomic polarizations, which experience dis- persion in the visible and infra-red but which are independent of fre- quency in the electrical range, and the contribution of this polariz- ability to the dielectric constant is therefore frequently calculated from refractive index measurements. A second type of polarization results if we assume that the dielectric we are considering contains a class of particles for which mo:'^ in equation (12) is negligible by comparison with rw and with/, but in which rw is of the same order of magnitude as/ in the electrical range of frequencies. Let W2, ^2, /2 be a typical member of this class, the number of such particles per unit volume of the dielectric being n-i. Then for this class of particles equation (12) becomes {tr2w + /?) This expression represents the type of variation with frequency to which the name anomalous dispersion is given, and in the preceding paper the type of polarization which produces it was called an absorp- tive polarization. It can readily be seen also that neglecting the ni's term in (10) or the mP term in (11) leads to the same expression for k, i.e., equation (16), as does neglecting the mw^ term in the denominator of (12). So for any member, (j^, ji, n^, of the class of particles which produces DIELECTRIC PROPERTIES OF INSULATING MATERIALS 655 anomalous dispersion, equation (10) reduces to r,s-^f,s = eF (17) and equation (11) becomes r2p +hP = n,e'F. (18) Decombe's theory, which has been mentioned earlier, was based upon an equation equivalent in most respects to (18), while Drude's expres- sions for dispersion were obtained by a method equivalent to neglecting wco^ in (12). Each term in equations (17) and (18) has an evident dynamical significance. Consequently, a physical picture of the essential nature of the anomalous dispersion process is given by equations (17) and (18) even though the values of constants r2,/2, «2 and e are not specified in terms of independently measurable properties of the dielectric. Thus the term f2S represents a restoring force tending to return the particles displaced by the impressed field to their initial positions, the constant /2 acting as a stiffness coefficient; the term r2S acts as a fric- tional force, r being a measure of the friction experienced by, for exam- ple, a moving ion or a rotating polar molecule; and, finally, eF is the driving force tending to displace a particle of charge e. Evidently conditions which are sufficient to produce anomalous dispersion exist whenever the motion of charged particles in an applied field is suffi- ciently specified by considering the effects of a restoring force pro- portional to the displacement of the typical particle and of a frictional force proportional to the velocity of the particle in the direction of applied field, as in equation (17). Or, putting it in more general terms, we may say that anomalous dispersion occurs whenever the relation between the polarization per unit volume and the force due to the internal electric field is given by an equation which can be reduced to (18). However, the possibility that anomalous dispersion may also occur under conditions which cannot be described by equation (18) is not excluded by the considerations given here. A third type of polarization which can be obtained by selecting suitable sets of values for the constants of equation (12) is that in which none of the terms in the denominator of (12) can be neglected in the electrical range of frequencies. Let ks be the polarizability for this type of polarization which can then be represented by affixing the subscript 3 to the constants m, r, f and n of equation (12). This type of dispersion includes both the normal and the anomalous types but, as has already been indicated, in the radio and power ranges of fre- 656 BELL SYSTEM TECHNICAL JOURNAL quency examples of a dispersion of this kind have not as yet been observed in dielectrics which are not piezo-electric.^® It follows then that dielectrics behave as though the inertia of the particles which contribute to dielectric polarization is small enough that the inertia reaction mor can be neglected in the electrical frequency range. This is an empirical result; the possibility of a polarization of the type kz occurring in the electrical frequency range is not excluded by the general theory of dispersion. The higher the frequency of an im- pressed field the greater should be the likelihood of encountering the type of frequency-variation described by kz (or equation (12)), because the prominence of the moP' term increases with the square of the frequency. The preceding discussion shows that we can write equation (14) in the form k = k, + ka, (19) where k is the total polarizability, ki is the sum of the instantaneous polarizabilities and ka the sum of the absorptive polarizabilities, that is, of the polarizabilities which vary with frequency according to equation (16). If for simplicity we take the case in which the dielectric has only one representative of ki and one of ka, we obtain by substituting the values of ki and ka given respectively in (15) and (16), k = ^ + ,. ""', (20) /i (^^2W + /a) as an expression for the total polarizability. Defining r' by t' = r/f, and dropping the subscripts in (20) to make the notation simpler, we obtain K Ki ~\ p 1 1 + iu^r' (21) which is the total polarizability per unit volume for a dielectric having two types of polarization, the one represented in (21) by the instan- taneous polarizability ki and the other by the absorptive polarizability 1'^ Piezo-electric crystals such as quartz and Rochelle salt form exceptions, but for them dielectric polarization is coupled to macroscopic mechanical strains in the material and the mass reactance is due to the flexing or extension of the entire crystal. The dielectric constant of such a crystal as measured in almost any direction, shows an increase with increasing frequency, followed by anomalous dispersion. This is the behavior required by equation (12), or rather by an equation for the dielectric constant derivable from equation (12). This dispersion, however, depends upon the size and shape of the crystal, the nature of the electrodes and the manner of supporting the crystal during the measurements, and the exact interpretation of such measure- ments is a rather complex procedure. See, for example, W. P. Mason, Proc. L R. E., 23, 1252-1263 (1935). DIELECTRIC PROPERTIES OF INSULATING MATERIALS 657 specified by the second term on the right. The quantity t' is called the relaxation-time. On multiplying the left-hand side of equation (21) by (47r/3)(M/p) and the right-hand side by (4x/3)(7V/L) we obtain Air Mk _ ^ { ki .,, ne^ { 1 fL\\^ ic (22) which is the molar polarization. For dielectrics to which the Clausius-Mosotti relation applies, equation (8) shows that ^ir Mk M e- \ 3 p p € -\- 2 (22a) and in fact the expression on the right-hand side of (22a) is frequently called the molar polarization. Reference to equation (6) shows, how- ever, that for gases (22a) reduces to the simpler relation. 4^Mk M(e- 1) TV^7~1 — ^^^^^ And for Wyman's relation between dielectric constant and polariz- ability, which has been discussed earlier, the molar polarization be- comes 4TAIk M e -f 1 3 p p 8.5 (22c) Equations (22a), (22b) and (22c) are not the only relations between dielectric constant and molar polarization which have been proposed, but they apparently cover moderately well many of the conditions met in practice. For the right-hand member of equation (22) can be substituted whichever of the three expressions (22a), (22b), (22c) seems the most suitable for the type of dielectric under investigation. If in equation (21) w is set equal to zero we obtain the zero-frequency (or static) polarizability k^ = ki + ne'^lf (23) and if w is set equal to infinity we obtain ^oo = ki. (24) Subtraction gives ko — k^ — ne^lf. (25) Substituting (24) and (25) in (21) gives ko - TTi k = k^-^( ^\ .^'^, ) • (26) 658 BELL SYSTEM TECHNICAL JOURNAL The constants ne^ff and ki are not present in (26), being replaced by two special values of the polarizability, the zero-frequency value and the infinite-frequency value. However, it is not the polarizability but the dielectric constant which is directly observed in measurements on dielectrics, so it is desirable to replace ^o and ka, by their equivalents in terms of the dielectric constant. But, as the earlier discussion has indicated, the relation between dielectric constant and polarizability is different for different types of dielectrics; three alternative expressions analogous to (22a), {22b) and (22c) will therefore be derived. For materials to which equation (22a) (or the equivalent and simpler relation (5)) applies ^0 — ^oo = 60- 1 eo + 2 r + 2. 9( eo Coo) 47r(eo + 2){e^ + 2) (27) where eo is the zero-frequency dielectric constant and eco is the infinite- frequency dielectric constant. Then equation (26) can be replaced by 47r €0 1 + 2 + eo- 1 60 + 2 -f 2 J 1 + icor' (28) By rationalizing and using the second expression given for ^o — ^od in equation (27) we can write equation (28) in the alternative form 47r^ + 2 + 3(60 .) (60 + 2)(6oo + 2) — i 1 1 + coV' 3(€o — 6oo) (eo + 2)(€^ + 2)J l-fcoV' (29) Equation (29) is the complex polarizability per unit volume multiplied by the factor 47r/3 and expressed in terms of observable values of the dielectric constant and the relaxation-time t'. The relaxation-time can also be expressed in terms of the reciprocal of a special value of the frequency; this permits all of the theoretical constants such as ne^/f and t' to be replaced by certain special values of the dielectric constant and a critical value of the frequency, A simpler expression for the polarizability is obtained in the case of gases, or whenever equation (6) gives the relation between polariz- ability and dielectric constant. Equation (26) then gives 4.Trk 1 - 1 + 60 — 6o (30) 1 + iur' And for materials to which the relation (cf. equation (9)) proposed by DIELECTRIC PROPERTIES OF INSULATING MATERIALS 659 Wyman applies the procedure followed above yields On multiplying equations (29), (30) and (31) by Mjp three alterna- tive formulae for the molar polarization of a dielectric having polariza- tions of the type specified by equation (21) are obtained ; the constants in these formulae include only special values (eo and fa,) of the dielectric constant and the relaxation-time, all of which can be obtained from dispersion curves. The quantity ko — ^00 is a constant of the material, which, as equa- tion (26) shows, represents the largest value which the absorptive part of the total polarizability, i.e., the ka term in (19), can have for a given material ; it may be described as the zero-frequency or static value of the absorptive part of the polarizability. Evidence as to the nature of a polarization can be obtained by investigating experimentally the dependence of (^0 — k^)lp on temperature; for example, if the polariza- tion is due to the changes of orientation of polar molecules according to the Debye theory this quantity should increase linearly with the reciprocal of the absolute temperature. It is useful, therefore, to express (^0 — ^co)/p in terms of observable values of the dielectric constant so that it may be plotted against temperature. In this con- nection there is, however, the same complication which has appeared in other places in this discussion regarding the relation between dielec- tric constant and polarizability. The three relations which have been discussed here yield for (^0 — ^co)/p the following expressions: 3 r eo - 1 600-1 (^0 - U/P = 47rp [ eo + 2 ecx, + 2 (Clausius-Mosotti) (32a) (for gases) (32^) (Wyman). (32c) 47rp \ 3 — 3 / eo — €0 ~ 47rp \ 8.5 The Complex Dielectric Constant As the dielectric constant (e) is the quantity directly measured in experimental investigations it is desirable to determine how it should vary with frequency for the type of dielectric polarization described in equation (21) or (29). Solving equation (5) for e we obtain 1 +8^^ e = —. i^i) l-4f. 660 BELL SYSTEM TECHNICAL JOURNAL By substituting the expression for 4(7r/3)^ given in equation (28) into (33) we obtain + t eo + 2 + 2 eo + 2 €oo + 2 (34) eo eo + 2 Coo , Eo + 2 \ Coo + 2 eo 1 eo + 2 1 + t p^ COT Coo + 2 Then, by setting we obtain + 2 , + 2^ = T, e = (l +i — cor eo / eo 1 + iiiiT (34a) (35) (36) and transforming this into polar form to faciHtate division gives £ _ Plg"^l ^ pl eo P2e'*'2 P2 where Pl = cos (v?! — v'2) + ^sin (v?! — (^2) 1 + I I w-r- eo P2 = [1 + coV2]s (37) ^Pl = tan~^ — cor and ) and between t' and t (fco - feco) t' ^, • 1>T . , • 13 , .• 3 3(€0 — €co) €„, "t" 2 Llausius-Mosotti Relation — - • -. , ^. . ; — — - ■ ■ — r 47r (€0 -h 2)(ecD -|- 2) to + 2 Gases —- • (to — «co) t 47r ■2 Wyman's Empirical Relation -— • — "" t 47r 8.5 and T in Table I. The resulting expressions can then be substituted for (^0 — ^co) and r' in equation (26) yielding expressions for the polariz- abilities of the different types of polarization listed in Table I. The molar polarization can then be obtained by multiplying (26) by (47r/3)(M/p). However, in general it is not likely that any useful purpose is to be served by calculating the molar, polarization for inter- facial polarizations; a more significant quantity would be the polariza- tion per conducting particle, when the polarization is of the type {M), Table I, and the number of conducting particles per unit volume can be estimated. We have pointed out that a number of theories which have been pro- posed for the explanation of the variation of dielectric constant with frequency may be expressed in the forms (41a) and (416) when the expressions listed in Table I are substituted for (eo — eco) and r, but we have not yet indicated how these formulae agree with experimental data. For such materials as ice (see Fig. 3, for example) and for 666 BELL SYSTEM TECHNICAL JOURNAL certain alcohols and glycols, the experimental points agree fairly closely with the curves obtained by plotting equations (41a) and (416) for a suitable choice of the values of the constants. But for many other dielectrics, particularly non-homogeneous systems or disperse systems such as those listed under Item 3, Table I, the simple dispersion formulae (41a) and (416) often fall very far short of adequately representing the experimental data. Von Schweidler ^o and Wagner ^^ have attempted to explain the form of dispersion curves obtained for such materials by postulating that the polarizations ^ 70 =v ^, \ s \ e" \ y \\ \, \ \ \ \ \ \ \ -12.0' ^ \ *- -7.1 = DEGREES ' 1 CENT. ^ \ /> < \ s > \ \ \ \ \ • • • t / / \ \ > \ \ '\ / \ V X \cS> \ y \ V \\ \ V V \ \ ^v \ \ > N s. V ^ \^ -"^ •V N. ~ >4.3 >. ^ "^^^ ^^ ;^ ^ ^^ ^V. 1 — ,~ - — — —T ■■»-~ ^. _ 50 100 500 1000 5000 10,000 FREQUENCY IN CYCLES PER SECOND Fig. 3 — Experimental dispersion curves for ice. 100,000 induced in the dielectric have a wide range of relaxation times at any given temperature, instead of a single relaxation time, as for the polarizations listed in Table I. A further contribution to the theory of the distribution of relaxation times has recently been made by Yager .^^ However, in spite of the existence of many materials which do not show the type of dispersion described by (41a) and (416), the value of these formulae in interpreting experimental data is consider- able, particularly as applied to pure materials. Table I emphasizes the point that mere agreement of experimental data for dielectric constant and dielectric loss with the theoretical 20 E. V. Schweidler, Aym. d. Phys., (4) (24), 711 (1907). 21 K. W. Wagner, Archivf. Elektrotechnik, 2, 371 (1914j. 22 W. A. Yager, Physics, 7, 434 (1936). DIELECTRIC PROPERTIES OF INSULATING MATERIALS 667 curves obtained by plotting (41a) and (416) for suitably adjusted values of the constants only places the type of mechanism to which the observed dispersion can be attributed within the rather large catagory which includes at least the seven types of mechanism listed in the table. Data showing the dependence of (^o — ^co)/p on temperature allows a further specialization of the processes which could account for the observed behavior; and of course a number of possibilities can be discarded on general grounds of physical improbability . And finally, agreement of the constants calculated from dielectric measure- ments with the values calculated from independent estimates of the sizes and other characteristics of the molecules or other elementary units which contribute to the polarization provides the most convincing evidence of the nature of the polarization. Such agreement is fre- quently obtained in the application of the Debye theory to gases and liquids. The characteristics which can be deduced from equations (41a) and (416) without substituting for the constants theoretical expressions, such as those given in Table I, are of considerable value in interpreting electrical measurements upon dielectrics. It may be convenient to describe these as the general characteristics of anomalous dispersion, distinguishing them thereby from the special characteristics peculiar to particular kinds of dielectric polarization which share the property of producing anomalous dispersion in the radio and power range of frequencies. Appendix The following list contains the definitions of the quantities which appear in Table I : ei, €2, 7i, 72 are respectively the dielectric constants and conductivities of two materials designated by subscripts 1 and 2, the unit of conductivity being such that 7 = 367r X 10'^ X, where X is in (ohm-cm)~^ eo, Coo are respectively the dielectric constant at the lower and upper extremities of dispersion curves; they are called the zero-frequency (or static) dielectric constant and the infinite-frequency dielectric constant, L is the number of molecules per unit volume, y] the viscosity of a liquid containing polar molecules, k Boltzmann's constant, T the absolute temperature, IX the permanent electric moment of a polar molecule, 668 BELL SYSTEM TECHNICAL JOURNAL a the radius of a polar molecule, assumed to be spherical, h the radius of a colloidal particle, d the thickness of a conducting skin on the particle of radius h, r a frictional resistance coefficient of unspecified origin, / an elastic restoring force coefficient of unspecified origin, n the number per unit volume of elementary charged particles subject to certain specified conditions, p the ratio of the volume occupied by the spherical particles in (3c, d, e) to the total volume, C\ the capacity of the blocking layer of (36), Table I, R the resistance of the dielectric, exclusive of the blocking layer. The following list contains the definitions of quantities which appear in other parts of the article. CO is 1-K times the frequency of alternation of the applied field, V the applied voltage, R the intensity of the applied field, P the polarization per unit volume induced by a field £, F the internal or local field, p the density of the dielectric, M the molecular weight of the material of which the dielectric is composed, m the mass of a molecule; in another context, the mass of any charged particle considered in the discussion, N is Avogadro's number, 6.06 X 10'^ molecules per mole, 5 the displacement of a charged particle from an equilibrium position by an applied field, k the velocity of the charged particle in the applied field, s the acceleration of the particle in the applied field. References Relating to Table I 1. P. Debye, "Polar Molecules," New York (1929). 2. P. Drude, Ann. d. Physik, 64, 131 (1898); L. Decombe, J. d. Physique (5) J, 215 (1912); and the present article. 3. K. W. Wagner, Chap. I of Schering's "Die Isolierstoffe der Elektrotechnik," Springer, Berlin (1934). 4. A. Joffe, "The Physics of Crystals," New York (1928). 5. K. W. Wagner, Arch. f. Elektrotechnik, 2, 371 (1914). 6. J. B. Miles and H. P.'Robertson, Phys. Rev., 40, 583 (1932). 7. A. Geniant, "Die Elektrophysik der Isolierstoffe," Berlin (1930). General reviews of the theory of dielectric behavior as it concerns dispersion for power and radio frequencies are included in the following places, among others: 1. E. Schrodinger, " Dielektrizitat," Graetz, Handb. d. Elek. u. d. Magn., Leipzig (1918), pp. 157-229. DIELECTRIC PROPERTIES OF INSULATING MATERIALS 669 2. E. V. Schweidler, " Die Anomalien der dielektrischen Erscheinungen," ibid., p. 232 ; Ann. d. Phys. (4) 24, 711 (1907). 3. J. B. Whitehead, "Lectures on Dielectric Theory and Insulation," McGraw-Hill (1927). 4. L. Hartshorn, Jour. I. E. E., 64, 1152 (1926). 5. P. Debye, "Polar Molecules," Chem. Cat. Co., New York (1929). 6. Schering's "Die Isolierstoffe der Elektrotechnik," Springer, Berlin (1924). 7. A. Gemant, "Die Elektroph)sik der Isolierstoffe," Berlin (1930). Abstracts of Technical Articles from Bell System Sources Electron Microscope Studies of Thoriated Tungsten} Arthur J. Ahearn and Joseph A. Becker. Many past experiments have shown that the thermionic activity of a thoriated tungsten filament is deter- mined by the concentration of thorium on its surface. This concentra- tion is in turn determined by the rate of arrival and rate of evaporation of thorium. Typical published values of these rates are given in Fig. 1. An electron microscope used to obtain electron images of thoriated tungsten ribbons is described. Comparison with photomicrographs shows that the active and inactive patches composing an electron image agree in size, shape and number with the exposed grains of the tungsten. The electron microscope shows that thorimn comes to the surface in ''eruptions'' at a relatively small number of randomly located points. From a comparison of photomicrographs showing thoria globules and electron images of thorium eruptions, it is deduced that all the thorium in a globule comes to the surface when an eruption occurs. Factors such as a high temperature flash and sudden heating and cooling of the filament affect the frequency of eruptions. Thorium eruptions are the only observed manner in which thorium arrives at the filament surface. They are repeatedly observed in the early stages of thoriation. Erup- tions are not observed in the later stages of thoriation where con- ditions are unfavorable for their observance. Electron images of a Pintsch single crystal filament reveal alternate active and inactive bands parallel to the filament axis. Thorium eruptions occur only on the active bands. With a polycrystalline ribbon the surface migration of thorium from the eruption centers is isotropic; with a single crystal ribbon there is a strongly preferred direction of migration. X-ray analysis shows that the surface is a (211) plane and that the preferred direction of migration agrees with the (111) direction in this plane. During the process of thoriating a filament the relative emissions from different grains change by substantial amounts; in many cases the change is so great that the relative emissions are reversed. Measure- ments of work function differences between grains gave values ranging up to 0.6 volt. The Mechanism of Hearing as Revealed through Experiment on the Masking Effect of Thermal Noise.^ Harvey Fletcher. In an electri- 1 Phys. Rev., September 15, 1938. ''Proc. Nat'l. Acad. Set., July 1938. 670 ABSTRACTS OF TECHNICAL ARTICLES 671 cal conductor there is a statistical variation of the electrical potential difference between its two ends which is due to the thermal agitation of the atoms, including the electrons. This electrical noise is amplified by means of a vacuum tube amplifier and then converted into an acoustical noise by means of a telephone receiver held on the ear. When this noise is present it reduces the capability of the ear to hear other sounds. The intensity per cycle of the acoustical noise compared to the intensity of a pure tone which can just be perceived in the pres- ence of a noise was determined experimentally using a group of ob- servers. This relative intensity for a given frequency range was constant throughout a wide variation of intensity. However, its value does vary with the position in the frequency spectrum and it is the amount of this variation which enables one to calculate the relation between the frequency of the tone and its position of maximum stimula- tion along the basilar membrane. The results of such a calculation are given and shown to be in good agreement with determinations from animal experimentation. Transcontinental Telephone Lines. ^ J- J- Pilliod. A fourth trans- continental line has just been created by the completion of four pairs of open wire between Oklahoma City and Whitewater, California. This open-wire line connects at its eastern terminus with the already existing toll cables from the east, and at its western terminus with a toll cable running into Los Angeles. In a cross-section of the United States just west of Denver, there are now 140 through telephone circuits and about the same number of telegraph circuits carried by four open-wire routes. The four new pairs which constitute the transcontinental line carry, in addition to the usual voice frequency channels, three channels of carrier. But their design throughout has been such that twelve additional carrier circuits can be superimposed upon the four channels now provided by each wire pair. The wires of each pair are spaced 8 inches apart with the nearest spacing between pairs being 26" while crossarms are 36" apart. New transposition systems have also been used to further reduce crosstalk. Application of Statistical Methods to Manufacturing Problems."^ W. A. Shewhart. The application of statistical methods in mass produc- tion makes possible the most efficient use of raw materials and manu- facturing processes, effects economies in production, and makes possible the highest economic standards of quality for the manufac- tured goods used by all of us. The story of the application, however, ^ Electrical Engineering, October 1938. * Jour. Franklin Institute, August 1938. 672 BELL SYSTEM TECHNICAL JOURNAL is of much broader interest. The economic control of quahty of manufactured goods is perhaps the simplest type of scientific control. Recent studies in this field throw light on such broad questions as: How far can Man go in controlling his physical environment? How does this depend upon the human factor of intelligence and how upon the element of chance? Observational Significance of Accuracy and Precision} W. A. Shewhart. Two of the most common terms used in pure and applied science are accuracy and precision. When such terms are used, as in the specification of quality of manufactured products, it is desirable that they have definite and, in so far as possible, experimentally veri- fiable meanings. It is, therefore, important to determine how far one can go towards attaining this end by applying with rigor the principle that only that which is observable is significant. In the application of the concepts of accuracy and precision, it is customarily assumed that the available data constitute a random sample. Hence, the first step in attaining experimentally verifiable meaning of these terms is to choose an operationally verifiable criterion of randomness. One such criterion is the quality control chart. In order to give experimental definiteness to any measure of either accuracy or precision derived from a random sample, it is also necessary to specify the way any statement involving the measure may be experimentally verified. To do this it is necessary to make at least four empirical choices as to the details of taking and analyzing the data in the process of verification. Hence, it appears that the meaning of either precision or accuracy is verifiable. Hence, it appears that the meaning of either precision or accuracy is verifiable only in a limited sense subject in any specific case to the choice of empirical criteria of verification. The Time Lag in Gas- Filled Photoelectric Cells} A. M. Skellett. In commercial gas-filled photoelectric cells there is a lag in response which becomes appreciable above frequencies in the neighborhood of 10,000 cycles. If this lag is due to the transit times of the ions across the cell, it should be possible to set up resonance conditions by varying the frequency of modulation of the incident light intensity. This has been accomplished in a cell of special design and the resonance condi- tions agree with the theory, thereby demonstrating that the transit time of the ions is the simple cause. The paper also discusses the flow of the ions and electrons across the cell and their impacts in relation to the flow of current in the external circuit. 5 Jour. Wash. Acad. Sciences, August 15, 1938 (p. 381). ^ Internat'l. Projectionist, September 1938; Jour. Applied Physics, October 1938. Contributors to this Issue Charles R. Burrows, B.S. in Electrical Engineering, University of Michigan, 1924; A.M., Columbia University, 1927; E.E., Univer- sity of Michigan, 1935. Research Assistant, University of Michigan, 1922-23. Western Electric Company, Engineering Department, 1924-25; Bell Telephone Laboratories, Research Department, 1925-. Mr. Burrows has been associated continuously with radio research and is now in charge of a group investigating the propagation of ultra-short waves. Arthur B. Crawford, B.S. in Electrical Engineering, Ohio State University, 1928. Member of Technical Staff, Bell Telephone Laboratories, 1928-. Mr. Crawford has been engaged chiefly in work relative to radio communication by ultra-short waves. Carl R. Englund, B.S. in Chemical Engineering, University of South Dakota, 1909; University of Chicago, 1910-12; Professor of Physics and Geology, Western Maryland College, 1912-13; Laboratory Assistant, University of Michigan, 1913-14. Western Electric Com- pany, 1914-25; Bell Telephone Laboratories, 1925-. As Radio Research Engineer Mr. Englund is engaged largely in experimental work in radio communication. L. A. Meacham, B.S. in Electrical Engineering, University of Wash- ington, 1929. Cambridge University, England, 1929-30. Bell Tele- phone Laboratories, 1930-. Mr. Meacham's work has been concerned with the generation and distribution of constant reference frequencies. S. O. Morgan, B.S. in Chemistry, Union College, 1922; M.A., Princeton University, 1925; Ph.D., 1928. Western Electric Company, Engineering Department, 1922-24; Bell Telephone Laboratories, 1927-. Dr. Morgan's work has been on the relation between dielectric properties and chemical composition. William W. Mumford, B.A., Willamette University, 1930. Bell Telephone Laboratories, 1930-. Mr. Mumford has been engaged in radio receiving work, chiefly on the problem of propagation and measurement in the ultra-short-wave region. E. J. Murphy, B.S., University of Saskatchewan, Canada, 1918; McGill University, Montreal, 1919-20; Harvard University, 1922-23. 673 674 BELL SYSTEM TECHNICAL JOURNAL Western Electric Company, Engineering Department, 1923-25; Bell Telephone Laboratories, 1925-. Mr. Murphy's work is largely con- fined to the study of the electrical properties of dielectrics. A. C. NoRWiNE, A.B., University of Missouri, 1923; B.S. in Elec- trical Engineering, 1924; E.E., 1925. Bell Telephone Laboratories, 1925-. Mr. Norwine has been principally engaged in studies of the effects of transmission delay and voice operated devices on toll tele- phone circuits. A. J. Rack, B.S. in Electrical Engineering, University of Illinois, 1930; M.A. in Physics, Columbia University, 1935. Bell Telephone Laboratories, 1930-. Starting with radio research, Mr. Rack has more recently been engaged in the analysis of special problems arising in amplifier circuits. E. F. Watson, M.E., Cornell University, 1914. American Tele- phone and Telegraph Company, Engineering Department, 1914-19; Department of Development and Research, 1919-34. Bell Telephone Laboratories, 1934-. Mr. Watson has been concerned with the de- velopment of various types of telegraph equipment, particularly teletypewriters, telephotograph equipment, telegraph maintenance and testing equipment, grounded telegraph systems and regenerative telegraph repeaters. His present work as Teletypewriter Engineer is along these same lines. S. B. Wright, M.E. in Electrical Engineering, Cornell University, 1919. Engineering Department and Department of Development and Research, American Telephone and Telegraph Company, 1919-34; Bell Telephone Laboratories, 1934-. Mr. Wright is engaged in trans- mission development of radio systems. Index to Volume XVII Alniqidst, M. L., H. J. Fisher and R. H. Mills, A New Single Channel Carrier Tele- phone System, page 162. Amplitude Characteristics of Telephonic Signals, Devices for Controlling, A. C. Nor wine, page 539. Amplitude Range Control, S. B. Wright, page 520. Analyzer, An Optical Harmonic, H. C. Montgomery, page 406. Anti-sidetone Subscriber Set, Common Battery, An Explanation of the, C. 0. Gibbon, page 245. B Best, F. H., New Transmission Measuring Systems for Telephone Circuit Main- tenance, page 1. Blanchard, Julian, Hertz, the Discoverer of Electric Waves, page 327. Bode, H. W., Variable Equalizers, page 229. Burrows, Chas. R., The Exponential Transmission Line, page 555. Cable Carrier System, Crystal Channel Filters for the, C. E. Lane, page 125. Cable Carrier Telephone System, Crosstalk and Noise Features of, M. A. Weaver, R. S. Tucker and P. S. Darnell, page 137. Cable Carrier Telephone Terminals, R. W. Chesnut, L. M. Ilgenfritz arid A. Kenner, page 106. Cable System for Television Transmission, Coaxial, M. E. Slriehy, page 438. Cables, Toll, A Carrier Telephone System for, C. W. Green and E. I. Green, page 80. Carr, J. A. and F. V. Haskell, Studies of Telephone Line Wire Spacing Problems, page 195. Carrier Telephone System for Toll Cables, A, C. W. Green and E. I. Green, page 80. Carrier Telephone System, Cable, Crosstalk and Noise Features of, M. A. Weaver, R. S. Tucker and P. S. Darnell, page 137. Carrier Telephone System, A New Single Channel, H. J. Fisher, M. L. Almquisi and R. H. Mills, page 162. Chesnut, R. W., L. M. Ilgenfritz and A. Kenner, Cable Carrier Telephone Terminals, page 106. Clarke, Beverly L. and A. E. Ruehle, Spectrochemical Analysis in Communication Research, page 381. Coaxial Cable System for Television Transmission, M. E. Strieby, page 438. Crawford, A. B., C. R. Engl und and W. W. Mumford, Ultra-Short-Wave Transmission and Atmospheric Irregularities, page 489. Crystal Channel Filters for the Cable Carrier System, C. E. Lane, page 125. Darnell, P. S., M. A. Weaver and R. S. Tucker, Crosstalk and Noise Features of Cable Carrier Telephone System, page 137. Darrow, Karl K., Radioactivity — Artificial and Natural, page 292. Davisson, C. J., The Discovery of Electron Waves, page 475. Dielectric Properties of Insulating Materials, The, E. J. Murphy and S. 0. Morgan, page 640. Diodes, Effect of Space Charge and Transit Time on the Shot Noise in, A. J. Rack, page 592. E Echo Suppressors, Two, The Occurrence and Effect of Lockout Occasioned by, Arthur W. Horton, Jr., page 258. 5 BELL SYSTEM TECHNICAL JOURNAL Electric Waves, Hertz, the Discoverer of, Julian Blanchard, page 327. Electron Waves, The Discovery of, C. J. Davisson, page 475. Englund, C. R., A.B. Crawford and W. W. Mumford, Ultra-Short- Wave Transmission and Atmospheric Irregularities, page 489. Equalizers, Variable, H. W. Bode, page 229. Fay, C. E., A. L. Samuel and W. Shockley, On the Theory of Space Charge between Parallel Plane Electrodes, page 49. Feedback Oscillators, Stabilized, G. H. Stevenson, page 458. Filters, Crystal Channel, for the Cable Carrier System, C. E. Lane, page 125. Fisher, H. J., M. L. Almquist and R. H. Mills, A New Single Channel Carrier Tele- phone System, page 162. Gibbon, C. O., An Explanation of the Common Battery Anti-sidetone Subscriber Set, page 245. Green, C. W. and E. L, A Carrier Telephone System for Toll Cables, page 80. Gustafson, W. G., Magnetic Shielding of Transformers at Audio Frequencies, page 416. Haskell, F. V. and J. A. Carr, Studies of Telephone Line Wire Spacing Problems, page 195. Herriott, W., High Speed Motion Picture Photography, page 393. Hertz, the Discoverer of Electric Waves, Julian Blanchard, page 327. Horton, Arthur W., Jr., The Occurrence and Effect of Lockout Occasioned by Two Echo Suppressors, page 258. I Ilgenfritz, L. M., R. W. Chesnut and A. Kenner, Cable Carrier Telephone Terminals, page 106. Impedance Concept and its Application to Problems of Reflection, Refraction, Shielding and Power Absorption, The, 5. A. Schelkunoff, page l7. Inglis, A. H., Transmission Features of the New Telephone Sets, page 358. Insulating Materials, The Dielectric Properties of, E. J. Murphy and S. 0. Morgan, page 640. J Jones, W. C, Instruments for the New Telephone Sets, page 338. K Kenner, A., R. W. Chesnut and L. M. Ilgenfritz, Cable Carrier Telephone Terminals, page 106. L Lane, C. E., Crystal Channel Filters for the Cable Carrier System, page 125. M Magnetic Shielding of Transformers at Audio Frequencies, W. G. Gustafson, page 416. Maintenance, Telephone Circuit, New Transmission Measuring Systems for, F. H. Best, page 1. Meacham, L. A., The Bridge Stabilized Oscillator, page 574. Mills, R. H., H. J. Fisher and M. L. Almquist, A New Single Channel Carrier Tele- phone System, page 162. Montgomery, H. C, An Optical Harmonic Analyzer, page 406. Morgan, S. 0. and E. J. Murphy, The Dielectric Properties of Insulating Materials, page 640. Motion Picture Photography, High Speed, W. Herriott, page 393. 6 BELL SYSTEM TECHNICAL JOURNAL Mumford, W. W., C. R. EnglundandA. B. Crawford, Ultra-Short-Wave Transmission and Atmospheric Irregularities, page 489. Murphy, E. J. and S. 0. Morgan, The Dielectric Properties of Insulating Materials, page 640. Murphy, 0. J. and A. C. Norwine, Characteristic Time Intervals in Telephonic Conversation, page 281. N New Telephone Sets, Instruments for the, W. C. Jones, page 338. New Telephone Sets, Transmission Features of the, A. H. Inglis, page 358. Norwine, A. C, Devices for Controlling Amplitude Characteristics of Telephonic Signals, page 539. Norwine, A. C. and 0. J. Murphy, Characteristic Time Intervals in Telephonic Conversation, page 281. O Oscillator, The Bridge Stabilized, L. A. Meacham, page 574. Oscillators, Stabilized Feedback, G. H. Stevenson, page 458. Rack, A. J., Effect of Space Charge and Transit Time on the Shot Noise in Diodes, page 592. Radio: Ultra-Short-Wave Transmission and Atmospheric Irregularities, C. R. Englund, A. B. Crawford and W. W. Mumford, page 489. Radioactivity — Artificial and Natural, Karl K. Darrow, page 292. Research, Communication, Spectrochemical Analysis in, Beverly L. Clarke and A. E. Ruehle, page 381. Ruehle, A. E. and Beverly L. Clarke, Spectrochemical Analysis in Communication Research, page 381. S Samuel, A. L., C. E. Fay and W. Shockley, On the Theory of Space Charge between Parallel Plane Electrodes, page 49. Schelkunoff, S. A., The Impedance Concept and its Application to Problems of Reflection, Refraction, Shielding and Power Absorption, page 17. Shockley, W., C. E. Fay and A. L. Samuel, On the Theory of Space Charge between Parallel Plane Electrodes, page 49. Short-Wave, Ultra-, Transmission and Atmospheric Irregularities, C. R. Englund, A. B. Crawford and W. W. Mumford, page 489. Shot Noise in Diodes, Effect of Space Charge and Transit Time on the, A. J. Rack, page 592. Space Charge between Parallel Plane Electrodes, On the Theory of, C. E. Fay, A. L. Samuel and W. Shockley, page 49. Space Charge and Transit Time on the Shot Noise in Diodes, Effect of, A. J. Rack, page 592. Spectrochemical Analysis in Communication Research, Beverly L. Clarke and A. E. Ruehle, page 381. Stevenson, G. H., Stabilized Feedback Oscillators, page 458. Strieby, M. E., Coaxial Cable System for Television Transmission, page 438. Telephonic Signals, Devices for Controlling Amplitude Characteristics of, A. C. Norwine, page 539. Teletypewriters Used in the Bell System, Fundamentals of, E. F. Watson, page 620. Television Transmission, Coaxial Cable System for, M. E. Strieby, page 438. Time Intervals in Telephonic Conversation, Characteristic, A. C. Norwine and 0. J. Murphy, page 281. Transformers at Audio Frequencies, Magnetic Shielding of, W. G. Gustafson, page 416. Transmission Features of the New Telephone Sets, A. H. Inglis, page 358. Transmission Line, The Exponential, Chas. R. Burrows, page 555. 7 BELL SYSTEM TECHNICAL JOURNAL Transmission Measuring Systems for Telephone Circuit Maintenance, New, F. H. Best, page 1. Tucker, R. S., M. A. Weaver and P. S. Darnell, Crosstalk and Noise Features of Cable Carrier Telephone System, page 137. W Watson, E. F., Fundamentals of Teletypewriters Used in the Bell System, page 620. Weaver, M. A., R. S. Tucker and P. S. Darnell, Crosstalk and Noise Features of Cable Carrier Telephone System, page 137. Wire Spacing Problems, Telephone Line, Studies of, J. A. Carr and F. V. Haskell, page 195. Wright, S. B., Amplitude Range Control, page 520. A.V.EMMOTT&SONSI BOOKBINDERS. Inc. 1101 HAMILTON HOUSTON 3, TEXAS