!)^^)iJiasa'^i^tJia;xga!Jia!>sai®j!ji«!«<)i«<):«i ^lln Kansas (Ettg publtr ICtbrarg This Volume is for REFERENCE USE ONLY c 1U29— 6m— H From the collection of the re THE BELL SYSTEM TECHNICAL JOURNAL A JOURNAL DEVOTED TO THE SCIENTIFIC AND ENGINEERING ASPECTS OF ELECTRICAL COMMUNICATION EDITORIAL BOARD Bancroft Gherardi F. B. Jewett H. P. Charlesworth W. H. Harrison E. H. Colpitts L. F. Morehouse H. D. Arnold O. B. Blackwell Philander Norton, Editor J. O. Perrine, Associate Editor INDEX VOLUME IX 1930 AMERICAN TELEPHONE AND TELEGRAPH COMPANY NEW YORK « ^ J w • • • •J . •• • Mr 1 6] (iS7H77 THE BELL SYSTEM TECHNICAL JOURNAL VOLUME IX, 1930 TABLE OF CONTENTS January, 1930 Telephone Communication System of the United States — Bancroft Gherardi and F. B. Jcivett 1 Structure and Nature of Troostite — Francis F. Lucas 101 Radio Broadcasting Transmitters and Related Transmission Phe- nomena— Edzvard L. Nelson 121 Wire Line Systems for National Broadcasting — A. B. Clark 141 Notes on the Heaviside Operational Calculus — John R. Carson ... 150 Contemporary Advances in Physics, XIX — Karl K. Darroiv 163 Wave Propagation Over Continuously Loaded Fine Wires — M. K. Zinn 189 Theory of Vibration of the Larynx — R. L. Wegel 209 April, 1930 Developments in Communication Materials — William Fondillcr . . . 237 Transoceanic Telephone Service — Short- Wave Transmission — Ralph Bown 258 Transoceanic Telephone Service — -Short-Wave Equipment — A. A. Oszvald 270 The \\^ords and Sounds of Telephone Conversations — Norman R. French, Charles W. Carter, Jr., and Walter Koenig, Jr 290 The Reciprocal Energy Theorem — John R. Carson 325 The Approximate Networks of Acoustic Filters — W. P. Mason . . 332 Contemporary Advances in Physics, XX — Karl K. Darrow 341 Motion of Telephone Wires in Wind — D. A. Quarks 356 Economic Quality Control of Manufactured Product — W. A. Shewhart 364 Optimum Reverberation Time for Auditoriums — Walter A. MacNair 390 3 BlilJ. SYSTEM TECHNICAL JOURNAL July, 1930 Radio Telephone Service to Ships at Sea — William Wilson and Lloyd Espcnschicd 407 A General Switching Plan for Telephone Toll Service — //. S. Osborne 429 Image Transmission System for Two- Way Television — Herbert E. Ives, Frank Gray and M. W. Baldwin 448 Synchronization System for Two-Way Television — H. M. Stollcr 470 Sound Transmission System for Two-Way Television — D. G. Blattner and L. G. BosHvick 478 Transmitted Frequency Range for Telephone Message Circuits — W. H. Martin 483 Some Recent Developments in Long Distance Cables in the United States of America— .4. B. Clark 487 Phase Distortion in Telephone Apparatus — C. E. Lane 493 Measurement of Phase Distortion — H. Nyqnist and S. Brand .... 522 Effects of Phase Distortion on Telephone Quality — John C. Steinberg 550 Long Distance Cable Circuit for Program Transmission — A. B. Clark and C. W. Green 567 October, 1930 Chemistry in the Telephone Industry — Robert R. Ullliains 603 The Trend in the Design of Telephone Transmitters and Receiv- ers—fi^. H. Martin and W. F. Davidson 622 Mutual Impedances of Ground-Return Circuits — A. E. Boiven and C. L. Gilkeson 628 A Survey of Room Noise in Telephone Locations — W. J. Williams and Ralph G. MeCnrdy 652 Contemporary Advances in Physics, XXI — Karl K. Darrozv 668 A Study of Telephone Line Insulators — L. T. IVilson 697 The Transmission Characteristics of Open-Wire Telephone Lines — E. L Green 730 Transients in Parallel Grounded Circuits, One of Which is of In- finite Length — Liss C. Peterson 760 Impedance Correction of Wave Filters — E. B. Payne 770 A Method of Impedance Correction — //. W. Bode 794 Index to Volume IX Approximate Networks of Acoustic Filters, The, JV. P. Mason, page 332. Auditoriums, Optimum Reverberation Time for, Walter A. MacNair, page 390. B Baldwin, M. W., Herbert E. Ives and Frank Gray, Image Transmission System for Two-Way Television, page 448. Blattner, D. G. and L. G. Bostivlck, Sound Transmission System for Two- Way Television, page 478. Bode, H. W., A Method of Impedance Correction, page 794. Bostzmck, L. G. and D. G. Blattner, Sound Transmission System for Two-Way Television, page 478. Bowen, A. E. and C L. Gilkeson, Mutual Impedances of Ground-Return Circuits — Some Experimental Studies, page 628. Boivn, Ralph, Transoceanic Telephone Service — Short-Wave Transmission, page 258. Brand, S. and H. Nyquist. Measurement of Phase Distortion, page 522. Broadcasting, National, Wire Line Systems for, A. B. Clark, page 141. Cable Circuit, Long Distance, for Program Transmission, A. B. Clark and C. IV. Green, page 567. Cables, Long Distance, in the United States of America, Some Recent Develop- ments in, A. B. Clark, page 487. Carson, John R., Notes on the Heaviside Operational Calculus, page 150. Carson, John R., The Reciprocal Energy Theorem, page 325. Carter, Charles W., Jr.. Norman R. French, and Walter Koenig, Jr., The Words and Sounds of Telephone Conversations, page 290. Chemistry in the Telephone Industry, Robert R. Williams, page 603. Clark, A. B., Wire Line Systems for National Broadcasting, page 141. Clark, A. B., Some Recent Developments in Long Distance Cables in the United States of America, page 487. Clark, A. B. and C. JV. Green, Long Distance Cable Circuit for Program Trans- mission, page 567. Contemporary Advances in Physics, XIX. Fusion of Wave and Corpuscle The- ories, Karl K. Darrozv, page 163. Contemporary Advances in Physics, XX. Ionization of Gases by Light, Karl K. Darrozv, page 341. Contemporary Advances in Physics, XXI. Interception and Scattering of Elec- trons and Ions, Karl K. Darrozv, page 668. Control, Economic Quality, of Manufactured Product, W. A. Shezvhart, page 364. Corpuscle Theories, Fusion of Wave and, Karl K. Darrozv, page 163. Darrozv, Karl K.. Contemporary Advances in Physics, XIX. Fusion of Wave and Corpuscle Theories, page 163. 5 BELL SYSTEM TECHNICAL JOURNAL Darrotv, Karl K., Contemporary Advances in Physics, XX. Ionization of Gases by Light, page 341. Darrozv, Karl K., Contemporary Advances in Physics, XXI. Interception and Scattering of Electrons and Ions, page 668. Davidson, W. F. and IV. H. Martin, The Trend in the Design of Telephone Trans- mitters and Receivers, page 622. Developments in Communication Materials, William Fondillcr, page 237. E Economic Quality Control of Manufactured Product, W. A. Shewhart, page 364. Effects of Phase Distortion on Telephone Quality, John C. Steinberg, page 550. Electrons and Ions, Interception and Scattering of. Contemporary Advances in Physics, XXI, Karl K. Darrozv, page 668. Espcnschicd, Lloyd and William Wilson, Radio Telephone Service to Ships at Sea, page 407. Filters, Acoustic, The Approximate Networks of, W. P. Mason, page 332. Filters, Wave, Impedance Correction of, E. B. Payne, page 770. Fondiller, William, Developments in Communication Materials, page 237. French, Norman R.. Charles W. Carter, Jr., and Walter Koenig, Jr., The Words and Sounds of Telephone Conversations, page 290. Frequency Range for Telephone Message Circuits, Transmitted, W. H. Martin, page 483. G General Switching Plan for Telephone Toll Service, A, H. S. Osborne, page 429. Gherardi, Bancroft and F. B. Jezvett, Telephone Communication System of the United States, page 1. Gilkeson, C. L. and A. E. Bozven, Mutual Impedances of Ground Return Circuits — Some Experimental Studies, page 628. Gi-ay, Frank, Herbert E. Ives, and M. W. Baldzmn, Image Transmission System for Two-Way Television, page 448. Green, C. JV. and A. B. Clark, Long Distance Cable Circuit for Program Trans- mission, page 567. Green, E. L, The Transmission Characteristics of Open-Wire Telephone Lines, page 730. H Heaviside Operational Calculus, Notes on the, John R. Carson, page 150. Image Transmission System for Two-Way Television, Herbert E. Ives, Frank Gray, and M. W. Baldzvin, page 448. Impedances, Mutual, of Ground-Return Circuits — Some Experimental Studies, A. E. Bozven and C. L. Gilkeson, page 628. Impedance Correction of Wave Filters, E. B. Payne, page 770. Insulators, Telephone Line, A Study of, L. T. Wilson, page 697. Ionization of Gases by Light, Contemporary Advances in Physics, XX, Karl K. Darrozv, page 341. Ives, Herbert E., Frank Gray, and M. W. Baldzvin, Image Transmission System for Two-Way Television, page 448. BELL SYSTEM TECHNICAL JOURNAL Jewett, F. B. and Bancroft Gherardi, Telephone Communication System of the United States, page 1. K Koenig, Walter, Jr., Norman R. French, and Charles W. Carter, Jr., The Words and Sounds of Telephone Conversations, page 290. Lane, C. E., Phase Distortion in Telephone Apparatus, page 493. Larynx, Theory of Vibration of, R. L. Wegel, page 209. Loaded, Continuously, Fine Wires, Wave Propagation Over, M. K. Zinn, page 189. Long Distance Cable Circuit for Program Transmission, A. B. Clark and C. W. Green, page 567. Lucas, Francis F., Structure and Nature of Troostite, page 101. M McCurdy, R. G. and IV. J. Williams, A Survey of Room Noise in Telephone Loca- tions, page 652. MacNair, Walter A., Optimum Reverberation Time for Auditoriums, page 390. Martin, W. H., Transmitted Frequency Range for Telephone Message Circuits, page 483. Martin, W. H. and W. F. Davidson, The Trend in the Design of Telephone Trans- mitters and Receivers, page 622. Mason, W. P., The Approximate Networks of Acoustic Filters, page 332. Materials, Communication, Developments in, William Fondiller, page 237. Measurement of Phase Distortionj H. Nyquist and 5". Brand, page 522. Method of Impedance Correction, A, H. W. Bode, page 794. Motion of Telephone Wires in Wind, D. A. Quarks, page 356. Mutual Impedances of Ground-Return Circuits — Some Experimental Studies, A. E. Bozvcn and C. L. Gilkcson, page 628. N Nelson, Edivard L., Radio Broadcasting Transmitters and Related Transmission Phenomena, page 121. Networks of Acoustic Filters, The Approximate, JV. P. Mason, page 332. Noise, Room, in Telephone Locations, A Survey of, W. J. Williams and R. G. McCurdy, page 652. Notes on the Heaviside Operational Calculus, John R. Carson, page 150. Nyquist, H. and 6'. Brand, Measurement of Phase Distortion, page 522. Open-Wire Telephone Lines, The Transmission Characteristics of, E. I. Green, page 730. Optimum Reverberation Time for Auditoriums, Walter A. MacNair, page 390. Osborne, H. S., A General Switching Plan for Telephone Toll Service, page 429. Osivald, A. A., Transoceanic Telephone Service — Short-Wave Equipment, page 270. Payne, E. B., Impedance Correction of Wave Filters, page 770. Peterson, L. C, Transients in Parallel Grounded Circuits, One of Which is of In- finite Length, page 760. 7 BELL SYSTEM TECHNICAL JOURNAL Phase Distortion in Telephone Apparatus, C. E. Lane, page 493. Phase Distortion, Effects of, on Telephone Quality, John C. Slcinberg, page 550. Phase Distortion, Measurement of, //. Nyquist and S. Brand, page 522. Physics, XIX, Contemporary Advances in. Fusion of Wave and Corpuscle The- ories, Karl K. Darroiv, page 163. Physics, XX, Contemporary Advances in. Ionization of Gases by Light, Karl K. Darroiv, page 341. Physics, XXI, Contemporary Advances in. Interception and Scattering of Elec- trons and Ions, Karl K. Darroiv, page 608. Q Quarles, D. A., Motion of Telephone Wires in Wind, page 356. R Radio Broadcasting Transmitters and Related Transmission Phenomena, Edward L. Nelson, page 121. Radio Telephone Service to Ships at Sea, William Wilson and Lloyd Espenschied, page 407. Receivers, The Trend in the Design of Telephone Transmitters and, W. H. Martin and W. F. Davidson, page 622. Reciprocal Energy Theorem, The, John R. Carson, page 325. Reverberation Time for Auditoriums, Optimum, Walter A. MacNair, page 390. Shewhart, W. A., Economic Quality Control of Manufactured Product, page 364. Ships at Sea, Radio Telephone Service to, William Wilson and Lloyd Espenschied, page 407. Short-Wave Equipment, in Transoceanic Telephone Service, A. A. Oswald, page 270. Short-Wave Transmission, in Transoceanic Telephone Service, Ralph Bown, page 258. Some Recent Developments in Long Distance Cables in the United States of Amer- ica, A. B. Clark, page 487. Sounds, and Words, of Telephone Conversations, Norman R. French, Charles W. Carter, Jr., and Walter Koenig, Jr., page 290. Sound Transmission System for Two-Way Television, D. G. Blattner and L. G. Bostwick, page 478. Steinberg, John C, Effects of Phase Distortion on Telephone Quality, page 550. Stollcr, H. M., Synchronization System for Two-Way Television, page 470. Structure and Nature of Troostite, Francis F. Lucas, page 101. Study of Telephone Line Insulators, A, L. T. Wilson, page 697. Survey of Room Noise in Telephone Locations, A, W. J. Williams and R. G. Mc- Curdy, page 652. Switching Plan for Telephone Toll Service, A, H. S. Osborne, page 429. Synchronization System for Two-Way Television, H. M. Stollcr, page 470. Telephone Communication System of the United States, Bancroft Ghcrardi and F. B. Jcwett, page 1. Television, Two-Way, Image Transmission System for, Herbert E. Ives, Frank Gray, and ^L W. Baldwin, page 448. 8 BELL SYSTEM TECHNICAL JOURNAL Television, Two-Way, Synchronization System for, H. M. Stoller, page 470. Television, Two-Way, Sound Transmission System for, D. G. Blattner and L. G. Bostwick, page 478. Theory of Vibration of the Larynx, R. L. Wcgel, page 209. Toll Service, Telephone, A General Switching Plan for, H. S. Osborne, page 429. Transients in Parallel Grounded Circuits, One of Which is of Infinite Length, L. C. Peterson, page 760. Transmission Characteristics of Open-Wire Telephone Lines, The, E. L Green, page 730. Transmitted Frequency Range for Telephone Message Circuits, W. H. Martin, page 483. Transmitters and Receivers, Telephone, The Trend in the Design of, W. H. Mar- tin and IV. F. Davidson, page 622. Transoceanic Telephone Service — Short-Wave Equipment, A. A. Oszvald, page 270. Transoceanic Telephone Service — Short-Wave Transmission^ Ralph Sown, page 258. Trend in the Design of Telephone Transmitters and Receivers, The, IV. H, Martin and W. F. Davidson, page 622. Troostite, Structure and Nature of, Francis F. Lucas, page lOL W Wave and Corpuscle Theories, Fusion of. Contemporary Advances in Physics, Karl K. Darrow, page 163. Wave Propagation Over Continuously Loaded Fine Wires, M. K. Zinn, page 189. Wegel, R. L., Theory of Vibration of the Larynx, page 209. Williams, Robert R., Chemistry in the Telephone Industry, page 603. Williams, W. J. and R. G. McCurdy, A Survey of Room Noise in Telephone Loca- tions, page 652. Wilson, L. T., A Study of Telephone Line Insulators, page 697. Wilson, Willia})! and Lloyd Espenschied, Radio Telephone Service to Ships at Sea, page 407. Wind, Motion of Telephone Wires in, D. A. Quarles, page 356. Wire Line Systems for National Broadcasting, A. B. Clark, page 141. Words and Sounds of Telephone Conversations, The, Norman R. French, Charles IV. Carter, Jr., and Walter Koenig, Jr., page 290. Z Zinn, M. K., Wave Propagation Over Continuously Loaded Fine Wires, page 189. The Bell System Technical Journal January, 1930 Telephone Communication System of the United States ^ By BANCROFT GHERARDI and F. B. JEWETT This paper presents the results which have been obtained up to the present time in developing telephone communication in the United States of America, this development having been worked out in a form to meet the particular conditions which present themselves in that country. The paper first deals with a brief description of the general structure and organization of the telephone communication system giving the organization of the Bell System which handles the greater part of the telephone service of the country and the reasons for and advantages of this organization. _ In this connection some figures are presented with respect to the technical personnel who are continuously engaged in studies to develop the art and to provide new methods and facilities for improving the service. Local service, that is the service within the limits of a single telephone exchange area, is next discussed. Figures are given with respect to the vol- umes of telephone calls handled in the Bell System, the speed with which the connections for these calls are completed and the operating force re- quired. Reference is also made to the standards of transmission given and the various problems encountered in meeting these standards. Figures are given with respect to station growth, to the increased efficiency of sta- tion apparatus and to the improvement in types of instruments. Various types of private branch exchanges provided to meet the needs of customers using a large amount of telephone service are discussed. The cable plant is considered mainly from the construction standpoint and typical illus- trations are given of some of the construction practices. The various types of central office switcning systems in common use are described, including magneto, common battery and dial systems, the latter including both the step-by-step and panel systems which ate being provided in increasing amounts in the Bell System. The subject of buildings to house these various equipments as well as the operating forces and headquarters staffs in many cases is briefly discussed, also standardized layouts and floor plans. The problem of giving telephone service in the rural com- munities, which is a very important one in the telephone development in the United States, is also biiefly treated. The toll service is considered, first with respect to the shorter haul toll business and the problems involved and then with lespect to the long distance toll service. Figures are given showing the speed of service and the amount of traffic handled. For the short distance toll service, two important methods of handling tne business are described, namely, manual straightforward tandem and dial tandem. The long distance service, which has developed most rapidly in recent years, is described in some detail in the paper. Among the important features of this service is noted the recently developed method of com- pleting toll calls with sufficient speed so that on most of the calls the calling subscriber remains at the telephone. The various types of toll circuits are described including open wire circuits operated both at voice frequencies and by carrier systems and long toll cable circuits. The operation of these long circuits requires a large number of repeaters in tandem and the design and maintenance problems which this arrangement requires are pointed out in the paper. 1 Presented by Dr. F. B. Jewett before the World Engineering Congress, Tokio Japan, October, 1929. 1 1 2 BELL SYSTEM TECHNICAL JOURNAL Information is given with respect to international telephone connections in North America, between North America and Europe and other inter- national connections. In covering this subject some of the important items relating to the operation of the transatlantic radio channels are given and reference made to the projected transatlantic telephone cable. Various forms of special services closely allied with the message tele- phone service are described. These include telegraph service, telephone circuits provided for private use, foreign exchange service, telephone net- works for program transmission to radio broadcasting stations, electrical transmission of pictures, telephony in connection with aircraft operation, ship to shore telephony, telephony to mobile stations such as railroad trains, telephone services of railroads and other public utilities, telephone public address systems and television. Reference is also made to some of the by-products of the telephone development work which include im- provements in submarine cable telegraphy brought about by the discovery of the alloys known as "permalloy and oerminvar," the development work in the reproduction of sound and in tne talking motion pictures. In concluding, the paper points out that careful studies of the future development of the telephone industry indicate a somewhat accelerated rate of development of the services required to meet the demands of the customers and a continuing very rapid technical development of telephone plant and systems to prov-ide the necessary facilities. In treating such a large subject in a paper of this kind it has been neces- sary to deal with technical problems in rather general terms and as an attachment to the paper references are made to numerous articles in the technical press for the more technical information. General THE purpose of this paper is to give a general description of the telephone communication system of the United States of Amer- ica, outlining briefly some of the more important engineering problems involved and indicating the service results obtained. At the begin- ning of this paper it seems important to give a brief description of the general structure and organization of the telephone communica- tion system. The commercial telephone system of the United States is entirely owned and operated by corporations, partnerships, and individuals. A group of 24 closely associated Bell Telephone Operating Companies owns and operates 14.8 million telephones and the telephone lines used for toll service within their territories. In addition there are in the country about 4.7 million telephones owned by several thousand independent telephone companies which have operating agreements with the Bell Companies providing for the interconnection of lines, thus permitting the operation of 19.5 million telephones as a single system. There are in addition about 140,000 telephones in the coun- try not connected with the Bell System. The 24 Bell Operating Companies cover the entire area of the United States and are responsible for all Bell Telephone operations within their respective areas. A number of the larger companies are subdivided into autonomous operating units, there being at the TELEPHONE SYSTEM OF THE UNITED STATES present time a total of 34 such units in the country. In many cases the area within corporate limits or within the limits of an operating unit is identical with that of a major political subdivision of the United States (a State) and this simplifies the application of govern- mental regulation. A typical organization of a Bell Operating Com- pany is indicated in Fig. 1. 'Accounting . . General Auditor 'Building and Equip- ment Engineer Plant Extension Engi- neer Financial. . . .Treasurer f Chief Engineer J Outside Plant Engineer Operation . . .Vice President Board 1 of ^ President^ DirectorsJ Personnel . . .Assistant to President ■< General Commer- cial Manager General Traffic Manager General Plant Manager Transmission Engineer Costs and Inventory Engineer General Supervisor of Methods and Results General Employment Supervisor ij General Supervisors — 1 1 Other Functions Staff Engineers Division Superintend- ents Legal General Counsel .^Secretary Fig. 1 — Organization of typical Bell Telephone Operating Company. In order to facilitate the best possible handling of the long distance service between points in different operating companies and to avoid the problems which would arise from divided responsibility, the long distance business involving territories of two or more Associated Companies is handled throughout the country by the Long Lines Department of the American Telephone and Telegraph Company. These operations are, of course, in the closest cooperation with the operations of the Associated Companies without duplication of con- struction or of operating effort. An important feature of the Bell Telephone System is the general 4 BELL SYSTEM TECHNICAL JOURNAL departments maintained by the American Telephone and Telegraph Company, including the Bell Telephone Laboratories. These depart- ments, constituting about 7,500 engineers, scientists, business experts and assistants, are continuously engaged in studies to develop the art and to provide methods and facilities for improving the service. They also provide consulting advice to the operating companies on all phases of the telephone business and render to them a large variety of services. One of these services is making available to all the companies rights under all patents necessary for the fullest and most economical development of the business. It is the intention, in general, that work which can best be done once for all the entire telephone system rather than individually by the several operating companies shall be done by these general departments and that the specific solution of the telephone problems in each area shall be the responsibility of the operating company involved, who, however, are free at all times to get the advice and assistance of the general staff. In all of the work of the general staff close contact is maintained with the various operating telephone companies of the Bell System. The experiences of these companies are studied and analyzed to make available for all the companies the valuable results to be derived in this way, and the advantages to be obtained by comparing the ex- periences of different companies under similar conditions. The organization of the American Telephone and Telegraph Com- pany and the Bell Telephone Laboratories is indicated in Fig. 2. Another very important feature of the Bell System is the very close relation between operating and manufacturing branches of the work through the ownership by the American Telephone and Tele- graph Company of the Western Electric Company, Inc., and arrange- ments between that company and the operating companies for the supply of telephone apparatus and materials. This permits the manufacture of apparatus and the purchase of materials from outside suppliers to be done on the basis of the large quantities required for the entire Bell System resulting in great economies. The organization of the Bell Telephone System is such as to result in close cooperation between the companies dealing with different branches of telephone work. This brings about the conditions nec- essary for universal service, for the development of the art along orderly and non-conflicting lines, and for the standardization of all apparatus, communication systems and operating methods to the extent that such standardization is helpful. New types of telephone plant, operating methods, methods of maintenance and business methods are standardized by the general TELEPHONE SYSTEM OF THE UNITED STATES 5 departments of the American Telephone and Telegraph Company, and are adopted and placed in use by all of the Associated Operating Companies to the extent that they apply to their local conditions. Special arrangements are, of course, made available to meet special requirements. The specifications for all standardized apparatus and rPersonnel Vice President Board of Director? Operation and Engi- neering. . .Vice President and Cliief Engineer Assistant Vice President Benefit and Medical Work J Assistant Vice President Employee Relations Assistant Vice President Special Employment and Training Commercial Engineer .■Assistant Vice President Plant Operation Engi- neer General Operating Results Traffic Engineer Plant Engineer Legal . f Attorney .\'ice President J Ta.x Attorney and General 1 Patent Attorney Counsel |^ General Solicitor Transmission Develop- ment Engineer Outside Plant Develop- ment Engineer Research Engineer 'Vice President President*! President Development and Research Assistant Vice-, Equipment Development Engineer Electrical Interference Engineer Technical Representative in Europe Bell Telephone Laboratories Chairman of Board President Vice Presi- dent Director of Research Director of Appara- tus Development Director of Systems Development General Patent At- torney Director of Publica- tion Personnel Director Accounts, Finance and Business Vice President 'Comptroller Treasurer .\ssistant \'ice President General Policies and Contracts Assistant Vice President General Service Bureau Information. Secretary r.^dvertising Manager . VicePresident-i .\ssistant Vice President L Publicity Fig. 2 — Organization of the general departments of the American Telephone and Telegraph Company, including Bell Telephone Laboratories. materials are prepared by the Bell Telephone Laboratories, and the Western Electric Company is enabled to concentrate on the task of purchasing and manufacturing standardized supplies, materials and apparatus in accordance with these standard specifications. Stand- ardization also has great operating advantages in minimizing stocks 6 BELL SYSTEM TECHNICAL JOURNAL of materials and providing interchangeability both of materials and working forces and is very important in making possible the operation of the entire interconnected network as a single system with uniform grades of service. As has been stated by Mr. Gifford, President of the American Telephone and Telegraph Company, "The ideal and aim today of the American Telephone and Telegraph Company and its Associated Companies is a telephone service for the nation, free, so far as humanly possible, from imperfections, errors, or delays, and enabling at all times any one anywhere to pick up a telephone and talk to any one else anywhere else, clearly, quickly and at a reasonable cost." With this aim in view, continuous effort is made further to improve and to extend the service within the nation and also the telephonic con- nections to other nations. It is recognized also that changes in business and social conditions bring about repeated changes in the services desired by the people of the nation and in the character and appearance of facilities furnished to them. These facts, in addition to the onward march of the application of science, form an important basis for the continued study by the general staff of the development of all phases of the telephone system. A few figures relative to the size and growth of the Bell System are helpful in an understanding of the more specific telephone prob- lems which are discussed below. Such figures are included in the statistical summary appended to this paper and include data regarding telephone messages, numbers of telephones, miles of wire and amount of telephone plant. In accordance with the general organization of the Bell System, the engineering problems involved in the design, construction and maintenance of the plant of each operating telephone company are the responsibility of the engineering department of that company. General studies of methods of improvement of service and the devel- opment of new apparatus and systems of communication, together with consulting engineering advice, are provided by the general de- partments. For the provision of new plant to meet additional demands for service, in the case of the more important items, often one year, and sometimes more, is required between the completion of detailed engi- neering plans and completion of construction. Furthermore, to ob- tain maximum economy it is necessary that much of the new construc- tion provide for expected increases in demands for service for a number of years to come. This applies particularly to telephone buildings and to runs of underground conduit and to a lesser extent to cables, TELEPHONE SYSTEM OF THE UNITED STATES 7 pole lines and many other very important parts of the telephone plant. The engineering of the additions to the Bell Telephone System, now aggregating over 500 million dollars a year is, therefore, neces- sarily based on careful forecasts of the amount and type of business to be expected for a number of years in the future and good engineer- ing judgment must be applied in determining the types, quantities and design of plant. These must take into account not only the expected amount of service required but also expected future changes in the character and standards of service demanded and in the appa- ratus and materials expected to become available. In view of the capital expended in extensions and the large amount of plant already in service, the engineering work involved is considerable. There are now approximately 10,000 engineers engaged in the work of the Bell System of which approximately 6,300 are in the operating companies, 2,200 in the headquarters departments and 1,500 in the Western Electric Company. These figures apply to men doing work of engi- neering grade, and inclusion of assistants of all kinds, stenographical, clerical, laboratory, etc., would more than double these figures. Local Service General Service within the limits of a single telephone exchange is spoken of as local service. This generally includes service within a large met- ropolitan area, a city with its surrounding suburbs or a town or village. During 1928 customers of the Bell System originated approximately 24,000 million local calls of which approximately 19,000 million originated from manual and 5,000 million from dial telephones. This represents an average daily usage of approximately 5.5 calls per telephone station per day. The speed of service is illustrated by the following average figures. In the smaller cities with manual operation where the operator who takes the call completes it herself without trunking, the average time from the start of the call to the answer of the called station is 19 seconds. The corresponding figure for manual calls in large cities based on about three million observations made in the year 1928 in 38 large cities of the country is 28.8 seconds. The same observations indicate that when fully converted to the dial system the speed of service in the large cities will be about 22.5 seconds. As to the accuracy of service, 98 per cent of all calls are handled without error. The most serious errors are those resulting in wrong numbers. The mistakes made by the subscribers and equipment 8 BELL SYSTEM TECHNICAL JOURNAL under the dial system are about the same in number as those made by subscribers and operators under the manual system. Calls resulting in busy reports amount to 10 per cent. This is something which is not directly under the control of the telephone company since the subscriber determines the telephone facilities which are provided. Records are kept, however, in both manual and dial offices of the lines responsible for the greatest number of busy reports and efforts are made to have the subscribers take additional facilities. Standards of transmission are applied to the design of the plant to insure that transmission will be clear between the most remote parts of the exchange area. This depends on the design of station equipment, wire lines and switchboard equipment, and is expressed in terms of the combined electric and acoustic efficiencies of the cir- cuits from the mouth of the talker to the ear of the listener. This overall efficiency is expressed in terms of the adjustment of a standard reference circuit. The standards in use in the United States refer to the maximum transmission loss permitted between any two sub- scribers and vary in magnitude between equivalents of 18 decibels and 22 decibels, depending on the circumstances of different cases. In order to meet these transmission standards the Bell Companies have standard requirements regarding the efficiency of transmitters and receivers and other station equipment, and these are made the basis for engineering the wire plant. Transmission losses in switch- boards are kept as low as practicable and within specified limits. The wire plant for subscribers lines and trunks is designed to be within the limits required for meeting the transmission standards. If under special conditions it appears desirable to exceed these limits, this is done only with the approval of responsible engineering authorities. To handle calls at the local switchboards there was in the Bell System in 1928 an average operating force of about 122,000 young women. In addition an average force of approximately 36,000 were employed at the toll boards of the Bell System. This made a total operating force of 158,000. In order to make up for losses and for growth, 75,000 women were employed, and to select this number approximately 300,000 applicants were interviewed. One of the important administrative problems is the scheduling of the operating forces so that an adequate number may be available in each central office throughout every period of the day. A method has been worked out whereby all types of operating work are equated to a common unit of measurements and the number of such units that an operator should handle to give the best service most efficiently has been determined. Frequent counts are maintained of the num- TELEPHONE SYSTEM OF THE UNITED STATES 9 ber of calls handled throughout each hour of the day and in this way the forces are adjusted to the work to be done. In order that the demand for telephone service may be met promptly as it develops and further that plant additions may be along sound and economic lines, calls for careful planning. To this end the funda- mental plans prepared for the different exchange areas forecast the telephone development from 15 to 25 years in the future. Such fundamental plans show the proposed central ofifice locations, the boundaries of the districts to be served by each office, and the plan of the underground conduit system. They are based on analysis of the existing market for telephone service; the forecasted market at a future date, considering both growth and distribution of population; expected changes in wage levels; estimates of the amount of service that will be sold under probable future rate conditions; and other factors. Station Apparatus One of the most important parts of the telephone plant is the appa- ratus installed on the subscribers' premises known as the station apparatus. Of this equipment the telephone transmitter and tele- phone receiver are fundamentally important elements and continued research work has been carried out to improve the efficiency, clarity of reproduction and reliability of these instruments. As a result of improvements in transmitters, receivers and induction coils the over- all efficiency, for example, of the station apparatus has since 1912 increased by a factor of 6.5. At the present time commercial trans- mitters when fully energized by direct current, are capable of deliver- ing electrical energy in the form of voice currents 200 times as great as the acoustic energy of the voice of the speaker by which the trans- mitter is actuated. For the most important part of the frequency range used in speech this ratio of output power to speech power is considerably greater. That is to say, the transmitter acts as a high ratio amplifier. In the Bell System the type of station equipment most generally in use is the desk stand. As the result of extensive development work it has been possible to produce a hand set which has transmission characteristics equal to those of the desk stand equipped with the best instruments heretofore in use. The hand set development in- volved the solution of difficult problems, the principal of which were to prevent singing or distortion of quality on account of the rigid connection between receiver and transmitter and to make the trans- 10 BELL SYSTEM TECHNICAL JOURNAL mitter efficient through the wide range of positions in which it is placed by the user. The latest form of this instrument is shown in Fig. 3. In addition to the usual black finish, this telephone as well as the bell box and other station apparatus have recently been made available in five colors, statuary bronze, old brass, oxidized silver, ivory and French gray. Practically all the service for business purposes is provided by individual lines or by private branch exchanges as discussed later. CSi^ Fig. 3 — The latest form of hand set. For residences, howeyer, there is in the United States a large develop- ment of two-party and four-party lines. The two-party stations are provided with selective ringing so that each station is signaled only for its own telephone messages, and the four-party stations are pro- vided in some places with selective ringing and in others with semi- selective ringing. Party lines have furnished a satisfactory means of providing service to small users and have been an important factor in the development of new fields of service in residences. To care for situations where something more than a single line TELEPHONE SYSTEM OF THE UNITED STATES 11 with one or two telephones is needed, but where an inter-communi- cating system or private branch exchange is not justified, so-called wiring plans are used which provide various arrangements for associ- ating the station equipment with the telephone lines. For the most part the customers' needs are satisfactorily met by one of the ten standard arrangements in general use. A specific example is that of Fig. 4 — -Telephone booth provided for public telephone stations. two central office lines with two main and two extension stations. Calls to or from either telephone line may be made from any one of the four telephones. Answering at a main station provides privacy by cutting off all the other telephones. There are in the United States a considerable number of extension stations. At the present time there are in the Bell System over 1.3 12 BELL SYSTEM TECHNICAL JOURNAL million of such stations. This number is rapidly increasing particu- larly for residence use as people appreciate further the advantages of having telephones in a number of convenient locations. The best residences are more and more being equipped to have telephones available in all parts of the house. In order to make telephone service possible for those people whose sense of hearing is more or less deficient, special sets are installed. By means of a vacuum tube amplifier which the user can adjust, the receiving may be amplified so as to bring the range up to the point giving best results, this point depending on the degree of im- pairment of his hearing. Public telephone stations constitute an important part of telephone development in the United States, there being at present more than 275,000 of such stations in service. Whereas residence and business service is largely given by contract, the customers contracting to pay a definite amount per month or a certain amount per call, a great many of the public pay stations are supplied with coin boxes by means of which the money is collected at the time the call is made. These installations are also for the most part in booths to insure quiet and privacy. Fig. 4 shows a form of booth furnished by the telephone companies, provided with a seat and with lighting and ventilation as well as a convenient location for the necessary telephone directories. Private Branch Exchange For customers who use a large amount of telephone service, one of several types of private branch exchange is provided which not only permits distribution of the incoming calls to the particular station desired but also makes it possible for one extension to call another without going through the central office. In the Bell System Private Branch Exchange Stations 2,740,000 Per Cent of Total Bell Stations 19.0 Private Branch Exchange Boards Cordless '. 53,300 Cord 60,900 Total 114,200 Private Branch Exchange Cord Positions Manual 68,600 Dial 1,700 Total * 70,300 Private Branch Exchange Attendants Cord Board Attendants 75,000 Cordless Board Attendants 53,000 Total 128.000 Fig. 5^Private Branch Exchange Statistics for the Bell System as of Feb. 1, 1929. TELEPHONE SYSTEM OF THE UNITED STATES 13 there are 36 million telephone connections handled each day by about 128,000 private branch attendants. About 17 per cent of all local calls originate at these boards. Equipment of both the manual and dial type is installed, the latter being particularly adapted to extension- to-extension calling. Further data regarding private branch exchanges are given in Fig. 5. The smallest manual installation is the cordless board illustrated in Fig. 6 where connections are established by means of keys, and the capacity is limited to three trunks and seven stations. A larger type, Fig. 6 — -Cordless private branch exchange installation for 7 stations and 3 central office trunks. using cords for the completion of connections, is illustrated in Fig. 7 with a capacity for fifteen trunks and 200 stations. For the largest private branch exchanges the equipment is of much the same type as that used at central offices. A large switchboard for one of the public utilities having 1,600 stations is shown in Fig. 8. Connecting this private branch exchange with the central office there are 148 lines and in addition 151 tie trunk lines extend to other private branch exchanges having a business association. There are a total of 42 switchboard positions. 14 BELL SYSTEM TECHNICAL JOURNAL •m - ■"^^Mi % i ■ f 1 ••'ti '■-•■••':,L^ III III! II nil k Fig. 7 — Private branch exchange switchboard arranged for 15 trunks and 200 stations. Fig. 8 — Private branch exchange— 1600 stations, 148 lines to Central Office, 151 tie trunks to other private branch exchanges, 42 switchboard positions, 60 operators and 24 hour service. TELEPHONE SYSTEM OF THE UNITED STATES 15 At the end of 1928 in the Bell System there were about 650 dial P.B.X. installations with about 100,000 lines. The smaller sizes of this equipment are designed to meet the needs of the larger residences and the larger sizes are adequate for any business office and large industrial plant. Typical equipment arrangements are shown in Fig. 9. In general the private branch attendants are in the employ of the subscribers having this type of equipment. It is essential that the attendants be recruited and trained with the same care as central office operators, and to this end, the telephone companies maintain employment bureaus and training courses for the benefit of private branch exchange attendants. Subscribers are encouraged to send their attendants to these training courses for retraining wherever this appears advisable. Instructors highly trained in local and toll central office operation and in the best methods for handling private branch exchange work constantly visit private branch exchanges in order to be of assistance both to the subscribers and to the attendants in giving the best pos- sible service. Cable Plant While open wires are occasionally used in limited quantity at out- lying points, 96 per cent of the exchange area wire plant is in cable. Of this 74 per cent is underground. An outstanding development is the steady increase in the number of pairs of conductors which it is possible to place in a single lead sheath of 6.7 cm. outside diameter. From the early use of 30 to 60 pair 19 gauge conductors there has been a continual increase in the number of pairs and a decrease in the size of the wires until at the present time 1,800 pairs of 26 gauge conductors are placed under a single sheath for use in the denser areas. The significance of this development is indicated in Fig. 10 which shows the year in which each important step was taken and the relative cost per pair of con- ductors resulting from each step in the development. In urban development main cables, called feeder cables, usually of the maximum size, radiate from each central office through under- ground conduit to the various parts of the area served. These feeder cables in general are run full size for considerable distances rather than being diminished at branch cable points, and the flexibility thus obtained is found advantageous for conditions in this country. Each main feeder cable has smaller branch cables bridged to it at intervals. The main feeder cable may continue all the way through the area 16 BELL SYSTEM TECHNICAL JOURNAL Fig. 9 — -Typical small dial private branch exchange installation. TELEPHONE SYSTEM OF THE UNITED STATES 17 or it may divide into two or more smaller cables branching out either underground or aerially in the area. A type of distributing plant located along the rear properties of a residential block is illustrated in Fig. 11. This represents the usual O o I < a. 00 < — a: CO -I u < < 0. cc. UJ a. 10 I- tn O u UJ > 16 15' 14 13 12 n 10 1 1 1.0 MM.i 50 PAIF 100 PAIR 1 132 PAIF ? .152 PAIf R IE 0 P AIRI di. Jb P MRt 303 PAli b. .c 1 Mh A. 46 5 P \\R- 404 fir PAI 6 Pi MR < 6 4 j^f^ 90 9 PAIR < — rsTTKT .4 0 MM 00 CO OD 00 o o o 01 o tn en 10 o (VI 01 00 (« o> YEAR Fig. 10. Fig. 11 — -Aerial cable distribution along rear property line. Poles used jointly for electric lighting power and telephone distribution. 18 BELL SYSTEM TECHNICAL JOURNAL type of construction followed in such areas. With continuous build- ing construction the distributing cables are often attached to the rear walls as in Fig. 12 or extended through the basements of the buildings. Underground cables are carried in conduit consisting of various combinations of multiple tile duct. A typical duct run, shown in Fig. 12 — Distribution telephone cable attached to rear wall of building showing terminal boxes and entrance by twisted pair into cellar. Fig. 13, illustrates the materials and methods of construction generally employed. At the central office the conduit system is designed to meet the ultimate requirements of the building and terminates in a cable vault as shown in Fig. 14. With this entrance arrangement the main TELEPHONE SYSTEM OF THE UNITED STATES 19 cables are spliced in the cable vault to smaller units of silk and cotton insulated conductors which extend up through the building in slots or ducts to the main distributing frame. Switching Systems The outstanding development in switching systems for a telephone communication has, of course, been the rapid trend toward an increase Fig. 13 — -72 duct underground cable run under construction. in the extent to which the operations are performed by automatic machinery. The general characteristics of the different types of switching systems in use in the United States and their extent of use are briefly discussed below. Magneto Magneto switching arrangements are used in small places and scattered rural areas. They vary in size from an arrangement to interconnect two or three lines up to a switchboard handling 300 to 400 lines. The average size of the magneto switchboards in the Bell 20 BELL SYSTEM TECHNICAL JOURNAL System is 170 lines. At the end of 1928 there were about 3,500 mag- neto offices with approximately 5,500 operators' positions, serving 1.1 million telephones. Common Battery At the end of 1928 there were 2,036 common battery offices, the maximum size office serving 10,500 lines and the average being 3,700 Fig. 14 — Cable vault in Central Office, St. Louis, Missouri, showing entrance of cables through ducts and connecting to silk and cotton insulated cables extending up through the building. lines. There was a total of 46,000 switchboard positions where the subscribers' lines are answered. In addition, there were 13,000 so- called trunk positions which are required where it is necessary to trunk calls from one office to another in areas having more than one central office. The trend in development in manual switchboard has been toward performing more of the necessary switching and signaling operations automatically by means of somewhat more complicated circuit and equipment arrangements and less and less by the operator. These changes have resulted in less manual operating labor and in an im- TELEPHONE SYSTEM OF THE UNITED STATES 21 provement in the service. Some of the more important of these changes are as follows : 1. Automatic ringing which continues automatically at regular inter- vals until the subscriber answers. 2. Ringing tone, very much reduced in volume, to the calling sub- scriber automatically advising him when ringing is in progress. ,1. The audible busy signal, a tone placed on the calling line when the called line is busy. An important recent change in manual central ofhce equipment relates to the trunking methods employed in completing a connection from one central office to another. In most of the larger cities the so-called "straightforward" method is used. With this operating plan the number that is desired in the distant office is passed by the originating or "A" operator to the completing or "B" operator over the trunk that is used for completing the trunk connection. This is in contrast to the "call circuit method" where all orders between operators are passed over a separate wire known as a call circuit. The principal equipment changes at the "B" positions have to do with the different circuit plans for connecting the trunk operator's telephone set to the trunk. This is done either by means of a key, by means of plugging the trunk into a listening jack or automatically by means of suitable relays. At the "A" end the principal change is the arrangement for testing whether or not an outgoing trunk is busy. This is done either by means of a lamp indicating a free trunk or by a lamp or tone indicating that a part of a trunk group is free. Dial Equipment Dial equipment of two types known as the step-by-step system and the panel system respectively are used in about equal amounts in the Bell System. The change from manual to dial operation pre- sented a very large problem from an engineering, a manufacturing, an installing and an economic standpoint. At first the dial installa- tions were to care largely for growth but they have been followed by installations for the replacement of existing manual equipment where, all factors considered, this was clearly justified. In this way an or- derly program has been developed. Figure 15 indicates the total number of stations on a dial and on a manual basis for each year since 1921 and the expected program up to 1933. Under the present contemplated dial program it is estimated that the areas employing step-by-step equipment will be on a complete dial basis by 1937 and that all areas employing panel equipment will be substantially com- pleted by 1942. BELL SYSTEM TECHNICAL JOURNAL 1921-1928 ACTUAL 1929-1933 ESTIMATED <0 z o H «0 ti. o ft 2 O _J 2 20 1 8 1 6 14 1 2 1 0 8 .^-' .^'^ ^'' -*- rs.V- ^'"' ^-^^ ^ 0^ f)^"^ . — NUAL ^^^ — — . . STATIONS MA p ^^' ,.-'' ** 0^ '■^ ^ '^-^ -* 1921 '22 '23 '24 '25 '26 '27 '28 '29 '30 '31 '32 1933 END OF YEAR 2.7 3,6 5.6 8.7 12.5 15.5 18.7 22 27 33 38 42 47 PER CENT DIAL STATIONcs OF TOTAL Fig. 15. Fig. 15 — Relation between manual, dial and total stations — -Bell owned stations. Step-hy-Step Dial System The step-by-step system is used in the Bell System in single office areas and in the smaller and medium sized multi-office areas where the number of central offices is limited and consequently the trunking problems are not complicated. Step-by-step equipment is in service in 194 offices to which are connected 1.6 million stations. The fundamental unit of the step-by-step dial system is the selector illustrated in Fig. 16. This switch has a capacity for 100 terminals placed ten on a level and ten levels high, thus making possible the selection of any one of a hundred lines. By placing several selectors in series a network of central offices may be built up, each office serving 10,000 telephones. The selectors are mounted on iron frameworks and the terminals are cabled to cross-connecting frames so that any grouping can be made as may be demanded by the number of calls which the par- ticular selector handles. A typical arrangement of selectors is shown in Fig. 17. TELEPHONE SYSTEM OF THE UNITED STATES 23 Fig. 16 — -Typical step-by-step selector showing relays which control the circuit opera- tion mounted at the top, and the selector banks of 100 terminals at the bottom. Fig. 17 — Installation of step-by-step dial equipment showing selectors mounted on iron framework. 24 BELL SYSTEM TECHNICAL JOURNAL Panel Dial System In the largest cities and the smaller municipalities around them making up the large metropolitan centers, a much greater degree of complexity is encountered in switching the calls due to the large number of offices of varying types to which calls are destined and due Fig. 18— Typical panel selector frame. Capacity 30 selectors in front and 30 selectors in rear. Motor driving mechanism is at the bottom of frame and controllinc apparatus at either side. TELEPHONE SYSTEM OF THE UNITED STATES 25 Fig. 19 — Panel dial office sender frame showing the apparatus for five senders. 26 BELL SYSTEM TECHNICAL JOURNAL to the routings involved in order to trunk economically either large or small volumes of trafific. The panel system was developed to meet these requirements and is now installed in a number of such metropolitan centers, notably, New York, Chicago, Philadelphia, Boston, Detroit, Cleveland, St. Louis, Pittsburgh, Baltimore, San Francisco, Buffalo, Kansas City, Seattle, Providence and Omaha. Panel equipment is in service in 128 offices to which are connected 1.6 million stations. Fig. 20 — ■Installation of panel dial equipment. The unused floor space is provided for future growth. In the panel system the fundamental switching unit is a large switch consisting of five banks of 100 terminals each. The selectors, by which contact is made with any one of the 500 terminals, move ver- tically on both sides of the terminal banks. A typical panel frame having capacity for 60 selectors is illustrated in Fig. 18. In the panel system the selectors do not follow in synchronism with the impulses of the dial as in the step-by-step system. Rather, a group of apparatus known as the " sender " records the impulses and in turn directs the operation of the several selectors in the train until the called terminal is reached. By this means the trunking arrange- ments and the numbering scheme can be designed independently of each other. This, combined with the large capacity of the panel TELEPHONE SYSTEM OF THE UNITED STATES 27 selectors, makes possible economies in interofifice trunks and a reduc- tion in the number of selectors involved in completing a connection in a large exchange. The selection may be either to a dial office or to a manual office reached direct or through a tandem switchboard. Figure 19 illustrates the apparatus for five senders. A switchroom in a typical panel office is shown in Fig. 20. Buildings The buildings in the Bell System at present number about 6,000, excluding those occupied by the Western Electric Company. All of Fig. 21 — Telephone office in the residential district of Silver Spring, Maryland. De- signed for small manual switchboard with a present capacity of 2,200 lines. the larger and many of the smaller are owned by the telephone com- panies. The range in size of the buildings is illustrated by Fig. 21 showing a small building for a single manual switchboard with a present capacity of 2,200 lines, and Fig. 22 showing a headquarters office and equipment building in a large city. This building has 66,000 square meters of floor space, in the lower 9 floors has space for dial equipment to serve 100,000 telephones, and in the upper floors in- cludes offices for 5,000 people. Figures 23 and 24 further illustrate 28 BELL SYSTEM TECHNICAL JOURNAL U 13111 1 imiiunw lliiliilliii! liiiiHiiit^' Fig. 22— Combined equipment and office building, New Yori< City, containing headquarters of the New York Telephone Company, 31 stories, 66,000 square meters of floor space. Lower 9 floors arranged for toll tandem equipment and for dial equipment with an ultimate capacity of 100,000 telephones. Upper 22 floors arranged for offices with a capacity of about 5,000 people. TELEPHONE SYSTEM OF THE UNITED STATES 29 some of the recent combined equipment and office buildings for large cities. The objective in connection with buildings of all types, including equipment and office buildings, garages and warehouses, is to provide Fig. 23 — Combined equipment and office building at Detroit, Michigan containing headquarters of the Michigan Bell Telephone Company. 19 stories, 28,000 square meters of floor space. 13 floors arranged for equipment including toll board and dial equipment to serve 60,000 telephones. 6 floors are arranged for offices. 30 BELL SYSTEM TECHNICAL JOURNAL Fig. 24 — Combined equipment and office building at Cleveland, Ohio containing headquarters of the Ohio Bell Telephone Company. 22 stories, 25,000 square meters of floor space. 13 floors designed for equipment including the toll board and ultimate dial equipment for 100,000 telephones. 9 floors arranged for office space. TELEPHONE SYSTEM OF THE UNITED STATES 31 1 n ■H V 5S A" =r 4= ft ^^ * 1 ■ r -■ 1 r -r 1 1 ■■, 1 It DC _ . a -- - o i +--t -t-r I I l-T L-l h , II ;i II n V rr -I— *- rt itn ^ ^ ^ U u n tin i! II ii i| II II uu A T + -(- + ^ ^ |l n n^ n II '11 ^1 nn n fi ' I' 1 11 ti ri 1 1 II rtn T-r WOOy 5NllVU3dO SOUV08K9ilMS E Z .J n o Ii bo (U l-> 3 I : *** C d c P bD C cS u & en Ii o «41 ^3 0) c *s -a o \B ■o "rt a (L) ■M en I I a < Fig. 26 — -The average time required from tliL' placing of a toll call to the response of the called party, or until a definite report is made by the operator. a definite report regarding the call. The service is sufficiently fast so that on 95 per cent of the calls, the subscriber stays at the tele- phone. This makes possible still more rapid service and simplified operation. There have also been very great improvements during this period in the clearness of speech transmission. The maximum permissible transmission loss between two subscribers on a toll connection has been materially reduced. The toll plant and subscriber plant are now so designed that most of the messages are handled with a maxi- mum transmission equivalent for the longest subscriber lines of 20 to 25 decibels overall referred to the standard transmission reference system. TELEPHONE SYSTEM OF THE UNITED STATES 35 Short Distance Toll Service General To a large and increasing extent the toll messages are completed and supervised by the local exchange operators who first answer the subscriber's call, providing in this way toll service with the same methods which are applied to local service and with comparable ..iT L A N T 1 C OCEAN 159 OFFICES IN SUBURBAN TOLL AREA SCALE : Fig. 27 — New York suburban toll area indicating the number of offices in Metro- politan and suburban districts. Population in the area 10.5 millions. Telephones in the area 2.5 millions. speed. This method of operation is used for most of the toll business up to a distance of 50 kilometers and to a considerable extent up to 100 kilometers. The use of this method includes the extensive sub- urban areas around the large cities. Calls handled by this method now amount to 650 million messages a year and its increasing use has 36 BELL SYSTEM TECHNICAL JOURNAL been one of the important ways in which increased speed of service has been brought about. The handling of this suburban telephone trafific adds greatly to the complexity of the transmission, trunking and operating problem of the larger cities. This is illustrated by Fig. 27 showing the sub- urban toll area surrounding New York City. It will be noted that the city itself includes 168 central offices and in the suburban areas in the metropolitan district there are in addition 159 central offices. In many cases of the shorter haul toll service, the volume of traffic between two offices is sufficient to warrant direct trunks and the calls are completed over these direct trunks by the usual local traffic meth- ods. In order to provide an efficient trunking arrangement for the smaller volume of traffic between widely separated offices, however, tandem trunking arrangements are provided, by which the calls are routed through a central switching point and from that point dis- tributed to the terminating offices. Either manual or dial central office equipment is used as outlined below, each type having its field of application depending upon the amount of traffic and the portion of traffic to and from manual and dial central offices. The trunks to the central switching point, or tandem office, are in general of somewhat larger gauge than interoffice trunks because of the greater distances involved and the correspondingly more severe transmission requirements. In some of the longer trunks telephone repeaters are used. It has not been found generally economical to use repeaters at the tandem switching point although this is done in certain in- stances and it is possible that in the near future the more general use of repeaters in this way may become an economical means of meeting the transmission requirements. Manual Straightforward Tandem The manual straightforward tandem is used in those medium sized areas in which most of the suburban calls are between manual switch- boards. The arrangement of the equipment is shown in Fig. 28. The tandem trunks from the originating office terminate on the plugs located at the rear of the keyboard and the tandem completing trunks to the various terminating offices appear in jacks in the face of the switchboard. The completing trunks are provided with lamp signals indicating idle trunks. When a call comes in on a tandem trunk the operator is advised by a flashing lamp signal on that trunk and her telephone set is automatically connected to it. The work of the tandem operator is limited to making the connection at the tandem board and making the disconnection when advised by lamp signal TELEPHONE SYSTEM OF THE UNITED STATES 37 that the conversation is over. Her work is thus greatly simpUfied and the operation of the board is extremely rapid. Fig. 28 — Manual straightforward tandem board for handling suburban toll calls. Trunks from originating offices terminate on plugs at the rear of the keyboard. Tandem completing trunks appear in jacks in the face of the switchboard. Incoming calls indicated by flashing of lamp associated with trunk cord. Dial Tandem Dial tandem systems have been designed for handling suburban toll trafific in areas where a large proportion of the central offices are of the dial system and also to facilitate handling the complex sub- urban traffic around the large metropolitan areas, such as New York, Boston, Chicago and Philadelphia. The type of dial equipment, in general, corresponds with the type used for the local traffic in the same area. For use in connection with calls from dial offices, the dial tandem usually requires no operators at the tandem office. The originating operator in the dial office who handles the short haul traffic controls the selection of the trunk to the terminating office. In some of the large metropolitan centers, dial tandem apparatus of the panel type is employed also for handling calls from manual 38 BELL SYSTEM TECHNICAL JOURNAL Fig. 29 — -Panel type dial tandem office in New York City. Fig. 30 — -Panel type dial tandem office. Arrangement of keyboard at tandem positions. Incoming trunks appear as lamps and associated keys on upper sloping part of keyshelf. Calls are completed by setting up the called office and called number on keys on lower part of keyshelf. TELEPHONE SYSTEM OF THE UNITED STATES 39 offices. With this arrangement operators are, of course, required at the tandem board. The tandem operator receives a request from the originating operator for the office and number called and by means of keys establishes the connection by dial switching apparatus to the called subscriber in case he is in a dial office or transmits the required information to the terminating office operator in the case of a manual office. Figure 29 shows an installation of panel tandem operators' equipment and ¥\g. 30 shows one section of this equipment in greater detail. Fig. 31 — ^Step-by-step dial tandem board. Calls completed to the terminating dial subscriber station by means of the 10 button key set shown on the keyboard. Incoming calls automatically distributed to an idle operator. A modified form of step-by-step tandem equipment using operators has been installed in step-by-step areas for handling calls from manual offices to dial offices in cases in which it was not advisable to equip all the subscriber operators' positions with dials. This equipment includes, in addition to the selectors, a simplified type of switchboard as shown in Fig. 31. Long Distance Service General For the longer hauls the subscriber is connected to a toll board operator who completes and supervises the toll message. This method is called the toll board method of operation, and is used for most all 40 BELL SYSTEM TECHNICAL JOURNAL the messages over about 100 kilometers. For the purpose of this paper this service will be referred to as long distance service. The messages handled by this method total about 300 million messages a year. The amount of long distance business at New York City, for example, requires the use of 1,275 operators' switchboard positions. During the past three years an important change has been generally applied in the methods of handling long distance service. Formerly the toll operator first receiving the call recorded the necessary infor- mation on a ticket and forwarded this ticket to another operator provided with facilities for completing calls to the particular part of the country involved in each case. An increase in speed has been brought about by providing the operators with arrangements both for recording calls and for completing calls to all points so as to avoid, in a large proportion of the cases, the necessity for transmitting the information to a second operator. By means of this change in method and other improvements, the average speed of service for all long distance messages has been de- creased from 6.9 minutes in 1925 to 2.6 minutes in 1928. Also in 1928, 90.7 per cent of the calls made by the customers resulted in completed messages. In placing a long distance call, the telephone subscriber in the United States may give simply the telephone number and city desired. This has some advantages in speed of service. At present, 50 per cent of the long distance messages are handled in this way and this per cent is increasing. About 15 per cent of the messages are handled in this same way, the called telephone number, however, being sup- plied by the operator. In addition, the telephone system offers, for a somewhat greater charge, what is called a " particular person " serv- ice. This means that the subscriber may, if he wishes, ask to talk with a specified person at a distant point, giving such information as he can regarding how that person may be located. The telephone operator then undertakes to complete this message by locating the desired party, following him up to points other than that designated, if necessary, and if the calling subscriber wishes. The percentage of messages handled on this basis increases with the length of haul. When a subscriber wishes he may transmit to the telephone com- pany in advance information regarding a number of calls which he wishes to have completed in sequence, beginning immediately or at a specified time. These sequence calls, as they are termed, are used particularly in connection with selling by long distance telephone. At the present time at the New York long distance office, for example. TELEPHONE SYSTEM OF THE UNITED STATES 41 about five per cent of the business is in the form of sequence calls, some of the sequence lists including as many as 1,000 calls. Except during times of emergency conditions, it is not the practice 800 600 400 200 CO -t lO U3 r^ 00 CVI (\1 (\l fVJ OJ CVJ o 0> Ok o^ a> a* Fig. 32 — Toll board messages per year in thousands, New York-Boston. in the United States to limit the length of conversations on toll con- nections. As a result it is very general for conversations, particularly on longer hauls, to exceed the initial three-minute period, the average length for transcontinental conversations being, for example, six 42 BELL SYSTEM TECHNICAL JOURNAL minutes. Conversations which run one half hour or an hour are not unusual, and in one case a transcontinental telephone conversation was eight hours in all. A striking feature of the long distance service is the more rapid 3 50 300 250 fo ■ o» a> O) O) Fig. 2>i — -Toll board messages per year in thousands, New York-Chicago. growth of very long haul business than business of moderate length. Figures Zl, 33) and 34, for example, show respectively the growth in messages for the last five years between New York and Boston, 370 kilometers. New York and Chicago, 1,380 kilometers, and the trans- continental business between New York and Chicago at one end and TELEPHONE SYSTEM OF THE UNITED STATES 43 Los Angeles and San Francisco at the other end, an average of 4,700 kilometers. It will be noted that while the toll business as a whole has increased as noted above, 67 per cent in this period, the New York-Boston business has increased 62 per cent, the New York- 60 50 40 30 30 10 n Tt ir> U7 r^ 00 (VI rvj f\i (M (Vl C\l o> 0^ a> a> a\ Fig. 34 — Toll board messages per year in thousands, New York and Chicago to Los Angeles and San Francisco. Chicago business 194 per cent and the transcontinental business 430 per cent. In the attached statistical summary is given the basis used for determining long distance toll rates, including the practices in the 44 BELL SYSTEM TECHNICAL JOURNAL United States in the offering of reduced rates in evening and night hours. The effect of these reduced rates is in some cases temporarily to slow down service at the hours when the reduced rates first go into effect, because of the large demand for long distant business at those hours. Jp-f-^'^f''^' o TELEPHONE SYSTEM OF THE UNITED STATES 45 Telephone Toll Lines To handle the toll business of the United States has required com- pleting a network of toll telephone lines completely covering the country. This network is shown in a general way in Fig. 35. It tc u t- u O -J O a. o z o 18 16 14 IS \0 / > y ^ CABLE >/^ r / ^ y X ^ — — OPEN WIRE, CARRIER 1928 1929 1930 1931 1932 1933 Fig. 36 — Estimated toll circuit kilometers in plant Bell Operating Companies. consists at the present time of about 14 million kilometers of wire on about 300,000 kilometers of toll route. The toll circuits are partly open wire, supported on insulators and are partly in cable. Both the open wire and cable circuits are, in general, phantomed, giving three independent circuits for each two pairs of wires, and in addition on the open wire is superposed a considerable amount of carrier current telephone circuits. The distribution at the present time between 46 BELL SYSTEM TECHNICAL JOURNAL cable, open wire and carrier and the expected increase of each during the next four years is shown in Fig. 36. Open Wire and Carrier Circuits The standard construction for open wire telephone circuits in the United States is indicated by the diagram of Fig. 37 and a typical pole line built in accordance with this construction is shown in Fig. 38. 3-C 3-C BH 3-C 3-C PH r. -.., PH &V& &V& srioi e^e""d"va l-C PH l-C &v & & V & T 13 Tj Jp T l-C PH l-C 3J 6 Cr ti" 0 vO l-C l-C PH & v""8""'""&"^v a iff^&j f\ v"g"""""ff"v fl Fig. 37 — Pole Line Configuration, phantomed construction, 12 inch spacing between wires of non-pole pairs. Symbol Facility Total Circuits V Voice frequency-physical 20 PH Voice frequency-phantom 10 3-C Carrier system furnishing 3 telephone circuits 12 l-C Carrier system furnishing 1 telephone circuit 12 T D-C telegraph 40 BH Carrier telegraph (10 channel) 40 Total telephone 54 Total telegraph 80 The wires are of copper and the sizes and weights are shown in the following table. Diameter — mm. 2.6 3.2 4.2 Weights — kg. per km. 47 74 118 Bronze and aluminum are not, in general, used in the United States for telephone lines, being not as economical or as generally satis- TELEPHONE SYSTEM OF THE UNITED STATES 47 factory as copper, taking into account transmission efficiency and construction conditions. The wires are placed on 10 pin cross-arms and supported, in gen- eral, by double-petticoated glass insulators. The grouping of wires to form phantoms is indicated in Fig. 37. This arrangement has been found desirable for conditions in the United States and transposition systems have been designed by which are obtained satisfactory opera- tion of the phantoms and side circuits, the mutual induction between the various circuits being sufficiently neutralized to prevent mutual interference. Fig. 38 — -Open wire pole line construction. Four 10-pin cross arms. On the longer circuits telephone repeaters are installed at an aver- age distance of about 175 to 300 kilometers, providing in that way for adequate transmission efficiency. The number of circuits which it is practicable to provide by means of open wire lines has during the past decade been very greatly in- creased by the extensive use of carrier telephone for superposing on the telephone circuits additional channels of communication carried by currents above the voice range of frequencies. These systems now form a network covering the entire country and in some areas a large proportion of the circuit growth on open wire lines is taken care of by carrier systems. The systems range in length from a minimum of 75 kilometers to a maximum of 3,800 kilometers. Two types of carrier telephone systems are standard for use in 48 BELL SYSTEM TECHNICAL JOURNAL the United States. One of these, designed for the longer hauls, pro- vides on one pair of wires three telephone circuits in addition to the voice frequency circuits. These three carrier circuits are provided by the modulation of frequencies between about 6,000 and 28,000 CHANNEL I NUMBER CHANNEL I NUMBER FIRST SYSTEM 2 EAST I EAST 3 EAST SECOND SYSTEM ^ a EAST 1 EAST 3 EAST 8 la 16 ao 24 28 32 FREQUENCY- KILOCYCLES PER SECOND Fig. 39 — -Frequency allocation of two long haul carrier systems. The blocks indicate the range of the transmitted side band. The arrows are located at the carrier frequencies and indicate the direction of transmission. UJ >■ UJ O -I m o — o o o^ 5e^ O z a: u 1^ _] z$ 9.3 < u > UJ Q M 12 10 2 - -2 400 800 laOO 1600 2000 CYCLES PER SECOND 2400 2800 3200 Fig. 40 — Average overall transmission-frequency characteristic of Long Haul Carrier telephone system. cycles, different frequencies being used for transmission in the two directions. The different conversations are amplified together in a common repeater at intermediate points and at the terminals sep- arated by electrical filters providing for each circuit a band of approxi- mately 3,000 cycles. The frequency allocation for two varieties of TELEPHONE SYSTEM OF THE UNITED STATES 49 the long haul system in common use are shown in Fig. 39 and typical transmission characteristics for a carrier channel are shown in Fig. 40. The long haul carrier systems give very satisfactory service and form a part of some of the longest circuits in the country. For ex- ample, the direct circuits between New York and Los Angeles, Cali- fornia, 5,100 kilometers in length, are made up of cable circuits from New York to Pittsburgh connected permanently to a Pittsburgh- St. Louis carrier system, and a St. Louis-Los Angeles carrier system. These two carrier systems connected together total 4,550 kilometers in length with 13 intermediate repeaters. Similarly the New York- San Francisco circuits are in cable from New York to Chicago and there permanently connected to the Chicago-Sacramento carrier sys- tem 3,800 kilometers long with 10 intermediate repeaters. The short haul carrier system is similar in its general character- istics but is simplified and provides a single carrier circuit for each pair of wires. In the case of both systems, single side-band carrier suppression circuits are used. In Fig. 37 showing the standard arrangement of open wires on pole lines are indicated the carrier telephone channels and also the carrier telegraph channels which can be superposed on these circuits without mutual interference, after the installation of suitable transpositions which have been designed to neutralize the mutual induction between the circuits. It is noted that with this arrangement it is possible to obtain from 40 wires 54 telephone circuits. Also 80 telegraph circuits are obtained, used for special contract service as described later in this paper. On a number of the open wire toll routes carrying very long circuits, it has become important to provide arrangements for using a larger number of long haul carrier telephone systems, thus obtaining a larger number of circuits. Whereas a number of arrangements using the standard spacing of wires have been tried out, it is found ex- tremely difficult to continue the use of phantoms and to so transpose the wires as to provide adequate freedom of interference between the higher frequency carrier channels if these are used on all pairs. The difficulty of doing this is evident in considering that in order to avoid overhearing it is necessary that the power transfer between different circuits should not exceed one part in a million even though they are parallel to each other for long distances. In order to make possible the maximum use of long haul carrier systems where desired, trials have been made with the arrangement of wires shown in Fig. 41 and these trials have shown very satisfactory results. With this arrangement the spacing of the two wires of each 50 BELL SYSTEM TECHNICAL JOURNAL pair except the pole pairs is reduced to 23 centimeters and the phan- toms on these wires are abandoned. Type "C" systems can then be used on all of the pairs with this spacing. The result as indicated on the diagram is that a 40 wire toll line provides 70 telephone and 80 telegraph circuits. 3-C T T 3-C 3-C ~ ~ T T<^ T 88 T y^ T 3-C 8v8 3-C T T T T 3-C T T 3-C 3-C T T 3-C 3-C T T T T .fivs iom&i ava "^ ^^, i "^^ ° ."^1 ,<^ T 3-C ML T T vv^ Fig. 41 — Pole Line Configuration, non-phantomed construction, 8 inch spacing between wires of non-pole pairs. Symbol V PH 3-C T BH Facility Total Circuits Voice frequency-physical Voice frequency-phantom Carrier telephone D-C telegraph Carrier telegraph (10 channels) Total telephone Total telegraph 20 2 48 40 40 70 80 Toll Cables In spite of the great extension in the use of open wire circuits brought about through the application of carrier systems, as indicated above, it would be extremely difficult with the present rapid growth in toll business to provide by open wire toll lines the large numbers of telephone toll circuits now required on many routes. It is ver}- fortunate that the development of means for providing satisfactory long distance circuits through telephone cables has matured in time to enable this method of construction to be widely used to meet the TELEPHONE SYSTEM OF THE UNITED STATES 51 present demands. Also, the toll cables provide practical immunity from the effects of storms, including the sleet storms, which are a hazard to open wire construction in nearly all parts of the United States. The first long distance toll cables in the United States were placed in service in 1906 between New York and Philadelphia and between Chicago and Milwaukee. These cables were both placed underground in multiple duct and are each about 150 kilometers long. The next step in the extension of toll cables was the completion in 1914 of an underground toll cable route between Boston, New York, Philadelphia and Washington, a distance of 730 kilometers. Cable running west from New York was completed to Chicago, a distance of 1,380 kilo- meters, in 1925 and to St. Louis, a distance of 2,150 kilometers, in 1926. This permitted placing in service circuits entirely in cable between New York and St. Louis. The present major toll cable routes together with the extensions which it is expected to complete during the next five years are indi- cated in Fig. 42. It is to be seen that in accordance with these plans toll cable will, within five years, extend entirely across the continent and up and down the length of both Atlantic and Pacific Coasts, will extend north into Canada and south almost to Mexico. In the north- eastern portion of the country where the development is the heaviest, there is already a multiplicity of toll cable routes and on some of these routes the rate of growth is high enough to require additional cables at successive intervals of one or two years. The amount of toll cable added to the network this year will be about 8,000 kilo- meters and this amount is expected to be increased materially in the following years. In the early toll cables before the extensive development of tele- phone repeaters, it was necessary, in order to provide satisfactory transmission, to use relatively large conductors and conductors up to a maximum size of 2.6 mm. diameter (No. 10 B and S gauge), were provided in the Boston-Washington cable. With the perfection of telephone repeaters for use with toll cable circuits, the transmission limitations on the extension of toll cable were removed and the economy of such circuits greatly increased by making it possible to use small conductors. The longest toll cable circuits at the present time are carried over conductors of 0.9 mm. diameter (19 B and S gauge). For the shorter circuits each path is used as a two-way circuit, while the longer circuits use separate paths for transmission in opposite directions. In order to improve the trans- mission characteristics, the circuits are provided with loading coils 52 BELL SYSTEM TECHNICAL JOURNAL / V < O 2 ; ' \ 5 ^ I ' » • \ 1 5 < 5^ \ * \ z 1 »c; A ~ >-". < 1 — ' ^» * -; * ' 1 z 'i z 5 1 ) 1 • -K, o 5 / < ' O ./ CO ct) a> c IJ a> 3 o c c cr c ^ c UJ 01 3 T3 ^ xt C K nj o UJ O TELEPHONE SYSTEM OF THE UNITED STATES 53 at intervals of 1,830 meters and at an average interval of about 70 kilometers are provided with telephone repeaters which renew the power of the attenuated voice currents. A single standard full size cable 6.7 cm. in diameter when so equipped is capable of providing between 250 and 300 long distance telephone circuits. The toll cable system includes various types of construction. For the routes having the most rapid growth, multiple duct subway is used. At the present time with the development of very heavy toll demands in many parts of the country, this type of construction is Fig. 43 — ^Typical aerial toll cable construction showing loading point. being extended very rapidly on a number of important routes. Mul- tiple tile duct with small splicing manholes located at intervals of 229 meters and large manholes for loading coil pots at intervals of 1,830 meters are generally used. For routes on which the growth is relatively light, for example, 40 or 50 circuits a year and where underground construction is de- sirable, two other types of construction have been used to a limited extent. In one type the cable is placed in a single duct of fibre and in the other type of construction cable covered with a double layer of steel tape is placed directly in the earth. With both of these types of construction, manholes are built only at loading points. 54 BELL SYSTEM TECHNICAL JOURNAL In many places the character of the country is such that under- ground construction would be very expensive. In such cases, and in other cases where it seems desirable, aerial toll cable construction has been used extensively in the United States. With this type of construction the cable is suspended from a steel messenger wire sup- ported on poles. Figure 43 shows typical aerial cable construction, including a loading point, the pots of loading coils being supported on an angle iron pole fixture. Long circuits in toll cables have some extremely interesting elec- trical characteristics. Figure 44 shows the net transmission charac- U1 _l m O u Q — llJ > UJ UJ Q -I o >- o o o o 1 2 10 8 6 4 2 0 - ■? \ / \ / \ / \ i / \ ^ y \ ^ ^ 400 800 1200 1600 2000 CYCLES PER SECOND Fig. 44. 2400 2800 3200 teristic over a range of frequencies of a New York-Chicago toll cable circuit 1,380 kilometers in length. It will be seen that the voice frequencies are transmitted with nearly the same net efficiency over a sufficiently wide band to give a high grade of telephone transmission. The net characteristic indicated, however, is obtained by almost wholly neutralizing with telephone repeaters the very large trans- mission loss in the circuit. The New York-Chicago circuit, for ex- ample, would have an attenuation loss at 1,000 cycles of about 470 db, which means that without amplification the ratio of output power at one end to input power at the other end of the circuit would be 10"'*^ The combined gain of the 19 telephone repeaters in the circuit is about 461 db, giving about 9 db net equivalent, lender these condi- tions, it is evident that a careful regulation of the circuit is essential. For example, variations in the temperature of a circuit in the course of a day could make as much as 30 db or 1,000 fold difference in the TELEPHONE SYSTEM OF THE UNITED STATES 55 electric power received at the end of the circuit. To prevent such vari- tions affecting the net equivalent the long circuits are all provided with automatic regulators which adjust the gains of the telephone repeaters to compensate for the effect of temperature variations on the equivalent of the circuit. The effects of transmission delay are also very interesting and important. Voice waves travel considerably more slowly over cable circuits than they do over open wire circuits. For example, the velocity is about 30,000 kilometers per second for " longdistance" type cable circuits as compared to nearly 300,000 kilometers per second for non-loaded open wire circuits. One important result of delaying the speech waves is the "echo" effect. The transmitted currents are in part reflected at the distant terminal due to variations in the impedance of the receiving circuit. If the reflected currents transmitted back to the other end are delayed enough they may be heard by the talker as echoes of his voice. They may be again reflected at the sending end of the circuit and returned to the listener as an echo following the directly transmitted speech. The effects of these echoes are largely eliminated by devices known as "echo suppressors" by means of which the transmission of voice waves in one direction over the circuit causes interruption of the path over which the echo currents are transmitted in the opposite direction. However, the effectiveness of echo suppressors is limited by the necessity that they shall not be operated by noise currents of extraneous origin as this would interrupt conversations. The echoes, therefore, are an important factor to be taken into account in deter- mining the type of toll cable circuit to be provided to meet the trans- mission limitations imposed on the long distance circuits. In cable circuits introducing considerable transmission delay, the fact that the delay is not exactly the same for waves of different fre- quencies is also important, tending to give rise to what have been sometimes referred to as "transient" effects. In loaded cable circuits the waves of higher frequency are delayed more than those of lower frequency because of the fact that the loading is applied in lumps. The coils and condensers in the repeaters and auxiliary apparatus on the other hand, tend to delay the waves of lower frequency. The result is that the waves of intermediate frequency arrive first, fol- lowed by the waves of higher and lower frequency. Devices known as " phase compensators " can be used to reduce the effects, particularly those caused by the line. To improve the situation at the low end of the frequency scale special attention has been given to the design of the repeaters and auxiliary apparatus. 56 BELL SYSTEM TECHNICAL JOURNAL Still another effect of the transmission delay is to somewhat slow up and perhaps otherwise interfere with conversations due to the delay which is added to the ordinary time elapsing between question and answer. F"or example, if a cable circuit is 5,000 kilometers long Fig. 45 — Thirty 4-wire repeaters and associated testing equipment. The repeaters are arranged in groups of 3 with a minimum of cabling, each group being associated with a phantom circuit and its 2 side circuits. and the voice waves travel 30,000 kilometers per second, the time required for the waves to travel from one end of the circuit to the other is }/^ second and to make a complete round trip, 3^3 second. This }y^ second delay is evidently added to the ordinary time which elapses between question and answer. In the United States cable connections somewhat longer than 5,000 kilometers will be used in TELEPHONE SYSTEM OF THE UNITED STATES 57 the future, while for international connections, of course, very much longer distances than this will be involved. In the United States considerable study is, therefore, being given to the effects of trans- mission delay and to methods of avoiding difficulties on the very long connections including the development of cable circuits of higher speed. The toll cable circuits today include two principal types, one, discussed above, for the longer distances having a transmission speed of about 30,000 kilometers a second, and the other for the shorter distances, transmitting a narrower band of frequencies and having about one-half the transmission velocity. In view of the superior transmission characteristics of the long distance type circuits it is the present practice in the design of new toll cable circuits in the United States to limit the use of the short distance type facilities to circuits about 160 kilometers in length if they are to be used for switched business, and about 280 kilometers in length if used only for terminal business. Toll Circuit Equipment The apparatus required for the operation of toll circuits has been developed in the form of panels mounted on standard bays of angle iron, thus bringing about a great reduction in the space required compared with earlier forms of mounting. Figure 45 shows a bank of 30 four-wire repeaters arranged in groups of three, each group being associated with a phantom and its two side circuits. Figure 46 shows the panels containing complete terminal equipment for two type "C" carrier telephone systems (six circuits) with associated testing appa- ratus. The equipment is housed in fire-proof buildings. Figure 47 shows a typical telephone repeater station, this one being located at Prince- ton on the cable route between New York and Philadelphia. This building now contains 1,100 repeaters. Some of the telephone re- peater stations now being built are designed for ultimate capacity with extensions of 10,000 repeaters. An interesting feature of the long telephone circuits is the use of 1,000-cycle current for signaling rather than the lower frequencies which have been general in the past. This higher frequency has the advantage of being efficiently transmitted by the telephone circuit without change in the amplifying apparatus and hence does not re- quire intermediate ringing apparatus. At the terminals it is rectified and caused to operate relays which actuate the desired signal. 58 BELL SYSTEM TECHNICAL JOURNAL Fig. 46 — -Complete terminal repeater apparatus for two long haul carrier telephone systems (6 circuits) with associated testing equipment. TELEPHONE SYSTEM OF THE UNITED STATES 59 Fig. 47 — Telephone repeater building at Princeton, New Jersey on New York- Philadelphia cable route. Building now houses 1 100 repeaters. Ultimate capacity 2200 repeaters. Fig. 48 — Earth boring machine and derrick. Will bore 60 centimeter hole 2 meters deep in loam or clay soil in about one minute and in stone or frozen soil in 5 or 10 minutes. Derrick operated by power driven winch for setting poles. Truck provided with four wheel drive. 60 BELL SYSTEM TECHNICAL JOURNAL Fig. 49 — Trenching maciiine. Digs trench 1.7 meters deep and 55 centimeters wide, at speeds varying between 0.2 and 1.2 meters per minute, and is carried from job to job upon a trailer drawn by 23^ ton truck. i-^^Stf^^. Fig. 50 — ^Automobile truck equipped with tracks for hauHng cable on private right of way. With tracks, speed about 16 kilometers per hour. Can carrv 4500 kilogram reel up 40% grade. Tracks can be removed using special equipment provided for that purpose; without tracks speed 27 kilometers per hour. TELEPHONE SYSTEM OF THE UNITED STATES 61 Toll Line Construction The construction of toll lines under a wide variety of conditions has required the solution of many interesting problems. The rela- tively high cost of labor in the United States contributes to the ex- tensive use of labor saving machinery, a large amount of which has been developed to meet the particular conditions of telephone con- struction. Figures 48, 49 and 50 illustrate some of the more inter- esting types of labor saving machinery used extensively for both open wire and toll cable construction. Numerous special construction problems are, of course, met in specific situations. One of the interesting river crossings is illustrated in Fig. 51. m^^»sf*:*afjvsi»^;wftpi^'»il^^ ■^ /tl^i liW: 'Mm^i Fig. 51 — Special aerial cable construction across a river. Cable and messenger secured to a catenary suspension wire. 2-spans each about 180 meters long. Switching of Toll Circuits As far as is economically practicable the toll business is handled by direct circuits without intermediate switching. At the present time this includes 80 per cent of the toll messages. Of the remaining 20 per cent, 17 per cent have one intermediate switch and 3 per cent more than one intermediate switch. It is the purpose of the Bell Telephone System to design the toll 62 BELL SYSTEM TECHNICAL JOURNAL 'J"Wf^. r' in _aj 'S a g o u tao G TELEPHONE SYSTEM OF THE UNITED STATES 63 telephone system in the United States to give satisfactory service between any two points in the country. In order to accompHsh this it is necessary to make arrangements for a minimum number of switches between any two points. Also the toll circuits which will be used as parts of the built-up connections must be designed for a very high standard of transmission so that the overall efficiency of the built-up connection will be satisfactory. Arrangements have recently been worked out in the United States for meeting requirements of switched traffic more satisfactorily than has heretofore been possible. These arrangements may be briefly described as follows: At different points in the country there have been selected a group of eight very important switching points shown in Fig. 52. These eight regional centers will all be interconnected by high grade groups of circuits directly, that is, without intermediate switch. Through- out the country there are selected about 147 important switching points known as primary outlets also indicated in Fig. 52, each of which is directly connected to at least one of the regional centers. Each of the remaining 2,576 toll offices in the country will be con- nected to at least one of these important switching points. Further- more, within limited areas, such for example as a State, all important switching points will be directly interconnected. Within such an area, therefore, any two toll offices can be connected together with not more than two intermediate switches. Also, every toll office can be connected to a regional center with not more than one switch and through that center can reach any other toll office in any part of the country with a minimum number of switches. To insure adequate transmission on the switched connections, each of the important switching points will be provided with means for automatically inserting gain in the connection when two toll circuits are switched together so that the overall connection may be operated at the highest possible efficiency. This will, in general, be done by automatic adjustment of the gain of terminal repeaters permanently installed in circuits which must, in general, because of transmission limitations be operated at a lower efficiency when used for terminal business than when used as parts of a built-up connection. Maintenance of Toll Service With the present network of long distance lines in the United States, it is common to have 20 or more repeaters installed on each of the longer circuits and this number will increase greatly with the further extension of toll cable. The maintenance of service over these long 64 BELL SYSTEM TECHNICAL JOURNAL and complicated circuits is a very considerable problem both from the standpoint of technique and of organization. In this paper, these problems will not generally be discussed, but certain features will briefly be indicated. The service maintenance of the circuits includes periodic tests of transmission efficiency with transmission measuring sets designed for rapid and efficient use by the plant maintenance forces. The fre- quency of tests varies according to the requirements of each circuit group. To expedite the testing and adjustment of the circuits the longer cable circuits are subdivided into circuit units, these units usually being in cable about 160 to 240 kilometers in length, including the conductors and equipment involved in one section arranged for the automatic compensation of temperature variations. When trouble occurs on a long circuit, the circuit unit in which the trouble is located is immediately replaced and the location of trouble within the circuit unit then can be carried out without further interruption of service. The responsibility for establishing and maintaining each circuit group is given to a control office which is provided with private communica- tion channels to all parts of the circuit. An important feature in the maintenance of long toll circuits is the physical relations between the telephone circuits and circuits for the transmission or distribution of electric power. The Bell Tele- phone System and the power companies of the United States as repre- sented by the National Electric Light Association are very actively cooperating in a study of the best means of so coordinating the plant of telephone and power companies as to avoid interference under the various types of conditions important in practice. By means of this work it has been possible to find in every case a satisfactory solution permitting each utility to extend and increase its service along natural lines and providing proper protection of the telephone service. International Connections General The connections between the telephone systems of the United States and the telephone systems of other countries are indicated in Fig. 53. The territory of the United States has direct contact with only two other nations, Canada on the north and Mexico on the south. The common language and the close commercial relations between Canada and the United States have naturally resulted in a well de- veloped arrangement of lines connecting the telephone systems of the TELEPHONE SYSTEM OF THE UNITED STATES 65 two countries. Telephone connection between the cities of the United States and Mexico was not made until 1927, due to the unsettled political conditions which obtained for some years in Mexico. The many close commercial, political and social relations between the peoples of Europe and America have naturally drawn the attention of telephone men for many years to the possibility of establishing Fig. 53. telephone communication between these two continents. It was a great satisfaction, therefore, to be able to inaugurate such a service in 1927. The transatlantic telephone circuits already connect over 20,000,000 telephone stations in North America to over 7,000,000 telephone stations in Europe, thus joining together over 85 per cent of the total telephone stations of the world. In somewhat more detail, the present status of the connections of the United States telephone system to the telephone systems of other countries is as follows: 5 66 BELL SYSTEM TECHNICAL JOURNAL Connections in North A merica Practically all the telephone stations in Canada have communica- tion to the telephone stations in the United States. There are ap- proximately 100 long distance circuits extending from cities in the United States to important Canadian centers, including Halifax, St. Johns, Montreal, Toronto, Hamilton, Winnipeg, Regina, Calgary and Vancouver. The remaining cities are reached either directly or by switching through the important centers. In addition to long distance circuits there are, of course, many short distance circuits connecting points on opposite sides of the boundary which have local relations with each other. The various companies and provinces in Canada cooperate very closely with the Bell Companies in the United States in the maintenance of international service and, in general, telephone practices are very similar or identical in the two countries. Telephone communication is extended from the United States to Mexico by means of a telephone line crossing the border near Laredo, Texas. Direct long distance circuits extend from points in the United States to Mexico City, Tampico and Monterey and through these centers to about one-half the telephone stations in Mexico. Local toll circuits cross the border at a number of places. Telephone communication was established between the United States and Cuba in 1921 by the placing of three telephone cables between Key West and Havana. Each of these cables furnishes one telephone circuit and a maximum of four telegraph circuits. The requirements for the cables were exacting since a length of about 190 kilometers is combined with a depth of water having a maximum of 1,860 meters. Each cable consists of a central conductor mag- netically loaded with a wrapping of fine iron wire and insulated with gutta percha compound. A metallic return path for the telephone currents is furnished by heavy copper tape wrapped outside of the insulation and, therefore, in contact with the surrounding water. Three of the telegraph circuits in each cable are obtained by using " carrier currents " at frequencies slightly above the voice range. The fourth is obtained by using frequencies below the voice range. Connections to Europe In 1915 the Bell System experiments on radio reached the point where telephone messages were transmitted by radio from the United States and were successfully received by engineers sent for the purpose to Paris and to the Hawaiian Islands. While the Great War delayed technical and commercial development, in 1923 the Bell Companies were able to carry out a successful demonstration of radiotelephone TELEPHONE SYSTEM OF THE UNITED STATES 67 transmission from a group of telephone officials in New York to a group of people interested in communication assembled for the pur- pose in London. The success of these experiments led to cooperation with the British Post Office and the establishment in 1927 of telephone service between New York and London. This service has now been extended to include the greater part of the telephones of North America and Europe. As indicated in Fig. 53 there now exist one long-wave and one short- wave telephone circuit between the two continents. A second short- wave circuit will be placed in service about June 1 of this year and a third in December. By the end of 1933 it is expected that there will be in service between New York and London a group of six cir- cuits consisting of three short-wave radio circuits, two long-wave radio circuits, and one cable circuit. Our best information indicates that the short-wave circuits will be suitable for service at least 60 per cent of the time, the long-wave, 90 per cent, and the cable, 100 per cent. Since the beginning of 1929, the average number of messages handled per week has been 275. For this period the average number of mes- sages per day, omitting Saturday and Sunday, has been 44. Eighty- nine messages were handled on Christmas Day, 1928. Certain technical features of these circuits are particularly inter- esting. The long-wave circuit operating at a frequency of approxi- mately 60,000 cycles employs the "single side-band carrier suppres- sion " method. This appears to be the only use of this method in radio, although it is widely used in " carrier " circuits over telephone wires. The energy saved by the suppression of the carrier and the increased selectivity permitted by the narrow band of frequencies which is transmitted gives this system a transmission efifectiveness as great as a system of three or more times as much power using the ordinary transmission method. At both ends the receiving stations are situated as far north as can conveniently be reached and use is made of highly directive receiving. It is estimated that at the United States end these two factors represent an improvement equivalent to an increase in power of five thousand times as compared to a non- directive receiving station located at the same latitude as the trans- mitting station. The short-wave transmitting and receiving stations located not far distant from New York and London employ highly directive antenna systems. The design of such antennas must take into account eco- nomic factors and possible reactions on receiving effects other than power efficiency such as fading. The improvements effected by such 68 BELL SYSTEM TECHNICAL JOURNAL systems depend on wave-length and transmission conditions. Under favorable conditions the improvement effected at each end is approxi- mately equivalent to a transmitted power increase of 100 times. The most useful wave-lengths for this service have proved to be in the vicinity of 16 meters, although wave-lengths of about 22 and H meters are also provided to increase the amount of time these circuits are satisfactory for service because at certain seasons and times of day they are more effective than the 16 meters wave-length. Service over the transatlantic facilities is carried on from 6.30 in the morning to 10.00 at night in New York, corresponding to 11.30 in the morning and 3.00 A. M. London. During the winter months the long waves give nearly continuous service over this period. Under summer conditions considerable difficulty is frequently experienced in maintaining the long waves during the afternoon period in New York, corresponding to the evening period in London. At these times, however, the short waves are usually effective. The projected transatlantic telephone cable will use new magnetic loading materials and new insulating compounds for submarine cables recently developed by the Bell Telephone Laboratories. It will have at least one intermediate repeater point at Newfoundland. A circuit of this kind, differing radically from radio circuit in its characteristics will add both to the message capacity and to the reliability of the transatlantic service. Connections to South America Figure 53 indicates a short-wave radiotelephone circuit from New York to South America which, it is expected, will be in service early in 1930. The South American transmitting and receiving stations, which will be in the vicinity of Buenos Aires, will be owned and oper- ated by the companies who operate the local telephone service in Buenos Aires and the wire lines extending to other points in South America. Special Services Telegraph Circuits While the Bell System handles practically no commercial telegraph message business, it plays an important part in meeting the communi- cation needs of the United States by furnishing a large mileage of telegraph circuits for the private use of individuals and institutions, and for the use of governmental departments. Over two million kilometers of such circuits are now in use. One-third of this amount is used by newspapers and press associations. The greater part of TELEPHONE SYSTEM OF THE UNITED STATES 69 70 BELL SYSTEM TECHNICAL JOURNAL TELEPHONE SYSTEM OF THE UNITED STATES 71 the remainder is used by commercial, financial and other organiza- tions. Between New York and Chicago, a distance of approximately 1,400 kilometers, there are slightly over 300 such circuits now in operation. Figure 54 shows the system of telegraph circuits furnished by the Bell Companies to one of the press associations. An indication of the importance of private communication systems to commercial and financial institutions is given in Fig. 55 which shows the telegraph circuits furnished by the Bell Companies to a single brokerage com- pany. The greater part of such telegraph circuits have in the past been operated by hand-speed Morse telegraph. At the present time, how- ever, nearly a third of the mileage is operated with telegraph printers and this method of operation is rapidly increasing. Two speeds of service employing printers are offered, one operating at 40 words per minute and the other operating at 60 words per minute. At the present time, in view of the use to which this service is put, no demand has arisen for multiplex operation, but this method of operation is possible and will be used if it should become desirable. The telegraph circuits were originally all obtained as a by-product of the telephone business by compositing or otherwise superposing them on the telephone wires, using direct current for the telegraph circuits. At the present time approximately two-thirds are obtained in this way. The remaining third are obtained by " carrier current " methods. The carrier current system of open-wire lines uses fre- quencies above the voice range and provides ten duplex telegraph circuits on each pair of wires. The carrier current system used on cable circuits employs frequencies within the voice range, the currents being transmitted over an ordinary telephone four-wire cable circuit. This system gives twelve duplex telegraph circuits on each such circuit. Telephone Circuits Provided for Private Use In addition to the usual telephone message business, the Bell Companies furnish telephone circuits for the private use of individuals and organizations. So-called " special contract " telephone circuits are set up between particular parties for their private use at definite times specified in the contract. Approximately 2,000,000 circuit km. hours of such facilities are now in use during each complete business day. This is the sum of the figures obtained by multiplying the length of each such special circuit by the number of hours per day it is continued in use. 72 BELL SYSTEM TECHNICAL JOURNAL About three quarters of this total is accounted for by circuits where the contract calls for 12 hours operation per day, nearly all the re- mainder is accounted for by circuits which remain in vService 24 hours per day. A remaining small fraction is made up of shorter period contracts which are permitted to be as short as 30 minutes per night one night per week, or 10 minutes per day five days a week. As an illustration of the extent of use of this service, there are at present 158 full-time special contract circuits between New York and Philadelphia and 89 of such circuits between New York and Boston. Foreign Exchange Service Closely related to the above is the furnishing of what is called foreign exchange service. This consists of an arrangement whereby a customer in one exchange area is provided with a circuit for his exclusive use to another exchange area, this circuit being associated with a telephone number in a distant exchange so that other telephone stations in that exchange can be connected to the special line without toll charge. By this means, a business office in Boston, for example, can be given a New York telephone number, all New York calls for that number being treated as local calls but being actually completed over the special line to Boston. This type of service has a considerable popularity, there being over 1,000 such lines in service at the present time. Most of them are for relatively short distances, but some are for material distances, the longest being between Cleveland and New York, a distance of about 900 kilometers. Telephone Networks for Program Transmission to Radio Broadcasting Stations Radio broadcasting has resulted in the development of networks of telephone circuits for transmitting programs from studios or other places at which they are picked up to the radio station or system of stations from which they are broadcast. By such telephone wire systems the ceremonies of the Presidential Inauguration on March 4, 1929, were simultaneously transmitted to 118 radio stations located all over the United States. A statement regarding these interesting telephone networks, the requirements which they must meet and their importance in program broadcasting in the United States is given in a separate paper presented to this Congress (see paper on Wire Systems for National Broadcasting by A. B. Clark). TELEPHONE SYSTEM OF THE UNITED STATES 73 74 BELL SYSTEM TECHNICAL JOURNAL Electrical Transmission of Pictures A commercial service for the electrical transmission of pictures between the cities of New York, Chicago and San Francisco was inaugurated in April, 1925. The eight cities now connected to this service and the routes of the lines used in connecting them are shown in Fig. 56. In addition, a portable transmitter is provided which may be moved to any desired point. At present this is located in the city of Washington, D. C. The pictures as transmitted are of about twelve centimeters by seventeen centimeters. Any size picture, of course, can be photo- graphed to come within these dimensions. The detail of each picture corresponds to 39.4 lines per centimeter in each direction, that is, each picture is composed effectively of about v300,000 independent elementary areas. The line time of transmission with the present commercial system is about 7 minutes. Pictures may be sent from any of the cities shown to one or to more of the other cities which are reached by this service. Newspapers use this service for the transmission of pictures of events of national importance or where matters arise in any part of the country of large news interest. For example, pictures of the inauguration of President Hoover were sent in this way to the newspapers in San Francisco. In view of the three hours difference in time between Washington and San Francisco the pictures were published in newspapers sold at a time of day earlier than that at which the event took place. The majority of the pictures transmitted are for business or social purposes including pictures of legal documents, advertising material to be simultaneously released at a number of separate points, pictures showing new styles in ladies' wearing apparel, personal greetings in the handwriting of the sender and finger-prints of criminals. The Western Union and Postal Telegraph Companies now have a service in which they will accept telegraph messages for " facsimile " transmission over this picture system between those cities which the system reaches. This service has not yet been offered long enough to show how much it will be used. Telephony in Connection with Aircraft Operation Telephony promises to play a very important part in the practice of commercial aviation. The Bell Telephone Laboratories are carry- ing out a large amount of development work on all phases of telephony for this purpose. One-way receiving sets have been developed per- mitting an airplane pilot to receive weather reports and to determine the direction of radio beacons. Experimental radio sets suitable for TELEPHONE SYSTEM OF THE UNITED STATES 75 two-way conversations between a moving plane and the wire telephone system have been developed and demonstrated. Safety of airplane travel depends a great deal on the rapid accu- mulation and dissemination of meteorological data. An experiment on a promising method of handling such data is being carried out on an airplane route between San Francisco and Los Angeles in the State of California. At each terminal landing field and at two intermediate fields meteorologists are located. At six periods during the day each of these is rapidly connected in succession by telephone to outlying weather observation points varying in number at the different points from three to sixteen. The information thus accumulated and co- ordinated at each of the four landing fields is rapidly transmitted to the other three fields by means of printer telegraph circuits connect- ing them. This constant rapid observation of weather conditions along the airplane route and over a considerable territory around it permits very accurate prediction of the weather conditions which any plane will meet in its travel over the route. Such weather pre- dictions may be communicated to the airplanes before starting or by radio during their flight. Printer telegraph circuits appear to be a particularly convenient means of interchanging information among important landing fields along airplane routes. Ship-to-Shore Telephony The Bell System development work on ship-to-shore telephony was originally started with wave-lengths in the neighborhood of 400 meters, which were later taken into the broadcasting range. In 1920 shore transmitting and receiving stations in northern New Jersey were equipped to operate simultaneously three separate telephone channels in this range. Through these radio stations any telephone subscriber could be connected experimentally to the steamships " Gloucester" and " Ontario " which were engaged in coastwise shipping from Boston southward. In October, 1920, a talk to one of these ships furnished an interesting part of a demonstration at a banquet in New York City tendered to the delegates to the " Preliminary International Communication Conference" which was meeting in Washington at that time. Development of ship-to-shore telephony has been delayed because of uncertainties regarding the commercial situation and wave-length assignments. At the present time the work is again being actively pushed using wave-lengths under 100 meters. A transmitting and a receiving station will shortly be in course of construction near the 76 BELL SYSTEM TECHNICAL JOURNAL seacoast of northern New Jersey and a radio-telephone set is about to be installed on the steamship "Leviathan" to operate with these shore stations. As this ship approaches or leaves New York it is expected to be possible to talk from it to any telephone in the Bell System. This is intended not only as a demonstration of the technical features of such a service but to afford an indication of the extent to which such a service will be used under commercial conditions. Radiotelephony is being used from shore stations to coastal boats in a number of cases in the United States, but not connected to the commercial telephone system. These include particularly certain boats of the U. S. Coast Guard Service. A careful study, including tests, has been made of telephone service to tugboats operating in New York harbor for the purpose of controlling and thus making more efficient the operation of such craft. So far, it is not clear that this service will be commercially justified. Telephony to Other Mobile Stations Consideration has been given to telephone connections for types of mobile stations other than ships and airplanes. Communication with moving trains can technically be carried out with facilities now available. Active studies are under way to determine the practica- bility of providing such service at a cost which would be attractive commercially and with apparatus which can be limited to a reason- able space on the train. Telephone Services of Railroads and Other Public Utilities The operation of railway systems requires a large amount of com- munication service. The dispatching of trains was, until recent years, carried out largely by the use of telegraph. This has been rapidly changed until at the present time on over 60 per cent of the total railway mileage the train dispatching is by telephone. The railroads' telephone service to stations in the Bell System is through P.B.X.'s leased to them by the telephone companies. In addition to this, the railroads frequently own private telephone circuits extending along their rights of way which connect to and are switched through these same P.B.X.'s. Similar arrangements are provided for meeting the special com- munication needs of electric power companies, oil pipe-line companies, and other utilities. Telephone Public Address Systems Experience in many cases has shown that with the public address system used by the Bell Companies it is possible to amplify speech TELEPHONE SYSTEM OF THE UNITED STATES 77 or other sounds so that they can be heard by an audience of prac- tically unlimited size. Such public address systems as they are called are used very extensively in large auditoriums and at large public gatherings. For example, the ceremonies of inauguration of President Hoover held on the steps of the Capitol in Washington were amplified by the public address system so as to be heard by a gathering esti- mated at a hundred thousand persons, gathered within a radius of about 300 meters. Furthermore, by using the public address system with suitable long distance telephone circuits, it is possible to convey the proceed- ings of such occasions simultaneously to audiences in all parts of the country. The local distribution of such proceedings is, however, now done largely by radio broadcast rather than by use of the public address system. A use of the public address system which so far has been taken advantage of only on a few special occasions is by providing two-way operation to interconnect two or more meetings held simultaneously in different places. A notable example of this usage is the joint meeting of the American Institute of Electrical Engineers and the Institution of Electrical Engineers in London on February 16, 1928, interconnected by the transatlantic telephone circuit. In this meet- ing, addresses were heard by both audiences and a resolution made in London and seconded in New York was jointly and unanimously carried. It is possible that this may foreshadow a future important use of a public address system. Television The possibility of transmitting pictures of a scene over electrical circuits at so high a speed that the effect is given of seeing at a dis- tance has naturally interested telephone people for a considerable while. However, the large amount of detail which is taken in by the human eye and the resulting broad band of frequencies required to transmit this detail as well as the necessary complexity of the terminal apparatus has, so far, prevented the development of a prac- tical service of this kind. In 1927 the Bell engineers demonstrated to a large number of interested people a television circuit which extended from New York to Washington, a distance of about 440 kilometers. The television pictures so demonstrated had a detail corresponding to 50 lines in each direction, that is 2,500 elementary areas and 18 such pictures were shown each second. Two circuits especially corrected for volume and phase distortion over a band width of about 20,000 cycles were 7H BELL SYSTEM TECHNICAL JOURNAL employed between the two cities. These circuits were, for the most part, in open wire although approximately 13 kilometers of specially loaded cable were necessary at the ends in entering the cities. By means of a separate talking circuit a person at one end of the system could talk to, as well as see, a person at the other end. Systems of approximately twice the detail and also systems adapted to the view- ing of larger scenes such as athletes in action have since been developed and demonstrated. Time Service Arrangements have been made in many parts of the country to furnish subscribers who desire it, accurate information as to the time of day. A subscriber wishing the information asks for or dials a particular number assigned for this purpose and is connected either to an operator who advises him individually as to the time or is switched across a bus-bar to which is connected the amplified speech of an operator repeating at fifteen second intervals the exact time of day. In the present development of this service it is the practice to localize in one place the time service for an entire exchange area. By-Products Certain interesting and important by-products of the telephone development work justify a brief mention. Three arts separate from the telephone art have been radically changed by such by-product developments. These include submarine telegraphy, phonographs and motion pictures. The changes in submarine telegraphy have resulted from develop- ment by the Bell Laboratories of the materials known as " permalloy " and " perminvar " which have unusual magnetic properties at low flux densities. Submarine cables so loaded can transmit approximately 10 times as many words per minute in one direction as compared to cables of the same weight as previously constructed. As such loaded cables are not duplexed the effective increase in speed of transmission is approximately five times. Development work in connection with the faithful recording and reproduction of sound has greatly improved phonographs and their records. The "Orthophonic Victrola" is an example of such devel- opment. An extension of this work led to the development of the "talking " motion picture. The systems known under the names " Vitaphone " and "Movietone" followed from this work. Great interest has been aroused in such systems in the amusement field in the United States, TELEPHONE SYSTEM OF THE UNITED STATES 79 Moving picture houses in the important cities and towns are already equipped to show pictures of this type and it appears destined to revolutionize the motion picture art. A study of speech and hearing in connection with telephone service has led to the development of various devices of value to those having abnormal hearing or speech. This work has been carried out in close cooperation with interested members of the medical profession. One of these devices, the "audiometer," is useful in determining the con- dition of hearing of individuals by determining the smallest volume of sound at a considerable number of different frequencies which the individual can hear. This device, in rapidly testing large groups of people such as in the public schools, is believed to be of consider- able importance. Sound amplifying devices are provided for those hard of hearing. Another interesting by-product is an artificial larynx for those who have lost their natural larynx as a result of pathological conditions. Apparatus has also been constructed to permit the totally deaf to understand speech sounds by holding their fingers against a moving diaphragm. In one form the individual fingers and thumb are held against separate vibrating bodies and the important range of speech sounds is divided by electrical filters and one part of it applied to each of these five vibrating bodies. This partial electrical analysis of sound appears to be of considerable help in this tactual apprecia- tion of sound. Other tools of interest to the medical profession include electrical stethescopes and electro-cardiographs. The first of these permits any desired number to listen to chest or other sounds in medical patients. Electrical filters may be interposed in such arrangements to exaggerate or subordinate certain part of the sound. The electro- cardiograph, by permitting the amplification and recording of slight differences of electrical potential between selected points of the skin of a patient give an indication of the condition of his heart beat. Conclusion In the above discussion, while emphasis has been placed upon engi- neering matters, it has naturally been impossible in the discussion of results to separate engineering considerations from many other important phases of the telephone communication problem. While engineering is essential to the results that have been obtained, they are due also to these other factors, commercial and general in their character, and to the policies as regards service and operations which guide the Bell Telephone System. Furthermore, the solution worked 80 BELL SYSTEM TECHNICAL JOURNAL out has been designed specifically to meet conditions in the United States, conditions which in many respects are different in the different countries. It is, of course, not possible in a paper of such broad scope to give technical details of the engineering problems involved. These have, however, been quite fully set forth in numerous articles in the tech- nical press of the United States. For the convenience of those who may wish to refer further to these matters, a bibliography containing a selected list of some of the more impoitant articles is attached to this paper. In looking forward, there seems to be no doubt that the develop- ment of telephone communication in the United States, commercially and technically, will be more rapid than in the past, not less rapid. There are strong indications that in the future very much larger amounts of telephone service, both exchange and toll, will be de- manded than at the present time, and in fact that for a number of years at least the rate of growth will continue to increase. The type and extent of services supplied will be modified to meet the broaden- ing and multiplying demands of the changing business and social structure of the country. Finally, it is evident that the rapid advance of science will continue to bring forward new possibilities by means of which new and improved forms of communication systems, apparatus, and materials, can be developed. These facts all indicate that the engineering work for the telephone communication system of the United States is not complete nor de- creasing in magnitude or importance, but on the contrary it is increas- ing in volume and complexity and in the importance of the problems to be undertaken and solved. Authors' Note The authors wish to acknowledge their indebtedness to a large number of members of the organization for their assistance in the preparation of this paper. It is impracticable to mention all who have been of assistance but they wish to express their appreciation par- ticularly to Messrs. O. B. Blackwell, W. E. Farnham, W. H. Harden, H. S. Osborne, and W. A. Stevens. TELEPHONE SYSTEM OF THE UNITED STATES 81 Partial Bibliography of Papers Relating to the Bell Com- munication System General Ideals of the Telephone Service. Bell Telephone Quarterly, Vol. 1, Oct. 1922, pages 1-11. Science, Vol. 57, Feb. 23, 1923, pages 219-224, Annual Report of Smithsonian Institution, 1922, pages 533- 540. Semi-Centennial of the Telephone. Bell Telephone Quarterly', Vol. 5, Jan. 1926, pages 1-11. Telegraph and Telephone Age, Vol. 44, March 1, 1926, pages 98-101. Fifty Years of Telephone Progress, 1876-1926. Telegraph and Telephone Age, Vol. 44, Feb. 1, 1926, pages 51-53. Building for Service. Bell Telephone Quarterly, Vol. 7, April 1928, pages 69-81. General Engineering Problems of the Bell System. Electrical Communication, Vol. 4, Oct. 1925, pages 111- 125. Bell System Technical Journal, Vol. 4, Oct. 1925, pages 515-541. Bell System Research Laboratories. Electrical Communication, Vol. 2, Jan. 1924, pages 153-163. Development and Research in the Bell System. Bell Telephone Quarterly, Vol. 4, Oct. 1925, pages 266- 280. The Budget Plan of the Bell System. Bell Telephone Quarterly, Jan. 1923, pages 32-42. Electrical Communication, April 1923, pages 64-68. Service in the Making. Bell Telephone Quarterly, Vol. l,Oct. 1922, pages 26-33. Functions and Management Problems of the Traffic Depart- ment. Bell Telephone Quarterly, Vol. 5, Oct. 1926, pages 203- 218. Standardization in the Bell System. Bell Telephone Quarterly, Vol. 8, Jan. 1929, pages 9-24, and April 1929, pages 132-152. J. J. Carty J. J. Carty J. J. Carty H. P. Charlesworth H. P. Charlesworth E. B. Craft E. B. Craft C. A. Heiss K. W. Waterson K. W. Wateison H. S. Osborne Local Service General Selection of Central Office Names. Bell Telephone Quarterly, Vol. 6, Oct. 1927, pages 231- 237. The Planning of Telephone Exchange Plants. American Institute of Electrical Engineers, Transac- tions, July 1928, pages 809-817. Cable Plant Development of Cables Used in the Bell System. Bell Telephone Quarterly, Vol. 2, Apr. 1923, pages 94- 106. 1800-Pair Cable Becomes a Bell System Standard. Bell Telephone Qu arterly, Vol. 8, Jan. 1929, pages 25-29. A. E. \'an Hagan W. B. Stephenson F. L. Rhodes F. L. Rhodes 82 BELL SYSTEM TECHNICAL JOURNAL Switching Systems Machine Switching Telephone System for Large Metropolitan E. B. Craft Areas. L. F. Morehouse Bell System Technical Journal, Vol. 2, Apr. 1923, pages H. P. Charlesworth 53-89. Ameiican Institute of Electrical Engineers, Transac- tions, Vol. 42, Feb. 1923, pages 187-201. Machine Switching Private Branch Exchanges and Their \V. H. Harrison Application to Railroad Service. In American Railway Association, Telegraph and Tele- phone Section, Papers, 1924, pages 418-440. Panel Type Machine Switching System in the United States. H. P. Clausen Electrical Communication, Vol. 4, Oct. 1925, pages 91-97. Telephone Switchboard — ^Fifty Years of History. Bell Telephone Quarterly, Vol. 7, July 1928, pages 149-165. Buildings Housing the Bell System. Bell Telephone Quarterly, \^ol. 5, July 1926, pages 131-139. Post Office Electrical Engineers' Journal, \'ol. 19, Jan. 1927, pages 325-334. F. B. Jewett H. P. Charlesworth Toll Service Short Distance Toll Service Tandem System of Handling Short-Haul Toll Calls. American Institute of Electrical Engineers, Transac- tions, Jan. 1928, pages 9-20. Long Distance Service General Engineering the Long Lines. Bell Telephone Quarterly, Vol. 2, Jan. 1923, pages 18-31. Advance Planning of the Telephone Toll Plant. Ameiican Institute of Electrical Engineers, Transac- tions, Vol. 47, Jan. 1928, pages 1-8. Telephone Toll Lines Telephone Transmission Over Long Distances. Electrical Communication, Vol. 2, Oct. 1923, pages 81-94. American Institute of Electrical Engineers, Transac- tions, Vol. 42, Oct. 1923, pages 984-995. Some Very Long Telephone Circuits of the Bell System. Bell System Technical Journal, Vol. 3, July 1924, pages 495-507. Transmission Features of Transcontinental Telephony. American Institute of Electrical Engineers, Transac- tions, Vol. 45, Sept. 1926, pages 1159-1167. Open Wire and Carrier Circuits Carrier Current Telephony and Telegraphy. American Institute of Electrical Engineers, Transac- tions, Vol. 40, Feb. 1921, pages 205-300. Electrician, Vol. 36, May 6, 1921, pages 551-554. Practical Application of Carrier Telephone and Telegraph in the Bell System. Bell System Technical Journal, Vol. 2, Apr. 1923, pages 41-52. F. O. Wheelock E. Jacobsen J. J. Pilliod J. N. Chamberlin H. S. Osborne H. H. Nance H. H. Nance E. H. Colpitts O. B. Blackwell A. F. Rose TELEPHONE SYSTEM OF THE UNITED STATES 83 Making the Most of the Line. Electrical Communication, Vol. 3, July 1924, oages 8-21. Carriei Systems on Long Distance Telephone Lines. Bell System Technical Journal, Vol. 7, July 1928, pages 564-629. American Institute of Electrical Engineers, Transac- tions, Vol. 47, Oct. 1928, pages 1360-1386. Carrier Telephone System for Short Toll Circuits. American Institute of Electrical Engineers, Transac- tions, Vol. 48, Jan. 1929, pages 117-139. Toll Cables Boston to Chicago Telephone Cable — Section of Largest and Longest Cable Line in the World Being Completed to Pittsburgh, Pa., by A. T. & T. Co. Telephony, Vol. 81, Dec. 31, 1921, pages 15-18. Philadelphia-Pittsburgh Section of the New York-Chicago Cable. Bell System Technical Journal, Vol. 1, July 1922, pages 60-87. American Institute of Electrical Engineers, Transac- tions, Vol. 41, June 1922, pages 446-456. Development of Cables Used in the Bell System. Bell Telephone Quarterly, Vol. 2, Apr. 1923, pages 94- 106. F. B. Jewett H. A. Affel C. S. Demarest C. W. Green H. S. Black M. L. Almquist L. M. Ilgenfritz R. W. King J. J. PiUiod F. L. Rhodes Bancroft Gherardi William Fondiller Thomas Shaw William Fondiller Toll Cables — Loading Commercial Loading of Telephone Circuits in the Bell System. American Institute of Electrical Engineers, Transac- tions, Vol. 30, pt. 3, June 1911, pages 1743-1764. Commercial Loading of Telephone Cable. Electrical Communication, Vol. 4, July 1925, pages 24-39. Development and Application of Loading for Telephone Cir- cuits. Bell System Technical Journal, Vol. 5, 1926, pages 221-281. American Institute of Electrical Engineers, Transac- tions, Vol. 45, Feb. 1926, pages 268-292. Electrical Communication, Vol. 4, April 1926, pages 258-276. Permalloy; the Latest Step in the Evolution of the Loading F. L. Rhodes Coil. Bell Telephone Quarterly, Vol. 6, Oct. 1927, pages 239-246. Toll Cables — Transmission Telephone Transmission Over Long Cable Circuits. A. B. Clark Ameiican Institute of Electrical Engineers, Transac- tions, Vol. 42, Feb. 1923, pages 86-97. Electrical Communication, Feb. 1923, pages 26-40. Bell System Technical Journal, Vol. 2, Jan. 1923, pages 67-94. Building-Up of Sinusoidal Currents in Long Periodically J. R. Carson Loaded Lines. Bell System Technical Journal, Vol. 3, Oct. 1924, pages 558-566. Distortion Correction in Electrical Circuits with Constant O. J. Zobel Resistance Recurrent Networks. Bell System Technical Journal, Vol. 7, July 1928, pages 438-534. 84 BELL SYSTEM TECHNICAL JOURNAL Toll Circuit Equipment Telephone Repeaters. American Institute of Electrical Engineers, Transac- tions, Vol. 38, Oct. 1919, pages 1287-1345. Practical Application of the Telephone Repeater. The Western Society of Engineers Journal, Vol. 27, May 1922, pages 129-142. Telephone Repeaters. Electrical Communication, Vol. 1, Aug. 1922, pages 6- 10; Nov. 1922, pages 27-36. Telephone Equipment for Long Cable Circuits. American Institute of Electrical Engineers, Transac- tions, Vol. 42, June 1923, pages 742-752. Echo Suppressors for Long Telephone Circuits. Ameiican Institute of Electrical Engineers, Transac- tions, Vol. 44, Apr. 1925, pages 481-490. Electrical Communication, Vol. 4, July 1925, pages 40-50. Bancroft Gherardi F. B. Jewett H. S. Osborne Bancroft Gherardi C. S. Demarest A. B. Clark R. C. Mathes Toll Line Construction Poles. Bell Telephone Quarterly, Vol. 1, Oct. 1922, pages 34-44. Bell System Sleet Storm Map. Bell System Technical Journal, Vol. 2, Jan. 1923, pages 114-121. Specializing Transportation Equipment in Order to Adapt It Most Economically to Telephone Construction and Maintenance Work. Electrical Communication, Vol. 1, Feb. 1923, pages 50-59. Bell System Technical Journal, Vol. 2, Jan. 1923, pages 47-66. Open Tank Cieosoting Plants for Treating Chestnut Poles. Bell System Technical Journal, Vol. 4, Apr. 1925, pages 235-264. Bell Telephone Quarterly, Vol. 4, Jan. 1925, pages 132- 142. Recent Toll Cable Construction and Its Problems. Telephone Engineer, Vol. 32, Sept. 1928, pages 31-33. Switching of Toll Circuits Toll Switchboard No. 3. Bell System Technical Journal, Vol. 6, Jan. 1927, pages 18-26. Electrical Communication, Vol. 5, Apr. 1927, pages 255-259. Maintenance of Toll Circuits Measuring Methods for Maintaining the Transmission Effi- ciency of Telephone Circuits. American Institute of Electrical Engineers, Transac- tions, Vol. 43, Feb. 1924, pages 423-433. Electrical Tests and Their Applications in the Maintenance of Telephone Transmission. Bell System Technical Journal, Vol. 3, July 1924, pages 353-392. Practices in Telephone Transmission Maintenance Work. American Institute of Electrical Engineers, Transac- tions, Vol. 43, 1924, pages 1320-1330. Bell System Technical Journal, Vol. 4, Jan. 1925, pages 26-51. F. L. Rhodes J. N. Kirk J. N. Kirk T. C. Smith H. S. Percival John Davidson, Jr. F. H. Best W. H. Harden W. H. Harden TELEPHONE SYSTEM OF THE UNITED STATES 85 International Connections Connections in North America Key West-Havana Submarine Telephone Cable System. American Institute of Electrical Engineers, Transac- tions, Vol. 41, Feb. 1922, pages 1-19. Comiections to Europe Telephoning to England. Radio Broadcast, Vol. 2, March 1923, pages 425-426. Transatlantic Radio Telephony. Bell System Technical Journal, Vol. 2, Oct. 1923, pages 116-144. American Institute of Electrical Engineers, Transac- tions, Vol. 42, June 1923, pages 718-729. Transatlantic Radio Telephone Transmission. Bell System Technical Journal, Vol. 4, July 1925, pages 459-507. Institute of Radio Engineers, Proceedings, Vol. 14, Feb. 1926, pages 7-56. Radio Telephone Developments of the Bell System. Bell Telephone Quarterly, Vol. 5, Oct. 1926, pages 219- 237. New York-London Telephone Circuit. Bell System Technical Journal, Vol. 6, Oct. 1927, pages 736-749. \'oices Across the Sea. North American Review, Vol. 224, Dec. 1927, pages 654-661. Transatlantic Telephony — The Technical Problem. American Institute of Electrical Engineers, Journal, Vol. 47, May 1928, pages 369-373. Bell System Technical Journal, Vol. 7, Apr. 1928, pages 161-167. Transatlantic Telephone Service — Service and Operating Fea- tures. American Institute of Electrical Engineers, Journal, Vol. 47, Apr. 1928, pages 270-273. Bell System Technical Journal, Vol. 7, Apr. 1928, pages 187-194. W. H. Martin G. A. Anderegg B. W. Kendall R. W. King H. D. Arnold Lloyd Espenschied Lloyd Espenschied C. N. Anderson Austin Bailey J. O. Periine S. B. Wright H. C. Silent Bancroft Gherardi O. B. Blackwell K. W. Waterson Special Services Telegraph Circuits Metallic Polar-duplex Telegraph System for Long Small-gauge Cables. American Institute of Electrical Engineers, Transac- tions, Vol. 44, Feb. 1925, pages 316-325. \'oice-Frequency Carrier Telegraph System for Cables. Electrical Communication, Vol. 3, Apr. 1925, pages 288- 294. Telephone Networks for Program Transmission to Radio Broad- casting Stations Telephone Circuits Used as an Adjunct to Radio Broadcasting. Electrical Communication, Vol. 3, Jan. 1925, pages 194-202. Telephoning Radio Programs to the Nation. Bell Telephone Quarterly, Vol. 7, Jan. 1928, pages 5-16. How Chain Broadcasting is Accomplished. Radio Broadcast, Vol. 12, June 1928, pages 65-67. J. H. Bell R. B. Shanck D. E. Branson B. P. Hamilton H. Nyquist M. B. Long W. A. Phelps H. S. Poland A. F. Rose L. N. Stoskopf C. E. Dean 86 BELL SYSTEM TECHNICAL JOURNAL Electrical Transmission of Pictures Transmission of Pictures Over Telephone Lines. Bell System Technical Journal, Vol. 4, Apr. 187-214. 1925, pages Telephone in Connection with Aircraft Operation Airways Communication Service. Bell System Technical Journal, Vol. 7, Oct. 1928, pages 797-807. Aviation, Vol. 25, Oct. 6, 1928, pages 1090-1091, 1136, 1138, 1140, 1142, 1144, 1146. Ship-to-Shore Telephony Radio Extension of the Telephone System to Ships at Sea. Institute of Radio Engineers, Proceedings, Vol. 11, June 1923, pages 193-239. Telephone Services of Railroads and Other Public Utilities Telephone Equipment for Train Dispatching Circuits: A dis- cussion of the Requirements, Development and Design of Latest Types of Equipment for High Grade Train Dispatching Systems Including Vac- uum Tube Amplifiers and Loud Speakers. Electrical Communication, Vol. 2, Oct. 1923, pages 111-140. Recent Developments in Telephone Train Dispatching Cir- cuits. Railway Signaling, Vol. 17, Feb. 1924, pages 73-75; May 1924, pages 208-211; June 1924, pages 253- 256; Aug. 1924, pages 320-322. Telephone Public Address Systems Use of Public Address System with Telephone Lines. Bell System Technical Journal, Vol. 2, Apr. 1923, pages 143-161. Electrical Communication, Vol. 1, Apr. 1923, pages 46-56. American Institute of Electrical Engineers, Transac- tions, Vol. 42, Feb. 1923, pages 75-85. High Quality Transmission and Reproduction of Speech and Music. Electrical Communication, Vol. 2, Apt. 1924, pages 238-249. American Institute of Electrical Engineers, Transac- tions, Vol. 43, Feb. 1924, pages 384-392. Television Television. American Institute of Electrical Engineers, Transac- tions, Vol. 46, June 1927, pages 913-917. Production and Utilization of Television Signals. American Institute of Electrical Engineers, Transac- tions, Vol. 46, June 1927, pages 918-939. Synchronization of Tele\asion. American Institute of Electrical Engineers, Transac- tions, Vol. 46, June 1927, pages 940-945. Wire Transmission System for Television. American Institute of Electrical Engineers, Transac- tions, Vol. 46, June 1927, pages 946-953. Radio Transmission System for Television. American Institute of Electrical Engineers, Transac- tions, Vol. 46, June 1927, pages 954-962. H. E. Ives J. W. Horton R. D. Parker A. B. Clark E. B. Craft H. \V. Nichols L. Espenschied W. H. Capen W. H. Capen W. H. Martin W. H. Martin Harvey Fletcher H. E. Ives Frank Gray R. C. Mathes H. M. StoUer E. R. Horton D. K. Gannet E. I. Green E. L. Nelson TELEPHONE SYSTEM OF THE UNITED STATES 87 R. W. King O. E. Buckley By-Products By-Products of Telephone Research. Bell Telephone Quarterly, Vol. 7, Oct. 1928, pages 304-312. Loaded Submarine Telegraph Cable. Bell System Technical Journal, Vol. 4, July 1925, pages 355-374. Electrical Communication, Vol. 4, July 1925, pages 60-70. American Institute of Electrical Engineers, Transac- tions, Vol. 44, June 1925, pages 882-890. Telegraph and Telephone Age, Vol. 43, Nov. 16, 1925, pages 524-525. Permalloy Loaded Cable. Electrical Communication, Vol. 2, Apr. 1924, pages 232-234. Man-made Ears for the Deaf; Why Many Deaf People Hear Normally in Noisy Places and Over the Telephone. Scientific American, Vol. 8, Nov. 1925, pages 320-321. Recent Advances in Wax Recording. Bell System Technical Journal, Vol. 8, Jan. 1929, pages 159-172. Sound Recording with the Light Valve. Bell System Technical Journal, Vol. 8, Jan. 1929, pages 173-183. Synchronization and Speed Control of Synchronized Sound Pictures. Bell System Technical Journal, Vol. 8, Jan. 1929, pages 184-195. A Sound Projector System for Use in Motion Picture Theatres. Bell System Technical Journal, Vol. 8, Jan. 1929, pages 196-208. Miscellaneous Telephone Transmission. Sibley Journal of Engineering, Vol. 31, Apr. 1917, pages _ 177-180. Transmission Unit and Telephone Transmission Reference Systems. Bell System Technical Journal, Vol. 3, July 1924, pages 400-408. American Institute of Electrical Engineers, Transac- tions, Vol. 43, June 1924, pages 797-801. Statistics of the Telephone Industry of the United States Figure 57. Number of Telephones in the United States. Figure 58. Telephone Development in the United States. Figure 59. Percentage Distribution of Bell Stations in Fifteen Large Cities in the United States. Figure 60. Telephone Conversations — Average Number Daily in Millions in the United States. Figure 61. Average Daily Number of Toll Messages in the United States. Figure 62. Yearly Telephone Messages per Capital in the United States. Figure 63. Kilometers ot Telephone Wire in the United States. Figure 64. Kilometers of Exchange and Toll Wire in the United States. Figure 65. Telephone Wire in the Bell System. Figure 66. Growth of Various Classes of Physical Property in the Bell System. Figure 67. Bell System Revenues. Figure 68. Table Showing Initial Period Toll Rates. Figure 69. Map of the Bell System Showing Territories of the Associated Com- panies. Figure 70. Telephone Employees in the United States. Figure 71. Table Showing Population and Telephones which may be Connected by Transatlantic Telephone Service. F. B. Jewett Harvey Fletcher H. A. Frederick D. MacKenzie H. M. Stoller E. O. Scriven Bancroft Gherardi W. H. Martin 88 BELL SYSTEM TECHNICAL JOURNAL 8.000.000 16.000.000 14.000.000 12,000.000 0,000.000 8,000.000 6,000.000 4.000,000 2,000,000 OfVJ^000 ^ eOeOoO<0<00>0)0>00000 — — — — — c\jfVJ<\lf\jf\j (Ocooo>0)0>o>a)o>o>a>o>o>a>a)a>o> Fig. 57 — 'Number of telephones in the United States. TELEPHONE SYSTEM OF THE UNITED STATES 89 z o CL o CL o o oc u Q. Ul U z o r Q. U _l u I- Jjnujry lit o^ eich yeir Fig. 58 — Telephone development in the United States. 90 BELL SYSTEM TECHNICAL JOURNAL < •A < tn W H < a w H w a CO U a o < H O H -l w b O iz; O H H tn o <: H u u Oh C3 O CI BQ re <^' ^' -^ O r^' O O 00 00 O 00 rc r' oc' oc' l~~' C~ t^ C'" C^ OC C" O OS o CI g C rt n ^ u OvoO'-iTt^OCC'l'^rrj^O"^-*-^-^'* •^' sd rf -- ■* "S ■^' >-< O "-^ oo' oo' r^ O 0\ f*^ CN r^i i>i c^i CN ro CN t^i ^, — ^ --H r^ '-I 3 '■3 00^'O'*'*"^ONv0Orc-+'O'^'*00 roOiOCO'+-rfCir;C^t^C>r^'-iOOO ■^ ir> ir> o O O- Sv O >-< J>- O "-^ fO O O d O O ^ O) C'oj 01 09 Osirj'O'-iON'^^'^ioior^-H-^ooo oo-H"ooo<^Ooo-*CNCsr^-^-*oO'* 00Ot^t^'*"0'-'O'-iCNOO0000iO (^4'cN~lO T}<"'-rtC-o"ro'>o"c-ro Tf '-H \0 ■^ q ca ? flj o cu OOOOOOCOOOOOOOO ooooooooooooooo •^ O CN "* CN O 00_^lO^O_lO 0_"T_0_0_0_ 0~ O" CT '-H 00~^-^ LO fO^ ■^ ■—< >0 lO On tN lO •.-iio'*icNt^mrr>ossOLOt^cN-HOv-^ r0CNO00O<~0'-iOCNt^OiO"^'^t^ VO~ fO CN ^~ -4" T^" -^ •rt .2 a rjl TELEPHONE SYSTEM OF THE UNITED STATES 91 80 70 60 50 ■40 30 20 10 O 1898 1908 1918 1928 Fig. 60 — Telephone conversations — Average number daily in millionsin United States. 00 CO 0> o 00 0> 00 5> 00 if) u o < to (0 UJ li. o (O z o _J _l Fig. 61 — -Average daily number of toll messages in the United States. 92 BELL SYSTEM TECHNICAL JOURNAL o in o o 0> o> a\ in o in 00 (M f\J t\j 0^ 0^ a> 200 150 100 50 Fig. 62 — Yearly telephone messages per capita in the United States. TELEPHONE SYSTEM OF THE UNITED STATES 93 120 110 100 90 in CL 80 UJ t- u 2 O 70 _l ^ u. 60 O eo z 50 o -J _l 40 i r 1 ^^ ^^ n « ^ ^ OTHE R COMP/i ^NIES 1 w ■ VWWYVV ■ ■ ■ 30 20 10 1920 1921 1922 1923 1924 1925 1926 1927 1928 Fig. 63— Kilometers of telephone wire in the United States. 94 BELL SYSTEM TECHNICAL JOURNAL 120 to UJ \- u Q U- O V) z o 1920 1921 1922 1923 1924 1925 1926 1927 1928 Fig. 64 — Kilometers of exchange and toll wire in the United States. TELEPHONE SYSTEM OF THE UNITED STATES 95 100 in z o J 1920 1921 1922 1923 1924 1925 1926 1927 1928 Fig. 65 — Telephone wire in the Bell System in millions of kilometers. 96 BELL SYSTEM TECHNICAL JOURNAL 4000 4000 — — — — — — — — — — CMt\JCy(\JC\Jt\Jf\IC\J(\JC\Jfr) Fig. 66 — Growth of various classes of physical property of the Bell System. o — f\in^«'>n(£)h-o— (\j(0^ inioi^oo OOOOOOOOOO— — — —— — — — — — (M(\|(\j(\J(\|(\j(\)(\j(\| Fig. 67 — -Bell System revenues in millions of dollars. TELEPHONE SYSTEM OF THE UNITED STATES 97 Report Charge O O lO o >o O LO lO lO NO NO t^ •o o o to o o t^ GO On ON O O o o c c o o ooc c o o oo c oc o O' C O C O C' o ooooo O O C' O 3 O oo oooo oo oooo «^ Person-to- Person All Hours ID lO LO O lO O -;f lO t^ O <>1 to to o o to o to t^ C to r^ O f^l lo o >o o o to t^ O '^1 to O i>l o to to o to o to t^ r^i to r— C O >o o to lo o to r^ C '^l i~~ C to o o to o to ^J to o r^i to l^ CNl CN fN ro ro rr) ro Tt -rh Tj- to to to NO NO vO J^ t^ r^ t^ 00 00 00 C^. O^. On C O O -H ^H -H CM C^l oo o o o •r-i ■— 1 CNi ro Ttl LO to to O O to O NO t^ On O r^ to o "^ c o o to to r^ O O O fN '>^ oi ro ro fO ro C C to O O to to to t^ o o o) CO ro" ro -^ >* •* o o to o o to to to t^ O O o to o to to to t^ f^ o oi to to lO lO o o" m 1 ■ 8^ lO lO o to o to NO «^ 00 On -H tv) to to o o to c '* no 00 o <>i to lO O to to to o t^ o '>) ^j rv) to lo o to o o to r- O CN) to to t^ O to O to to c o <~Ni to t^ t-- o to o o to o to (vj to to r^ O oi ■^ •■-H ^ ■« tN CN) *>! r^ f^l ro '^ ro r^ 't ^ "* -f •* 'Tf to to to to to O NO O nO nO r^ t^ t^ t^ t^ 00 GO 1 . lO) to to to to to On O ■" ro to t^ O to O to C to O tN to t-^ O fN o to o to o to to t^ O '^1 to t^ O to o to O to O f^ to t^ O f^l o to o to o to to t~~ O ^1 to t^ o to o to o to O ^1 to t-. O <^J •■— 1 tN r^ CN CNi r^i ^ ro ro ro -* -^ Tf" •<* to to to to NO O nO nO f^ t^ t— t^ GO 00 00 00 On On ON ONO O 1-1 i-H Air Line Kilometers (Fractions Omitted) 711- 763 763- 814 814- 866 866- 969 969-1072 072-1175 175-1303 303-1432 [432-1561 561-1689 689-1818 818-1947 NO •* (^ J^ tN O r^ O ^ On NO On O tN r^ "^ "* to r^ r-j r^i o] r-1 CN 1 1 1 1 1 1 t^ O ^ ^O t^ Ol Tj< t^ O 'T) On NO On O O) f^ "^ ^ 2590-2719 2719-2848 2848-2977 2977-3105 3105-3234 3234-3363 3363-3492 3492-3620 3620-3749 3749-3878 3878-4006 4006-4135 4135-4264 4264-4393 4393-4521 4521-4650 4650-4779 4779-4907 C, CO tj Si Pi'O cooooo o o to to to to O O O O to to CN) fN tN CN C-) rNI lo too o o o (M ^ to lO »o o *o o to lO NO r— t-~ 00 00 o to o to to o On ON O O -^ -^I lo o o to o to "^ rO -f -* to to to C to O O LO O J^ I— 00 On ON o to to o to o O O -^ i^l '^1 ^ €«?^ rs (>4 cNi fvi tN r^i ^3 6 c '-t~t B 8:30 P.M.- 4:30 A.M. o "o o 'o o to •--< .r-i CN C^l (^ fO to to to to >o to ro r^ rO ro rO rO loco to o o rO '^^ •* -^ to to to to c c to o to to O NO O t^ o to o o to o r^ t-~. 00 00 00 On O to o O lo to ON Ov OO o o &^ ^H T^ 1— 1 ^H 7:00 P.M.- 8:30 P.M. o >o o 'o o to T-^ ■^ r>i CN (-O f^ to to o O 'o c ro rr> ^ ^ 'Tf iri lO to O to o to to to no o t^ t^ o o o c o 00 00 oo ON o c to O to O 'O C O -- — r^i '^1 oo lo o to o to o -O -f -f to lO o 4^ o to o "o o to — I— 1 (TNI r-j ro fO o to c to o to -:f -+ to to O nC O to O to O VO t-~. t^ GC 3C ON O^ O to O to O "O O O ■^ >— 1 c^l (M o to o to o to -O oo 'f ^ to to o to o to o to O O t— t^ 00 00 €r% 11 c js.uonieters (Fractions Omitted) On On On 00 00 00 "-I r^I ro -f to NO O On On On 00 00 -H "-l '^, ^ to 68- 77 77- 90 90-103 103-116 116-129 129-145 ^ t^ rO >0 — 1 — nO t^ O^ i-^ ^ o ^H T-i ■^ 1^1 rs) (>i 1 1 1 1 1 1 lO ^H f^ ro to ^H "* NO J— O^ CNl -Tf -H ..-H ^ .rt tV| CVI O On On l-~ t^ NO 00 On -^ to I-^ On ITS 1^1 ro ro "^ oo 1 1 1 1 1 1 '-^ O O^ On «^ J^ O oc On >— ' to I^ "^ CN CN ro ro ro 396-415 415-454 454-473 473-492 492-531 531-550 o GO '^ o to — 1 r^ O oo O 00 ^ in ^ ^ ^ ^ t^ 1 1 1 1 1 1 O O 00 rt" O to to t^ o 'o o 00 lo to o o o o CO NO o 98 BELL SYSTEM TECHNICAL JOURNAL X w E ^1 i - ' - '" s' 4 ie \\ n '.■Si V t* t' CO 'S a o cj c 4) a o •a o o CO m (U 01 o be c o CO >. C/3 i $ a S 9 » C O o lU 3 c Oh 00 OS O J3 tn oi c o bfl c a to O a; O s -4-J CO bfl WIRE LINE SYSTEMS FOR NATIONAL BROADCASTING 143 c o c 'tn 00 O M Z c o bo c a CO J*! O (U s O Ui Oh s CO 144 BELL SYSTEM TECHNICAL JOURNAL adequate, however, for program transmission because of the different character of the transmitted material. The bulk of present-day broadcast programs consists of musical selections, including a fair amount of high-grade material. To reproduce music, and particu- larly high-grade music, in a pleasing manner calls for a materially widened band. This wider band also gives a high degree of natural- ness to speech which is particularly desirable when loudspeakers are used for reception. At the present time in the United States the frequency band which is transmitted over the long distance program chains extends from about 100 cycles to about 5,000 cycles. It is, of course, possible to transmit an even wider band than this, although the cost of the circuits will, of course, increase as the band is widened. In consider- ing how wide the band should be, the complete system, including pickup apparatus, wire transmission line, radio transmitters, radio transmission paths through the ether, radio receiving apparatus and loud speakers must be considered. It seems probable that as the art progresses a band wider than the above will be found desirable. On the wire line systems, development work is going forward looking toward the possibility that such wider bands may be found desirable in the future. At the lower frequencies, where most people consider that improvement is particularly desirable, consideration is being given to the possible extension of the band down to 50 cycles and possibly lower. Consideration is also being given to the possible addition of two or three thousand cycles to the top of the band. In addition to this broad band transmission requirement, program transmission circuits must be designed to handle wide ranges of vol- ume, particularly for the transmission of musical programs. Much of the enjoyment in listening to good music appears to come from the ranges of volume, so that in order to deliver such musical programs properly these ranges of volume must be preserved in large part at least. At the present time the volume ranges are "compressed" somewhat by adjustment of amplification under control of an operator at the pickup point. This tends to make easier the radio trans- mission problem as well as the wire transmission problem. The range of volume which is now delivered, as read by a "volume indi- cator" (a meter which roughly indicates the peaks), is of the order of 30 decibels (3.4 nepers), which means that during the fortissimo parts of programs the power which is transmitted is about 1,000 times as great as it is during the pianissimo portions. The designer of the wire circuits must be concerned lest during those periods when the program power is strong, the program circuits WIRE LINE SYSTEMS FOR NATIONAL BROADCASTING 145 produce an undue amount of disturbance in neighboring circuits which may be transmitting other programs or telephone messages. The designer is also concerned lest when the program power is weak the programs be unduly interfered with by noise or crosstalk from other circuits. He must particularly consider the noise and cross- talk which may be heard during pauses in programs. During such pauses it is very annoying to the listeners to hear a background of noises of various sorts and it is essential that the listeners be unable during such pauses to pick up intelligible speech from telephone message circuits crosstalking into the program circuit. At the present time generally satisfactory results are being obtained in transmitting the volume range of about 30 decibels (3.4 nepers). Considerably more must be done both in the radio and in the wire systems, however, before there can be transmitted volume ranges comparable with those put out by symphony orchestras, high-grade artists, and the like. Having indicated in a general way the requirements of program transmission circuits, there will next be described the wire systems which are now in use in the United States. The present-day program transmission circuits in the United States are "on a voice-frequency basis," which means that the waves trans- mitted over the circuits are essentially copies of the sound waves impinging on the microphones. Most of the circuits now being pro- vided are carried by the familiar open wires, usually copper wires 0.165 inch (4 mm.) in diameter spaced about 1 foot (30 cm.) apart on the crossarms. The transmission properties of an open-wire pair without loading are well suited for program transmission purposes since the distortion is comparatively small although it is far from negligible. Spaced at intervals on these circuits, averaging roughly 150 miles (240 kilometers) apart, are one-way repeaters or amplify- ing devices. Along with these amplifiers are other electrical devices for counteracting the distortion introduced by the open-wire circuits, incidental cables involved, etc. Other one-way repeaters are pro- vided at the terminals of the circuit. Considerable technical refine- ment is, of course, involved in the design of these amplifiers and of the auxiliary apparatus associated therewith which cannot be gone into here. In setting up the program transmission circuits, an important part of the work consists in making measurements at different single frequencies within the band which it is desired to transmit over the circuit. Before making such overall measurements, the amplifiers and auxiliary apparatus are so adjusted locally as to compensate for 146 BELL SYSTEM TECHNICAL JOURNAL the amount of distortion which theory and experience indicate should be expected. Then, final adjustments are made by certain specially provided adjustable parts in accordance with the overall measure- ments. Such overall tests and adjustments are, in general, made daily. In setting up these circuits, another important consideration is that each amplifier carry its proper load or, in telephone parlance, each amplifier deliver to its associated line the proper output level. To insure this, diagrams are prepared in advance, showing the desired transmission levels at each repeater, a typical diagram being shown in Fig. 3. In setting up the circuits, the repeater gains are first set ^ TO LOCAL .BROADCASTING STATION MONITOR'S i __ _ AMPLIFIER -Sdb LEVEL 3D LLVLL I FROM LOCAiTI CIRCUIT ~-f.l_, KEY TO SYMBOLS LONG DISTANCE CIRCUIT [r] FILTER LOCAL CIRCUIT ^ AMPLIFIER ^ REPEATING COIL ^ LINE EQUALIZER P TRANSMITTING INPUT EQUALIZER [|] 600<-J ARTIFICIAL LINE [E] LOCAL EQUALIZER VOLUME INDICATOR ? Fig. 3 — -Typical Circuit Layout and Transmission Level Diagram of Program Network Circuits. to values which theory and experience indicate should result in con- ditions as shown in the prescribed transmission level diagram. Test- ing current is then applied to the sending end of the circuit and sensi- tive measuring devices are applied at the output of each repeater. If the results of these measurements do not accord with the trans- mission level diagram, suitable adjustments are then made. In building up the large chains which tie together a considerable number of radio transmitters, wire distributing centers are provided at strategic points. Figure 4 shows the circuit layout of the various chains which have been referred to and indicates in a general way how noMO 5tx; TLe.WAjn naw tugene ^ ^ f^Odeburq npo SAN 2AN FKMC nCO, CAUF Y Y Y SAN LWIS oetspo V^- KFl 5] Y 7 BA«CR3 f RESMC Molly wo 1> 0€ LOS ANGCLES. CALIF ra ff? '*m ^^^oi la — 1 ~t^ — gcs= WISNb lyVVThU nOUlTQN .TEJAS Fig. 4— Layout of Permanent and Recurring Program Network Circuits with Associated Repeater Equipment Stiown on the Basis of Normal Direction of TranBausaiun as of January 15, 1921>. KtV • IVmonent Circuit ■ Recurnnq Circuit Repeater and dsaociaced equipmenc 1 Radio Scatlon Oni Cham (PPA) is not shown. This ch«sin is usedonlyonehuwi* awecK and consists of 3460 ctrcvit miles involving 19 ridio stations. WIRE LINE SYSTEMS FOR NATIONAL BROADCASTING 147 the various chains are interconnected and arranged for switching at certain distributing centers. In the United States the largest distributing center is, naturally, in New York City, since the bulk of the program material originates at that point. At such a distributing center a special collection of various forms of equipment is provided consisting of one-way ampli- fiers, loud speakers, multifrequency oscillators, various forms of trans- mission measuring devices and miscellaneous apparatus. The photo- graph of Fig. 5 shows a portion of the program layout in the New Fig. 5 — Portion of Program Apparatus Layout in New York Long Distance Telephone Office as of January 15, 1929. York long distance telephone office as of January 15, 1929. The various bays at the left carry the line apparatus associated with branches of various chains. In the rear are located the transmission measuring apparatus and multifrequency oscillators. In the fore- ground are the terminals of various telegraph order wires. In transmitting programs over a wire network, as has been pointed out above, it is important that the volume range be held within WIRE LINE SYSTEMS FOR NATIONAL BROADCASTING 147 the various chains are interconnected and arranged for switching at certain distributing centers. In the United States the largest distributing center is, naturally, in New York City, since the bulk of the program material originates at that point. At such a distributing center a special collection of various forms of equipment is provided consisting of one-way ampli- fiers, loud speakers, multifrequency oscillators, various forms of trans- mission measuring devices and miscellaneous apparatus. The photo- graph of Fig. 5 shows a portion of the program layout in the New Fig. 5 — Portion of Program Apparatus Layout in New York Long Distance Telephone Office as of January 15, 1929. York long distance telephone office as of Januar}^ 15, 1929. The various bays at the left carry the line apparatus associated with branches of various chains. In the rear are located the transmission measuring apparatus and multifrequency oscillators. In the fore- ground are the terminals of various telegraph order wires. In transmitting programs over a wire network, as has been pointed out above, it is important that the volume range be held within 148 BELL SYSTEM TECHNICAL JOURNAL proper limits. It is one of the obligations of the one who " picks up " the program to hold his range of volume between proper limits. At the central distributing point those in charge of the wire circuits usually find it desirable to make checks from time to time to insure that the proper range of volume is maintained. This checkup is made by means of a device known as a "volume indicator" similar to the one which the program supplier uses for purposes of regulating his volume range. Other volume indicators are provided at various strategic points in the wire network in order to insure that the proper range of volume is reaching these points. In addition to regularly making these observations by means of volume indicators, loud- speaker monitoring observations are continually made at practically all repeater points. The results of these observations are transmitted back to the con- trol points periodically by means of telegraph order wires so that the control operator knows at all times the condition of transmission at every point in his territory. With the network chains grown to such vast proportions as indi- cated in Figs. 1 and 4, it is essential that the system for controlling the networks be such that all points involved be in instant commu- nication with certain designated control points. To accomplish this, the United States has been divided into four areas, each area of which is under the control of a distributing center or control station. The four control stations in the United States at present (January 15, 1929) are. New York covering the eastern section, Chicago the western section, Cincinnati the southern section, and San Francisco the Pacific Coast section. Each of these control points is connected to ever>- repeater point in its area by means of telegraph order wires and in addition is connected to every radio station in the area served by the networks under its control. The various control points are also connected together by means of order wires and arrangements are provided so that New York can be placed in communication with any of the radio stations in the United States which are served by the chains. The total telegraph wire mileage employed for this service is now approximately 43,000 miles (70,000 kilometers). A large corps of specially trained telephone men is needed to properly supervise the transmission performance of the chains as well as to take care of the switching and general coordination work involved. At present, about 300 men are employed in the United States for this service, these men, of course, being in addition to those who care for the regular wire and equipment maintenance. WIRE LINE SYSTEMS FOR NATIONAL BROADCASTING 149 Acknowledgment Acknowledgment is made to Mr. H. S. Hamilton for considerable assistance in connection with the preparation of the text and par- ticularly of the drawings, and to Mr. G. S. Bibbins and Mr. H. C. Read for furnishing most of the statistical data. Notes on the Heaviside Operational Calculus By JOHN R. CARSON This paper briefly discusses the following topics: (1) the asymptotic solu- tion of operational equations; (2) Bromwich's formulation of the Heaviside problem, and its relation to the classical Fourier integral; and (3) the existence of solutions of the operational equation. The paper closes with some general remarks on the interpretation of the operator and the opera- tional equation, emphasizing the purely symbolic character of the latter. THE large amount of work done in the past thirteen years, start- ing with important papers by Bromwich ^ and K. W. Wagner,^ has served to remove whatever mystery may have surrounded the Heaviside operator, and has placed his operational calculus on a quite secure and logical foundation. However, certain phases of the prob- lem still do not appear to the writer to have as clear or adequate treatment as perhaps might be desired ; these it is the object of the present paper to discuss. The topics dealt with are (1) the asymp- totic solution of operational equations; (2) Bromwich's very important formula and its relation to the classical Fouiier integral; and (3) the existence of solutions of the operational equation. In the following it will be assumed that the reader has a general acquaintance with the Heaviside operational calculus as well as the Fourier integral, but a brief sketch of the former may not be out of place. It will be recalled that the Heaviside processes were originally developed in connection with the solution of electrical problems:' more precisely, the determination of the oscillations of a linearly connected system specified by a set of linear differential equations with constant coefficients or a partial differential equation of the type of the wave equation. This system is supposed to be in a state of equilibrium at reference time / = 0, when it is suddenly acted upon by a 'unit' force (zero before, unity after time / = 0) ; the subsequent behavior of the system is required. In the solution of this problem, Heaviside's first step was the purely formal and symbolic one of replacing the differential operator d/dt by the symbol p, thereby 1" Normal Coordinates in Dynamical Systems," Proc. Lond. Math. Soc. (2), 15, 1916. 2"Uber eine Formel von Heaviside zur Berechnung von Einschaltvorgange," Archiv. Elektrotechnik, Vol. 4, 1916. ^ Since this paper is addressed largely to physicists and engineers, we shall employ to some extent the language of circuit theory rather than pure mathematics; no loss of essential generality is involved. 150 NOTES ON THE HEAVISIDE OPERATIONAL CALCULUS 151 reducing the differential equations to an algebraic form, the formal solution of which we shall write Here h = hit) is the variable with whose determination we are con- cerned and H{p) is the Heaviside function, derived as stated from the differential equations of the problem. This equation is as yet purely symbolic, and its conversion into an explicit solution for h, as a function of t, constitutes the Heaviside problem. Bromwich ^ formulates the problem as the infinite integral hit) =^. TWrr^^P- (2) The writer's formulation of the problem is, that h is uniquely determined by the integral equation ^ ^W^"'* = WW) <'^ This equation is valid for all values of p, for which its real part is greater than some finite constant c; c must be at least large enough to make the infinite integral converge. In the majority of physical problems this constant may be taken as 0; in some, however, the equation is valid only when c is greater than some finite constant. The equivalence of (2) and (3) is very easily established in a num- ber of ways; perhaps the simplest is to show, following March,^ that (2) is the formal solution of (3). Either can be deduced from the other. The Bromwich solution can, of course, be derived directly from the Heaviside problem, as shown below. I One of the most interesting and perhaps the least generally under- stood of Heaviside's methods of solving the operational equation is the process whereby he derives a series solution, usually divergent and asymptotic, in inverse fractional powers of t. What I have termed the Heaviside Rule ^ for deriving this type of solution may be formu- lated as follows: * "The Heaviside Operational Calculus," B. S. T. J., 1922; Bulletin Amer. Math. Soc, 1926. *"The Heaviside Operational Calculus," Bulletin Amer. Math. Soc, 1927. * In terming this process the Heaviside Rule I do not in any sense imply that Heaviside himself would have applied it incorrectly. In fact in one case he adds an extra term which contributes to numerical accuracy although the series itself is 152 BELL SYSTEM TECHNICAL JOURNAL If the operational equation h — \/II{p) admits of formal series expansion in the form h = aa -\- aiVp + a^p + a^p^p + a^p- + . . . , (4) a solution, usually divergent and asymptotic, results from discarding the /— ^" 1 terms in integral powers of p, and replacing />"\/> by -7— — -^ , whence ;.'^ao + |ax + a3|^ + a.|^,+ ---}-i=- (5) As stated in a forthcoming paper, this divergent series is a true asymptotic expansion, as defined by Poincare, if and only if, the singularities in 1/H(p) all lie to the left of the imaginary axis in the complex plane. Otherwise the series may require the addition of an extra term or factor, or even be quite meaningless. An excellent illustration of the preceding principle is furnished by the operational equation, V/> + X For convenience and without loss of essential generality we take |X| = 1 and X = e*^; that is, the parameter X may lie anywhere on a circle of unit radius in the complex plane. Now the solution of (6) is easily derived by well known processes of the operational calculus: it is h{t)=- \ ' dr (7) ^ Jo Vr-V/ - r ^ Jo ^T^r^ = dT. (8) T The solution is also known to be '^ h{t) = e-O^^l'U, (^], (9) where /o(X) is the Bessel function Jo(ix). a true asymptotic expansion. On the other hand Heaviside in his frequent appli- cations of the Rule gives no hint or indication of the restrictions imposed on its applicability. Fortunately in most applications of the operational calculus to physi- cal problems, the Rule leads to correct results. ^ See formula (p) of the table of integrals in Chap. IV, "Electric Circuit Theory and Operational Calculus." NOTES ON THE HEAVISIDE OPERATIONAL CALCULUS 153 Now return to the operational equation (6), and expand as follows, without reference to convergence, h = 1 /, , p\-''-' r -P 1 +Y V^ = \2l (^)(fr- ■■■■)* Application of the Heaviside Rule ; now gives the divergent solution hit) ~ S{\t). p. 32 2! (iif- ] VttX/ (: (10) We have now to distinguish three cases: 1. Xfi > 0. (Real part of X > 0.) In this case it can be shown from (7) that ^ ;z(/) ~ 5(X/) (11) and that the Heaviside Rule leads to a true asymptotic expansion, as defined by Poincare. When X = 1, by the known expansion of the right hand function in equation (9) we find that the error com- mitted by stopping with any term in the divergent series is less than that term. This property, however, does not characterize the series for all complex values of X for which the real part is positive. 2. X« < 0, X = - M, Mfl > 0. In this case, comparison of (8) with (7), gives by aid of (11), h{t) ~ e'"5(M/), (12) which again is a true asymptotic expansion. The expansion differs, however, from that given by the Heaviside Rule, by the factor g''^ and the alternation in sign of the odd terms of the series. 3. Xfl = 0, X = iw. In this case it is easily shown that ^ A(/) = ^-"'-'/^>/o(f ), (13) where /o is the Bessel function of order zero. From the known asymptotic expansion of this function, we find that hit) ~e-"'-'/^T^^*"'''^-5(^'c«^0]Rea.Part (14) with an error less than the last term included. * L.c. by the process described in Chap. V. ^ L.c. formula (w) of table of integrals. Chap. IV. 154 BELL SYSTEM TECHNICAL JOURNAL Perhaps the simplest way of establishing the Heaviside Rule for the asymptotic solution of the operational equation h = l/H(p) and the conditions under which it is valid, is as follows: We start with the integral equation r h(t)e~p'dt = l/pH(p) (15) Jo and specify that the singularities of llpH(p) and its derivatives are all confined to the left hand side of the complex plane, except at the point ^ = 0, in the neighborhood of which ao - + ai + a•2^1p + a^p + a^p^fp + • • • . (16) pH{p) ^jp In other words, l/pH{p) admits of expansion in powers of V/? Now since •M p-pt 1 we have from (15) r h^dt= ^ (17) Jo r( "-^)'-"-pm-w <'«^ By virtue of the restrictions imposed on l/pH{p), equation (18) is valid at ^ = 0, whence by (16) r("-s)*="- ''"' Now differentiate (18) with respect to ^; we get Now add H ^ ^ dt to the left of (20) and its value a2/2-yJp to Jo 2 ^j^^t the right hand side; we have NOTES ON THE HEAVISIDE OPERATIONAL CALCULUS 155 Now set ^ = 0; from (16) we have r( "-^+14'^'=-- '''' a formula which again is valid by reason of the restrictions imposed on l/pH(p). Proceeding in this manner we get the formula Jo (h - Sn)-t-dt = (- l)"w!a2n+i, (23) '0 where + (- 1)"1.3 ••• (2« - 1) '^'" (20" = first {n -\- \) terms of the divergent Heaviside series. (24) Also since S.» = 5. + (- !).+■ '■'•;3//.: + '^ ^ (25) we have from (23) by changing w to (« + 1), f Jo h-s.- (- i)"+^^-^"(J^!: + ^^^)/"+w/ = (- iy+'{7i + l)!a2„+3. (26) Equations (23) and (26) establish the fact that (h — 5„) converges, for indefinitely great values of t, at least as rapidly as l/t"+^-y[t, since otherwise the integrand of (26) would diverge; stated in mathematical notation h - Sn= 0(1/ 1"+^'^). (27) Consequently the series S when divergent is a true asymptotic ex- pansion, as defined by Poincare, of the function h. The foregoing says nothing, it will be noted, regarding the error committed when 5„ is employed to compute the function h. Nothing, in general, can be said about this question, which requires an inde- pendent investigation in every specific problem. In some cases the error will be less than the magnitude of the last term of Sn, but this is the exception rather than the rule. In other exceptional cases the series may even be absolutely convergent. 156 BELL SYSTEM TECHNICAL JOURNAL The foregoing results can undoubtedly be derived by integration of the Bromwich integral (2) along the contour suggested by March {I.e.). Wiener in his paper on "The Operational Calculus" {Math. Ajinalen, Bd. 95, 1925) gives an entirely different treatment of the problem. The operational calculus he deals with, however, differs under some circumstances from that of Heaviside, as Wiener himself remarks. A paper by Tibor v. Stacho on "Operatoren Kakiil von Heaviside und Laplaceshe Transformation" (publication 1927 VI 15 by the Hungarian University, Francis Joseph) may also be consulted. n Subject to certain well known restrictions a function f{t) can be expressed as the Fourier integral /W = T"; HP)e^''dp. (28) the path of integration being along the imaginary axis. We assume for the moment that this equation is valid. Now suppose that f{t) represents a force applied to an electrical or dynamic system whose "steady state" or forced response to an applied force F{p)eP^ is H{p)' • Then the forced response g{t) of the system to the applied force /(/) is given by However, in applying the foregoing to the Heaviside problem we encounter an initial difficulty. This is that if /(/) is taken as the unit function (zero before unity after, / = 0) it does not admit of formu- lation as the Fourier integral (28). The unit function, however, when multiplied by e~'' when c is a positive real constant, does admit of such formulation, and it is easy to show that the unit function itself is given by c-fioo , — dp c > 0. (30) c-i« P Consequently, if the unit function is the force impressed on the sys- tem, the forced response is iTTiJc-io, pn{p) NOTES ON THE HEAVISIDE OPERATIONAL CALCULUS 157 If now all the singularities of the integrand lie to the left of the imag- inary axis, then k{t) = /?(/) and (31) is the formulation of the Heaviside problem. Suppose, however, that the electrical or dynamic system specified by II{p) is "unstable"; that is, it contains some internal source of energy which makes its transient oscillations in- crease with time / instead of dying away. In such a case H{p) will have zeros to the right of the imaginary axis, and in order that (31) shall be the solution of the Heaviside problem, c must be taken so large that all the singularities of the integrand lie to the left of the path of integration. Consequently h{t) -^ I -^TTTT^P' J_ r+'" g^' ^^ (2) iiriX-i^ pn{p) pro\ided c is so chosen that all the singularities lie to the left of the path of integration in the complex plane. This is Bromwich's formu- lation of the Heaviside problem. ^° From the foregoing it follows that the Fourier integral -f is, in general, the formulation of the Heaviside problem if and only if, all the singularities of the integrand lie to the left of the imaginary axis. If there are singularities on the imaginary axis, the integral is ambiguous, while if there are singularities to the right of the im- aginary axis, the integral gives an incorrect solution of the Heaviside problem. ^^ As a simple example consider the operational equation h = \/H{p) = ^ P-is' where the real part /Sr of /3 is positive. The correct solution as given by either (2) or (3) is // = 0 / < 0 = e^' t > 0, ^° The appropriate mathematical methods of solving the infinite integral (2) are dealt with in great detail by Jeffreys in his "Operational Methods in Mathematical Physics" (Cambridge University Tracts). " To prevent misunderstanding it should be stated that the application, when permissible, of the classical Fourier integral (2a) to the Heaviside problem, was known long prior to the work of Bromwich. Bromwich's essential and important contribution lay in showing that the path of integration must be shifted to the right of all the singularities, together with a verification of an important form of solution, first given by Heaviside, of the operational equation. 11 158 BELL SYSTEM TECHNICAL JOURNAL whereas the Fourier integral (2a) gives h = - e^' t < 0. = 0 / > 0. There is another reason why care must be exercised in applying the classical Fourier integral to the Heaviside problem. This is that in solving the operational equation, h = \/H{p), the appropriate expansion of \/H{p) may introduce singularities on or to the right of the imaginary axis in the component terms. This offers no difficulty if either (2) or (3) is employed, but renders the Fourier integral (2a) inapplicable. As an example consider the equation V^+ 1 One form of solution is gotten by multiplying numerator and denomi- nator by V^ — 1 , whence Ji = _ 4p p - 1 p - 1 and each term has a singularity at ^ = 1. A physical interpretation of the foregoing may not be without interest. Suppose that an elementary force F(p)eP^dp, where p = c + io), is applied at an indefinitely remote past (negative) time to a system specified by H{p) . The response of the system is then ^''^^ e^'dp-^ T^{t)dp, H{p) where Tp{t)dp is the concomitant transient or characteristic oscillation of the system. If c is chosen sufficiently large then at least for t > 0 the transient term can be made as small as we please compared with the first term. Finally if the impressed force is the unit function (zero before, unity after, time t = 0) and it is written as 1_ r+*"£!!^ 5W Jc-i 00 P the total response and therefore h{t) is given by J_ f <^+<«' .tp e' pllip) dp, NOTES ON THE HE AVI SIDE OPERATIONAL CALCULUS 159 provided c is sufficiently large to make the transient term /^C-|-lQO T„{t)dp ^ C—iaa negligibly small. Analytically this requires that c be so large that the zeros of pllip) shall all lie to the left of the axis pji = c. Ill The foregoing discussion tacitly assumes the existence of an unique solution of the operational equation. On the part of the physicist this assumption is entirely proper because if the operational equation is the symbolic formulation of a correctly set physical problem an unique solution must and does exist. When approached from the purely mathematical standpoint, however, the case is different and there is no assurance of the existence of a solution. As an example consider the operational equation h = e" The corresponding integral equation f = r P Jo h{t)e-p'dt pR > 0 has no solution, while Bromwich's formula c-ioo P 1 r'+"^ ov gives h = 0 t < — 1 - 1 / > - 1 which is obviously incorrect. As a matter of fact the operational equation itself has no solution. To formulate the necessary and sufficient conditions for the exis- tence of a solution we may proceed as follows: If a solution exists it is given by either of the equations h{t) = ^ r j\P)e"'dp, (2) KP) = f 00 hit)e-p'dt pR > c, (3) 0 . 160 BELL SYSTEM TECHNICAL JOURNAL where f{p) denotes \/pH{p). Substitution of the value of //(/), as given by (2), in (3), gives the transform f(p) = -L e-,i(U J\z)e''dz. (32) IlTl Jo Jc-lx In addition, since //(/) = 0 for / < 0, we must have • C+ioo -^ I f{p)e'Pdp = 0 when / < 0. (33) Equations (32) and (33) formulate the necessary and sufficient restrictions on f(p) for the existence of a solution of the operational equation h = pfip) = l/H{p). To correlate the transform (32) more closely with the classical Fourier transform, write p — n -\- ioo and f{ii + 7co) = 0(w) i( and co real. Then the transform (32) becomes 0(co) =— c'""(It (l>{x)e''Hx (34) ZrJo J -00 for all values of u. > c. Also since //(/) = 0, for / < 0, the lower limit of integration with respect to / in {33) may be replaced by — oo , whence 0( w = ^ e-^'hll I 0 162 BELL SYSTEM TECHNICAL JOURNAL or 1 /•« }i{t)e-p^dt pit > 0 _L- r* V^ Jo More broadly stated, the operational equation is the shorthand state- ment of true equations in which p has lost its original significance and is simply the complex argument of functions which obey all the laws of algebra and analysis. Failure to recognize these simple principles is responsible for a large amount of confusion, loose reasoning and profitless discussion of so called 'fractional differentiation,' a term which, to the writer at least, is quite meaningless. On the other hand, their recognition should go far towards removing whatever mystery may have surrounded the Heaviside operator and the Heaviside processes. Contemporary Advances in Physics, XIX. Fusion of Wave and Corpuscle Theories. By KARL K. DARROW. In this article certain of the simple and familiar phenomena of optics and of electronics — for instance, refraction at a boimdary between two media, and diffraction by a grating — ^are interpreted by both of the theories, undu- latory and corpuscular, which have so often been condemned as incom- patible with one another; the attitude being, that the theories may be brought into concordance by modifying one at least in ways which, extra- ordinary as they seem, do not quite destroy its character. NOT quite five years ago I published in this journal an article entitled Waves and Quanta, expounding there the data which invited a corpuscular theory of light, regardless of the great array of classical phenomena of optics which demanded with no less insistence the long-triumphant undulatory theory. Today, not only are those data still extant and undeniable; they have been reinforced by obser- vations on electron-streams which have compelled a wave-theory of free negative electricity, despite the very abundant evidence for free corpuscular electrons. Most physicists expect that not only light and negative electricity, but whatever other fundamentals there may be — meaning, probably, positive electricity and nothing else — will be found to conform in some ways to simple wave-theory, and in some to simple particle-theory. Most physicists, I think, would concede that the two ideas must be forced into one scheme, whatever violence it may entail to others of our preconceptions, inborn or inbred. We must stretch the theories and our minds, so that corpuscles and waves shall appear no longer as alternatives of which election must be made, but as complementary aspects of one reality. To make a beginning with this process of stretching, I propose to treat some of the very simplest and most familiar of the phenomena, which up to lately have been interpreted by ofie only of the theories: phenomena such as the refraction of light in passing from air to water, the bending of the paths of electrons in passing from vacuum into metal, the diffraction of light and electrons from a ruled ditifraction- grating. (None of these examples, incidentally, involves a theory of the structure of the atom.) Each of them shall be interpreted by the other theory — not in order to substitute the other for the one, but in order to practice the art of using both theories in alliance. 163 104 BELL SYSTEM TECHNICAL JOURNAL Refraction of Waves and Refraction of Corpuscles. I presume that every textbook of optics and every history of physics informs its readers that anciently there was a controversy between a wave-theory of Hght (attributed to Huyghens) and a corpuscular theory (accredited to Newton) which was totally decided in 1850 by an experiment of Foucault. Light is refracted toward the normal in passing from air to water, and should therefore move more rapidly in water than in air if it consists of particles, but not so rapidly if it consists of waves — so runs the argument. Foucault and Fizeau discovered that light does move less rapidly in water than in air.^ Let us analyze the argument more closely before deciding what was proved. The reasoning from the "wave-theory" is usually made in graphic fashion by showing "Huyghens' construction" (Fig. 1) which should remind many a reader of his high school days ! This is a very crude form of wave-theory, much too primitive to account for most of the phenomena which the physicist has in mind when he says that light (or electricity, or matter) is of the nature of waves; but for the present purpose it will do. In Fig. L A A' is the trace, on the plane of the paper, of a wavefront moving through air (say) in the direction LM toward the boundary between air and water. It is the trace of the wavefront at a par- ticular moment, say /; at a later moment, say /', the front has moved on to another position BB' . Denote by v the speed of the wave- front in air; then the perpendicular distance between BB' and A A' is equal to v{t' — t). While the wave is advancing through this distance, its intersection with the boundary of the water sweeps over the distance AB, which we will denote by D. Designate by d the angle between wavefront and boundary, the "angle of incidence." From the diagram one sees immediately: sin 6 = v{t' - t)ID. (1) Now in Huyghens' view, whenever the oncoming wavefront passed over an atom in the boundary-surface it incited that atom to emit a "wavelet." The circles drawn around various points on the line AB are the traces on the plane of the paper, of halves of those spherical wavelets — the halves expanding downwards into the water. Accord- 1 Foucault usually gels all the credit, but Fizeau and Breguet were working at the same time, incited by the same suggestion of Arago, and using the same method with differences in detail; and they announced their result only six weeks later. Indeed, at the meeting of the Academic des Sciences (May 6, 1850) at which Foucault reported his success. Fizeau said that if the sun had shone that day or the day before, they too would have had data to present. CONTEMPORARY ADVANCES IN PHYSICS 165 ing to "Huyghens' Principle" the ongoing wavefront in the water is the envelope of these spheres. In Fig. 1 they and the ongoing wave- front are represented for the moment t' when the wave in the air reaches B. The radius ^C of the wavelet expanding from A is then the distance which light traverses in water during time (/' — /), for that wavelet started when the wave in the air reached A. Denote by v' the speed of light in water and by 6' the angle between the new WATER Fig. 1. wavefront and the boundary, the "angle of refraction " ; then from the diagram : sin d' = v\t' - t)ID (2) and from (1) and (2) together, we obtain: sin ^/sin d' = vjv'. (3) From this familiar equation it follows in general, that the ratio (sin 0/sin d') is independent of the angle of incidence. (It is called the index of refraction of the second medium with respect to the first ; I denote it hereafter by N.) Also it follows in particular, that when light is refracted towards the normal the wavefronts must move more slowly in the second medium than in the first, which is what Foucault verified, or rather, thought he had verified. Now try it by the corpuscle-theory. In Fig. 1, I have the line LAIN redrawn as a heavy line, and the lines at right angles to it left out; for the line LAIN, one of the "rays" of light, is now to be interpreted as the path of a corpuscle, and there are no wavefronts. So long as the corpuscle is too far from the boundary-surface to feel any force from the water, it moves in a straight line with unchanging momentum; for the forces exerted on it by the air, being equally applied in all directions, balance one another out. In the region near 166 BELL SYSTEM TECHNICAL JOURNAL the boundary, this remains the truth for the components of force parallel to the surface; but the components along the normal, applied respectively from the direction towards the air and the direction to- wards the water, need not be perfectly equal. After the corpuscle has gone through the transition region and reached the depths of the water, it continues in a straight line with a momentum of which the component parallel to the boundary — the "tangential" component, say — is still the same as it was in the air, while the normal component is changed. Denote by pt and pn these two components of the original momentum of the particle through the air, by p the magnitude of their resultant which is the original momentum ; by p/, p„' and p' the corre- Fig. 2. spending quantities for the final flight of the corpuscle through the water. From Fig. 2 we see : sin d = Ptfylpt'' + pn" = pt.lp, sin 0' = p/jp, and since pt = p/ : sin djsin 0' = p'/p. (4) (5) The corpuscle-theory therefore leads to the statement that the sines of the angles of incidence and refraction stand to one another as the momenta of the corpuscle in the first medium and in the second; and when light is refracted towards the normal, the corpuscles must move with a greater momentum in the second medium than in the first. CONTEMPORARY ADVANCES IN PHYSICS 167 Comparing the equations (5) and (3) to which the two conceptions lead, one sees that far from contradicting one another, they are both acceptable, provided that : PIP' = v'lv. (6) We may hold both the theories simultaneously, we may interchange the two at will, provided we assume that the momentum of the cor- puscles varies inversely as the speed of the wavefronts. In spite of the outcome of Foucault's experiment, we may adopt either the wave- theory or the corpuscle-theory or both at once to describe refraction, provided we assume that when a beam of light is refracted toward the normal, the speed of the wavefronts diminishes but the momentum of the corpuscles grows greater. Why then did everyone concede that the corpuscular theory of light was killed by the experiment of Foucault? Because everyone was making two assumptions which seemed so obvious as to be hardly worth the stating, and so certain that it would have been regarded as absurd to call either into question: (A) It was being assumed, that the momentum of a corpuscle must always be strictly proportional to its velocity; in other words, that the mass of a corpuscle must be invariant. (B) It was being taken for granted that in a wave-theory of light the speed of the waves, and in a corpuscle-theory of light the speed of the corpuscles, must be identified with the actual speed of light as measured in any actual experiment. When these assumptions are made, equation (5) goes over into the form, sin e/sln d' = p'/p = v'/v, (7) which is contradictory to equation (3) and disproved by the experiment of Foucault. But it no longer seems radical to change the first of these assump- tions, for it is known from observation that there are particles — elec- trons, for example — of which the mass is not invariant, but depends upon the speed. For such a particle the momentum is not exactly proportional to the velocity. It is then not quite so revolutionary to go further, and suppose that the corpuscle of light is of so strange a nature that its velocity and its momentum are in magnitude inversely proportional to one another. If one made this supposition then one could accept the second assumption, and still explain the refraction of light by the corpuscle-theory. Even the second assumption, however, is not sacred. It may seem absurd to set up a wave-theory of light, and then say that the speed of 168 BELL SYSTEM TECHNICAL JOURNAL the wavefronts is not to be identified with the measured speed of light. It does seem absurd to set up a corpuscle-theory, and then say that the speed of the corpuscles is not necessarily the same as that of light. Yet it may turn out in the end that a theory of either kind is strength- ened, and made more competent to account for a variety of facts, by abandoning that easy and natural identification. I will try to prove by actual examples that it does so turn out. Meanwhile I summarize this section in a sentence: If we wish to interpret light, or electricity, or matter, by both a corpuscle- theory and a ivave-theory, the momentum of the corpuscles must be supposed to vary inversely as the speed of the ivaves. I have omitted the special reference to refraction, for any more general theory must include that particular case, or fall down com- pletely; I have added allusions to electricity and matter, for the test of any alteration of the two classical assumptions will depend chiefly on whether it helps in understanding the wavelike properties of these two, and not of light alone. We now carry the wave-theory a great step beyond the primitive form in which Huyghens left it, by introducing the ideas oi frequency and wave-length. Wave-length of Waves and Momentum of Corpuscles Instead of the single "wavefront" of Fig. 1, suppose a train of sine- waves of frequency v, period T{— 1/f), wave-length X and wave number /x(= 1/X) travelling through air along the course LMN. For definiteness, think of sound-waves. The condensation ^ of the air conforms to the equation: p = Po sin 2ir {vt — p.s -{- a), (8) wherein 5 stands for distance measured from some arbitrary plane perpendicular to LM, and a for some constant. I write the equation down because one like it (or more than one) occurs in every wave- theory. In that of light there are six such equations, with components of electric and magnetic field strength replacing p; but it will be sufificient to think of one. In the wave-theory of matter there is one, with a quantity of very abstract meaning replacing p. Now when the wave train passes through into the water, its fre- quency remains the same. With sound-waves, or any mechanical vibrations of matter, this is obvious; two pieces of matter in con- tinuous contact must vibrate in unison, or not at all. We generalize this statement to cover light-waves, and waves of other varieties later ^ The excess of the density over the normal vahie, dulded by the normal value. CONTEMPORARY ADVANCES IN PHYSICS 169 to be considered. Using primes to designate the values which things have in the second medium, we put: / = p. (9) The speed of the waves is the product of their wave-length by their frequency : V = v\ v' = v'\'\ (10) consequently: v'jv = y/\. (11) The wave-lengths of the wave train on the two sides of the boundary vary directly as the speeds. Return now to the last section, and introduce this result into equa- tion (6) ; one gets: P'/P = V^' (12) which means: we can interpret refraction of light (or of electricity, or of matter) by both the wave-theory and the corpuscle-theory, provided that we make the momentum of the corpuscle vary inversely as the wave-length of the waves. Write accordingly, p\ = constant. (13) Now there are several remarkable experiments which show that this relation actually holds, and moreover that the constant which appears in it is the universal constant h of Planck: p = h/X. (14) For instance, one may pour a stream of X-rays — that is to say, high-frequency light — into a gas, after having measured its wave- length in the known and reliable way depending on one of the phenom- ena in which X-rays behave as waves. A certain portion of the rays is scattered; it is scattered as though it consisted of corpuscles, each of which strikes an individual free electron and bounces off, the electron meanwhile recoiling from the blow.- Further analysis of the data shows that there is conservation of momentum — that the momentum which the electron gains is equal to that which the corpuscle of light has lost, provided that the momentum of this latter is equal to the quotient of h by the wave-length of the rays. For the wave-length of the scattered X-rays, measured in the same way as that of the primary rays was measured, is not the same as theirs; and the difference between the values of /;/X, before and after scattering, is equal to the momentum which the electron received. - The Compton effect (cf. the seventh article of this series). 170 BELL SYSTEM TECHNICAL JOURNAL Again, one may pour a stream of electrons against a crystal or an optical ruled grating, after having measured the speed of the electrons in one of the well-known ways depending ultimately on the deflection of such a beam in known electric and magnetic fields.^ The mass of the electrons being known, one knows also their momentum. Now the crystal or the grating, whichever it may be, forms from the primary beam a diffraction-pattern of new beams. Well! the formation of a diffraction-pattern is the primary reason for saying that light is wave- like, and it gives the primary way of measuring wave-length of light. One is equally obliged to admit that a stream of free negative elec- tricity is wavelike, and to accept the value for its wave-length which the diffraction-pattern gives. Again it turns out that the wave-length is equal to the quotient of h by the momentum of the electrons. It may be objected that in all of those experiments, the corpuscles were observed in a vacuum. Compton measured X-rays before and after scattering, but during the measurements they were in vacuum or at any rate in air. Davisson and Germer, Thomson and Rupp, observed electrons returning through the same evacuated space as they had crossed on their way to the diffracting lattice. One might emphasize that all these savants compared momenta and wave-lengths for different beams in the same medium instead of comparing them for the same beam in different media. The distinction is certainly worth noticing; but happily there are experiments which bear directly on refraction. Davisson and Germer measured, not precisely the refraction of an electron-stream passing from vacuum into nickel, but a minor perturbation of the diffraction-pattern which is due to that refraction. We will analyze their result, for nothing shows more clearly the relations — or lack of relation, the reader may think — between speed of waves, speed of corpuscles and measured speed of stream. Davisson and Germer came to values of the index of refraction (sin 0/sin d') which were greater than unity — which corresponded therefore to a bending of the stream towards the normal, as it passed from vacuum into nickel — which therefore signified that the speed of the waves is not so great in nickel as in air. On the other hand, it is known that when an individual electron passes from vacuum into a metal, its kinetic energy and its velocity increase as it goes through the surface. We have in fact the situation described in the corpuscle-theory picture of refraction, a few pages back. Return to equations (4) and (5), and consider a corpuscle for ^ The experiments of Davisson and Germer, of G. P. Thomson, and of Rupp (cf. the eighteenth article of this series). CONTEMPORARY ADVANCES IN PHYSICS 17 1 which the momentum p, the velocity u, the kinetic energy K, the mass m are related to one another as in Newtonian mechanics — properties which are practically those of electrons except when these are moving much more rapidly than any involved in these experiments: p = mil, K = y^mu^. (15) Use Ui and Un to denote tangential and normal components of speed; use primes to designate the values which things have in the second medium (nickel). Starting from equation (5), we continue: sin djsm 9' ^ N = M'jM = u'/u; iV2 _ 1 = (m'2 _ ^^2)1^^2 = (^z _ x)/K. (16) The quantity {K' — K) is the gain in kinetic energy which the electron wins on passing into the nickel ; and this gain, as I have said, is positive; hence by equation (16) the index of refraction must be greater than unity. This is in agreement with the result of Davisson and Germer; the agreement, in fact, appears to be quantitative.'* It is always pleasant to get an agreement; but note how we got this one. We got it by dropping the assumption that the speed of the corpuscles and the speed of the waves must be the same. Or rather, by not making that assumption. For though the fact of experience is always the same — the swerving of the electron-stream toward the normal as it enters the nickel — it is interpreted by the two theories in opposite ways; the waves are slowed down, but the corpuscles are speeded up, in passing from the vacuum to the metal. Even if wave- speed and corpuscle-speed were the same in empty space, they could not be the same in any other medium. This is more serious than it may appear at first. It amounts in effect to saying that a beam of free negative electricity has two dif- ferent speeds; one when we visualize it as a jet of particles, another quite different when we visualize it as a train of waves. But is not one of these "the right one" and the other "a wrong one," and can we not settle between them by measuring the actual time which the electricity takes to pass a measured distance? Let us examine this possibility. We shall find that after all it is not so easy to evade the ambiguity in such a fashion. Phase-Speed and Group-Speed Suppose an endless train of perfect monochromatic sine-waves marching along through space. For definiteness, think again of sound- ^ There is a remarkably interesting correlation between these results and the new statistical theory of the electron-gas inside the metal (cf. my article in the October 1929 number of this Journal, pp. 710-716). 172 BELL SYSTEM TECHNICAL JOURNAL waves. It might seem as if we could measure their speed by picking out one crest, as A of Fig. 3, and checking off with a stop-watch the moments when it passes two fixed markers placed a known distance apart. Not so; for we cannot see or hear or in any way perceive the individual crests. The wave train produces a perfectly uniform tone in the ear which it strikes. If two listeners are stationed at different points along the path of the sound, neither can recognize the moment at which any particular crest glided by. All they can recognize, all they can compare, is the moment of passage of a perhirbation of the wave train ; a sudden beginning, a sudden ending, a transient swelling of the sound. Most measurements of the speed of sound, in fact, are measures of the speeds of something violent — the crack of a pistol or an electric spark, the roar of an explosion — something very unlike a uni- form train of sine-waves.^ Now a sine-wave with a perturbation is in effect a sum of two or more sine-waves each of endless extent and constant amplitude, but having different wave-lengths and different amplitudes. This state- ment is the content of Fourier's principle from which the method of Fourier analysis is derived. One might represent even the sudden and violent pulsation of air due to an explosion, or the electrical spasm due to an outburst of static, by a summation of properly-chosen endless monochromatic sine-wave trains. I take however the simplest con- ceivable case: the wave train composed of only two sine-waves of dif- ferent wave-lengths. The reader will probably recall that when the difference between the wave-lengths is only a small fraction of either, this composite wave train resembles a sine-wave with regular fluctuations of amplitude — that is to say, with "beats" (Fig. 3). The maximum or centre of a beat occurs where a crest of one sine-wave coincides with a crest of the other — the minimum between beats, where crest falls together with trough. Denote the two wave-lengths by X and X -|- AX. One sees by inspection that a wave-length is the same fraction of the dis- tance D between two consecutive beat -maxima, as the discrepancy AX is of the wave-length -.^ D/\ = X/AX. (17) Of course this statement is exactly true only in the limit of vanishingly small AX. We shall always stay close to this limit, though some of the following statements would be valid even otherwise. ^ I except so-called measurements of the velocity of sound which are really measures of frequency and wave-length in stationary' wave-patterns, these being then multi- plied together. ^ The principle of the vernier. CONTEMPORARY ADVANCES IN PHYSICS 173 Fi^. 3. 12 174 BELL SYSTEM TECHNICAL JOURNAL Now if the two component waves advance with equal speed, the beats are simply carried along with a speed equal to theirs. But if the velocities of the two component waves are not the same, then the velocity of the beats is not the same as either, nor the mean thereof. It is in fact something totally different. To see this, imagine that you are moving along with one of the sine- waves; for definiteness, that you are riding on the crest B of the train with the shorter waves (Fig. 3). At a certain moment, say / = 0, it coincides with a crest A of the other sine-wave, and you are at the top of the beat. Meanwhile the other train is moving relatively to the first; for definiteness suppose that the longer waves move faster, so that relatively to the shorter they are gliding upward. After a cer- tain time they have gained on the shorter waves by a distance AX, the difference between the two wave-lengths. But when this time has elapsed, the top of the beat is no longer where you are, but where the crest B' of the first train coincides with the crest A' of the second. It has dropped back through the distance X, while the second wave train was getting ahead by the distance AX. Perhaps it will be easier to realize that while the second wave train is gaining on the first by X, the beat is dropping back by the distance D between consecutive beats; by equation (17) this comes to the same thing. Therefore when the longer waves travel faster than the shorter, the beats travel more slowly than either. If the longer waves were the slower, the beats would travel more rapidly; but this case is never realized in nature, not at least with light-waves '' and waves of elec- tricity^ and matter. We now deduce the formula for the actual value of the speed of the beats. Denote by v and v -\- f^v the speeds of the two sine-waves of which the wave-lengths are X and A -f- AX, respectively; by g the speed of the beats. It is sufficient to put into notation what has just been said in words. Relatively to the former wave train, the velocity of the latter wave train is Av, that of the beats is {g — v). Relatively to the former wave train, the latter moves a distance AX while the beats are moving a distance X in the opposite sense, therefore with a minus sign. Hence: {g - v)/M' = -X/AX (18) " The exception to this statement — the case of light having wave-lengths lying within a region of anomalous dispersion of t he transmitting substance — has been an- alyzed by Sommerfeld and L. Brillouin {A7in. d. Phys 44, pp. 177-202,203-240; 1^14) who find that in this case the group-speed defined by (20) loses its physical im- portance, and a segment of a wave train is transmitted with a speed never exceed- ing the speed of light in vacuum. This appears to be related to the absorption which always gees with anomalous dispersion. CONTEMPORARY ADVANCES IN PHYSICS 175 and solving for g, .?=^'-X^, (19) or going over to the differentia! notation, which will not only look more natural but will signify that the result which we have just attained is strictly valid in the limit for infinitesimal differences of wave-length: g^ V - X(dvi'd\). (20) This is the formula for the group-speed; for the term "group-speed" is the usual one for what I have been calling "speed of beats." Like- wise phase-speed is commonly used to denote the speed of the individual sine-wave trains. The term "group-speed" is in one respect unfortunate; for it implies that any "group," that is to say any sequence of uneven and irregular wave-crests and troughs, is propagated with a perfectly definite speed. However this is true only for the simplified group which we have been considering, the beat formed of no more than two wave trains; and even for this it is exactly true only in the limit, where the wave-length- difference between the trains approaches zero. All other groups change in form as they advance. Now there is always something arbitrary in defining "speed" for something which changes as it goes, like a puff of smoke or a cloud. The arbitrariness is nil in only the limiting case which I have just been formulating. However, it must not be exaggerated. A bunch of irregular crests and troughs may retain enough of its form and compactness, as it travels over a distance many times as great as its width, to justify the statement that it has a speed of its own. And if such a group turns out, on being analyzed in Fourier's way, to consist mainly of sine-waves clustered in a small range of wave-lengths, then its speed will not be far from the value of g computed by equation (20) for a wave-length in that range. Now these deductions explain a very remarkable experiment by Michelson, which otherwise might have disproved — indeed I do not see how it could have been interpreted otherwise than as destroying — both the wave and the corpuscle theory of light. I will preface the account of this experiment by saying that for light in empty space the speed of all wave-lengths is the same,** so that there never is any dif- * The chief evidence for this statement is astronomical. If light of one color traveled faster than light of another, a luminous star emerging from behind a dark one would be seen first in the faster-travelling hue; in fact there would be a .se- quence of colors, the same for every emergence of every such star, and spread out over a time-interval proportional to the distance of the stars. Nothing of the sort has ever been observed, although there are plent>- of luminous stars revolving around dark ones which regularly occult them. 176 BELL SYSTEM TECHNICAL JUURXAL ference between velocity of groups and velocity of wave-crests; they both have the same universal constant value c. However this cannot be true for light in transparent material media such as glass, water, or carbon bisulphide; for the refractive index of all these media varies from one wave-length to another — they are said to be dispersive. Now Michelson measured the time taken by a flash of light to cover a measured distance, first through air (very nearly the same as vacuum) then partly through air and partly through carbon bisulphide. The source of light shines continuously, and an incessant beam falls on a revolving mirror and is reflected in a continuously-changing direction; a second, stationary mirror receives this reflected beam during a very small fraction of each complete revolution and sends it back, so that the twice-reflected beam is a series of segments cut from the primary beam. It was the time taken by the segments to travel a known dis- tance which Michelson measured.^ Reasoning back from the data, he computed that they took (1.76 ± 0.02) times as long to go a given distance in carbon bisulphide as in air. But the refractive index of carbon bisulphide, in the range of the spectrum where Michelson's source of light was brightest, is about 1.63; so that the primitive wave- front-theory gives 1.63 for the ratio of the speeds in air and CS^, and the corpuscle-theory gives (1.63)~^ Foucault and Fizeau, be it remembered, had done the experiment with water. It happens that for water the derivative dvld\ is much smaller, and the group-speed therefore much closer to the wave-speed, than for carbon bisulphide. Also their experiments, though performed by the same method as Michelson was later to adopt and adapt, were less accurate than his. But if they had performed the Michelson experiment in 1850, the result would have been astounding. For Arago had asked, in effect: is it the speed of the wave-fronts in the wave-theory, or the speed of the corpuscles in the corpuscular theory, which agrees with the measured speed of a piece of light? Arago had said: "These experiments . . . will permit no further hesitation as between the rival theories. They will settle mathematically (I employ this word on purpose) they will settle mathematically oneof the greatest and most disputed questions of natural philosophy." He had proposed a question to Nature, and had written down two and only two answers. Everyone thought that Nature must reply by ratifying one of the 9 When the segments returned from the second to the first mirror they found that the latter had revolved a little further beyond the oiientation which it had when they left it, so that it reflected them onward not quite along the path to the source of light, but along another path inclined to that one at an angle twice as great as that through which it had revolved. Michelson measured the angle, and knowing the rate of revolution of the revolving mirror he then knew how long the light had taken to go from it to the stationary mirror and back. CONTEMPORARY ADVANCES IN PHYSICS 177 answers. Foucault and Fizeau reported that she had repHed: the former. But they had not heard distinctly ; for her actual response was : tieither. Michelson's experiment however came after the idea of group- velocity as distinp;uished from wave-velocity had been invented and established. The refractive index of carbon bisulphide varies with wave-length. On determining the wave-speed or phase-speed v from the refractive index (by the equation N = c/v) and then the derivative dv/d\, it is found '" that in the region of the visible spectrum, the term \{dvld\) amounts to about seven per cent of the term v, on the right- hand side of equation (20) — that is, the group-speed should be some seven per cent lower than the wave-speed in carbon bisulphide. In air, however, group-speed and phase-speed are sensibly the same. The ratio of the group-speeds in air and CSo falls close to Michelson's value." Coming as it did, therefore, the Michelson experiment merely showed that those who had subtilized the Huyghens' theory by introducing sine-waves had incidentally invented something able to move with the measured speed of a light-flash, though nothing of the sort had been available in the original form. Had it come earlier — well, there is no way of knowing what would have been inferred; but people might have come to think that after all a wavefront-theory or a corpuscle-theory of light may have some use and value, even though the speeds assigned to the waves or the corpuscles do not agree with those actually meas- ured. Such an attitude of mind would be rather advantageous, today. As a corollary for the present I submit: in picturing a jet of free negative electricity as a beam of waves or a stream of corpuscles, we should not be too confident that either the speed of the waves or the speed of the corpuscles is the speed with which a segment dissected from the jet would move from place to place, until someone succeeds in making actual measurement of this last. Fundamental theory has something to say on this point, which we will presently consider. '" I take all the numerical values in this section from a re\'ie\v of Michelson's work by J. Willard Gibbs {Am. Jour. Sci. 31, pp. 62-64; 1886) which so far as I know is the latest critical discussion of the data. '1 The problem is more complex than I have intimated, not only because Michelson observed light covering a very wide range of wave-lengths so that i' and dvjdX both extend over wide ranges of values, but also because different parts of a wave-front are reflected from different parts of the mirror at different moments, and therefore from differentl y-incl ined parts. Quite a controversy went on during the eighteen- eightiesin the pages of " Nature" as to what it was that Foucault had really measured. Rayleigh at first {Nature 24, p. 382; 1882) thought it was g; then changed his mind, (25, p. 52; 1882) and decided it was t'-/g; then was convinced by Schuster {3i, pp. 439-440; 1886) that it was really i'V2(t' - g). J. W. Gibbs then took a hand {Zi, p. 582; 1886) and contended that after all it was really g. The contro\"ersy seems to have rested there. It may be added that Michelson's data eliminate v'-jg, but do not quite discriminate between g and Schuster's expression. 17cS BULL SYSTEM TECIISICAL JOURNAL r'.ROrP-Sl'EED AXn rORI'lSCI.K-SPEED Thus far I have said thai if we wish to use wave-theory and corpuscle- theory alternatively, we must make the momentum of the corpuscle equal to the quotient of the constant h by the wave-length of the waves; but I have said nothing about the energy of the corpuscle. Let us adopt the universal assumption — based on a multitude of experiments, for instance those on the photoelectric effect — that the energy £ of a corpuscle of light is equal to the product of its frequency V b>- the same universal constant //; and let us extend it to the other kinds of corpuscles which we may associate with other kinds of waves, and vice versa. Then the complete description of the particles associated with waves of wave-length X is as follows: p = ///X, E= hv^ hv/X. (21) Here, as before, v stands for the phase-speed of the waves (not the particles). Returning to the formula (20) for the group-speed, we now can write it thus: r^= V - \{dv:d\) = v\ - \dip\)ld\ , = -XHdp/dX) = - i\'lh)idE/d\). ^ ' Suppose next that the energy and the momentum of the corpuscles in question are related to each other and to their speed in the well- known fashion of ponderable bodies, to which it is known that electrons conform. Thus for sufficiently low speeds, the relations are practically those of the "classical" mechanics: p = Wo/^ E = i^Wo«-, whence E = p-,2n!Q. (23) Here nio stands for the constant mass, u for the speed of the corpuscles (not the waves). The energy of the corpuscles is a function of the momentum only, and continuing to develop the formula (22) for the group-speed, we find: o = {-}^jh)(dE;dp){dp;d\) = dE;dp .^^. — p^i — I3~lm() = II. The group-speed of the waves is equal to the speed of the corpuscles. The same conclusion follows if we use the relativistic definitions for the energy and the momentum of a particle, E = WocVVl - (3-, p = Wn^f/Vl - f3- (J3 = II 'c), E = c^niifc- -\- p- as the reader mav test for himself. CONTEMPORARY ADVANCES IN PHYSICS 179 Summarizing: if the corpuscles associated with the waves have the properties of ordinary material bodies — if, let us say, for short, the corpuscles are material particles, their speed is equal to the group-speed of the waves. This is a ver\' happy and agreeable result. It compensates very largely for our having been forced to concede that if we want both waves and corpuscles, the wave-speed and the corpuscle-speed must be different. The wave-theory has supplied another velocity which is equal to that of the corpuscles. Moreover it is precisely the \'elocity with which we should expect an isolated segment of a wave train to move from place to place. If someone were to cut a piece out of an electron-jet and measure the time it took to traverse a known distance, the speed which he would deduce from his data would probably agree both with the corpuscle-speed and with the group-speed, and disagree with the wave-speed. It would be interesting to try this out. In the equations ^23) I have taken account only of the kinetic energy of the corpuscles; in the equations (25), only of their kinetic energy and of the 'rest" energy associated with their mass. But the explanations of refraction by the two theories will no longer be con- cordant, unless the potential energy also is admitted. Let us denote the potential energy of a corpuscle by U; and, since as yet these theories have been verified only for negative electricity, let us immediately write eV for U, e standing for the charge of an electron and V for the electrostatic potential in the region where it is. For the total energy of the corpuscle, then, we have instead of (25) the relati\'istic expression, E = nl,c^i^l - ff' -\- U = WocVVF^^^ + eV, (26) which for small values of the corpuscle-speed u (= 3c) reduces to the classical expression, E = hnuu- -\- U = hnuii" + eV. (27) In an earlier section we interpreted the refraction of an electron beam passing from vacuum into metal by thinking of the metal and the vacuum as being two regions in which different values of electrostatic potential prevail, the potential thus changing sharply from one value to the other at the surface which bounds the solid. Xow when the beam considered as a stream of corpuscular electrons passes across such a surface, the energy of each electron as expressed by (26) or (27) remains the same, though the proportion which is kinetic energy is changed; and therefore the frequency Ejh of the equivalent wave-train remains the same. If then we keep the assumption that the wave- ISO BULL SYSTEM TECIIXICAL JOCRXAL length of the waves is equal to // divided by the monientuin of the particles, we have the following value for the ratio between the wave-speeds v' and v on the two sides of the surface: „y_x;.v.(|^)/(|*).,V,, (28) and the speed of the waves varies inversely as the momentum of the corpuscles, which is just what is required in order that we may hold both the theories simultaneously. But how about the theorem that corpuscle-speed is equal to group- speed? Returning to the equations (25), we see that the introduction of the potential energy has altered the relation between energy and momentum; we now have: E = c^Jnioc' -f p- + eV. (29) But so long as we are comparing different electron-streams in the same medium (vacuum, for instance), the potential energy is the same for all and does not depend on the momentum; and differentiating E with respect to p to obtain the value of the group-speed g, we get: .? - dE/dp = ^^ = c^m.nHX_-^ ^ E - eV nioc-l^l - (3- and thus group-speed and corpuscle-speed are equal, as before. I will write down the expression of the phase-speed, although for the physicist it is of minor importance, not being measurable — a fact which exempts us, temporarily at least, from pondering over the curious feature that it depends on the value of the potential energy of the corpuscles, and therefore (for electrons) on the value accepted for the electrostatic potential of the region where the wave-train is, even though in practice it is generally assumed that electrostatic potential may be measured from an arbitrary zero. The formula is v = E/p = -^^-h'^^i±U Wo«/Vl - /3- (30) = c-ju + U/p, and if we put the potential energy of the corpuscles equal to zero, we find the phase-speed varying inversely as the corpuscle-speed,^'- and greater than the speed of light. 1- There is a paradox here which, as I can testify from personal experience, is a dangerous source of confusion. The formula v = c'-jii sounds like an approximation to tlie formula v = const' p which I have gi\'en as the retpiisite relation lictwecii CONTEMPORARY ADVANCES IN PHYSICS 181 Stationary Waves and Oscillating Particles We have tried out, separately and in tandem, two alternative ways of interpreting a beam of radiation advancing through space; first as a stream of corpuscles, then as a train of waves. We will now try out two alternative ways of interpreting radiation enclosed in a box; first as a system of stationary waves, then as a quantity of corpuscles rush- ing to and fro and bouncing from the walls. To simplify the case as much as possible, think only of motions parallel to one side of the box; or to make the pictures more graphic, think of a tube or pipe like those often used in experiments on sound, in which the waves travel along the axis. Now it is well known that when a train of sound-waves is sent through a tube, or generated by vibrations somewhere in the tube, it is partially reflected from the far end, then again partially reflected from the near end, and so on over and over again; the overlapping wave trains passing to and fro interfere with one another; and when the wave-length is related in a certain way to the length of the tube, the overlapping wave trains form a stationary ivave-pattern of alternating loops and nodes — the tube is said to be in resonance. If the two ends of the tube are alike (both open, or both closed) so that reflection takes place in the same way as both, the waves which admit of resonance are those of which the half-wave-length or an integer number of half- wave-lengths fits exactly into the tube; denoting by d the length of the tube, these wave-lengths are given by the formula-: wQj= d, n= 1,2,3, . . . (41) This equation defines what may be called the characteristic wave- lengths of the tube. The tube distinguishes these, or the wave trains possessing these wave-lengths, from all the others. Suppose on the other hand we had particles rushing back and forth along the axis of the tube, and rebounding without loss of energy whenever they struck either wall. Denote by ii the speed of a particle ; it takes a time-interval Id/u to describe a complete round-trip with two rebounds, and one might say crudely that it has a frequency u/ld. I say "crudely" because the corpuscle is not moving with a sinusoidal motion, like a pendulum-bob; its speed does not vary as a sine-function wave-speed and momentum. However the two relate to entirely different situations. The first is a comparison between wave-speeds and corpuscle-speeds for different beams in the same medium. The second is a comparison between wave-speeds and corpuscle-momenta for the same beam in different media. The resemblance between the two is accidental and misleadin;^. I am ind(-l)ted to Professors C. 11. lukarl and E. C. Kemblc for elucidation of liiis point. 182 BELL SYSTEM TECHNICAL JOURNAL of time, but retains the same value throughout except for the change of direction; if we were to apply a Fourier analysis to this motion, we should find not only the frequency ///2c?, l)ut all of its overtones. Let us think however only of this fundamental frequency. Now it is evident that there is nothing, in our ordinary conceptions of particles rushing back and forth and rebounding from walls, to distinguish any value of speed or frequency above any others. The phenomenon of resonance sets certain wave-lengths apart from others, but there is nothing to correspond to resonance in this latter case, and set certain speeds apart from others. But instead of sound, think of some kind of radiation which we have interpreting both as corpuscles and as waves — light, for example. Light enclosed between parallel reflecting walls forms stationary waves,''' provided that its wave-length is related to the distance d between the walls by the equation (41), which I rewrite: X = Id/u, ;/ = 1, 2, 3 . . . (42) The parallel reflecting walls, or the limitation which they set upon the space accessible to the light, thus single out certain characteristic wave-lengths and distinguish them from all others. How interpret this fact by corpuscle-theory? Well, we have been associating waves of wave-length X with cor- puscles of momentum p = h/\; let us continue to do so. The reflecting walls, then, single out certain characteristic values of momentum given by this equation, derived straight from (42): p = nh/ld, (43) which I proceed to rewrite thus, 2d-p = uh ;/ = 1, 2, 3 . . . (44) These values of momentum, I have said, are set apart from all the rest. If waves and corpuscles are interchangeable as bases for a theory of light, then the feature of wave-motion known for short as "resonance" obliges us to make that supposition. But in what way, and to what extent, are they set apart? According to modern quan- tum-theory, they are actually the only possible values. A particle describing a cyclic motion of this character, in which it moves a fixed distance with a fixed momentum and then moves the same distance backward with the same momentum reversed and so forth ad infinitum, is constrained by something in the order of nature to have one or '^Interference f)atleriis are essentially of this type, thoiii'li iisiiali\- they are formed between mirrors ol)liciiie to one another. CONTEMPORARY ADVANCES IN PHYSICS 183 another of the "permitted" momenta defined by equations (43) and (44). Examining equation (44), one sees how this definition of the per- mitted momenta may be stated. The quantity on the left of (44) is the product of the momentum of the particle, by the distance which it traverses each time it performs its cycle. '^ This product must be equal to an integer multiple of the Planck constant //. Now the quantum-theory of the atom developed fifteen years ago by Bohr, Sommerfeld and W. Wilson — the first and greatest of the forward steps in the contemporary conquest of the problem of atomic structure — was based on the assumption that an electron perfomiing a cyclic motion must perform it in such a way, that its momentum conforms to a condition of which equation (44) is but a special case. This is the condition alwavs written thus: / pdq — 7ih, w = 1, 2, 3 . . . (45) If the electron is oscillating to and fro in a straight line through a position of equilibrium, q stands for its distance from that position and p for its momentum, and the integral is taken once around a complete oscillation. It is evident that (44) is the special form of this equation for the case in which the force acting on the electron is vanishingly small until it hits the wall and then suddenly becomes enormous. If the electron is revolving in an orbit in two or three dimensions, there are two or three equations like (45) all postulated at once; but I shall not take up such more complicated cases. Summarizing the outcome of this section in a phrase: if we associate waves of wave-len»th X with corpuscles of momentum hi\, and stationary waves in an enclosure with corpuscles flying hack and forth between its walls, then the condition that the waves must fulfil to form a stationary system is equivalent to the quantum -condition imposed upon the corpuscles. This is an illustration of wave-mechanics. How extraordinarily fruitful and valuable such comparisons have proved in the hands of Louis de Broglie, of Schroedinger, Rose, Fermi and Sommerfeld — to name only a few — I have shown in part, in earlier issues of this journal. Here it must suffice to say that Schroedinger developed the principle into a form suitable for predicting the stationary states of atoms; Bose constructed out of it a competent theory of radiation in themial equili- '^ It travels a distance d in the forward sense with a momentum p, and then an equal distance in the backward or negative sense with a momentum of equal amount but reversed sign, so that the total product of distance by momentum is t>d + (- p)(-d) = 2dp. 1S4 BULL SYSTL.M TliCIINICAL JOURNAL liriiim, considered as a gas of which the atoms are corpuscles of hght; while Fermi. Dirac and Sommerfeld between them used it to make a powerful theory of the free negative electricity in metals, conceiving this alternatively as a gas of which the atoms are electrons, and a system of stationary waves enclosed within the surface of the metal as in a box with reflecting walls. Diffraction' of Waves axd Diffraction' of Corpuscles The effect of a diffraction-grating upon a beam of light projected against it has always been considered the most striking evidence that light is of the nature of waves and not of corpuscles. Indeed it is considered to suffice in itself to prove the corpuscle-theory untenable. With any common understanding of the term corpuscle-theory, this statement is correct; but we had better put it in the softer form, that the effect of a diffraction-grating on a beam of light proves that if we adopt a corpuscular theory we must endow the corpuscles with some very strange property which nobody ever thought that particles could possess, and which may even seem to be in contradiction with their nature. W'e had better put the statement in this milder way, because it now is known that in spite of all the evidence for individual electrons, a beam of negative electricity is affected by a grating in much the same way as a beam of light. Take then almost the simplest conceivable case of diffraction; a plane-parallel beam of light falling perpendicularly on a wall containing many equally-spaced parallel slits, and a part of the light passing through the slits to a screen infinitely far away. On this infinitely- distant screen — which may in practice be brought up to a convenient nearness, b}- means of a lens — one sees a peculiar pattern of light and shade. I single out one particular feature of this pattern : the fact that there are maxima of illumination along certain lines parallel to the slits. One of these, for instance, is straight ahead from the slits, along the direction of the incident beam prolonged; another is oft' to one side, in a direction making a certain angle (say ) with that of the incident beam; another is equally far off to the other side. These two last- named, the Jirst-order maxinia, are those we shall consider; it will be enough to speak of one. By the wave-theory, a first-order maximum is explained as follows. Each of the slits is the source of a secondary wave train of spherical wave crests, stimulated by the primary wave train, and having the same frequency and wave-length. Consider any two adjacent slits. Secondary wave crests start from the two at the same moment. At any point equally distant from the slits, they arrive simultaneously, and CONTEMPORARY ADVANCES IN PHYSICS 185 reinforce each other; this is the explanation of the central bright fringe. At any point not quite equally distant from the slits, they do not arrive quite simultaneously, and the reinforcement is impaired. But at a point which is further from one slit than from the other by just the wave-length X, the wave crest arriving from the latter meets the next previous crest from the former, and the reinforcement is re- stored. The first-order maximum is located at these points. Fig. 4. From Fig. 4 one sees ''^ that when the screen is very far away, the points distant from the slit 6\ by one wave-length more than they are distant from S-i are situated in the direction inclined at 0 to the straight- ahead direction, the angle 0 being given approximately by the formula sin 4> = ^,<3, (46) where a stands for the distance between the slits. When the screen is infinitely far away, the formula is exact. (I must admit that it is somewhat disingenuous to simplify the problem by solving only the special case in which the screen is infiniteh' far away, for the general case opposes much more serious difiiculties to the corpuscle- theory-; but this is the special case of greatest physical importance, and one has to make a beginning somewhere.) We have now explained the presence of a first-order maximum in the pattern of light and shade on the screen, though it cannot be said that we have "verified" formula (46), for that formula serves as the practical definition of wave-length : wave-lengths are measured by '■' From the figure wq see that for di and d>, the distances from .S'l and 6- to the point P on the screen, we have: d{- = D' + X-, dy = D' + (.V - a)\ d, = D sec <^, .v = D tan <^ and hence (d, - d-i)(d, + and using equation (46). Let us now try the corpuscle-theory on the problem. Putting as heretofore the value h/X for the momentum of the cor- puscles, translate (46) into the language of the alternative theory; one gets: sin (/) = h/ap. (47) In words: a corpuscle of momentum p, passing through any slit, is particularly likely to bend around through an angle (/> of which the sine depends in a certain way on its momentum and on the distance to the next slit. Which is to say: the likelihood that a corpuscle entering a slit will bend its course through a certain angle depends on the presence of other slits in the same wall, and on the distance between these slits. But the reader will inquire: how does the corpuscle entering one of the slits know that the other slits are there? If all the other slits were suddenly stopped up, the first-order maximum would vanish, the likelihood that the corpuscle would turn in the direction given by (47) would fall to zero; but how could it know that they had been stopped up? Well! this is precisely the strange and extravagant property with which we are forced to endow the corpuscles, if we want to use the particle-theory to explain diffraction. It must be supposed that when passing through a slit, a particle of light knows whether there are other slits and, if so, how they are spaced. It must be supposed that an X-ray particle striking an atom in a crystal knows that there are other atoms in a regular array, and knows moreover just the pattern and the scale of that array. It must be supposed that electrons enjoy a like omniscience. Or to express it in more technical language; the prob- ability that a corpuscle of light, of electricity or of matter shall be deflected through a given angle when it strikes an atom or passes through a slit must be supposed to depend on the arrangement of the other atoms or the other slits in the vicinity. This idea is not easy to accept; but it must be accepted, if one is to build up a complete cor- puscular theory of any of these entities. But if one accepts it, one finds that the stipulation (47) turns out to be another example of the general quantum-condition of which, in (44), we have already met one instance. For write it thus: ap sin (j) = apt = nh, « = 1, 2, 3 . . ., (48) the factor n being now introduced to take account of the maxima of second, third, and higher order which also occur on the screen, though CONTEMPORARY ADVANCES IN PHYSICS 187 I refrained from mentioning them earlier. I have used the symbol pt for the quantity p sin 0, for this, as one sees immediately, is the tan- gential component of momentum which the corpuscle acquires at the deflection, not having had any before. The wall containing the slits, or the row of atoms if we consider instead the difi^raction of X-rays by a crystal, receives an equal momentum in the opposite sense. We may therefore say that diffraction occurs in such a way, that the regularly- spaced series of slits or atoms receives a momentum pi given by the formula : apt = nh. (49) But now what is the product apt? It is the product of the mo- mentum of the row of atoms or slits by the distance a between any adjacent two; it is therefore the integral J'pdqoi the general principle (45), evaluated for the range of integration a. Now the general principle is supposed to apply when the range of integration covers a complete cycle of a periodic motion. There is nothing obviously periodic about a steady sidewise sliding of a row of atoms with a constant momentum. But in a sense, there is after all something periodic. For if the row of equally-spaced atoms (or slits) extends to infinity in both directions, then when it has moved sidewise through the distance a each atom lies exactly in the former place of another atom, and the original arrangement is to all appearances restored. The steady onward motion of the regular array is also a cyclic departure and return to a periodically-restored arrangement; and the maxima of the difl^raction-pattern are determined by applying the quantum-condition to this cyclic motion. The reader may ask: how about the component of momentum in the direction at right angles to the grating? Without precisely answer- ing that question, I will end the article by applying the corpuscular theory to a case in which all the components of momentum are duly taken into account: diffraction of X-rays or of electrons by a three- dimensional crystal. Suppose an "ideal" crystal extending infinitely far in all directions. It is composed of similar and similarly-oriented "atom-groups" — I will use the language and the symbols of the eighteenth article of this series — arranged upon a "space-lattice," of which the three spacings shall be denoted by a, a', a". If we start with one atom-group A, then along one direction from it there is an infinite sequence of such groups at distances a. la, 3a, . . . and also at distances (— a), (— 2a), (— 3a), ... in the opposite sense. Call that the .v-direction. Then along another direction through .1, say the y-direction, there is an 188 BELL SYSTEM TECHNICAL JOCRXAL infinite sequence of groups at distances a', 2a', 3a', . . . and (— a'), (— 2a'), etc.; and along a third or 2-direction through A, there is an infinite sequence of atom-groups spaced at intervals a". Now think of the atom-groups as hard particles, and the corpuscle of light or of electricity (the "X-ray quantum" or the electron) as a hard particle which rushes into the lattice, hits one of the atom- groups — A, say — and bounces oft". Denote by 0, ', 0" the angles which its original direction of motion makes with the x, y, z directions respectively; Ijy 0, C, 6" the angles which its final direction of motion makes with these three. Before the defiection, the corpuscle has a momentum of magnitude p, parallel to its original direction of flight; afterward it has a momentum of the same magnitude, but parallel to its final direction of flight. At the deflection, then, it loses — that is, it communicates to the lattice — a momentum of which the three components along .v, y, z have the values: /?(cos d - COS (p); picos d' - cos 0'); picos d" - cos (/>"). Now if, following the foregoing procedure, we equate the first of these to some integer multiple of h/a, the second to some integer multiple of h/a', and the third to some integer multiple of h/a", and then translate momentum of corpuscles into wave-length of waves by the now-familiar formula p = h/X, we get: a(cos d — cos (f)} = ii\, a'(cos d' - cos 4>') = ii'X, (50) a"(cos d" - cos 0") = ;/"X, where ;/, ;/', ii" stand for any three integers. Now these are the equations (numbered 3, 4, 5 in the eighteenth article) to which conform the " Laue beams," which is to say, the directions in which electrons and light are actually diff^racted by crystals. Perhaps I should close with two or three admonitions. To make the wave-theory and the corpuscle-theory equivalent for a few simple cases is of course not at all the same as making them equivalent universally. Also, the examples in this article are not always so elementary as they may seem. The first involved two distinct media with a sharp bound- ary between; and discontinuity is always less agreeable than continuity to the mathematician. The last but one involved a non-sinusoidal vibration, which is much more complex than a sinusoidal one. More- over, the concepts of light-waves and quanta are not nearly so beauti- fully welded together as those of electricity-waves and electrons. Nevertheless these illustrations may help to weaken the idea that there is no way out of the present situation but to abandon either waves or corpuscles; for decidedly, there is a way. Wave Propagation Over Continuously Loaded Fine Wires By M. K. ZINN The paper contains the resuhs of a theoretical investigation of wave prop- agation along a pair of wires that are "loaded" by enclosing each wire in a continuous sheath of magnetic material. The results of greatest practical interest are certain approximate formulas that are sufficiently simple to be adapted to engineering design studies, while having a high degree of precision for all practical dimensions and frequencies. THE purpose of this investigation is to define the character of wave transmission along a pair of wires each of which is loaded with a continuous sheath of magnetic material. Exact expressions for the propagation constants are developed from the general theory that applies to such a system. Also, simple approximate formulas are given for the sizes of wires that are generally used in paper-insu- lated cables. Wave Propagation Along a Pair of Wires with Magnetic Sheaths For the benefit of those who are not interested in following the theoretical work in detail, a general sketch of the method and a sum- mary of the mathematical results will be given first, together with a discussion of some numerical examples. Details of the theoretical work have been placed in the Appendices. The analysis here given follows closely the methods developed by John R. Carson ^ in a solution of the transmission of periodic currents along a system of coaxial cylinders. The analysis for the case where the outgoing and return conductors are coaxial is applied, with only small modifications, to the case where the two conductors are parallel and not coaxial. This application of the theory ignores the "proximity effect." - That is to say, it assumes that the electric and magnetic forces within each conductor are functions only of the distance from its axis and of the coordinate in the direction of propagation, which is strictly true where the cylindrical conductors are coaxial. 1 "Transmission Characteristics of the Submarine Cable," John R. Carson and J. J. Gilbert, Jour>ial of the Franklin Institute, December 1921. - This is the usual method of dealing with problems involving balanced parallel conductors. The alternating-current resistance of the s\-stem may be expressed as the product of the a.c. resistance, assuming a concentric return, and a "proximity effect correction factor," which takes into account the influence of the parallel return conductor. The "proximity effect" is in general negligible at voice frequencies for conductors of sufficiently small cross-section, such as those of paper insulated cables. References: "Wave Propagation over Parallel Wires: The Proximity Effect." John R. Carson, Phil. Mag., April 1921, and "Wave Propagation over Parallel Tubular Conductors," Sallie Pero Mead. Bell System Technical Journal, April 1925. 13 189 IQO BELL SYSTEM TECHNICAL JOURNAL The physical system contemplated is shown in Fig. 1. The out- going and return systems of conductors, each comprising a cyhndrical wire with insulated cylindrical sheath, are assumed to be identical in all respects. For the sake of generality, it is assumed that the magnetic sheaths may be insulated from the wires, as shown. The interesting practical case where wire and sheath are contiguous, form- ing a bi-metallic conductor, then appears as the limiting case of infinitesimally thin insulation. WIRE •E'; DIRECTION OF PROPAGATION I MAGNETIC SHEATH 3L| bg cl2 INSULATION (ADMITTANCE Yg) MAGNETIC SHEATH (X2,|J.2) ^\ E" *i XV, + 4^ dz \— INSULATION (ADMITTANCE Y,) — ^ » 7 — 1- iHl'+^'dz (I) ^Ii Y WIRE(X,.|I,) * oz ^ Fig. 1 — Illustrating various quantities involved in the analysis. The problem consists in finding a solution for the propagation con- stant of the system from Maxwell's equations. If the magnetic sheaths are in contact with the wires, the propagation constant is given in the usual form, V = \ Y^Z, where Y2 is the admittance across the insulation between the sheaths and Z is the series impedance of the system. The admittance is, in general, either a known, or an experimentally determined, quantity; so that for this case the theoretical problem resolves itself into that of finding the series im- pedance. An important part of the investigation is, however, to determine WA VE PROP A GA TION 0 VER CONTIN UO USL Y LOA DED WIRES 1 9 1 what the effect would be of introducing insulation between the wire and its sheath. In this more general system, shown by the sketch, the solution for the propagation constant has two values, because two layers of insulation are involved, and cannot be expressed in the usual form. It is found, however, that it can be expressed in terms of the propagation constant for the elementary case where wire and sheath are in contact by introducing two other known propagation constants that determine transmission along the separate pairs of conductors in the system. The expression for the propagation con- stant, when given in this form, shows directly the effect of insulating the wires from their sheaths. It is necessary first to define certain impedances. Let /] be the total current in one of the wires and lo the total current in its sheath. The tangential electric forces in the surfaces of wire and sheath are denoted by Ei" , £«' and E^" , as shown in Fig. 1. These electric forces are linear functions of the currents, as follows: -C,2 — -^21 1 1 I ■'^22 -'2) -C,2 ^^ •^21 J 1 ~r ^^22 ■*2) E," = Zn"h. (1) The impedances which appear in these equations as the coefihcients of the currents are functions of the electrical constants and dimensions of the wires and sheaths. Their values are given in Appendix A. Now let 7 = propagation constant determining transmission along the loaded wires if the wires and their sheaths were in contact = VKoZ. 7i2 = propagation constant determining transmission along one wire with its sheath as the return, when the sheath is insulated from the wire = VFiZio. 722 = propagation constant determining transmission along the two sheaths if the wires were removed = V ^2^22. Then, from (1) £2' - Ey" Z12 — Z22 = Z = I Ai — Zii — Z21 ~l~ Z22 -(- Ai, 1-2 = — /i 2E2" h IE." + A2 = 2Z22" + X: In these equations, + X2 = 2 Z22 7 'i Z]i — Z21 ~r Z22 /i = 0 y (2) + X2 192 BELL SVSTE.U TECHNICAL JOURNAL Xi = iwLio = reactance arising from the magnetic field between the outer surface of the wire and the inner surface of the sheath. A'o = iccLoc, = reactance arising from the magnetic field between the two sheaths. The terms in brackets in the equation for Z give the "internal impedance " of one of the loaded wires for the elementary case where wire and sheath are in contact, and Xo is the additional reactance that arises from the magnetic field outside the wires. With the elementary propagation constants, 7, 712 and 722 so defined, it is found that the propagation constant, F, of the general system can be expressed as follows: 2r~ = 712- + 722- ± \(7i2- + 722-)^ - 47-712-. {^) It is convenient also to express the two solutions for T in the form of series : -r -1 ^ 2 T12'- I J 7i2^ , r, r, 712*^ , 712" + 722- (712- + 722"/ (712- + 722-) To- = 7i2- + 722" — Ti'-'. (4) lation between the wire and sheath. For, if is small The solutions in the series form show the effect of introducing insu- 47-7i2"-^ (719- + 722")- compared to unity, as it would be in a continuously loaded wire with a thin magnetic sheath of high resistance, then, to a first order of approximation, the principal propagation constant Fi is less than 7, the propagation constant that determines transmission when wire and sheath are in contact, by the factor The other propagation constant, Fo, is, in this case, very large com- pared to Fi and plays no appreciable part in defining the character of transmission except at points very near to the terminals of the system. For practical purposes, the system may be considered to have only one significant mode of propagation. Case of a Wire with Contiguous Sheath The Internal Impedance The practical case where the magnetic sheath and the wire are contiguous, forming a bi-metallic conductor, is of special interest. WAVE PROPAGATION OVER CONTINUOUSLY LOADED ]VIRES 193 In this case, the propagation constant is uniquely determined from a knowledge of the admittance between the loaded wires and of their series impedance. The "internal impedance" of the loaded wires comprises the larger part of this impedance. For the purpose of engineering design work, it is convenient to have at hand approximate formulas for the "internal impedance." The exact expression for the impedance is given by the last of equations (2). \A'hen the magnetic sheath is thin, as compared to the radius of the copper wire, certain approximations can be made. These are explained in Appendix A. The result is the following formula for the "internal impedance": where F = wjULob Z i _\ — orF + ioiCi -^ + lull ^7rX,X,M-.>/'^ + Mi&(^^-^log^ (5) II = 2ir''-\,n.af \\t{\, - Xi) + Xi^l +2^/ /? /? R = p ,'p = d.-c. resistance of one of the pair of bi-metallic conductors, Ri = , ,., = d.-c. resistance of the inner part of the conductor TrKib- . (the wire), i?2 = ^ , ., 7T- = d.-c. resistance of the outer part of the irXMi- — 0~) , / , , 1 X conductor (the sheath), Lo = 2^2 log T = low-frequency inductance contributed by the sheath, b = radius of the wire, a = outside radius of the sheath, t = a — b = thickness of the sheath, ^1, Ml = conductivity and permeability of the wire, X2, M2 = conductivity and permeability of the sheath, CO = Itt times the frequency, I'M BELL SYSTEM TECHNICAL JOURNAL The total series "loop" impedance of the pair of loaded conductors per centimeter is Z = Zi -\- Xo-^ For the purpose of indicating the degree of precision of the approxi- mate formula, data are given in Fig. 2 on the internal resistance and inductance of various copper wires coated with loading material to 300 250 tf) I 200 a. us a. LU o z < in IS) LU a. (0 0.0248 > a. z UJ 0.0246 uj 0.0244 _1 cc LU a ai o z < I- o D a z 0.0242 0.0240 0.0238 O0236 '^^s^ o ( ) ( 1 26 GAUGE ^ "V. :^ 16 3 4 6 8 FREQUENCY — KILOCYCLES 10 12 pi^_ 2 — -Internal impedance of wires of various sizes with continuous loading ol approximately 25 millihenrys per wire mile (for very small currents, i.e., hysteresis losses not included). 3 All quantities are expressed in the electromagnetic c.g.s. system of units. To obtain the result in ohms per loop mile, multiply by 160,934 (10~»j. In the case of cable circuits, A'2 ( = zcoL22) is an experimentally determined quantity, L22 having a value of about .001 henry per mile. WAVE PROPAGATION OVER CONTINUOUSLY LOADED WIRES 195 such a depth as to give an internal inductance of about .025 henry per wire mile. The magnetic material in the sheath has been assumed to have a permeability of 3,000 and a conductivity of .77 x 10~^ in e.m.u. (resistivity 13 microhm-centimeters, in practical units). The data shown by the solid lines are exact while the points give the results obtained by means of the approximate formula (5). A com- parison of results is tabulated below for the largest wire (16 B. & S. gauge), where the errors of the approximate formula are greatest. Internal Resistance and Inductance (of One Wire) Ohms and Millihenrys per Mile Frequency — Kilocycles Exact Approximate Errors Res. Ind. Res. Ind. Res. Ind. 0 2 5 8 10 21.065 31.674 86.795 186.65 276.04 24.77 24.56 24.37 24.05 23.75 21.065 31.63 86.18 183.6 269.4 24.77 24.63 24.41 24.01 23.66 - .14% - .71 -1.63 -2.41 +.29% + .16 -.17 -.39 The errors are roughly proportional to the quantity, WajM2X2. For a loading material having, say, one-quarter the permeability and the same conductivity, the errors would be about twice as large, therefore, if the inductance and the wire size remain the same. Hysteresis Loss The real part of the internal impedance given by (2) or (5) is the effective internal resistance of the bi-metallic wire, taking into account the heat losses that arise from the electric current, namely, d.-c. resistance, eddy current loss and "skin effect loss." The formulas do not take into account hysteresis loss, which is a magnetic phe- nomenon as distinguished from these electric phenomena. The de- termination of hysteresis loss rests upon experimental data. If the energy loss due to hysteresis in the magnetic material per unit volume per cj'cle is Ji (ergs), then the resistance increment due to hysteresis is R, hrdr. (6) For the low values of magnetic force that obtain in telephony, it is found that // = ■qB'^, where r? is the hysteresis coefficient and B the induction density. Therefore Rk 1' X IPrdr. (7) 1<)6 BELL SYSTEM TECHNICAL JOURNAL Since the majjnetic coating is thin, and the "demagnetizing," or "screening," effect of eddy currents small, it may be assumed that // = 2//r. (It will not exceed that value, at least.) Using this approximate value for //, the resistance increment due to hysteresis is Ri, = 8r?co/x-'/ a ab = 2r]fJ.O)BaL2, (8) 0,8 0.6 Z UJ o < I- o H UJ I 0.4 Z u CC a. O 0.2 I I- < a: 4 6 FREQUENCY — KILOCYCLES Pig 3 — .Illustrating the fractions of the total current that are carried by the copper wire and by the magnetic sheath (19 gauge (B. & S.) wire with continuous loading of 25 millihenrys per mile). where Ba is the induction density at the outside boundary' of the sheath. The Distribution of the Current in Wire and Sheath It is a matter of interest to know how much of the current is carried by the magnetic sheath and hf)W the current is distributed over the cross-section of the wire and sheath at various frequencies. The WAVE PROPAGATION OVER CONTINUOUSLY LOADED WIRES 197 solution of this problem is not an essential part of the investigation, but it helps in understanding what takes place in the bi-metalhc conductor. The ratio of the currents in wire and sheath to the total current, as computed from (1), is plotted in Fig. 3. It will be noted that the fraction of the total current carried by the sheath becomes greater 800 5 o a s If) Q. Z < z u Q UJ CC u 600 400 200 FOR A TOTAL CURRENT OF I AMPERE IN THE WIRE AND SHEATH • DIRECT CURRENT ^ Uir\C.^ I v_-'-'r\r\ui"< I *M0 KILOCYCLES 10 KC. 2KC, UJ < O UJ o UJ < o ? < Q- J 90 r^ 60 10 KC IjzKC. JU n _^-^_^_- L 1 2KC.^ *^IOKC. " -30 0.01 0.02 0.03 RADIUS OF COPPER WIRE- CENTIMETERS 0.04 0.045580 0.046758 HSHEATHk- THICKNESS / SCALE \ I ENLAROEDJ Fig 4— Illustrating the current density throughout the cross-section of a Avue loaded with a continuous magnetic sheath— for direct current and 2 and 10 kilocycle alternating currents. (Same 19 B. & S. gauge wire as that of Fig. 3.) as the frequency increases. But the fraction carried by the copper nevertheless remains very nearly unity at all frequencies. This behavior is explained In^ the curves representing the phase angles involved. These show, of course, that at very low frequencies the 198 BELL SYSTEM TECHNICAL JOURNAL copper current and the sheath current are nearly in phase, but with increasing frequency, the copper current lags behind the sheath current, until at high fretjuencies the two currents approach a queidrature jihase relation. It may be said that at high frequencies the current in the loading material is practically all "wattless" current, in the sense that it contributes very little to the energy delivered to any receiving device connected to the line, but it dissipates energy, of course. At 10 kilocycles, for the 19-gauge loaded wire, the current carried by the magnetic sheath contributes only 2 per cent of the useful current (see Fig. 3) ; yet 75 per cent of the energy loss occurs in the sheath (see Fig. 2). The difference in phase between the component currents in wire and sheath is explained by the consideration that the reactance of a given filament of current is proportional to the magnetic flux external to it. In the copper, therefore, the elementary current paths have a small resistance, but a large reactance, due to the fact that nearly all the magnetic flux is in the loading material. Near the outer surface of the loading material, on the other hand, the current paths have less internal reactance, but the resistance is large. This brings the discussion to Fig. 4, which shows how the amplitude and phase of the current varies over the cross-section of the bi-metallic conductor for direct current and for 2 and 10 kilocycle alternating currents. For the 19-gauge loaded wire, illustrated, the "skin effect" in the copper is seen to be very small, the alternating current dis- tribution being practically uniform, as for direct current. At the boundary between the copper and the magnetic material, the current amplitude suffers a discontinuity, but the phase is continuous. The discontinuity in the current amplitude conforms to the law that the component of electric force along the conductor must be continuous at a boundary, which requires that the ratio between the current amplitudes on the two sides of the boundary must equal the ratio of the conductivities of the two materials. The current density dis- tribution over the cross-section of the magnetic sheath is uniform for direct current, of course, but for alternating currents, the density increases and the phase advances abruptly toward the outer surface of the sheath. APPENDIX A The impedances ' which appear in equation (1) in the bod\- of the paper as the coefficients of the currents are given b\-: ^ See abo\'e noted paper (reference i) tor ihc development of tliese forniulas. WAVE PROPAGATION OVER CONTINUOUSLY LOADED WIRES 199 7 " — 2io}iJ.2 Z/a — 1 X2 U,' ' 7 " — 2zCOyU2 X'2 U-2 u-r Z2/ = 2/CO/i2 .V2 f// • Z.J = 2iwfJL2 X2 1 z,/' = llCO/Jil 1 2/aJAi] /oGvi) (y) where ^'' ^ .V: L^' .vi /o'(.ri) (Note that Zoo' = Zoo" - Zoi"), t^y = - 3'y[/o(xy)i^o'(>'y) - /o'Cv/)A'o(xy)], V, = - ylMyi)K,{xj) - Mxj)Ko(yn, U/ = - ylJo'{xj)Ko'(yi) - Jo'{y^)Ko'(xjn V/ = - ylMyi)Ko'(xj) - Jo'(.r,)i^o(v/)]. (10) Jo and Ko are Bessel functions of zero order of the first and second kind, respectively, and Jo' and Ko' are their derivatives with respect to the arguments, which are given by X: = ad-^A:Tri(j)iXjKj, (11) 3'; = hji->^AiriwiXj\j, where w = It times the frequency, i — V— 1, a; and bj are the outer and inner radii respectively of conductor j, and ixj and Xy are its permeability and conductivity. Quantities wath the subscript 1 refer to the wire and those with the subscript 2 refer to the sheath. All quantities are expressed in the electromagnetic c.g.s. system of units. Writing Maxwell's Law, curl E = — -7- , around the contours indi- cated by the dotted rectangles in Fig. 1 gives where I'l, V2 are the potential differences between the surfaces of 200 Bh.LL SVSrF.M TECHNICAL JOURNAL the conductors, as shown, and $), $2 are the normal values of the magnetic flux that cuts the surfaces bounded by the contours. The term — 2E->" results from the symmetry of the system, which imposes the condition that the electric and magnetic forces at corresponding points in the outgoing and return conductors are equal and oppositely directed. Also, it is unnecessary to write a third equation for the field between the other wire and its sheath, because this equation would be the same as (12). Therefore, the transmission is charac- terized by only two modes of propagation. Since all the variables are propagated at the same rate, and since sinusoidal currents are being considered, djdz may be replaced by — F and didt by iw. Then £,' - E," + TV, = XJu (14) - lEo" + TFs = X.iU + /2), (15) where V is the propagation constant and A'l = i(joLi2 = reactance arising from the magnetic field between the outer surface of the wire and the inner surface of the sheath. Xo = ioiLoi = reactance arising from the magnetic field between the two sheaths. The potential dilTerences Fi, Fo can be expressed in terms of the currents by writing Maxwell's Law, curl II = 4tI, around contours in the outside surfaces of wire and sheath. (Such a contour for the wire is indicated by dotted lines in the sketch.) This gives 2T:a^-^= - 47rFiFi, (16) dz 27ra.^^= - 47rF2Fo. (17) az where Y\ = admittance across the insulation between wire and sheath. F2 = admittance across the insulation between the two sheaths. Smce ill = — and //■> = > a 1 a 2 r/i = \\Yu (18) r(/i + Li) = F2F2. (19) WAVE PROPAGATION OVER CONTINUOUSLY LOADED WIRES 201 Substituting (18) and (19) in (14) and (15), respectively, gives (20) (21) Xl — yT ] IX = £0' — El", X, - Y^ ) (/, + /2) = - 2E,", and substituting (1) in (20) and (21) gives the two equations of the currents. In order that they shall be consistent, the determinant of the coefficients must vanish. Therefore r2 X\ — ^ — Z'ji -\- Zii J 1 Xi — — + 2Zoi ■^ 2 A'o 7 ' + 2Z.J = 0. (22) The roots of this equation give the required solutions for the propaga- tion constant. First, however, it is convenient to introduce two known propagation constants. Let 712 = propagation constant determining transmission along one wire with its sheath as the return = -ylYiZn- 722 = propagation constant determining transmission along the two sheaths if the wires were removed = VFoZoa- Then, from (1), (20) and (21), Z12 — Zii — Z21 ~r Z22 r A 1) Z22 ^^^ 2/^22 I -^2> substituting (23) in (22) and rearranging, 712- - r2 (23) Fi - 2Z..0' — /v 22 T22" ^ " Fo = 0. (24) Expanding r - P(722- + 7l2-) + T12-722- - 2Z22''FlF2 = 0. (25) The remaining impedance can be eliminated by introducing 7, the propagation constant that would characterize transmission if the 202 BELL SYSTEM TECHNICAL JOURNAL wires were in contact with the sheaths. (In order not to disturb the dimensions, it may be imagined that the insulation between wire and sheath be replaced by an infinitely conducting material, which, how- ever, is assumed to conduct no current axially. Then E^' — Ei" — Xili.) To find 7, make Yi infinite and solve (25). Then r = Y.Z, and Z = Z,, - ^^ • (26) Zi2 Therefore 2Z22''J^ll^2 = T22^7]2' ~ 7"7l2"- (27) Finally, substituting (27) in (25) and solving the resulting equation gives the two solutions for the propagation constant, 2P = 712- + 722' ± V(7i2'' + 722'/' - 47-712-. (28) The arbitrary constants remain to be determined. The currents are, in general, 7i = ^iie-f-- + ^126-1- + ^iie^- + ^126^-% (29 The condition of principal practical interest is that of a long cable with connection made to the two wires and with the sheaths left free at the sending end. For this case, the conditions are (1) At s = 0, /i = Jo and h = 0, (2) At s = 00 , /i = 0 and A = 0, where Jo is the current delivered to the cable pair at the sending end. From the second condition, Bii = B12 = B21 — B22 — 0. (30) From the first condition, Au +-4 12 = /o, (31) A21 + ^22 = 0. But these constants must satisfy, for each of the two values of F, the equations of the currents, whose coefficients are given in the WAVE PROPAGATION OVER CONTINUOUSLY LOADED WIRES 203 determinant (22). Therefore w here A21 = KiAn, A 22 ^^ -'^2-'4i2i (32) ^12 — ^2*2 — T r~ ^22 — —■^22 — p 2 p 2 ZT~ ^22 — —^22 — ^, Z22' 7 , r,^ - Z.2 + 3r F - F.- Z7 ' ~ 7 0 7' - {2^2>) Zii ^ . r 2 -Z22+Y: Substituting (32) in (31) and solving A - T ^^ A - - T ^' -^11 — -' 0 ^? T^ } ^12 — -f 0 -A.2 — -A-i 7V2 — iV 1 ^21 — -'O'i? ^ — ~ ^22- A2 — Ai (34) Finally, the currents are given by /n (35) This completes the analysis for the more general system where the magnetic sheaths are insulated from the wires. For the special case where wire and sheath are contiguous, 712" is infinite and (28) shows that Fi = 7 and r2 = «> . The transmission is, therefore, defined by only one mode of propagation. The series impedance of the system is, from (23) and (26), Z - 2 |^Z22" - z,- - Zo/ + Z22' J + ^'' (36) where the terms in brackets give the internal impedance of one of the loaded wires, and X2 is the reactance that arises from the magnetic field between them. The internal impedance can be obtained also by 2£./' finding -j — f-f- directlv from the last two of equations (1), of course. l\ ~T Li 204 BELL SYSTEM TECHNICAL JOURNAL The constant K-2. becomes — 1 and the total current, / = /i + I2, is propagated in accordance with / = Iq€~'^^, where 7 — -sZY-y. The constant Ki, which is the ratio of the current in the sheath to tliat in the wire, is of interest. It becomes Jo T, /^ ]•> — i^oo ^]] — ioo] /2*"^ T-=-f^i = ^—f = ^—f (^0 1 I Z/22 ■'-'■22 The approximate formulas for the case where wire and sheath are contiguous are derived as follows: The arguments, xo and ,V2, of the Bessel functions differ by only a small amount when the magnetic sheath is thin. This situation is favorable to an advantageous use of Taylor's series. Joixn), for example, can be expressed in terms of Joiy^), its derivatives and the difference of the arguments in a Taylor series as follows: , /o(.r) = My) + rJo'iy) + ^j /o"(v) + fi M"(y) + • • • > (38) where r = .v — _v (x-i, y2 being written simply, x, v, here, for con- venience). Furthermore, Bessel functions are subject to recurrence formulas,^ which enable us to express each of the derivatives occurring in the series in terms of the function of zero order, its first derivative and the argument. Therefore, by applying the recurrence formulas to the Taylor series, we find functions U and V (see Appendix B) such that Mx) = UMy) + VJo'(y), (39) Ko{x) = UKoiy) + VKo'(y) (40) {U2, V2 being also written now, U, V). Differentiating (39) and (40) with respect to t, Jo'(x) = U'My) + V'Miy), (41) K,'(x) = U'Ko(y) + V'Ko'(y), (42) 3t7 , dV where U = — — > V = -z— • OT OT ''The two recurrence formulas required are: J,/{Z) = ^' /(=) - /„ + ,(3), J,/{z) = /„_,(=) - "/„(=). l*lie Bessel Functions ut" the second kind satisfy the same furmuius. WAVE PROPAGATIUX Ol'ER CONTINUOUSLY LOADED WIRES 205 If (39) to (42) be solved for U, V, U', V, it can be verified that the solutions are the definitions of these functions already given in equa- tions (10).*^ The exact formula for the internal impedance of a wire with con- tiguous sheath has been given in (36). In terms of the functions U and V, this formula becomes By using the series for these functions and discarding all terms of degree higher than w", the approximation given in the body of the paper (equation 5) may be obtained. APPENDIX B \^ hen the recurrence formulas are applied to the Taylor series, it is found that T- r'' rW 3 \ t'" / 2 1 2 \ ^='+2+6T. + 24('-/)-T2o(,v-7) ^ 2v 6V V-/ 24Vv v^ , rW . 7 , 24\ T« /3 X^ , 120\ , ,,., These series converge for < 1, which condition is satisfied by the sheath dimensions of any practical continuously loaded conductor. A considerable number of the terms in the series for U and F are ^ A relation that can be used to advantage at times is V h U'V - uv = -■- = --. X a This relation corresponds to the similar one for the Bessel functions themselves, namely: J„'{z)KJz-) - J„{z)K„'(z) =1. z 14 206 BELL SYSTEM TECHNICAL JOURNAL parts of well-known series that define certain elementary functions. It can be verified readily that U = cos T -\- '— F=sin.+,[log(l+^)-^]+^ (46) + TT7fl+ •••. (47) i2oy 2403; ' 24oy U' = — sin T + ^ 1- -— (above remainder of (3)), (48) T V d V = COS T 1- T- (above remainder of (4)). (49) 1+- ^' y The series (46) to (49) possess a certain advantage for computing in that the quantities in brackets are real numbers. ( Note that - = ^^-7 • \ They have been used also in obtaining the approxi- mate formulas given in the body of the paper. The quantities discussed above all pertain to the sheath. For finding Z//, involved in the last of the formulas (9), the series are not valid, of course. For this we have the well-knowii series, 1 /o(.rO 1 U,' Jo'{xO .Vi ^4 ^96 1536 ^ 23040 (50) — see, e.g.. Gray, Mathews and McRoberts, " Bessel Functions,' 2d edition, p. 170. Theory of Vibration of the Larynx^ By R. L. WEGEL The vibration in the larynx is caused by an automatic modulation by the vocal cords of the air stream from the lungs. Analytically the mechan- ism is the same, and physically, closely analogous to that of the vacuum tube oscillator. It depends principally on the resonance of the vocal cords, the modulation of air friction in the glottis by their motion and the attraction due to constriction of the air stream between them. When these forces exist in certain relative proportions and phases, sustained oscillation as in singing takes place. The whole mechanism may be rep- resented analytically b>' force equations, from which conditions for accre- tion or subsidence of the vibration or for sustained oscillation may be easily deduced. These equations also show the analogy with other types of oscillating systems. IT is customary in treating the theory of the voice to assume the glottis or space between the vocal cords to be a source of a steady stream of air with superimposed periodic impulses caused by the vibration of the vocal cords. The harmonic content of these impulses is modified by the "resonating" vocal cavities before being radiated into free air. It is the nature of this modification which receives most attention. The mechanism by which the vibration of the vocal cords is maintained has not been carefully studied. The vocal cords are maintained in a state of sustained vibration by the proper balance between the various mechanical constants of the complete system, which thus act as a transformer of a part of the non-vibratory power derived from the air stream from the lungs into the vibratory power resulting in sound. It is a simple theory of this mechanism which is considered here. The method used is to obtain the force equations, which describe the vibrations of the complete mechanical system, by means of the Lagrange equations, from expressions of the total instantaneous kinetic and potential energies, the instantaneous forces acting and rate of dissipation of energy. The resulting simultaneous equations relating to the displacements and velocities of the various parts are then studied to find the frequencies of free vibration and the relations which must obtain between the various mechanical parameters of the system in order that one of these frequencies be sustained. The method is an application of the theory of H. W. Nichols, published in Physical Revinv, August, 1917. The theory is reduced to easily workable form by the introduction of simplifying approximations which will be described in the progress 1 Presented before Acoustical Society of America, May 11, 1929. 207 208 BELL SYSTEM TECHNICAL JOCK SAL of the discussion. The principal one of these is the neglecting of all reactions of second or higher order, thus leaving a set of linear differential equations. Structure of the Vocal Tract The vocal tract consists of three principal parts, the lungs and associated respiratory muscles for maintaining a flow of air, the EPIGLOTTIS FALSE VOCAL CORDS GLOTTIS VOCAL CORDS TRACHEA Fig. 1 — Anterior-Posterior Section of the Larynx. larynx (see Fig. 1) for producing the periodic modulation and the upper vocal cavities, pharynx, mouth and nose for varying the rela- tive harmonic content of sound originating in the larynx. THEORY OF VIBRATION OF THE LARYNX 209 The capacity of the lungs in an adult man is capable of being varied from about two to five liters. The av'erage in quiet breathing is about 2.6 liters. The average expiration of air in quiet breathing is about .5 liter. The rate of expiration of air in medium loud singing varies from 40 to 200 cm. ^ sec, the lower values obtaining for trained singers. The larynx (see Fig. 1) consists of an irregularly shaped cartilaginous box at the top end of a tube, the trachea, about 12 cm. long by 2 cm. in diameter, leading from the lungs. The larynx contains the vocal cords, a pair of fibrous lips which in vibrating vary the width of the slit called the glottis, between them. The length of the glottis in the adult male averages about 1.8 cm. and in the female 1.2 cm. The width of the glottis varies widely with differing sounds. A few tenths of a millimeter may be considered representative. The tension and separation of the vocal cords are controlled by muscles. The principal upper vocal cavities are the pharynx, a space just over the larynx, the mouth and the nasal cavities. The first and second may be varied in size and shape at will, but the effect on the last is controlled only by varying the communicating aperture be- tween it and the pharynx. Equations of Motiox of the Larynx Fig. 2 shows a cross-section of a model which illustrates the essential details of the larynx in so far as it is necessary for this treat- ment, ^o represents the area of the opening to the trachea. The vocal cords are represented by elastically hinged members of com- bined effective area S-^. By effective area is meant the area of aper- ture which displaces the same Aolume of air as the vocal cords when it moves the distance 50 of the tips of the cords. This area is less than that of the vocal cords. The tips of the vocal cords are separated to form a gap, the glottis, of area Su A positive or up and outward displacement q-i of the vocal cords increases S^. It will be assumed that the air is not appreciably compressed in the neighborhood of the glottis, that is, any tendency to compression is relieved by flow into the trachea or pharynx. The pressure in the lungs forces a steady current of air through the glottis. Let the velocity in the trachea of this steady flow be /o and in the glottis /]. Small vibrations of the vocal cords superimpose additional small velocities, Zo and /i, in the trachea and glottis re- spectively. If the instantaneous velocity of the vocal cords be u and it be assumed that they are constrained to move in synchronism (/o + H)S, = {h + h)S, -f HS2. (1) 210 BELL SYSTEM TECHNICAL JOURNAL The above material is a description of a simple model of two degrees of freedom which simulates the principal characteristics from the standpoint of performance of the more complex larynx which has many degrees of freedom. It is this idealized model which will be considered in the subsequent treatment. Such points of performance of the actual larynx which may be due to the action of ignored and Fig. 2 — Schematic Larynx Model. presumably subsidiary modes of motion will, of course, not be pre- dicted by the theory. These are assumed to be of minor importance. The possible independence of motion of the two vocal cords will be considered later, however. The contraction of the air stream at the glottis introduces a rela- tively large concentrated kinetic energy in the air stream at this point similar to that at the mouth of a Helmholtz resonator. The inertia of a small plug of air between the vocal cords may then to a first approximation be treated as a mass L\. A concentration of frictional resistance also occurs at this point due to viscosity and to turbulence. A positive displacement q-z (outward) of the vocal cords causes an increase in the mass of the plug of air in the glottis and a change in the effective resistance, R, encountered by it. The inertia Z-/ and resistance R of the glottis are therefore both functions of q-i, the dis- placement of the vocal cords from a mean position, and of the width of the glottis. If further Q^ represent the average displacement of THEORY OF VIBRATION OF THE LARYNX 211 the vocal cords from an appropriately chosen reference position, L^ their inertia coefficient and K^ their effective stiffness, all measured at the tips, the total kinetic energy, T, and potential energy, V, of the larynx are T = hL,'(A + hy + H2^2^ (2) V = ^K,(Q2 + 92)-. (3) The Lagrange equation of forces for the nth coordinate of any system is _ddT_dT dV dtdtn dqn dqn, in which Fn is a reaction due to friction. The force equations for the glottis and vocal cords therefore become 17 P I ^ ^ dt dti hU'ih + hY (5) ^ = ^^^ + ^^ + ^^^^^ + ^^^ 2 ^ • ^^^ Nature of the "Constants" of the System It is quite safe to conclude that none of the coefficients (inertia, dissipation and stiffness) of the larynx are sensibly constant over the range of operation of the coordinates. Direct measurements are evidently impossible. It is conceivable that they may be arrived at indirectly by means of a comparison of experimental data, especially taken for the purpose, on voice curves and the results of dynamic analysis of the kind described here. The problem may also be studied by means of models. In order to solve equations 5 and 6 it is, how- ever, necessary to evaluate the space and velocity derivatives. A few simple experiments were performed on models for the sole purpose of determining the qualitative nature of variation of resistance of the glottis with displacement of the vocal cords. A diagram of the model used in the measurements is shown in Fig. 3. This con- sists of a brass tube, a, ^i" in diameter, beveled off on the top at an angle of 45° with the axis, and two 3 s" brass plates, h, fitted on these beveled surfaces so as to leave a slit, S, which was made adjustable in width. A cross-section of this model is shown in c. The bottom of the tube was attached to a large air chamber in which the pressure and velocity of air flow could be regulated and measured. Three shapes of "glottis" were measured. The first had square corners, as shown on Fig. 3f. The second, M, was the same as ic, 212 BELL SYSTEM TECHNICAL JOURNAL except that the corners of the Hps were rounded. The third, Fig. 3>e, had square corners as before, but the sUt was about .1 mm. wider in the middle than at the ends. Fig. 3— Glottis Models. The resistance R is given as the ratio of the product of pressure and slit area to the linear velocity of flow. Measurements were made in each case through a range of pressures such as to give fluxes through the slit through a range of 50 to 200 cm.^/sec. (Stanley and Sheldon values, see Sci. Am., Dec. 1924) and through a range of slit width W of .01 to .10 cm. The data can be represented approximately in this range for the three slits by the following formula?: i? = 3.6 PW'-' X 10-«, i? = 6.1 PW- X 10-«, R = 800 /-n^--3 X io-«. In these expressions / is the velocity of flow of air through the slit. More careful data taken through a wider range of / and W would undoubtedly have given i? in a power series. These formula^ are taken to indicate that the resistance of the actual glottis increases faster than a linear function of / and W due to turbulence and may be represented as a single valued function of either displacement of the vocal cords q-i (or glottis width) or of air velocity as expressed by a Taylor's series as follows: R = R, , d,R dq-i , d,R . + 1 d^R ., , 2d,fR . dqi" dqodti + do'R ~d7r ir + etc. (7) THEORY OF VIBRATION OF THE LARYNX 213 In this expression Rq is the resistance measured in the reference position at which point the derivatives are taken, where ii and q- are zero. The experiment mentioned above determines the signs ot the coefficients of q-i and i\ as positive. If the flow were purely lami- nar, i.e. due to viscosity only, the first would be negative and the second zero. The Reaction F^ By definition, £o = RJi, where £o is the force of the lung pressure on the glottis and /i a corresponding linear velocity of fiow of air. If a force Fi slightly greater than Eq act on the glottis and result in a velocity I = h -\- i], zh, = ^- («' A combination of (7) and (8) constitutes an evaluation of Fi for sub- stitution in the force equation (5). To a first order approximation then : /^, =J?,/, + (i?. + /,^)/, + /,^^,.. (9) The coefficient of g-^ is dimensionally a stiffness and that of /'i a re- sistance. In what follows they will be denoted by F, = RJ, + Rrii + K„q.. (10) Glottis Mass (L/) Reactions The kinetic energy of the air stream being proportional to the volume integral of the square of the velocity is largely concentrated in the glottis on account of the relatively high velocity at this point. On account of the irregularity in shape and turbulence in the stream it is impracticable to attempt an integration. If the velocity were so small that the turbulence were absent an approximate value of the air mass would be obtained by taking the mass of a cylinder of air having the length of the slit and a diameter equal to its width. This would make the mass L/ proportional to W~, or since the width is proportional to displacement of the vocal cords, to q-i'. Owing, how- ever, to turbulence and other non-linearities, the mass is probably more nearly described as a tongue of air issuing from the glottis, the inertia L/ of which varies as some power function of the width and also of the velocity. - - It has been found since experimentally that the mass reaction is very nearly that of a cylinder as described but reduced somewhat in diameter due to viscous or turbulent drag at the tips of tiie "socal cords. 21-1 BELL SYSTEM TECHNICAL JOURNAL It might be seen by carrying through an expression for this glottis mass involving a function of velocity similar to that for R of equa- tion (7) that only a quantitative change in effective mass would result in the final equations and that no new type of reaction would be introduced. This demonstration is not included here. In order to save space in this qualitative treatment it is ignored. For small displacements go from a reference position at which the velocity of the air is /o, the glottis inertia may be represented by the direct function: Ly = Li +-3— ^2 + ^^-^^2- + etc., (11) in which the coefficient of 92 is obviously positive. The second term of the second member of (5) may now be evaluated by performing the differentiations as indicated. Neglecting second and higher order terms and denoting dq^/dt by /n the reaction in question becomes The glottis mass of air, therefore, introduces two kinds of reactions: a simple inertia and a reaction proportional to the velocity of the vocal cords. For simplicity of notation (12) will be written Uf^ + Gi,. (13) This completes the evaluation of the terms (5), the force equation of the glottis, which may now be written £0 == Roh + RJ, + K,.q, + ^ + Gu. (14) Force Equation' of the \'ocal Cords The force equation (6) of the vocal cords contains four terms. The first is the inertia reactance of the vibrating lips. The mass Lo is the effective vibrating mass which, if multiplied by one-half the square of the velocity at the cord tip, gives the kinetic energy of their motion. If the distribution of the velocity in the vocal cords were known this might be found by integration. The second term Fo in equa- tion (6) represents the internal dissipation and is assumed propor- tional to the small velocity u. The third term is the elastic reaction which is proportional to displacement. The fourth term is a " gyrostatic " term. This term ma>- be written as follows: THEORY OF VIBRATION OF THE LARYNX 215 - T- = - HA + h)- -^ — h -^-^ g^ + etc. • (15) dq2 ^ dq-i dq-r ) Again by neglecting second and higher order effects this reaction becomes It will be seen that the first term of this expression represents a static force tending, since it is negative, to draw the vocal cords together. This is the BernoulU effect utilized in a venturi meter. This steady force is counterbalanced by an elastic reaction of the vocal cords with which it combines to determine an equilibrium position which obtains when the cords are not vibrating. This term may, therefore, be dropped from the fin'al equations representing only superimposed motions. The coefficient of i] is identical, except for a sign, with G of (13). It represents a force on the vocal cords due to a superimposed part of the Bernoulli effect caused by the small superimposed velocity ix in the glottis. The coefficient of qo is dimensionally a stiffness. This apparent stiffness is due to the nature of the air flow and is inde- pendent of any elastic members. It is negative if the second differ- ential of glottis mass with respect to cord displacement is negative, positive when this coefficient is positive and vanishes when this coefficient is zero. It simply adds or subtracts in effect from the stiffness K-i of the vocal cords. The first possibility is the more likely.^ These terms may then be written for simplicity -^= - F - Gh - K^q,. (17) oq-i Force Equations of the Larynx The force equations of the glottis and vocal cords with constants thus evaluated are E, = Lr^ + R,H + Gk + R,U + /v„?o, (18) 0 = Lo^ + R^u + A>, + K.Q. - F - Gi, - K^q.,. (19) As explained before, Eq = i^o^i and F = KoQn; so these cancel and are of no interest here. In the following it will be seen that the field ^ This coefficient has since been found to be negative. 216 BELL SYSTEM TECHNICAL JOURNAL stiffness Kj. is included in K^ to simplif\' notation. This leaves (18) and (19) finally: 0 = L, ^ + R,H + Gu + K.,q,, (20) 0 = Lo^ + Roi. + K.q. - Gi\. (21) It should be noted that these equations represent all first order internal reactions of the idealized model of the larynx. The series expansions have been carried out, to show to what approximations these equations hold. It should also be pointed out that the effects of mechanical hysteresis of the parts, which make the relative posi- tions of the parts dependent on the previous history of their motion, have not been considered. A consideration of hysteresis complicates the theory considerably and is ignored for the same reason and with the same justification and limitations that it is ignored in the ele- mentary treatment of electrical circuits containing coils with magnetic material and condensers with electrostatic hysteresis. External Reactions of the Trachea and \'ocal Cavith^s on THE Larynx So far the modifying effect of the trachea and lungs, as well as the upper vocal cavities, on the motion have not been considered. Before using the equations it is necessary to evaluate these reactions and add them in their proper places. Imagine a weightless piston fitted into the trachea just below the vocal cords such that the volume of air thus enclosed in the larynx is so small in comparison to that of the trachea and lungs that its compressibility may be neglected. If the vocal cords are held rigid and the plug or piston of air in the glottis is forced inward, a reaction in addition to the resistance and inertia of the glottis will be encoun- tered due to the impeding effect of the trachea piston, which impe- dance is determined by the constants of the lower chambers. If a small force /o act on the trachea, causing a small velocity, to, and we assume linearity of response /o = Zoio where Zo is a constant which may, due to a positive inertia reactance or a stiffness, contributed by air compression in the lungs, involve either a time derivative or integral of displacement. For the present consider it to be a gener- alized impedance operator. Due to the relative incompressibility of the air in the larynx, the volume displaced by the trachea piston is i^So = iiS]. Since also the instantaneous pressure inside the larynx THEORY OF VIBRATION OF THE LARYNX 217 is constant on all its walls, including the surface of the trachea piston f^/So = f.'Si. We then have f\ = zMn. (22) This reaction due to the trachea must be added to those of the glottis given in (20). In like manner if the effective area of the vocal cords is 52 a reaction h must be added to their force equation h = zMi,. (23) •Jo" Due to the steady component of air flow there is a static component of pressure tending to force the cords outward. This is counter to the static Bernoulli term and again, if second order effects of small quantities be neglected, serves only to alter the equilibrium position and may therefore be disregarded here. When the glottis plug of air is displaced inward a force is exerted on the vocal cords tending to move them outward which is relieved to a certain extent by a yield of the trachea piston. This force on the vocal cords may be shown by reasoning similar to that above to be Zo%i-^\. (24) Since this part of the system is linear, the reaction between glottis and vocal cords through this channel is reciprocal so a force is exerted on the glottis when the vocal cords are displaced of Zo%|i2. (25) It will be noticed that ^i is a variable because of the variation in width of the glottis while vibrating. The effect of this variation in these terms is obviously second order since ii is small and will therefore be neglected. The reactions of the upper cavities might be similarly added, but they are apparently relatively small and since they are at present not quantitatively known, are disregarded in the general equations be- cause of the increased complexity. Generally, however, Zo may be thought of as representing the additive effects of both upper and lower chambers. The complete force equations of the voice for small vibrations, 2 IS BELL SYSTEM TECHNICAL JOi'RNAL taking into account all major external as well as internal reactions, mav then be written: 0 dix 'dt A^ + R,H + Gh + K,^q,+ ^"'^'" -• ' ""^^ ^'o^ ZqSiSo ^1 H rr-TT— ^2 6'o^ di ■'^n^j'i" • , JunOiOt . i) = Lo-f-\- R2i2 + A'og. - Gi\ + -^~- k + t,. (26) (27) These equations may be put in a somewhat simpler form by virtue of the fact that they are linear differential equations with constant coefficients. In such a case the time differential may be replaced by an algebraic operator p such that i = pq, di/dt = p^q, where p is of the dimensions and nature of a frequency 0 = ifU + pR: + pZof,]qx +(pG + pZ,^+K„^q,, (28) 0 = (- pG + pZo^ jq, + ( p-U + pR-i + K. + />Zo|^; The determinant of this system is (calling pZ^ = Fo) (29) D p'L, + pR, + Yo-^^pG + Yo^ + A'„ - pG+ Yn^)U-'L. + PR, + Ao + F„|^n (.SO) This determinant represents the complete reactions of the larynx and the external effects of communicating air chambers. If the effects of the air chambers be disregarded the system is represented by placing Fo == 0, giving the simple form D (p'U + pRi) - pG ipG + K„) ip-Lo + pR.2 + K.) (31) Nature of This System The voice system represented by determinant (3) is very closely analogous to other vibrators, such as the microphone oscillator or door buzzer and the vacuum tube oscillator. The literature on the latter subject is now so extensive that the pointing out of the analogy should make the method of solution for sustained oscillation, as in singing, or for subsidence or accretion of the oscillation, as in speak- ing, clear to any one familiar with it. THEORY OF VIBRATION OF THE LARYNX 219 Fig. 4a is a schematic diagram of a three-element vacuum tube oscillator circuit known as the "tuned grid" circuit. This is one of many kinds. The transformer coupling between the plate and grid circuit is represented by an auto-transformer. Fig. 4b represents the same circuit schematically but with circuit elements only. In this R2 represents that part of the resistance of the coil which belongs to the grid circuit and any other associated dissipation, L2 is the in- ductance of the coil as seen from the grid mesh and Ko the reciprocal of the combined tuning capacity across the grid and that of the grid- filament. It is the electrical stiffness or elasticity of the grid mesh, K, ^ :^2 AAAAM- nmr^ Fig. 4—" Tuned Grid " Oscillator. in other words. Ri is a plate-filament resistance ("a.c") and Lj that part of the coil in the plate circuit. M is the mutual inductance of the transformer which is not part of the mesh impedance of either plate or grid. The element K,, is the " uni-lateral mutual impedance " (G. A. Campbell, 1914) between the plate and grid meshes and is numerically equal to (jlK^, where ^ is the amplification constant of the tube. Other internal tube impedances are as usual neglected. The impedance determinant may be written directly from the circuit diagram Fig. 3b. {p'-L, + pR,) {^M + K„) pHI ip-Lo + pR, + K.2) D (32) The quantities on the principal diagonal of this determinant, that is the first and last elements, are as usual in a circuit determinant the mesh impedances while the others are the mutuals. The principal features of the analogy may be seen by comparison of determinants (31) and (32). Except for the thus far undefined external or trachea impedance the mesh impedances are the same, from which it appears that the glottis is analogous with the plate-filament path in the vac- uum tube and the vocal cords with the grid-filament path. The air 220 BELL SYSTEM TECHNICAL JUl'RXAL velocity in the glottis 7i corresponds to the plate current. In the vacuum tube this plate current is modulated by varying charge, q-i, on the grid. In the larynx the glottis air velocity is modulated by varying displacement, q2, of the vocal cords. The charge on the plate (again neglecting internal capacities except the grid-filament) causes no effect on the grid mesh and in the larynx the position of any element of glottis air has no effect on the vocal cords. The uni- lateral mutual impedance, K,,, is the same in both. The analogy breaks down at the point where the " feed back " part of the mechanisms is compared. The " feed back" is the bilateral part of the mutual impedance between the two meshes. In the vacuum tube circuit this is p-M, the mutual of the transformer, while in the larynx it is pG, the " gyrostatic " mutual. The latter is a type of element which does not occur in electrical circuits, arising as it ^im(y^ mw ^^AAJW — I ^im^ ■^ww a b Fig. 5 — Tuned Grid and Wind Reed Circuits. does from a variation of a mass or inductance with a displacement. Inductance, being a function purely of the geometry of a circuit, can only vary with mechanical displacement and not with electrical displacement or charge. The gyrostatic mutual is common in the mechanics of rotating bodies whence it derives its name. It is also the mutual in an electromagnetic telephone receiver or relay be- tween the electrical circuit and the armature or diaphragm. In order to fix the rather useful concept of the analogy in mind. Fig. 5 is added showing the schematic circuit of the vacuum tube (5a) and a circuit diagram (5b), which represents determinant (31) the characteristic formulation of the dynamics of the larynx. Fig. 5b is represented by the conventions of an electrical circuit, except for the element G for which a different convention is necessary. The one taken here is that of a resistance enclosed in a rectangle. From (31) it will be seen to be similar to a resistance in its association with frequency p but different from resistance in that it occurs non-sym- metrically in sign in the determinant. It does not involve dissipation. THEORY OF VIBRATION OF THE LARYNX 221 Its occurrence here is the simplest possible for when there are appre- ciable concealed or ignored modes of motion it may have the form of a generalized impedance containing at least one element of resist- ance, but will always be non-symmetrical as a whole in sign in the determinant. The use of the circuit for representing the mechanical system is an extension of an old but recently popularized method of studying mechanical or electrical vibrating systems by the help of analogy, one with the other. The extension consists in the explicit representa- tion by diagram of the gyrostatic mutual which makes the deter- minant unsymmetrical in sign and of the unilateral mutual which makes the determinant unsymmetrical in magnitude. Fig. 6 is a K. ■m^5^ AAAAA ^m^ AAAAA- Fig. 6 — General Wind Reed Circuit. diagrammatic representation of the more general system of deter- minant (30). This includes the external Yq reactions as well as the internal. Having thus described the extended method of analogy the follow- ing study of the larynx with the help of the circuit diagram of its determinant should be clear. Sustained Vibration of the Simple Larynx In vibrating, the vocal cords do not receive excitation of the fre- quency at which they vibrate. The source of power is in the air stream /i which enters the equations in iv",,, the unilateral mutual impedance. Since this is treated as a constant circuit or dynamical element this air stream may be ignored as a drive and the resulting 15 222 BELL SYSTEM TECHNICAL JOURNAL vibration considered as the free oscillation of the system. The de- terminant (31) (or 30) is then used to determine the free frequencies and decrements of the system. The method is as usual to solve for p in the equation D = {). {32,) To simplify the demonstration the simple larynx without the load of the air chambers will be considered. Taking D of (31) then and expanding: ^^L,Lo + p\L,R, + L.Ri) + p\UK. + R,R. + G^) + p{R,K, + GK^,) = 0. (34) If this be divided by Z1L2 and the uncoupled decrements and natural frequency defined : -^1; ;77- = Ao; -7- = coo-, (35) = 0. (36) 2.Li\ 2.L11 Li then p"- + p\l^, + 2A2) ^ pi coo- + 4A,A2 + -^ ) One of the roots of this equation is zero and another is negative real since all coefficients are positive. This root is therefore the decre- ment of a mode of non-vibratory motion. The remaining two roots may be real, imaginary or generally complex, of the form Aija,. (37) If it is found that A = 0, then an oscillation once started will be sustained. If A be negative then any existing oscillation must sub- side or if A be found positive then an impulse will start an oscillation which of itself increases in amplitude to a point where its violence modifies the constants to such an extent as to make A vanish, leaving a sustained oscillation, or negative leaving the oscillation to subside to a lower amplitude or completely if sufficient permanent changes have been made. If now (36) be written Ap-' + Bp~ + Cp + D ^ Q (38) and the tirst root (37) be substituted, two equations result, one from the real and the other from the imaginary terms, as follows: ^A(A- - 3co-) + 5(A- - CO-) + CA + /^ = 0, (39) ^(3A- - CO-) + B{2^) + C = 0. (40) THEORY OF VIBRATION OF THE LARYNX 223 Now the condition for sustained oscillation is that A = 0 and if the value of o) when this obtains be wo then D C coo- = -g and '^°^""J' (^^) or the condition for sustained vibration in terms of the constants is AD = BC. (42) In addition to this if use be made of the fact that in an algebraic equation such as (36) the coefficient of p- is the negative sum of all the roots then this coefficient is the real root. Let this be Ao and then (36) may be written p^ + p'-^, + ;^coo- + Aocoo- = 0. (43) The coefficient of p in (36) is therefore the square of radial frequency at which sustained oscillation will take place and this is seen to be higher than the natural frequency wa of the vocal cords, the difference being increased when the damping of either mesh is greater or when the coupling mutual G is greater. It might be noted in passing that (43) is the free oscillation equation for any system which may be represented by a cubic equation and is not confined to the simple larynx. Such an equation always results when there is only one kind of reactive element in one of the meshes. It holds also for the tuned grid circuit. The condition for sustained oscillation to be fulfilled for the con- stants may from (42) be reduced to: R,R, = G'[^- ij- (44) It is rather difficult to place a simple physical interpretation on this formula. The qualitative import of it may however be seen by substituting the values of G and Ku from (13) and (10): R\R'i = Li' I]' doR/dqo doLi/dqo doLi/dqo ,.. The first term in brackets is in the nature of a resistance modula- tion constant, a fractional change in glottis resistance per unit dis- placement of the cords, to be designated by r and the second term similarly a glottis mass modulation constant, /. The quantity Lil^ is the momentum of the air in the glottis. This equation is then i?ii?2 = {UU)\r - /)/. (46) 224 BELL SYSTEM TECHNICAL JOURNAL Thus it appears that the resistance modulation must always be greater than the mass modulation and when the difference is small the air momentum must be increased to compensate. Owing to the physical limitation in accuracy of continuous maintenance of adjustment in the larynx, if a large momentum is depended upon to compensate for a small modulation difference, an unsteadiness or instability is likely to result. It is common experience that it is impossible to produce a sound with the voice with less than a certain minimum intensity. This corresponds, with the most favorable adjustment of the modu- lation constants which are physically possible, to a minimum momen- tum of air from the lungs which satisfies (46). It will be evident that this interpretation must not be taken too seriously quanti- tatively. Subsidence and Accretion of Vibration of the Simple Larynx Oscillograms made of the speaking voice show that, among other things, the amplitude of the oscillation and the pitch are in a con- tinuous state of change. This is also true in singing but not nearly to the same extent. It seems therefore that in singing the adjust- ment of the voice system for sustained oscillation as described in (44) above is of major importance, while in speaking conditions for variation are of most importance. The principle of the investigation of variation is simple enough but in all but the most elementary systems the algebra involved is impracticably awkward. If by solving {ii) directly for the roots of p, it be found that A is positive, then any existing vibration will tend to increase while if A is negative, then vibration will tend to subside. The algebraic difficulties arise in the general solution but these are largely obviated by making the assumption, which is most likely usually fulfilled in practice, that the real parts of the roots may be treated as small quantities when compared with the imaginary parts. A common frequency for a man's voice is 150 cycles per second for which coo is 1000 in round numbers. The decrement of a telephone receiver is ordinarily 100 to 200 in open air. The decrement of a tuning fork is represented by a fraction. Judging from variations in amplitude in an oscillogram (from which of course decrements may not be read directly) it would seem reasonable to assume that A is small compared with wo- The study of variation thus becomes an investigation of small departures from a condition of sustained oscil- lation, the reference condition being that critical adjustment for which the roots of interest of {ii) are pure imaginary. THEORY OF VIBRATION OF THE LARYNX 225 Suppose in (38) that A = \; then without loss in generahty: pz ^ Bp"' + Cp + D = 0. (47) In such an equation the roots are continuous functions of the co- efficients. The same is true of their derivatives except at the one point where transition occurs from pure real to complex values. The values of the roots of interest in this connection are in their complex region at the point where the real part of the root passes through a zero value. This is the point at which free oscillation of the oscil- lating mode occurs, the values of the roots of this mode being as shown before, ± jwo. If it now be supposed that one cause produces small variations, directly or indirectly on each of the coefficients and that the magni- tude of this cause be .r, then : (3,= + 2., + C)g + .= f + .f + f = 0. (48) The problem then is to determine dp resulting from any assigned cause dx when p = jooo- From (43) we have at this point B = Aq, C = coo" and D = Aqcoo^. .Ao\ dp ,dB , . dC , dD This is the frequency (complex) variation equation taken in the neighborhood of free oscillation. When any readjustment of the larynx takes place all of the "con- stants" entering the coefficients undergo change, in particular those of the glottis Ku, Ri, G. Suppose for simplicity that one only varies, then this variation dKu, dRi, or dG may be taken as the magnitude of the cause dx. In particular if Ku vary, dB = 0 = dC and dD/dx = G/L^L^, .Ao\ dp G coo / dKu IwrfLiLo If in addition Ao be small compared with coo, . GdKu /, , .Ao\ dp = -> or J- 1 +J— • (.-^1) This shows that if a condition of sustained oscillation is departed from by slightly increasing Ku, an increase in the amplitude of vibra- tion begins which is proportional to the logarithm, since {p -\- dp) 226 BELL SYSTEM TECHNICAL JOURNAL is the exponent, of the increment dKu and the frequency (imaginary part) of vibration increases sHghtly in proportion. If K,, were the only varying element the vibration would continue indefinitely to increase. If on the other hand K„ be assumed constant, the variation being in Ri, then it may be similarly shown that dp =1— T, - ( 4A, + -^ + Ao-^ ) + /""-^'^ Li\Li2 I COq (52) whence it appears that a small increase in glottis resistance dR\ (or (/Ai) introduces a subsidence of vibration but an increase in frequency of oscillation as before. A decrease — dR^ of course produces the opposite efifect. If the change be in G, it turns out that dG dp = ZcOo'LiyLiO A^-2GAo ) + /•( ^^^ + 2couG Wo (53) Here it appears that an increase in the gyrostatic mutual, G, may introduce either a subsidence or an accretion in amplitude but like the others makes for an increase in frequency of oscillation. Variation in other elements produces similar conflicting tendencies not only in damping but in frequency. The physical picture to be drawn from this is that in speaking the voice modulates from one amplitude and frequency to another by proper relative variations in adjustments in its constants, being con- stantly in a state of changing subsidence or accretion. It would seem also that the principal cause of change in frequency is in the vocal cords and that of amplitude variation in the glottis. Speaking is, in this respect, a more intricate process than singing. Other Types of " Feed Back" The detailed study of the larynx has so far been limited to the assumption that the " feed back" is entirely gyrostatic. This is of course actually not the case. How much influence is exerted by the general Yq is difficult to estimate. If the trachea were a long tube but still shorter than a quarter wave-length of sound at the frequency of oscillation and rather smaller in diameter, and substantially open at the end the mass of the air in it would then be appreciable and Fo in (30) would be written p''Lo. If in addition the gyrostatic term were negligible the system would then be exactly analogous with the tuned grid system and (32) rather than (31) should be the subject of detailed study. THEORY OF VIBRATION OF THE LARYNX 111 If on the other hand the lungs acted substantially as a solid walled chamber of comparatively small size, the elasticity of the contained air would be represented by taking K^ for Fo. The surface area in the lungs is very large compared with a regular chamber of equal volume so considerable dissipation must be encountered by vibration. If this were the most important reaction Fo should have been replaced by pR^. Unquestionably all three types of reaction enter. A more general treatment to include them is plainly not a subject for a short paper. It is interesting however to note that in the dynamical system of brass horns these latter Fo reactions exert controlling influences. In this case the lips of the player perform the same function as do the vocal cords of the voice while the external load, the horn, corresponds to the pharynx, the reaction of which is the same dynamically as the trachea. In this case the frequency of the horn is that of sustained oscillation and not that of the lips. The same is true of the wood- wind, in which case the reed or reeds replace the lips or vocal cords. In these cases Fo is proportional inversely to the hyperbolic tangent of the frequency or may be approximately represented by the im- pedance of an anti-resonant element. Abstracts of Technical Articles From Bell System Sources Notes on the Effect of Solar Disturbances on Transatla?itic Radio Transmission.^ Clifford N. Anderson. In 1923 when the relation between abnormal long-wave radio transmission and solar disturbances was first noted, the outstanding abnormality was the great decrease in night time signal field strength accompanying storms in the earth's magnetic field. There was a slight increase in daylight signal field but this was distinctly secondary to the efi^ect upon night field. Previous to 1927, data on signal fields were limited to one set of measurements a week, and although daylight signal field strengths were higher during periods of increased magnetic activity, it was somewhat difficult to determine the efi"ect of individual storms. The present notes show the efl^ects of individual storms of 60-kc transatlantic radio transmission and also give some indication as to their efi'ect on short-wave radio transmission. The Mutual Impedance Between Adjacent Antennas} Carl R. Englund and Arthur B. Crawford. The simple theory for the computation of reflecting or multibranch antenna systems is sketched. If the points at which observations of electrical quantities are to be made are definitely specified, a knowledge of the self and mutual impedances (properly defined) between antennas is sufficient to make the computations determinate. Of the circuit constants, the most use- ful and accessible is the antenna current ratio and in the work here reported 0 has been measured in the range 0.33 X to 1 X. Experiment has shown that in this range 0 is that theoretically calculable for a Hertzian doublet. Actually this range is equivalent to X/3 to 00 . The discussion of experimental procedure is purposely thorough. An Experimental Method for the Determination of the Ballistic De- magnetization Factor.''^ Donald Foster. A method is described for experimentally determining the ballistic demagnetization factor. By means of a double search coil ot novel design the magnetization and 1 Proceedings of the Institute of Radio Engineers, September, 1929. - Proceedings of the Institute of Radio Engineers, August, 1929. ^Philosophical Magazine, September, 1929. 228 ABSTRACTS OF TECHNICAL ARTICLES 229 the magnetic field intensity are determined from ballistic galvanometer deflections. While the discussion refers mainly to circular cylinders, the scheme is adaptable to specimens of other shapes. It is particularly designed to obtain accurate measurements of field intensity in cylinders of small diameter. Details of a special design are given. Curves are given which illustrate the variation of the demagnetiza- tion factor with the magnetization, as well as the dependence of this relation on the material and on the dimensional ratio. The Use of Continued Fractions in the Design of Electrical Netivorks.^ Thornton C. Fry. In U. S. Patent No. 1,570,215 and in several technical papers by Bartlett and Cauer it has been shown that con- tinued fractions can often be used in designing networks with pre- assigned impedances. The chief difficulty of the method has been that it frequently required the structures to contain negative resistances, inductances or capacities and therefore the results, though correct in theory, were often worthless in practice because the networks could not be constructed. The present paper removes this difficulty in virtually all cases w^here the analytic character of the desired impedance is known, that is, where it can be represented by a formula and not merely by a graph. In such cases the choice of a type of structure, as well as the assignment of values to the elements, becomes almost a matter of routine with the definite assurance in advance that no negative elements will be required. A Voltage Regulator for Gas Discharge X-Ray Tubes} F. E. Ha WORTH. This note describes a device used in connection with a gas discharge x-ray tube, to regulate the voltage across it by automatically adjusting a mercury valve between the tube and the pumps, thus con- trolling the pressure of the gas. It has been used with tubes of the Hadding and Shearer types and has operated satisfactorily for more than a year. It was designed to replace the regulator described by Bozorth, which is similar in principle but has certain disadvantages, for example the moving parts have high inertia and adjustment is required when the atmospheric pressure changes. The Significance of the Hydrogen Content of Charcoals/' H. H. LowRY. Most studies of the thermal decomposition of hydrocarbons ^ Am. Math. Soc. Bull., July-August, 1929. '" Journal of the Optical Society of America, August, 1929. s Journal of Physical Chemistry, September, 1929. 230 BELL SYSTEM TECHNICAL JOURNAL are confined to an examination of the composition of the Hquid and t^aseous products. Among exceptions to this generalization may be mentioned the interest in coke, carbon black, and charcoal. Even in these cases the physical properties rather than the chemical composi- tion are regarded as the factors which determine their suitability for specific uses. However, in an earlier paper it was pointed out that certain physical properties of a group of charcoals were rather simply related to the per cent hydrogen which was contained in them as de- termined by ultimate analysis. This group of charcoals was prepared in a gas-fired furnace from a single, specially-selected lot of anthracite coal. As stated in this earlier paper, careful consideration of the commercial records taken at the time of preparation indicated that the hydrogen content was probably determined by the maximum temperature to which the samples were heated during their preparation. The hydrogen contents ranged from 0.21 to 0.53%, while the probable range of maximum temperature was 900° to 1200°. The presence of hydrogen in these charcoals was shown to be consistent with a point of view that so-called "amorphous" carbons are hydrocarbons of low hydrogen content built up of polymerized residues from the thermal decomposition of hydrocarbons of greater hydrogen content. Since the significance of the hydrogen content of charcoals has been generally overlooked, the present study was undertaken in order to evaluate the factors which may ordinarily be varied in the preparation of charcoals for various purposes. The factors which were independently varied in this study were the maximum temperature, the time of heating, the atmosphere surrounding the sample during heating and the raw material. To a limited extent the effect of previous heat treatment was also determined. A later paper will give the results of the study of the correlation of hydrogen content and some adsorptive properties of charcoals prepared under carefully controlled conditions. Btginnings of TelepJiovyJ Frederick Leland Rhodes, Outside Plant Development Engineer, Department of Development and Re- search, American Telephone and Telegraph Company. It is only within the past decade or so that science and business have become subjects for literature. Somehow these great phases of human endeavor have been sadly neglected in the literary world until very recently, and now it seems as though, conscious of the lack of good literature in these fields, engineers, scientists and business executives are making up for lost time. Frederick Leland Rhodes has written a new book which undoubtedly will be of great assistance to those in the ' Harper & Brothers, New York and London, 1929. ABSTRACTS OF TECHNICAL ARTICLES 231 telephone industry, for it supplies them with an accurate picture of the technical background of a great industry. It is greatly to the ad- vantage of an individual to know the history of his own business, and Mr. Rhodes has supplied it in an interesting form, thoroughly accurate and readable. No effort has been made to set down the more recent achievements in the world of telephony, but only to carry each chapter to what might be termed the "middle period" in development. There are many phases of the telephonic art which have not been touched upon in the volume, but at the same time, one is not conscious of any lack in this respect as one reads through its interesting pages. Any volume is the better off for illustrations, and Mr. Rhodes' book- is generous in that it carries fifty-four illustrations scattered through 260 pages. The first portion of the book naturally deals with Alexander Graham Bell and occupies three chapters. Following this we have two chapters called "The Bell Patents." As General John J. Carty, Vice President of the American Telephone and Telegraph Company, says: "Never before had the claims of an inventor been subjected to such exhaustive litigation and judicious scrutiny, and never before d'd an inventor receive such a complete and dramatic vindication." The remainder of the fourteen chapters deals with the truly romantic progress of telephone plant, its improvements and expansion over a term of years when telephony was young and the road was fraught with difficulty. Of special interest are the numerous references to original and authentic sources, and in this regard the author has unquestionably used great care and much labor in order to give his reader the most accurate information possible, thus more truly gaining his end of supplying a concrete picture of the younger days of a great industry. Mr. Rhodes' volume is a great contribution, not only to the literature of telephony, but also to that rapidly growing library which contains in its pages the romance of business in America. As a library reference book it will be valuable to the technical student. Any member of the Bell System would do well to familiarize himself with this work, not only because it will help him in his job, but because he will find it a really interesting story. Further Note on the Ionization in the Upper Atmosphere.^ J. C. ScHELLENG. In this paper Mr. Schelleng records certain considera- tions that were omitted from a previous paper, which omission resulted in some difficulty. * Proceedings of the Inslilute of Radio Engineers, August, 1929. 232 BELL SYSTEM TECHNICAL JOURNAL Transmission Networks and Wave Filters.-' T. E. Shea. In this book is summarized the research and experience of the Bell System in the application of electric wave filters, equalizers, balancing networks and similar electrical systems. The preface discusses the nature of the signals transmitted over communication systems and a statement f)f the principal ways in which selective networks are used to modify sig- nal transmission. A detailed example of the application of selective networks to an actual long distance telephone circuit gives specific en- gineering requirements and limitations. The next portion of the book deals with some of the more general principles governing network analysis. The engineering terms used to evaluate network performance are described and a number of general theorems and equivalences which simplify the analytic treatment of networks are demonstrated. A considerable discussion is also given of the characteristics of the elementary two-terminal networks most used as constituents of larger structures. With this background the author is now ready to consider the proper- ties of wave filters. Conditions for free transmission and attenuation in ladder networks are set up and the particular networks of chief practical importance are described in detail. The various structures revealed by this listing differ widely among themselves as regards propagative and impedance characteristics even when they transmit the same frequency bands. Since the ideal network characteristics seldom correspond exactly to any one of these structures, filter re- quirements are usually met most efficiently by composite networks, containing sections of several different types. The author describes the conditions which must be satisfied before different sections are joined together and gives several examples of methods of computing the performance of such composite structures. This treatment of networks deals only with their response to steady single-frequency electrical impulses. It cannot be applied directly to communication systems, since signals are of more complicated wave forms and are transient in character. In the last portion of the book therefore, the author discusses the use of Fourier analysis in relating the characteristic of the network computed on a steady-state basis to its response to a transient impulse of arbitrary character. Some Principles of Broadcast Frequency Allocation}'^ L. E. Whitte- MORE. This paper discusses some of the technical factors which must be considered in the allocation of frequencies to broadcasting stations 8 D. Van Nostrand Company, New York. ^^Proceedings, Institute of Radio Engineers, August, 1929. ABSTRACTS OF TECHNICAL ARTICLES 233 in such a way as to provide the best possible coverage of a given country or continental area. A given frequency or channel can be used for either of two kinds of service; (1) by one station, exclusively, to give high grade service to the immediate locality and opportunity for service over broad rural areas when transmission conditions are good, and (2) by two or more stations simultaneously, to give local service to a number of separate regions, each of rather restricted area. The problem, therefore, in- volves a determination of (1) the proper balance between the two kinds of service, rural and urban, and (2) the proper basis for the apportion- ment of the assignments. Reference is made to the basis of apportionment of radio broad- casting assignments laid down in the U. S. Radio Act of 1927, and to certain suggestions which have been made for the apportionment of broadcasting frequency assignments among the countries of Europe. A brief discussion is given of the relation between field intensity, or signal strength, and distance of transmission at broadcast frequencies. The paper also discusses briefly the effects produced in the case of (1) a single station operating exclusively on a "clear" channel, and (2) two or more stations operating simultaneously on the same channel. It is suggested that the distribution of assignments on "clear" channels, in a given continental area be made proportional to the population of each of several large geographical units or zones and that the distribution of assignments on "multiple assignment" channels be made to comparatively small geographical units in proportion to their areas. Contributors to this Issue John R. Carsox, B.S., Princeton, 1907; E.E., 1909; M.S., 1912; American Telephone and Telegraph Company, 1914-. Mr. Carson is well known through his theoretical transmission studies and has published extensively on electric circuit theory and electric wave propagation. A. B. Clark, B.E.E., University of Michigan, 1911; American Telephone and Telegraph Company, 191 1-. Toll Transmission De- velopment Engineer, 1928-. Mr. Clark's work has been largely con- cerned with toll telephone and telegraph systems. Karl K. Darrow, B.S., University of Chicago, 1911; University of Paris, 1911-12; University of BerUn, 1912; Ph.D., University of Chicago, 1917; Western Electric Company, 1917-25; Bell Telephone Laboratories, 1925-. Dr. Darrow has been engaged largely in writing on various fields of physics and the allied sciences. Some of his earlier articles on Contemporary Physics form the nucleus of a recently published book entitled "Introduction to Contemporary Physics" (D. V^an Nostrand Company). A recent article has been translated and published in Germany under the title "Einleitung in die Wellen- mechanik." Bancroft Gherardi, B.Sc, Polytechnic Institute, Brooklyn, N. Y., 1891; M.E., Cornell University, 1893; M.M.E., Cornell University, 1894. New York Telephone Company, Engineering Assistant, 1895- 99; Traffic Engineer, 1899-1900. New York and New Jersey Tele- phone Company. Chief Engineer, 1900-06. New York Telephone Company, and New York and New Jersey Telephone Company, Assistant Chief Engineer, 1906-07. American Telephone and Tele- graph Company, Equipment Engineer, 1907-09; Engineer of Plant, 1909-18; Acting Chief Engineer, 1918-19; Chief Engineer, 1919-20; Vice President and Chief Engineer, 1920-. Mr. Gherardi is a Past President of the American Institute of Electrical Engineers. Frank B. Jewett, A.B., California Institute of Technology, 1898; Ph.D., University of Chicago, 1902. American Telephone and Tele- graph Company, Transmission and Protection Engineer, 1904-12. Western Electric Company, Assistant Chief Engineer, 1912-16; Chief Engineer, 1916-21; Vice President and Chief Engineer, 1921-22; Vice President, 1922-25. International Western Electric Company, 234 CONTRIBUTORS TO THIS ISSUE 235 Vice President, 1922-25. Manufacturers Junction Railway, Vice President, 1922-25. American Telephone and Telegraph Company, Vice President, and Bell Telephone Laboratories, President, 1925-. Dr. Jewett is a Past President of the American Institute of Electrical Engineers. Francis F. Lucas, Associated Bell Telephone Companies, 1902-10; Western Electric Company, 1910-25; Bell Telephone Laboratories, 1925-. Mr. Lucas has specialized in the development and appli- cation of microscopy. He has received international recognition and awards for the development of high power metallography and ultra-violet microscopy and for numerous scientific papers which he has contributed on the subjects of metallurgical and biological re- search. For several years he has been Consulting Technical Expert for the War Department, U. S. A., Watertown Arsenal. Edward L. Nelson, B.S. in E.E., Armour Institute of Technology, 1914; Western Electric Company, 1917-25; Bell Telephone Laborato- ries, 1925-. As Radio Development Engineer of Bell Telephone Lab- oratories, Mr. Nelson is responsible for the development and design of commercial radio apparatus, which includes radio broadcasting equip- ment. R. L. Wegel, A.B., Ripon College, 1910; Assistant in Physics, University of Wisconsin, M. A., 1910-12; Western Electric Company, 1914-25; Bell Telephone Laboratories, 1925-. Mr. Wegel has written several papers on theory of telephone receivers and on the theory of hearing. The article appearing in this issue is taken from lecture notes on Mechanics of Vibrating Systems by the author. It is planned to publish these notes in future issues of the Bell System Technical Journal. M.K. ZiNN, B.S. in E.E., Purdue University, 1918; American Tele- phone and Telegraph Company, 1919-. Mr. Zinn's work has been related particularly to the design of loading for telephone circuits. The Bell System Technical Journal April, 1930 Developments in Communication Materials^ By WILLIAM FONDILLER The subject of engineering materials is one of increasing importance, as is evidenced by the expenditure of over a half bilhon dollars annually in new construction by the Bell System. This has led to the concentration of the research and engineering work on materials in a group devoted particularly to this field of activity. Studies of the chemical and physical properties of materials must be combined by the materials engineer with a knowledge of the operating requirements of telephone apparatus. The paper covers broadly the materials used in communication engineer- ing and gives instances in which the needs of the telephone plant imposed requirements which were not satisfied by commercially available materials. Some of the instances cited are phenol fiber having improved resistance to arcing for use in sequence switches; a composite molded plastic for use in terminal strips; textile materials for central office wiring treated to improve their electrical insulating quality and non-ferrous metals of more uniform rViprarf prictiVc Prnhlfime inirnlA/incr tVip nisp n( duralumin fnr radin hrnarl- CORRECTION SLIP FOR ISSUE OF JANUARY, 1930 Page 153: Equation (10) should read //(/) ~ {^-h(D-^rii:r--}^' //(/) ^ S(\t) (10) of the signal, sound or scene to distant points, or their recording. Up to about ten years ago the average manufacturer left to his designing engineer the problem of selecting and testing the materials which were to be embodied in a design, and he in turn was dependent on the manufacturers of raw materials as to the variety and quality of the materials available. Without depreciating the ability or ini- tiative of manufacturers of engineering materials, it will be evident that the special needs of a particular industry would, in general, not be as fully appreciated by an outside manufacturer as by an engineer working 1 Presented before A. I. E. E. on November 13, 1929. 237 16 The Bell System Technical Journal April, 1930 Developments in Communication Materials^ By "WILLIAM FONDILLER The subject of engineering materials is one of increasing importance, as is evidenced by the expenditure of over a half billion dollars annually in new construction by the Bell System. This has led to the concentration of the research and engineering work on materials in a group devoted particularly to this field of activity. Studies of the chemical and physical properties of materials must be combined by the materials engineer with a knowledge of the operating requirements of telephone apparatus. The paper covers broadly the materials used in communication engineer- ing and gives instances in which the needs of the telephone plant imposed requirements which were not satisfied by commercially available materials. Some of the instances cited are phenol fiber having improved resistance to arcing for use in sequence switches; a composite molded plastic for use in terminal strips; textile materials for central ofilice wiring treated to improve their electrical insulating quality and non-ferrous metals of more uniform characteristics. Problems involving the use of duralumin for radio broad- casting transmitters and the light valve used in sound pictures are also de- scribed. Particular emphasis is laid on the benefits resulting from the con- tinuous research in magnetic materials which have produced successively — powdered electrolytic iron cores for loading coils, permalloy, and recently perminvar. Summing up, the work on materials has resulted in benefits along two general lines: 1. Improvement in quality of commercial materials. 2. Discovery or development of valuable new materials. THE subject of this paper, "Developments in Communication Materials," perhaps needs some definition with the rapid addition of new fields to the pioneer arts of telegraphy and telephony. Today we must include high frequency wire telegraphy and telephony by means of carrier currents, radio, telephotography, television and, in a sense, sound pictures. All of these modes of communication of intelli- gence are characterized by the use of electrical means for the transfer of the signal, sound or scene to distant points, or their recording. Up to about ten years ago the average manufacturer left to his designing engineer the problem of selecting and testing the materials which were to be embodied in a design, and he in turn was dependent on the manufacturers of raw materials as to the variety and quality of the materials available. Without depreciating the ability or ini- tiative of manufacturers of engineering materials, it will be evident that the special needs of a particular industry would, in general, not be as fully appreciated by an outside manufacturer as by an engineer working 1 Presented before A. I. E. E. on November 13, 1929. 237 16 238 BELL SYSTEM TECHNICAL JOURNAL on these problems. Thus it has come about in the Bell System, as with other large consumers of materials, that the investigation of materials has been organized as a distinct branch of research and engineering activity. Studies of the chemical, physical and metallurgical proper- ties of materials are embraced in this work. In general the materials engineer should not only be well versed in materials, but should also have a good knowledge of the operating characteristics of the apparatus to be designed. Thus he can discuss the materials side of the problem with the designing engineer on equal terms and make his contribution to the best advantage. The importance of a thorough knowledge of materials in the telephone business will be appreciated from the fact that, during 1929, it is estimated that about $590,000,000 will be spent for additions to the Bell System plant. In telephony the general introduction of the dial system has imposed more severe requirements than heretofore because of the need for the utmost in reliability of performance of the large number of switches, relays, etc., which are required to operate automatically with a mini- mum of maintenance. In the central ofihce small size of apparatus con- stitutes a very important consideration, not only because of building space required, but the mass and travel of the automatic switches have an important effect on the speed with which connections can be es- tablished and hence on economy of operations. Thus, close control of the quality of materials and the need for small, compact apparatus are important design considerations. In a brief survey of progress in the development of materials, it will be necessary to select a few typical items of interest. The items selected deal primarily with the telephone problem as this is, at the present time at least, the largest single factor in the communications group. The subject may be divided broadly into insulating materials and metallic materials. . Insulating Materials Phenol Fiber Considering first sheet insulating material, we have been using the term "phenol fiber" to cover such materials as bakelite-dilecto, mi- carta, formica and similar fibers made by various manufacturers. Phenol fiber is used extensively in telephone apparatus. One of its applications is in the sequence switch which has insulators alternating with conducting segments, as shown in Fig. 1. The sequence switch, which is used in the dial system, draws out an arc when in operation which sometimes causes carbonization of the insulators. In some cases a hole was burned through the insulator and in other cases the arc was DEVELOPMENTS IN COMMUNICATION MATERIALS 239 Fig. 1 — Sequence switch, used in dial system. Fig. 2 — Detail of apparatus for arcing test of phenol fiber. 240 BELL SYSTEM TECHNICAL JOURNAL sustained over the insulation to such an extent that the circuit was not broken at the proper moment. An examination of the various grades of phenol fiber commercially supplied indicated that they varied widely as to their resistance to arcing. Fig. 2 shows testing apparatus de- signed to evaluate this characteristic. The sample under test was made into a sequence switch cam and rotated on the fixture at a speed of 10 r.p.m. The set is wired to give a circuit condition comparable with that causing failure in service, except that slower speed and higher voltage are used to accelerate the test. The position of the rear brush is so adjusted that after the material has become carbonized through an arc of 15 degrees or a hole has been burned through the insulation, the machine would be stopped by means of a circuit breaker, shown in Fig. 3. This instrument makes the Fig. 3 — Assembly of apparatus for arcing test failure value independent of the operator's judgment, and has proven so satisfactory that it has been employed for specification purposes. Fig. 4 shows insulators tested by this instrument; those at the top having been rejected, and those at the bottom being satisfactory. An improvement of 20 to 1 in arcing characteristics was obtained. This was brought about by close cooperation with the Bakelite Research Laboratories, which developed a special grade of resin to be used in the manufacture of this material. In this case the materials engineer developed a method of test for evaluating the particular quality de- sired which enabled the supplier to improve his product in the desired respect. Even though resistant to moisture in the ordinary sense, phenol fiber absorbs a certain amount of moisture depending on the quality of the material furnished. As this moisture is given up, the material DEVELOPMENTS IN COMMUNICATION MATERIALS 241 Fig. 4 — Insulators subjected to arcing test. Top — F"ailure value, 20 rev. Bottom — Failure value, 1200 rev. % Fig. 5 — Telephone relay showing phenol fiber insulators between contact springs. 242 BELL SYSTEM TECHNICAL JOURNAL tends to shrink. If the fiber is not sufficiently hard as manufactured, it will also flow under pressure. In telephone relays of a commonly used type, illustrated by Fig. 5, the contact springs are insulated from each other by thin sheets of phenol fiber, and any material change in dimensions of these insulators, due to moisture absorption or cold flow, will alter the spacing of the contacts, thus throwing the relay out of adjustment. To measure these tendencies on materials used in spring pile-ups, we use the method illustrated by Fig. 6. It will be seen that a Brinell machine, usually Fig. 6 — Modified Brinell machine for flow-test of insulator laminations. employed for metals testing, has been modified to use a flat-ended plunger resting on a pile of insulating material. The test material is first cut into pieces 'jA" square and then subjected to atmospheric conditions which would cause it to take up an amount of moisture comparable to that expected under manufacturing conditions. The pieces are then stacked and a pressure of 2,000 pounds per square inch applied. The testing apparatus is installed in a heat insulated box DEVELOPMENTS IN COMMUNICATION MATERIALS 243 Fig. 7 — Flow-test apparatus of Fig. 6 enclosed for temperature control. 7 6 5 5 ^^ ^AP^ ^^ '^■——-^ Z 1- O lU i:^ 3 / ^ / f PRESSURE 2000 LB PER SQUARE INCH TEMPERATURE 120° F UJ o 2 / / P HENOL . FIBRE 1 U ^ 0 f 24 48 72 TIME IN HOURS 96 120 Fig. 8 — Flow-test results for hard rubber and phenol fiber. 244 BELL SYSTEM TECHNICAL JOURNAL shown in Fig. 7, so that the temperature throughout the 24 hour test may be maintained at 120° F. corresponding to the maximum Hkely to be experienced in service. The amount of shrinkage or flow is meas- ured on the dial previously shown. Fig. 8 shows the relative per- formance of hard rubber and phenol fiber under the conditions of this test. Molded Plastics In recent years there has been great activity on the part of manu- facturers of molded plastics to develop improved molding compounds, and we have endeavored to keep informed of new developments by examining new compounds as they became available. An interesting problem presented itself in the application of suitable molding com- pounds to a device known as a test strip, shown in Fig. 9. It will be Fig. 9 — 100 point test strip used in switchboards. seen that it consists of a number of metal terminals mounted flush on the face of the strip and projecting at the back to provide soldering lugs for the central office wiring. In operation it is necessary to touch a metal contact plug to the appropriate test strip contact which will pro- duce an audible signal in the operator's receiver. In passing the plug over "live" terminals an arc is drawn out, which is accentuated by a habit of some operators of running their pencils along the grooves leaving a conducting path. Such arcs caused permanent conducting paths in the surface of the bakelite, despite the adoption of strenuous cleaning routines. The need for a better insulating material for this use became even more urgent with a demand for a test strip having 200 terminals Instead of 100 in the same space. Studies of compounds having such base materials as cellulose-nitrate, shellac, hard rubber, casein, and cellulose-acetate showed the last mentioned to give desirable arcing resistance. Foreign conducting material on the surface was burned off by the arc; the products of combustion of the small amount of cellulose acetate actually burned by such an arc are largely volatile, and the residue is non-conducting. The compound used was found not to be sufficiently heat resistant to be satisfactory for the body of the test strip. The problem was solved by using it as a veneer on the test face of the bakelite strip. DEVELOPMENTS IN COMMUNICATION MATERIALS 245 This face is farthest from the heated ends of the terminals, is free from mechanical strain and is therefore not damaged by soldering operations. Since it was the practice to mold this test strip using several partially- cured preforms, the veneer construction was introduced with only a slight increase in cost. The cellulose acetate has nearly the same molding temperature as the phenol plastic, so that the composite test strip could be molded in one operation. Fig. 10 shows the appearance ■TERMINALS IN HOLES IN PREFORMS PREFORM OF PHENOL PLASTIC PREFORM OF PHENOL PLASTIC PREFORM OF PHENOL PLASTIC PREFORM OF CELLULOSE ACETATE Fig. 10 — Method of molding composite 200 point test strip, of the modified test strip and the method of molding. Textile Insulation Another development was in the improvement of textile insulation which was recently described before the Institute.^- ^ It is mentioned here only in passing, because of its great commercial importance. As a result of several years of study in the laboratory, it was found that the insulating quality of textiles depended on (1) the kind of fiber; (2) impurities present in the fiber; (3) moisture. The salts of sodium and potassium were found to be highly detrimental from an insulation standpoint. A very great improvement was effected by a washing treatment of the textile. Thus it has been possible to make ^ "The Predominating Influence of Moisture and Electrolytic Material Upon Textiles as Insulators," R. R. Williams and E. J. Murphy, Trans. A. I. E. E., Vol. 48, 1929. ^ "Purified Textile Insulation," H. H. Glenn and E. B. Wood, Trans. A. I. E. E., Vol. 48, 1929. 246 BELL SYSTEM TECHNICAL JOURNAL cotton an acceptable substitute for silk as wire insulation, as well as to improve greatly the insulating properties of silk. In one instance, central office distributing frame wire, of which the Bell System uses about five hundred million conductor feet annually, it was found pos- sible to use double silk insulated conductor of treated thread where formerly triple silk insulation was required. An actual improvement in insulation was effected at the same time that a considerable economy resulted. Metallic Materials N on- Ferrous Metals Telephone apparatus uses about 30,000,000 lbs. yearly of brass, bronze and nickel silver as structural members, springs and bearings. Because of space limitation the parts are necessarily small, many are formed into irregular shapes; spring parts must maintain accurate adjustment and have long fatigue life; certain other parts must resist wear. Experience with commercial grades of brass indicated wide variations under existing specifications and unsatisfactory means of testing the quality. At first blush there may not appear to be any connection between the temper of a metal spring and the grade of telephone service furnished, but looking at the matter broadly we were convinced that the stakes were large enough to warrant our launching an investigation of non-ferrous metals with the object of arriving at a better purchasing specification. Accordingly the Bell Telephone Laboratories initiated a joint study with the Western Electric Company and the American Brass Company which has extended over a period of several years. The results of this work have been described in considerable detail in appropriate papers before the American Society for Testing Materials.'' •" This has resulted — 1. In a more accurate knowledge of the physical properties of brass, phosphor bronze and nickel silver. 2. Development of improved methods of test. 3. Preparation of better purchasing specifications with resulting improved control of the quality of the materials. As an instance of the benefits derived, the work on hardness testing may be cited. For many years the scleroscope had been used as a rapid means of controlling the quality of sheet metal but trouble was fre- quently encountered because results could not be readily duplicated on * "Physical Properties and Methods of Test for Sheet Brass," H. N. Van Deusen, L. I. Shaw and C. H. Davis, Proc. Amer. Soc.for Testing Materials, 1927. *" Physical Properties and Method of Test for Sheet Non-Ferrous Metals," J. R. Townsend, W. A. Straw and C. H. Davis, Proc. A. S. T. M., 1929. DEVELOPMENTS IN COMMUNICATION MATERIALS 247 different instruments and it was necessary to allow rather wide limits on each temper resulting in considerable overlapping of the temper tolerances. While tensile strength is usually considered the reference test for cold worked metal, it is necessary to have a test which can be used for more rapid inspection. As a result of our study we were able 100 40,000 60,000 aO,000 100,000 120,000 TENSILE STRENGTH fPSl) Fig. 11 — Relation between tensile strength and Rockwell hardness — sheet brass. to develop means for maintaining the Rockwell hardness tester to an accuracy within 2 points compared with 8 or 10 points on the sclero- scope. Fig. 11 shows the relation between tensile strength and Rock- well hardness for a rolling series made up by the American Brass Com- pany under carefully controlled manufacturing conditions. This rolling series covered all ranges of hardness and thickness of sheet metal generally used in telephone apparatus. The tension test is used 248 BELL SYSTEM TECHNICAL JOURNAL as a reference test and is resorted to only when the Rockwell test indicates the material to be close to the limiting values specified. Work has been completed, resulting in the preparation of improved specifications for leaded brass, annealed brass, nickel silver and phosphor bronze and a similar investigation of rod stock in all grades of these metals is now under way. It is interesting to note in passing \^ 9* r /s %4 H Q> ^ Fig. 12 — Rotary selector used in dial system. that in the course of our investigations we determined that the endur- ance limit of non-ferrous metals is only half that established for ferrous metals, averaging approximately }i of the ultimate strength.^ For one of the rotary selectors used in the panel dial system we developed a leaded phosphor bronze sheet containing approximately 3 per cent of lead which proved very valuable in terms of increased 6 "Fatigue Studies of Non-Ferrous Sheet Metals," J. R. Townsend and C. H. Greenall, Proc. A. S. T. M., 1929. DEVELOPMENTS IN COMMUNICATION MATERIALS 249 life of the switch. This selector consists of an arrangement of closely spaced terminals referred to as the " bank " and a set of rotating brushes contacting with the bank terminals as shown in Fig. 12. Experience in the field indicated that under severe service conditions these selectors have a comparatively short life. As a result of our studies we replaced the brass brushes in the rotor with phosphor bronze, and the brass terminals of the bank with leaded phosphor bronze, a combination which has given approximately four times the life obtainable with brass parts, with corresponding maintenance savings. The reduced wear seems to be due in part at least to a lubricating effect of the lead constitutent in the bank terminals. Aluminum alloys have had considerable application to telephone apparatus not only in die castings but in sheet form as diaphragms in certain of the new developments in telephone transmitters and re- ceivers. One of the most interesting of the aluminum alloys is dur- alumin, an alloy of aluminum, copper, silicon and magnesium. This material has about one-third the specific gravity of steel and like steel can be increased in strength by heat treatment in the manufactured form. Our first application of duralumin was as a stretched diaphragm in radio broadcasting transmitters. Here it was necessary to obtain material with as small a mass as possible and with the necessary strength to allow stretching to give a high natural period essential for good quality transmission. The material used in this case was 1.7 mils thick and had a tensile strength between 70,000 and 80,000 pounds per square inch. Probably one of the most difficult applications of sheet duralumin is to the light valve used in the film method of sound picture recording. The light valve is an electromechanical device actuated by amplified speech currents, and consists of a loop of duralumin tape supported in a plane at right angles to a magnetic field. A view of the light valve is given in Fig. 13 which shows the tape held by two wind-lasses, AA"^, at one end, and wrapped over a spring-supported pulley B at the other. This places the tape under considerable tension. The tape is 6 mils wide and .5 mil thick. The central portion of the loop is supported on insulating bridges just above the face of the pole piece which constitutes the armature of an electromagnet. Viewed against the light, the valve appears as a slit 2 mils wide by 256 mils long. In operation the amplified speech current is passed through the duralumin tape which, reacting with the magnetic field of the electromagnet causes variations in the width of the slit controlled by the variations in the speech current. The light beam directed toward the film is thus modulated by the slit in accordance with the variations 250 BELL SYSTEM TECHNICAL JOURNAL of the speech current. In order to avoid distortion, severe require- ments were imposed on the straightness of the edges of the tape, and on the strength, in order to permit stretching to give a natural period in excess of 7,000 cycles per second. To obtain these properties special heat treatments and methods of rolling the material had to be de- veloped. Fig. 13 — Light valve used in recording of sound pictures. AA'^ — Wind-lasses. B — Pulley supported by spring. CC^ — Insulating bridges. Ferrous Metals Some interesting problems have been encountered in the use of ferrous metals in telephone apparatus, particularly in the operator's calling dial. Considerable trouble was encountered from slippage of the dial governor resulting from premature wear or breakage of the tips of the pawls or the teeth of the pinion. These parts had been made out of low carbon steel which had been found satisfactory for the subscriber's dial. The operator's dial, however, being used for a greater number of times, presented a more severe condition and case hardening was applied to obtain better wear resisting properties. This treatment was found to be unsatisfactory because the parts have thin sections and the combined weight of the two parts amounts to only 2 grams. Case hardening either produced too deep a case giving brittleness or too shallow a case which soon wore through. A nickel-chrome steel, originall}- developed for the automotive industry was finally adopted for the pawl and pinion combined with a special heat treatment. It was thus possible to obtain a useful life of 8 million operations as compared with an average of }4 million operations for the steel formerly used. This is another instance in which an increase DEVELOPMENTS IN COMMUNICATION MATERIALS 251 in first cost resulted in appreciable savings in annual cost of the device, considered from the operating companies' standpoint. Ferro- Magnetic Metals Up to about 15 years ago, telephone engineers used the magnetic materials in their designs which had been originally developed for the power industry, viz., magnetic iron and silicon steel. An exception was the use of 4. mil hard drawn steel wire for loading coil cores where extremely low permeability was desired. The increasingly severe requirements imposed by compositing and phantoming of telephone circuits and the introduction of vacuum tube Fig. 14 — Loading coils showing core rings of liighly compressed powdered iron. repeaters, made necessary the development of materials which would more adequately meet the new requirements. It was in 1915 that the Western Electric Company first produced compressed powdered electrolytic iron cores for loading coils. The construction of such powdered iron core coils is illustrated by Fig. 14. Electrolytically deposited iron is ground to a fine powder; the particles are covered with an insulating film and then compressed at a pressure of 200,000 lbs. per square inch to form rings as shown in the figure. This material was sensational in the improvements which it afforded over the core materials theretofore available as it combined with extremely high resistivity, high stability of A.C. permeability under conditions of powerful superposed or residual D.C. magnetization. The change in 252 BELL SYSTEM TECHNICAL JOURNAL A.C. permeability resulting from the temporary application of large magnetizing forces did not exceed 2 per cent as compared with changes of the order of 30 to 40 per cent commonly found in previously available materials. The next important step was the discovery of permalloy, a nickel- iron alloy having extremely high permeability which had its first application in the loading of submarine telegraph cables. This mate- rial with its extremely low hysteresis loss and high induction for feeble magnetizing forces, has since been applied extensively in the design of transformers, relays, receivers, and other telephone apparatus. Fig. 15 6000 4000 2000- -2000- -4000 -6000 1.^ »UCON STEE / A ,>-^ / / >■ o -I _l «»• / / / 2 a UJ a. y / / CO K / y ' ■^^ -0.10 -0.8 -0.6 -0.4 -0.2 O 0.2 0.4 0.6 0.8 H Fig. 15 — Hysteresis loops of silicon steel and permalloy. 0.10 shows comparative hysteresis loops for permalloy and silicon steel. The much smaller hysteresis loss of permalloy, approximately one- seventh of that of the silicon steel sample is indicative of its greatly reduced tendency to remain magnetized after the removal of a mag- netizing force, a property which is of great importance in the operation of quick release types of relays. In transformers and in continuously loaded cable, the very high permeability at small magnetizing forces of this material, strikingly shown in Fig. 16, is of great value. It is the high permeability of permalloy that made it possible to load telegraph cables successfully and thereby attain a threefold increase in telegraph speed. In transformers such as those used in vacuum tube amplifiers, the high permeability permits the designer either to achieve equivalent quality with a much smaller apparatus volume or, in the same space, to furnish equipment of better quality. The latter result is shown by the curves of Fig. 17 which indicate how transformer performance at DEVELOPMENTS IN COMMUNICATION MATERIALS 253 100,000 80,000 ^ 60,000 ffl < ui S a. UJ a i. 40,000 20,000 4 000 8,000 12,000 B Fig. 16 — Permeability curves of soft iron and permalloy. 16,000 < o u. 3 _l Q. <2 u o < I- > - 1 r ^' .^- ^ ^ ^ i A = SILICON STEEL CORE B = PERMALLOY CORE Tsl °/o 10 100 1,000 CYCLES PER SECOND 10.000 Fig. 17 — Showing improvement in quality of voice frequency amplifier due to permal- loy core transformers. 17 254 BELL SYSTEM TECHNICAL JOURNAL very low frequencies is improved by the use of permalloy. Sheet permalloy has been followed by compressed powdered permalloy^ and this by perminvar,^ the newest member of the magnetic alloy family. Compressed powdered permalloy has replaced the powdered iron as it has all of the desirable properties of the latter and to an even greater degree. By virtue of higher permeability combined with lower hys- teresis loss, it has made possible the design of smaller coils of superior performance characteristics. As an illustration the two loading coils of Fig. 18 are shown, the smaller of these being the electrical equivalent Fig. 18 — Relative size of powdered iron (left) and powdered permalloy (right) loading coils. of the larger in all respects and in some its superior. In general the reduction in coil size made possible by the use of powdered permalloy in place of powdered iron amounts to about 75 per cent giving very substantial savings in manufacturing costs, handling problems and installation space required. Perminvar is remarkable in an entirely unique respect. Its permea- ability is not exceptionally high, being of the same order as that of ordinary soft iron at moderately low magnetizing forces, but it is exceptionally constant with respect to magnetizing forces. This is shown in Fig. 19 from which it will be noted that there is substantially no change in permeability up to a force of about 2 gausses whereas over this same range, the permeability of soft iron undergoes a change of more than 2,000 per cent. Up to somewhat smaller magnetizing forces, perminvar has a vanishingly small hysteresis loss. Fig. 20 depicts this loss for perminvar. It is to a material of constant permeability and low hysteresis loss that the transformer designer turns when he has a difficult requirement as to low modulation to meet. Unfortunately, while perminvar has these properties over a limited range of magnetiza- ' "Compressed Powdered Permalloy, its Manufacture and Magnetic Properties," W. J. Shackelton & I. G. Barber, Trans. A. I. E. E., Vol. 17, 1928. * "Magnetic Properties of Perminvar," G. W. Elmen, Jotir. of Franklin Institute, Vol. 206, 1928. DEVELOPMENTS IN COMMUNICATION MATERIALS 255 3600r 3200 2600 2400- 2000- :i. 1600- 1200- 800- 400 1300 I i---. AIR QUENCHED / -ANNEALED / J / baked! y y y / y Fig. 19 — Permeability curves for perminvar. 1200 1000 800 CD 600 400 200 0 1200 1000 800 OQ600 400 200 0 A A *' / h y 4, / 1 y / J y Ofl 1.6 2.4 32 H 6000 5000 4000 3000 2000 1000 (0 0 -1000 -2000 -3000 -4000 -5000 -6000 0.8 1.6 2.4 3.2 -5-4-3-2-1 0 I 2 3 4 5 H H A 1 ^ y / y 1^ •y / h ^ K / 1 ^ [ } / y 1 y \ y y 1 / ' I / 1 Fig. 20 — Hysteresis loops for perminvar. A — Air quenched. 5— Baked. 256 BELL SYSTEM TECHNICAL JOURNAL tion, if this range is exceeded they are lost and so it is necessary that it be used within suitable limits. These materials have been described in technical papers before various scientific societies and are not, therefore, discussed in detail here. As indicating the wide scope of magnetic performance that is de- manded of materials for use in communication apparatus, some of the necessary properties are listed below : High permeability — at very feeble and at high inductions. High saturation value of induction. Low residual magnetization. Low hysteresis loss at feeble and moderate magnetizations. Low eddy-current losses over the frequency range from 0 to 80,000 cycles. High constancy of permeability over a wide range of magnetization. Small effect on A.C. permeability at feeble currents with superposed or residual D.C. magnetization. Certain of these requirements are imposed from the simultaneous transmission of D.C. telegraph currents, speech currents and carrier frequency telephone or telegraph currents through the transformers, loading coil or other iron-core apparatus in the circuit. Interference between channels, due to magnetic modulation in the cores, must be kept at an extremely low value for satisfactory quality of transmission. Summing up our work on materials, the results have been along two general lines: (1) improvement in quality of commercial materials and (2) development of new materials. As regards the first, we have worked in close cooperation with material suppliers whose progressive attitude has made possible certain of the advances described. The more striking advances have been due to the discovery of new or im- proved materials in our laboratories, the savings from which have amply justified the program of continuous research which has been the Bell System policy for a number of years. To take a single instance, the field of magnetic alloys — probably the first to which we applied intensive effort, — a single invention, the powdered electrolytic iron core resulted in savings of such magnitude as to far overshadow the cost of the investigational work. As already noted, this material has since been superseded by the powdered permalloy core which represents an equally great advance. There is one point which should be emphasized and that is, that the most economical material is not necessarily the cheapest one. Treated textiles cost more per pound than ordinary textiles; permalloy costs more per pound than silicon steel. In these particular instances so DEVELOPMENTS IN COMMUNICATION MATERIALS 257 much less material is required to obtain the desired result that there is a net saving in cost of manufacture. The true criterion of relative economy, however, takes into account not alone the cost of manufac- ture, but the serviceability of the device throughout its operating life. Hence the designer, if he be free to decide on purely engineering grounds, will make his decision as to the best materials to use on the basis of the lowest annual charge over a period of years, thus taking into account the important item of maintenance cost. Transoceanic Telephone Service — Short-Wave Transmission By RALPH BOWN i The discussion relates to the transmission problems involved in short-wave radiotelephony over long distances and the transmission bases for design of the systems used in commercial transatlantic service. Choice of operating frequencies, amounts of transmitter power, directive transmitting and re- ceiving antennas, automatic gain controls in receivers, and voice-operated switching devices are all factors which may be invoked to aid in solving these problems. The way in which they have been applied in the transatlantic systems and the results which have been obtained are set forth briefly. TRUNK circuits between London and New York which furnish telephone service between these two cities and also permit suc- cessful conversation by means of toll wire extensions between the United States and Europe more generally are being carried over both long waves and short waves. It is the purpose of this paper to consider the transmission side of the new short-wave circuits which the American Telephone and Telegraph Company and the British General Post Office have made available for this service. In doing this we shall proceed from the more general considerations, relating to wave-lengths and communication channels, through a discussion of the principles govern- ing the general design of the system, into a brief summary of practical performance results. The frequency range so far developed for commercial radio use is roughly 20 to 30 million cycles wide, extending from about 10 kilocycles to perhaps 25,000 kilocycles per second. There are two parts of this whole spectrum suitable for transoceanic radiotelephony — the long- wave range which is relatively narrow, extending roughly from 40 kilocycles to 100 kilocycles, and the short-wave range which in its entirety is much broader, extending from about 6000 kilocycles to 25,000 kilocycles. It is evident that the long-wave region, including perhaps only 50 kilocycles, offers opportunity for development of relatively few tele- phone channels, particularly in view of the fact that it is in use by a number of telegraph stations. Also it must be borne in mind that for telephony these waves are suitable for only moderate distances of the order of 3000 miles and for routes in the temperate zones where static 1 Presented at the Winter Convention of the A. I. E. E., New York, N. Y., Jan. 1930. 258 TRANSOCEANIC TELEPHONE SERVICE 259 interference is moderate. The first transatlantic radiotelephone cir- cuit opened in 1927 was a long- wave circuit (58.5-61.5 kilocycles). In providing the next few channels for the initial growth of the service the opportunity to determine the utility of short waves was embraced. The short-wave range is vastly wider in kilocycles but, nevertheless, has its limitations as to the number of communication facilities it afifords. For a given route of a few thousand miles a single frequency gives good transmission for only a part of the day. For example, from the United States to Europe a frequency of about 18,000 to 21,000 kilocycles (17 to 14 meters) is good during daylight on the Atlantic. But in the dawn and dusk period a frequency of about 14,000 kilocycles (22 meters) is better. For the dark hours something like 9000 kilo- cycles (33 meters) gives best transmission and for midnight in winter an even lower frequency near 6000 kilocycles (50 meters) is advantageous. Thus, in considering the short-wave range in terms of communication circuits, we must shrink its apparent width materially to take account of the several frequencies required for continuous service. At the present time the frequency spaces between channels are much greater than the bands of frequencies actually occupied by useful transmission. This elbow room is to allow for the tendency of many stations not to stay accurately on their nominal frequencies but to wander about somewhat. But in spite of this allowance, cases of interference are common and one of the activities which must be carried on in connection with a commercial system is the monitoring of inter- fering stations and the accurate measurement of transmitting fre- quencies to determine the cause of the conflict. To permit intensive development of the frequency space offered by Nature the greatest possible constancy and accuracy of frequency maintenance in trans- mitting sets will be required. The fact that channels have been assigned (within wide bands set aside for a particular service) with little regard to the geographical location of stations may result in neighboring channels having much stronger signals than those in the channel being received. When this is so, a severe requirement is placed on the selectivity of the re- ceiver to prevent interference. Interconnecting with Wire Circuit Extensions The skeleton of a radiotelephone circuit is in its essentials very simple. It consists merely of a transmitter and a receiver at each end of the route and two oppositely directed, one-way radio channels between them. These two independent channels must be arranged at the terminals to connect with two-wire telephone circuits in which 260 BELL SYSTEM TECHNICAL JOURNAL messages in opposite directions travel on the same wire path. The familiar hybrid coil arrangement so common in telephone repeaters and four-wire cable circuits might appear to solve this problem, were there not difficulties peculiar to the radio channels. In the short-wave case large variations in attenuation occur in the radio paths within short intervals of time. These would tend to cause re-transmission of received signals at such amplitudes that severe echoes and even singing around the two ends of the circuit would occur unless means were provided to prevent this. To overcome these fundamental transmission difficulties, an auto- matic system of switches operated by the voice currents of the speakers has been developed.^ These devices cut off the radio path in one UNITED STATES „ BOARD RECEIVING AMPLIFIER "LI — ' RECEIVING DELAY TOLL awn SWITCH- kSiXl TSW^ hw^ — TELEPHONE OPERATOR HYBRID COIL TESQ- RADIO RECEIVER ETCOr J- FROM ENGLAND a RECEIVING DETECTOR w. TRANS- MITTING DETECTOR TRANS- MITTING DELAY LAWRENCEVILLE CO 2 RADIO -► TRANS- MITTER J- TO ENGLAND Fig. 1^ — Circuit diagram illustrating operation of voice-operated switching device. direction while speech is traveling in the reverse direction and also keep one direction blocked when no speech is being transmitted. The operation is so rapid that it is unnoticed by the telephone users. Since this system prevents the existence of singing and echo paths, it permits the amplification to be varied at several points almost without regard to changes in other parts of the system, and it is possible by manual adjustment to maintain the volumes passing into the radio link at relatively constant values, irrespective of the lengths of the connected wire circuits and the talking habits of the subscribers. Fig. 1 gives a schematic diagram of the United States end of one of the short-wave circuits showing the essential features of a voice- operated device which has been used. This kind of apparatus is ^ For detailed description of this system see "The New York-London Telephone Circuit" by S. B. Wright and H. C. Silent, Bell System Tech. JL, Vol. VI, October, 1927, pp. 736-749. TRANSOCEANIC TELEPHONE SERVICE 261 capable of taking many forms and is, of course, subject to change as improvements are developed. The diagram illustrates how one of these forms might be set up. This form employs electro-mechanical relays. The functioning of the apparatus illustrated is briefly as follows: the relay TES is normally open so that received signals pass through to the subscriber. The relay SS is normally closed to short circuit the transmitting line. When the United States subscriber speaks his voice currents go into both the Transmitting Detector and the Transmitting Delay circuit. The Transmitting Detector is a device which amplifies and rectifies the voice currents to produce currents suitable for operating the relays TES and SS which thereupon short circuit the receiving line and clear the short circuit from the transmitting line, respectively. The delay circuit is an artificial line through which the voice currents require a few hundredths of a second to pass so that when they emerge the path ahead of them has been cleared by the relay SS. When the subscriber has ceased speaking the relays drop back to normal. The function of the Receiving Delay circuit, the Receiving Detector, and the relay RES is to protect the Transmitting Detector and relays against operation by echoes of received speech currents. Such echoes arise at irregularities in the two-wire portion of the connection and are reflected back to the input of the Transmitting Detector, where they are blocked by the relay RES which has closed and which hangs on for a brief interval to allow for echoes which may be considerably delayed. The gain control potentiometers shown just preceding the transmitting and receiving amplifiers are provided for the purpose of adjusting the amplification applied to outgoing and incoming signals. The relief from severe requirements on stability of radio transmission and from varying speech load on the radio transmitters which this system provides permits much greater freedom in the design of the two radio channels than would otherwise be possible. The Radio Channels One of the first questions which comes up in considering the design of a radio system is the power which can be sent out by the transmitter. The word "can" is used advisedly, rather than "should," since in the present art the desideratum usually is the greatest amount of power that is technically possible and economically justifiable. There are few radio systems so dependable that increased power would not improve transmission results. At very high frequencies the generation of large powers is attended by many technical difficulties but fortu- nately the radiation of power can be carried out with much greater 262 BELL SYSTEM TECHNICAL JOURNAL efficiency than is feasible at lower frequencies. At 18,000 kilocycles (about 16 meters) a single half- wave radiator or doublet is only about 25 ft. long and it is possible to combine a number of them, driven in phase by a common transmitter, into an antenna array which concen- trates the radiated power in one geographical sector. In that direction the effectiveness may be intensified 50 fold or more (17 db) and waste radiation in other directions reduced materially. Thus, one of the transmitters at Lawrenceville, New Jersey, used in the short-wave transatlantic circuits when supplying 15 kw. radiates in the direction of its corresponding receiving station as effectively as would a non- directive system of about 750 kw. The transmitting antennas also give some directivity in the vertical plane, increasing the radiation sent toward the horizon and decreasing that sent at higher angles. It is not yet certain that vertical directivity is always advantageous and this effect has not been carried very far. At the receiving station the radiated power has dwindled to a small remnant which must be separated from the static as far as possible and amplified to a volume suitable for use in the wire telephone plant. Here again directive antenna arrays are of value. A receiving antenna system sensitive only in a narrow geographical sector, and that lying in the direction from which the signal arrives, excludes radio noise from other directions and thereby scores a gain of perhaps 40 fold (16 db) in the power to which the signal can be amplified without bringing noise above a given value. It also scores against noise which arises in the tubes and circuits used for amplification, since the combined action of the several antennas of the array delivers more signal to the initial amplifier stage where such noises originate. Thus, it is evident that transmitter power, transmitting directivity, receiving directivity, and quiet receiving amplifiers are of aid in pro- viding signal transmission held as far as possible above the radio noise. In a well designed system the relative extents to which these aids are invoked will depend upon economic considerations as well as upon the technical possibilities of the art. There is one other type of noise than that provided by Nature which is of particular importance at short waves,- — electrical noise from the devices of man. One of the worst offenders is the ignition system of the automobile. The short-wave transoceanic receiving station at Net- cong. New Jersey, is so located that automobile roads are at some distance, particularly in the direction from which reception occurs. Service automobiles which produce interference cannot be allowed near the antenna systems unless their ignition systems have been shielded. Also, electrical switching and control systems incidental to the power, TRANSOCEANIC TELEPHONE SERVICE 263 telegraph, and telephone wire systems at the station are shielded or segregated. At both the transmitting and receiving stations at least three antenna systems are supplied for each circuit, one antenna for each of the three frequencies normally employed. The design and arrangement of these are dictated by the requirements flowing from their uses. The purpose of the transmitting antenna is to concentrate as much power as possible in one direction. The purpose of the receiving antenna is to increase reception from the desired direction and to cut down reception at all other angles. In the former the forward-looking portion of the characteristic is of greatest importance, while in the latter the rearward characteristics need greatest refinement. Transmission Performance In short-wave telephone systems the width of the sidebands is so small a percentage of the frequency of transmission that tuning charac- teristics of the antennas and high-frequency circuits are relatively broad and impose little constriction on the transmission-frequency characteristic. A flat speech band is easy to obtain over the range of approximately 250 to 3000 cycles employed for these commercial circuits. This relieves the short-wave circuits from many of the problems of obtaining sufficient band width which are troublesome in designing long-wave systems. Short-wave transmission is subject to one frailty which particularly hampers its use for telephony. This is fading. Where fading is of the ordinary type, consisting of waxing and waning of the entire trans- mitted band of frequencies, automatic gain control at the receiving station is of value and is employed in the transoceanic circuits under discussion. The amplification in the receiver is controlled by the strength of the incoming carrier and is varied inversely with this strength so as to result in substantially constant signal output. Ob- viously this control can be effective only to the extent that the signal seldom falls low enough to be overwhelmed by radio noise. When fading is of the selective type, that is, the different frequencies in the transmitted band do not fade simultaneously, the automatic gain control system is handicapped by the fact that the carrier or control signal is no longer representative of the entire signal band. Selective fading is believed to result from the existence of more than one radio path or route by which signals travel from transmitter to receiver. These paths are of different lengths and thus have different times of transmission. Wave interference between the components arriving over the various paths may cause fading when the path lengths change even slightly. 264 BELL SYSTEM TECHNICAL JOURNAL If the path lengths differ by any considerable amount, for example, a few hundred miles, the wave interference is of such a character as to affect the frequencies across a band consecutively rather than simul- taneously. With the presence of selective fading there comes into being the necessity of guarding against rapid even though small variations in the transmitted frequency, since if such variations are present a peculiar kind of quality distortion of the telephone signal results. The varying load which speech modulation places on the transmitter circuits tends to cause slight variations in the instantaneous equivalent frequency which are known as "frequency modulation" or "phase modulation" depending on their character. To prevent this effect the control oscillator must be carefully guarded against reaction by shielding and balancing of circuits and the design must be such as to preclude variable phase shifts due to modulation in subsequent circuits of the transmitter. It is apparent that if there are two paths of different lengths, two components which arrive simultaneously at the receiver may have left the transmitter several thousandths of a second apart. If the transmit- ter frequency has changed materially during this brief interval trouble may be expected. The trouble actually takes the form of a distortion of the speech as demodulated by the receiving detector.^ Defects in short-wave transmission due to radio noise, minor varia- tions in attenuation, fading, and distortion are nearly always present to some extent and, when any or all are severe, cause a certain amount of lost service time. These interruptions are of relatively short duration and, furthermore, there is enough overlap in the normal times of useful- ness of the several frequencies available, so that shifting to another frequency may give relief. There is, in addition, a kind of interruption which from the standpoint of continuity of service is more serious. At times of disturbance of the earth's magnetic field, known as " magnetic storms," short-wave radio transmission is generally subject to such high attenuation that signals become too weak to use and sometimes too weak to be distinguishable. These periods affect all the wave- lengths in use and may last from a few hours to possibly as much as two or three days in extreme cases. They are followed by a recovery period of one to several days in which transmission may be subnormal. Severe static may cause interruption to both long- and short-wave services at the same time but the short waves are relatively less affected by it and are usually able to carry on under static conditions which 3 For a discussion of this phenomenon see "Some Studies in Radio Broadcast Transmission" by Bown, Martin, and Potter, 7. R. E. Proc, Vol. 14, No. 1, p. 57. TRANSOCEANIC TELEPHONE SERVICE 265 prevent satisfactory long-wave operation. On the other hand severe fading or the poor transmission accompanying a magnetic disturbance may interrupt short-wave service without affecting the long waves adversely, — in fact magnetic disturbances often improve long-wave transmission in the daytime. The service interruptions on the two types of circuits are thus nearly unrelated to each other and have no definite tendency to occur simultaneously. This is the principal reason why both long-wave circuits and short-wave circuits appear essential to reliable radiotelephone service. On routes which are very long or which cross tropical areas which result in static sources facing the directive receiving antennas, long + 10 + 5 .0 in X o z If) -10 o y -15 >-20 I- < Ml a. -25 -30 -36 r\K\ IQ ayi Kkt- t 1& kJlC-r(rDC\ nCTAI M 1 Xr\ KJF\A/ o- o •- ■• SOUTHGATE, ENGLAND. TRANSMISSION ON60KC (5000 METERS) ROCKY POINT, L.I. TO CUPAR, SCOTLAND. HORIZONTAL COMPONENT OF EARTH'S MAGNETIC FIELD. k i L r ^^ 1 , J ^, c / A A 1 ! / V >-— — < ^ / y \ \ \ 1 ^ V \ \ I A y r- \ L \ \ I \\ Y y / / / 1 — < _,-' <' • / / f / / / / / / / / r / 1 r 18,740 -5^ < ■2. 2 18,730 < sl£- Q 18,720 H 18.710 p UJ z o 18,700 < 18,690 "i I- < 5 4 3 2 HAYS BEFORE 0 I STORM 3 4 5 6 DAYS AFTER 18,680 , 18,670 ^ z o a 18,660 2 O O 18,6501: Fig. 2 — Effect of magnetic disturbances on radio transmission. waves cannot as yet be successfully employed and short waves alone are available. However, experience tends to indicate that on North and South routes such as between North and South America, the interruptions associated with magnetic storms are less severe and of shorter duration. The cycle of events which accompanied a particularly severe mag- netic storm ^ in July, 1928, is shown graphically in Fig. 2. The light ■* Data regarding other magnetic disturbances are given in a paper by C. N. Ander- son, entitled "Notes on the Effect of Solar Disturbances on Transatlantic Radio Transmission," /. R. E. Proc, Vol. 17, No. 9, September, 1929. 266 BELL SYSTEM TECHNICAL JOURNAL NEW YORK TO LONDON 10 L n ONDON TO NEW YORK n i 1 miM ^^■tii 1 ' ^Hiii>y ^■tii II r n I^^^^^B^Hj [■ A. H Hi 1 IH Jill UJ 5 H| 1 II i^lii ^Hil ^^Hflili ^^^^■iHi ^^^^^^Hi nil ■{ I^^^^^^^IH '^^^^^I'^l'^^^li^l ^H 1 H Q B ■■ 1^1' i^^^^l tr S 2 Z < Jo 12 NOON Z CC HI 1- 10- 1 < n^ 1 8- ■ III! 1 1 i^^^^^^^l ■ M II ' i ■ 1 II 1 4 JUNE JULY 1929 AUGUST SEPT. COMMERCIAL ^ — KEY — :=3 UNCOMMERCIAL c =3 NO DATA Fig. 3 — Chart showing transmission performance of a short-wave transatlantic telephone circuit. TRANSOCEANIC TELEPHONE SERVICE 267 dotted curve shows the variation in the horizontal component of the earth's field. The heavy solid line follows the daily averages of the short-wave received signal field. It is apparent that the disturbance took two days to reach its peak and the recovery to normal took nearly a week. The heavy dotted line shows received field on long waves (60 kilocycles) and indicates that transmission was improved slightly at the same time the short waves were suffering high attenuation. The experience with transatlantic telephone service on short waves covers a period of nearly three years, there having been available a one- way channel from the United States to England used as an emergency facility for the first year and a half, a two-way circuit for the next year, and two circuits since June, 1929. It is only in this later period, how- ever, that a circuit has been available operating regularly with the amounts of transmitter power and antenna directivity which have been mentioned. The performance of the two one-way channels forming this circuit is charted in Fig. 3. The charts are plotted between hours of the day and days in the year so that each unit block represents one hour of service time. The solid black areas are time in which commercial operation could be carried on. The dotted strips are uncommercial time. The blank areas are for time in which, for one reason or another, the circuit was not operating and no data were obtained. Perhaps the most outstanding feature of these charts is the tendency of the lost time to fall in strips over a period of two or three days. These strips coin- cide approximately for both directions of transmission. The principal ones are about July 10 and 15 and August 2 and 17. These are charac- teristic of the interruptions accompanying magnetic disturbances of the kind which occur at irregular intervals of a few days to several weeks. They are, of course, not as severe as the disturbance illustrated in Fig. 2. It is apparent that for these three summer months this new circuit gave a good account of itself and furnished commercial transmission for something like 80 per cent of the time that service was demanded of it. In these same months the long-wave system suffered its greatest difficulty from static, and we have concretely illustrated the mutual support which the two types of facilities give each other. It should not be inferred from these data that the short-wave trans- atlantic radio links furnish 80 per cent of the time talking circuits as stable and noise free as good wire lines. Under good conditions they do provide facilities which compare favorably with good wire facilities. On the other hand they may at times be maintained in service and graded "commercial" under conditions of noise or other transmission defects for which wire lines would be turned down for correction, since 268 BELL SYSTEM TECHNICAL JOURNAL ^^ ^-^^ (v- NETCONG RECEIVING STATION -^^^ ■^^ PATERSON O MORRISTOWN O NEWARK O BOUNDBROOK O NEW BRUNSWICK O LAWRENCEVILLE TRANSMITTING STATION Fig. 4 — Map showing transmission considerations affecting location of stations. TRANSOCEANIC TELEPHONE SERVICE 269 the obviously undesirable alternative is to give no service at all until conditions have improved again. The present development effort is largely directed toward improvements which will insure not only a greater degree of reliability against interruptions but which also will improve the grade of service as a whole. In the foregoing little has been said about the stations and plant since a description of these and the operation of them are treated in two companion papers by Messrs. Cowan and Oswald. It may be well, however, to view the physical scene broadly as set forth on the accom- panying map, Fig. 4. The geographical arrangement of the transmitting and receiving stations was governed among other things by transmission considera- tions. The two stations were placed about 50 miles apart because this is approximately the distance for minimum signal and at a lesser or greater distance the signals from the American transmitter might be strong enough to offer some interference to receiving the English or South American stations on adjacent channels. For the same reason they were placed at considerable distances from the transmitters and receivers of other communication agencies. The Netcong receiving station lies to the north of the Lawrenceville transmitting station so as not to be in paths of strong signals from the directive antennas which face northeast toward England and southeast toward South America. This configuration also places the transmitter outside the sensitive angles of the directive receiving antennas. 18 Transoceanic Telephone Service — Short-Wave Equipment A. A. OSWALD! The application of short-wave radio transmission to transoceanic tele- phone circuits is developing apparatus and stations designed specifically to meet the needs of these services. This paper describes from the radio point of view the important technical features and developments incorporated in the new transmitting and receiving stations of the American Telephone and Telegraph Company located respectively at Lawrenceville and Netcong, New Jersey, and it outlines some of the radio problems encountered in the station design. ***** SHORTLY after transatlantic telephone service was opened in January, 1927 the long-wave radio circuit between New York and London was supplemented, first by an experimental short-wave radio link in the west-east direction and later by a short-wave link in the east-west direction.^ From this beginning, as an auxiliary to the long- wave circuit, the short-wave system has been improved steadily so that its average performance throughout the year now more nearly ap- proaches that of the long-wave system and it has become an important part of the transoceanic facilities. The relative merits of the two sys- tems, their combined usefulness, and their transmission features are the subject of another paper and will not be discussed here. For the present purpose it will be sufficient to note that there are now in opera- tion between New York and London, one long-wave and three short- wave two-way circuits and that within a few weeks a short-wave circuit will be available between New York and Buenos Aires. The radio transmitting units for the New York end of these four circuits are located at the new station which the American Telephone and Telegraph Company has recently established at Lawrenceville, New Jersey. The receiving units are concentrated at Netcong, New Jersey. The factors entering into the selection of these station loca- tions are outlined in another paper and therefore need not be men- tioned further. This paper is limited in scope to a necessarily brief description of the transmitting and receiving systems and apparatus, a discussion of technical features in the station layouts, and an outline of the major problems encountered in the station design. Comprehen- sive treatment of individual units is properly left for other entire papers. It will be convenient to deal with the transmitting and receiving sta- ^ Presented at the Winter Convention of the A. I. E. E., New York, N. Y., Jan., 1930. a O. B. Blackwcll, A. I. E. E. Journal, May 1928. B. S. T. J. April 1928. 270 TRANSOCEANIC TELEPHONE SERVICE 271 tions separately and in each case to consider briefly the system and apparatus of one channel before describing the general station plan. Transmitting System The four channels at Lawrenceville are equipped with independent transmitters using certain auxiliary apparatus in common. Each channel involves a radio transmitter with its associated power plant and wire equipment, and a group of directive antennas designed and adjusted for the specific wave-length assignments of the channel. The general method of transmission, with the exception of directional sending, is the same as that employed for program broadcasting sta- tions in that the radiated signal contains the carrier and both sidebands. Systems in which one or more of these components are suppressed at the transmitter appear to offer further means of improving short-wave transmission, and the necessary apparatus for the practical application of such systems when operating at frequencies in the order of 20,000 DIRECTIONAL ANTENMA TELEPHONE LINE LINE F TERMINA AND iEPEATEF DON Iv ONITORIN CONTROL DESK JG 1 i "WO STAC AUDIO ^MPLIFIEF ' TRANSMISSION LINE.^ CONTROLLED OSCILLATOR J I RADIO F A rWO STAG -REQUENC MPLIFIEF^ E .Y POWER PIEZO-ELECTRIC OSCILLATOR HARMONIC GENERATOR*! HARMONIC GENERATOR *2 Fig. 1 — Block schematic of transmitting system. kilocycles is undergoing development. However, throughout the development of the transmitters as now installed at Lawrenceville the possibility of future major modifications in the method of transmission has been kept in mind. For this reason the modulator-amplifier system was adopted. In this system the signal which is to be radiated, is prepared by modulation processes at relatively low power levels and thereafter amplified the requisite amount. The amplifier and its power plant, representing a large proportion of the investment in equipment, can be continued in service with no appreciable alterations, even though the system of transmission and the modulating apparatus undergo radical changes. The general scheme of transmission is shown in Fig. 1. After passing through the line terminal and control apparatus, which includes 272 BELL SYSTEM TECHNICAL JOURNAL Standard repeaters, the voice currents are further amplified and em- ployed to modulate the plate voltage of an oscillator consisting of two 250-watt tubes connected in a push-pull circuit and oscillating at the frequency of the carrier which is to be transmitted. The frequency of such an oscillator, if not carefully controlled, will wander outside of the assigned frequency band, thus causing interference with other services and it will also suffer variations during the modulation cycle which contribute to fading phenomena encountered at the distant receiving station. In order to reduce these effects the oscillator is held in step at the desired carrier frequency by means of a second oscillator which is electrically removed from the reactions normally influencing and tending to vary the frequency of the controlled oscil- lator. Every precaution is taken to maintain accurately the frequency of the second oscillator and among other things it is governed by a piezo-electric quartz crystal whose temperature is regulated closely. Since it is impractical to use crystals cut sufficiently thin to oscillate directly at frequencies in the range 10,000 to 20,000 kilocycles, thicker crystals of lower frequency are used in combination with harmonic generators which multiply the crystal frequency first by two or three and then by one or two as the case requires. By virtue of the wide differences between the input and output frequencies of the harmonic generators these intermediate steps tend to isolate the crystal oscillator from the other radio circuits and thus aid in stabilizing the frequency. The modulated radio frequency output of the controlled oscillator is applied to the grids of a two-stage power amplifier employing water- cooled tubes designed for operation at these frequencies. The first stage contains two tubes and the second stage contains six. The tubes are arranged in push-pull circuits, the entire system being care- fully balanced to ground. The carrier output power from the last stage is 15 kw. With 100 per cent modulation this corresponds to 60 kw. at the peaks of the modulation cycle. In other words, a radio tele- phone amplifier of this type, rated at 15 kw. when provided with a sufficiently large d-c. power source, could be used as a 10,000-kilocycle continuous wave generator of 60 kw. capacity. The radio signal delivered by the amplifier is conveyed to the antenna by means of a 600-ohm open wire transmission line. The antenna itself is both a very efficient radiator and a highly directive one. Transmitting Equipment At the transmitting station the apparatus for each channel comprises, (1) wire terminal equipment and repeaters, (2) a voice frequency control TRANSOCEANIC TELEPHONE SERVICE 273 desk, (3) the radio transmitting set containing the oscillators, modu- lators, and power amplifier, (4) a power control board, (5) rectifying apparatus and filters for supplying direct current at 10,000 volts, (6) motor-generators for providing various circuits with direct current, (7) water circulating pumps, tanks, and cooling units. The wire terminal equipment and repeaters at the transmitting station are standard units mounted on relay racks beside the voice frequency testing apparatus common for all channels. The voice frequency control desk provides facilities by which the attendant can monitor the incoming voice currents and the outgoing radio signal. Means are provided for observing the volume of these signals. Oscillators are provided for the purpose of quickly checking the performance of the system during line-up periods and for sending Morse signals over the radio link when required. The control desk Fig. 2- — Front view of short-wave radio transmitter of type used at Lawrenceville. is also equipped with apparatus for direct telegraph communication with the technical operator at New York. The radio transmitter consists of seven independently shielded units mounted on a common sub-base to form a single assembly, 4 ft. by 20 ft. by 7 ft. high. Some of the units are subdivided into several small shielded compartments. Very effective electrical screening or shield- ing between the various parts of a short-wave transmitter is essential. Otherwise stray fields introduce unwanted feedback couplings which produce distortion effects and spurious oscillations. Fig. 2 is a front view of the transmitter. Beginning at the left there are two units for speech amplification, one for radio frequency generation and modula- tion, one unit each for the first stage, the interstage circuit and the last stage of radio-amplification, and a double-sized unit for the output circuit. It is interesting to note that the over-all length of this as- 274 BELL SYSTEM TECHNICAL JOURNAL sembly is as much as five eighths of a wave-length at the highest fre- quency in its operating range, which is 9000 to 21,000 kilocycles. Each transmitter is required to operate at several assigned frequencies within this range and to change in a few minutes from one to another. This is done by changing coils and varying condensers in the oscillator and amplifier circuits and switching to different quartz crystals. Except in cases where two assigned frequencies are in harmonic relationship, it is necessary to provide a crystal for each of the frequencies. The crystals are mounted in an oven and continuously maintained at 50 deg. ± 0.05 deg. cent, by recording regulators. In order to avoid long inter- ruptions to service in the event of a crystal failure or other circum- stance requiring the opening of the oven and the subsequent re- establishment of temperature equilibrium, the ovens and crystals are provided in duplicate. The electrical problems which are encountered by the engineer designing a power amplifier for these high frequencies arise largely from the inherent stray or distributed capacities and inductances which are far less important at lower radio frequencies. For example, between the anodes of the amplifier circuit there exist capacities, which are composed of capacities within the tube itself, the direct capacities be- tween the tube water jackets, the mounting plates and the like. The total value of this composite capacity in the last stage is approximately 100 m.m.f. This value cannot be appreciably reduced by any change in design which now seems desirable. The reactance of 100 m.m.f. at 20,000 kilocycles is about 80 ohms. Thus the engineer is confronted at the outset with a generator (the tubes) which has an internal impe- dance in the order of 2000 ohms but across whose terminal is shunted inherently an 80-ohm reactance. Fortunately, this obstacle can be surmounted by introducing resonance effects but nevertheless it places very important limitations on the design of the associated circuits. These problems become more difficult with increase of either power or frequency. Increase in power requires higher voltages and currents and thus larger elements, spaced farther apart. The augmented bulk increases both stray capacities and unwanted inductance of leads. Higher frequencies increase the magnitude and therefore the relative importance of these effects. The power control board has nine panels equipped with the necessary instruments and apparatus for controlling and distributing all power to the transmitter. The motor-generators, pumps, fans, oil circuit breakers, and other apparatus are remotely controlled from this point. A system of relays and signal lamps provides protection and indicates the location and general nature of any trouble. With the exception of TRANSOCEANIC TELEPHONE SERVICE 275 the application of high-voltage direct current, the entire system starts up and shuts down in the proper sequence in response to the manipula- tion of a master control switch. Direct current at 10,000 volts is supplied to the anodes of the power amplifier tubes by a transformer and rectifier using six standard two- electrode thermionic tubes. The rectified current is filtered separately for each stage of the amplifier. This is necessary to prevent distortion by interstage modulation caused by the common impedance of the rectifier. Effects of this nature become important as the requirements placed on unwanted modulation products become more stringent. Transmitting Antennas The antennas at Lawrenceville all have comparatively sharp direc- tional properties. Such antennas are readily realized when dealing with radio waves of very short wave-lengths. Although the funda- mental principles involved in producing these directional effects have been known for many years, economic limitations effectively prevented their application to transmitting antennas for long wave-lengths. These limitations are altered immensely in the case of antennas for short wavelengths and, when the useful propagation properties of short waves became known, great stimulus was given to the development of antennas for directional sending and receiving. The same type of antenna can be used, of course, for both purposes but, since the objec- tives when sending and receiving are somewhat different, the tendency has been to develop arrangements adapted to each case. Directional transmission is a very large subject and will only be touched upon suflficiently to describe in a very general way the anten- nas at Lawrenceville. There are many possible arrangements and combinations and the engineers must choose from these the ones most suitable for their purpose. In general all of the schemes depend upon producing interference patterns which increase the signal intensity in the chosen direction and reduce It to comparatively small values in other directions. One of the methods of obtaining a sharply directive characteristic is to arrange a large number of radiating elements in a vertical plane array, spacing them at suitable distances and interconnecting them in such a manner that the currents in all the radiating members are in phase. A simple way of accomplishing this result and the one which is now being employed at Lawrenceville depends upon the manner in which standing waves are formed on conductors. It is generally known that current nodes and current maxima will recur along a straight conductor whose length is an exact multiple of one half the wave-length 276 BELL SYSTEM TECHNICAL JOURNAL of the exciting e.m.f. and that the phase difference between successive current maxima is 180 deg.^ Such a conductor when folded in a vertical plane as shown in Fig. 3 and with its length adjusted slightly to com- pensate for the effects of folding, satisfies the aforementioned require- ments for producing directional radiation. The arrows in Fig. 3 indi- cate the relative directions of current flow and the dotted line indicates Fig. 3- — Conductor bent to form one section of simple directive antenna. The type used for transmitting at Lawrenceville. the current amplitudes along the conductor. It will be noted that the instantaneous currents in all the vertical members are in the same direc- tion and that in the cross members their directions are opposed. Due to these current relations and the physical positions of the elements, the cross members radiate a negligible amount of energy whereas the vertical members combine their effects for the directions perpendicular 3 This assumes of course that the conductor is in space free frorn objects affecting its electrical properties and that the ends are free or properly terminated to produce reflections. TRANSOCEANIC TELEPHONE SERVICE 277 to the plane of the conductor. In other directions destructive inter- ference reduces the radiation from the vertical members. The system is equivalent to four Hertz oscillators driven in phase, and arranged in two groups one half wave-length apart, the two oscillators of each group being placed one above the other. Both computation and experiment have shown that with this system of radiation there is an improvement of approximately 6 db. In other words the same signal intensity in the chosen direction is obtained with one fourth of the power required by a one-element radiator. A second similar conductor system placed directly behind the first in a parallel plane one quarter wave-length away, will be excited parasitically from the first conductor and will act as a reflector, thereby creating a unidirectional system. It has been found that the reflector further reduces by 3 db the power required to maintain a given signal intensity in the desired direction, thus bringing the total gain for the system up to 9 db. This is also in agreement with the theoretical computations. It is obvious that the system in Fig. 3 can be extended vertically to include more radiating elements by increasing the length of the con- ductor and it can be enlarged horizontally by placing several units alongside each other, care being taken to obtain the desired phase rela- tions by transmission lines of the proper length. In this way large power savings may be effected. At Lawrenceville the maximum gain is about 17 db (a power ratio of 50) over a vertical halfwave oscillator. The enlarged system lends itself readily to mechanical support and forms so-called exciter and reflector "curtains" which are suspended between steel towers appropriately spaced. Aside from other con- siderations, which will be mentioned in connection with station layout, the size of the antenna is influenced by the complex and variable nature of the wave propagation through space. At present this determines the degree of directivity which is most useful for the average conditions.^ The closed loops of each unit corresponding to Fig. 3 greatly facilitate the removal of sleet. In addition to loading the antenna mechanically, ice, having a dielectric constant of 2.2 at these high frequencies, ad- versely affects the tuning. At Lawrenceville sleet is removed by heat- ing the wires with current at 60 cycles. This is accomplished without interfering with the service by employing one of the less familiar properties of a transmission line. The same property also is used to effect impedance matches wherever the transmission lines are branched. If a line, exactly one quarter wave-length long, of surge impedance Z^ is terminated with a load Z^, the sending-end impedance Z, is equal to *J. C. Schelleng, "Some Problems in Short Wave Telephone Transmission." Presented to the Institute of Radio Engineers at a meeting Nov. 6, 1929. 278 BELL SYSTEM TECHNICAL JOURNAL Zg-jZji. If Z^ is a pure resistance the sending-end impedance is a pure resistance. Hence a quarter wave-length line may be used to connect two circuits of different impedances and these impedances may be matched by controlling the value of Zg either by varying the diam- eter of the conductors or their spacing. Likewise, if Z,, is fixed and Z^ is made very small, then Zg will be extremely large. 60 CYCLE SOURCE RADIO TRANSMITTER Fig. 4 — ^Antenna sleet-melting circuit. In Fig. 4 two units of the type shown in Fig. 3 are excited through transmission lines 1 and 2 of equal length in order to give the correct phase relations in the radiating elements. The lines are joined in parallel by condensers of low impedance at radio frequencies and they are connected in series for 60-cycle currents by the quarter wave-length line A which, being short-circuited at the one end, presents a very high impedance to radio frequency currents at the other end and therefore behaves like an anti-resonant circuit. The quarter wave-length line B serves as a transformer and is adjusted to match the impedance at the junction of lines 1 and 2 with that of the radio transmitter. The quarter wave-length line C is effectively short-circuited for radio fre- quencies by the condenser D and acts the same as A. These quarter TRANSOCEANIC TELEPHONE SERVICE 279 wave lines consist of short lengths of pipe mounted on frames under the antenna curtains as shown in Fig. 5. Transmitting Station Among the first radio problems encountered in the design of a trans- mitting station for several channels are those concerning the size, shape, and number of antennas, their directions of transmission, their Fig. 5 — Section of antenna system at Lawrenceville, showing lower portion of curtains and quarter wave transmission lines used as transformers and anti-resonant circuits. relative positions from the point of view of mutual interference, and their grouping around the transmitters. The number of antennas required for each channel is determined by the hours of operation and the average grade of service which the sys- tem is expected to render. For service covering a large portion of each day several wave-lengths are necessary. Transmitters Nos. 1, 3, and 4 at Lawrenceville each are assigned three frequencies. No. 2 has five assignments in order to improve the likelihood of at least one channel being available throughout the entire day at all seasons. The size and shape of the antennas are, of course, determined by the directivity wanted, by the type employed, the frequency assignments, and by considerations of cost. They are governed also by the necessity of connecting several antennas to the same transmitting set. This involves both the spacing and arrangement of antennas to avoid 280 BELL SYSTEM TECHNICAL JOURNAL adverse mutual reactions and it requires that attention be given to the losses in the connecting transmission lines, which are by no means negligible. Operating economies suggest concentrating all the trans- mitters at one point but the cost per kilowatt hour of modulated high-frequency power must be taken into account when considering the use of long transmission lines. It should be recognized, of course, that in the early applications of a comparatively new art, it is impossible to approach anything like accurate evaluation of all the factors entering into economic balances and furthermore very considerable weight needs to be given to the probable future trend of developments. M4.4M TRANSMITTING STATION AMERICAN TELEPHONE & TELEGRAPH CO LAWRENCEVILLE N.J. 32.7 M GREAT CIRCLE TO LONDON 30.7 M 20.7 M 5.6 M Fig. 6 — ^Arrangement of antennas at Lawrenceville transmitting station. At Lawrenceville all of the antennas for the three channels to Eng- land are arranged in a straight line about one mile long. The direction of this line is perpendicular to the great circle path to Baldock, England, where the signals are received, (Fig. 6). The antennas for the fourth channel are similarly arranged in a line 1500 ft. long and they are directed for transmission to Buenos Aires, Argentine. Placing several antennas in a single line reduces the cost of the sup- porting structure, and all the antennas have a clear sweep in the direc- tion of transmission. By locating them in proper sequence with re- TRANSOCEANIC TELEPHONE SERVICE 281 spect to wave-lengths It is possible without objectionable interference, to place the antennas end-to-end and thus use supporting towers in common. Due to the wide difference in wave-length between adjacent antennas and their right-angle position with respect to the line of transmission, their proximity has no appreciable effect different from that of the towers. The proper selection of tower spacing in respect to wave-lengths makes it possible to erect a uniform supporting struc- ture. This has the advantage of flexibility and will permit future alterations of either the location or size of a given antenna. At present, each antenna occupies the space between three towers. In order to avoid undue loss In the transmission lines the radio trans- mitters are grouped In two buildings. The buildings each contain two transmitters and are identical In layout, in so far as the radio equipment is concerned. Building No. 1 has additional space for the central wire terminating and testing equipment. This apparatus is contained In an electrically screened room which effectively prevents high-frequency fields from interfering with the proper functioning of the apparatus. Receiving System Short-wave reception is characterized by less difficulty with static than that encountered with long waves. On the other hand it suffers interference from sources such as the ignition systems of passing air- planes and automobiles, which ordinarily do not disturb long-wave systems. Frequently the Incoming radio waves suffer wide and rapid swings in Intensity and there are variations in the apparent direction of arrival. On account of the extremely high frequencies the ap- paratus and antenna structures are very different from those for the long waves; otherwise the general schemes of reception are similar, directional effects and double detection methods being employed for both. The radio wave is collected by means of a directional antenna array whose prime function is to improve the ratio between the desired signal and unwanted noise or other Interference. This It does in two ways: viz., (1) by increasing the total signal energy delivered to the receiver and (2) by discriminating against waves whose directions of arrival differ from the chosen one. Increasing the total energy collected from the incoming message wave permits the detection of correspondingly weaker signals because there is an apparently irreducible minimum of noise inherent to the input circuits of the first vacuum tube in the receiver '" and this noise establishes a lower limit below which signals cannot be received satisfactorily. Since, under many conditions, the ^ J. B. Johnson, Physical Rev., July 1928. 282 BELL SYSTEM TECHNICAL JOURNAL directions of arrival of static and other disturbances including unwanted radio signals are random, it is obvious that sharp directive discrimina- tion aids very materially in excluding them from the receiver. On the other hand, the antennas are not sharply resonant systems and they do not distinguish between waves from substantially the same direction and closely adjacent in frequency. This duty is left to the circuits of the radio receiver. BEATING OSCILLATOR DIRECTIONAL ANTENNA TRANSMISSION ^ LINE LINE TERMINATION AND REPEATER FIRST DETECTOR SECOND DETECTOR TWO STAGES RADIO AMPLIFIER INTERMEDIATE FREQUENCY AMPLIFIER AND FILTER TELEPHONE LINE AUDIO AMPLIFIER AND MONITORING APPARATUS AUTOMATIC TWO STAGES GAIN INTERMEDIATE CONTROL FREQUENCY AMPLIFIER Fig. 7- — Block schematic of receiving system. Having collected the signal with a directional antenna the energy is conveyed to the receiving set by means of concentric pipe transmission lines of small diameter. The use of concentric conductors simplifies the prevention of direct signal pick-up by the lines, it reduces losses and prevents external objects from influencing the transmission proper- ties, thus allowing the line to be buried in the ground or placed a few inches above the surface where it will have no appreciable adverse effect on the antenna performance. Referring now to Fig. 7, the radio currents arriving over the trans- mission line are first amplified by two stages of radio amplification involving tuned circuits which discriminate further in favor of the wanted signal. The signal delivered by the radio amplifier is at a suitable level for efficient demodulation and is applied to the grid of the first detector. By means of a beating oscillator whose frequency is suitably adjusted, the first detector steps the signal carrier frequency down to a fixed value of 400 kilocycles from one in the range 9000 to 21,000 kilocycles which depends, of course, on the distant transmitting station assignment. The intermediate frequency signal at 400 kilo- TRANSOCEANIC TELEPHONE SERVICE 283 cycles then passes through a combination of amplifiers and filters which further exclude the unwanted interference. The wanted signal reaches the second detector where it is demodulated and the voice currents reproduced. The latter are then amplified and applied to the telephone lines. A portion of the output from the intermediate amplifier which would normally go to the second detector grid, is diverted and further am- plified. It is then supplied to a device which automatically tends to maintain the receiver output volume constant by controlling the bias potential of the first detector grid circuit. The time constants are adjusted so that this gain control does not respond to the normal varia- tion in signal power corresponding to speech modulation. Otherwise of course, there would be serious distortion effects. This device partially offsets the ill effects of wide fluctuations in signal intensity but it does not overcome the deterioration in signal quality which usually accompanies the low field strengths during such fluctuations. Receiving Equipment At the receiving station the apparatus for each channel comprises (1) the radio receiving set, (2) a power plant for the receiver, (3) wire terminating equipment and repeaters. The latter are located at a central point in the station along with certain voice frequency testing apparatus used in common by all channels and supplied with power from a common source. A radio receiving set which embodies the above described system and of the type installed at Netcong is shown in Fig. 8. It consists of a large number of individually shielded units mounted on panels and assembled on three self supporting racks of the type commonly em- ployed in the telephone plant. This permits the use without modifica- tion of certain standard pieces of equipment, such as jack strips, fuse panels, meter panels, audio frequency filters, and the like. It also permits the removal and repair or substitution of units with a minimum of delay. The set is required to receive signals at three fixed frequen- cies in the range 9000 to 21,000 kilocycles. This involves connections with three antennas through three separate transmission lines. The tuning of the antenna and transmission line terminations are rather lengthy processes requiring precise adjustments. In order to facilitate quick changes from one operating frequency to another without intri- cate tuning operations, the first stage of radio amplification is provided in triplicate and the switching is done between the first and second stage. Thus the antennas are permanently connected to the set and their adjustments remain undisturbed. The circuits of the second 284 BELL SYSTEM TECHNICAL JOURNAL ca u nS CQ > c o 1-4 > <1> CJ (U o 13 I +-> O in bO S TRANSOCEANIC TELEPHONE SERVICE 285 stage require tuning when the frequency is changed. Hence to tune the receiver on any one of the assigned frequencies the attendant merely moves the dials of the second stage to predetermined settings, switches the grid circuit to a first stage which is already tuned and connected with the proper antenna and he adjusts the beating oscillator to obtain an intermediate frequency of 400 kilocycles. Screened grid tubes are used for the first two stages of amplification. A key shelf is provided with telephone and telegraph facilities. The power plant consists of standard 24-volt and 130 batteries, rectifier charging units and automatic regulators. Receiving Antennas In discussing antennas for directional sending it was mentioned that an identical antenna could be used for receiving purposes, but since the requirements in the two cases are not the same, quite dif- ferent structures have been developed, although the methods of ob- taining directivity are alike. In the sending case the reduction of random radiation ceases to be profitable when the increment thus added to the energy, which is radiated in the direction of the distant receiving station, is a relatively small part of the total. In the receiving case, although the response to the wanted signal may not be increased appreciably by further improvement in the directive pattern, the reduction in noise and interference from random directions justifies additional improvement. Expressed another way, the objective in the transmitting case is a high gain compared to a nondirectional antenna, whereas in the receiving case the objectives are, first, a high average signal-to-noise ratio and, second, a gain sufficient to override the noise inherent to the receiving set. Satisfying the first accomplishes the second. Improvement of the average directional discrimination means a nearer approach to ideal conditions. Whereas steel towers, section- alized cables, guys and the like, when properly located relative to the conductors of a sending antenna, do not cause any appreciable power loss, their presence near the receiving antenna may prevent the realization of the extreme directive properties which are wanted. Moreover, there is need for much greater rigidity in the positions of the conductors. For this reason the antennas at Netcong are supported on wooden frames constructed like large crates. Due to the variable conditions surrounding the propagation of short waves in space, the vertical angle of arrival of the signal wave at the receiving station frequently changes considerably throughout a twenty- four hour period and is not always the same from day to day. In 19 286 BELL SYSTEM TECHNICAL JOURNAL order to combat this variable condition, it appears desirable to select an antenna arrangement which does not have sharp directional proper- ties in a vertical plane passed through the horizontal direction of arrival. The type of antenna selected for Netcong meets this requirement by having only a single horizontal row of quarter-wave vertical elements in one plane. Another solution, of course, would be to provide several antennas of different characteristics and to shift about from one antenna to another as the conditions warranted. Fig. 9 is a general view of one of the Netcong receiving antennas. Like the transmitting antennas, the conductors are arranged in two parallel planes one quarter wave-length apart in order to obtain a unidirectional system. The conductor in each plane is bent and ter- Fig. 9 — One of receiving antennas at Netcong. (24.7 meter wave-length.) minated as indicated in Fig. 10 but is much longer than that shown. The vertical members are marked A. As in the transmitting case the directional effect depends upon the manner in which standing waves occur along the conductor. A signal wave arriving broadside to the array, induces voltages in the vertical members which are identical in phase and amplitude. Because the vertical members are interconnected alternately at the top and bottom by members of one quarter wave-length and the last horizontal members are one eighth wave-length, the net effect of the induced voltages is the establishment of standing current and voltage waves along the conductor. The receiver is connected at a voltage anti-node and the current which flows through it is proportional to the sum of the voltages induced in the vertical members. In the case of TRANSOCEANIC TELEPHONE SERVICE 287 a signal wave arriving from the horizontal directions parallel to the plane of the array, the voltages in the vertical members are in succes- sive quarter-phase relationships, no standing waves are produced, and no current flows through the receiver. Because current nodes occur at the center of each horizontal member, the loss by reradiation from these members is negligible. This is an important feature which contributed to the selection of this type of antenna for Netcong. The size of the antenna is determined largely by the manner in which the signal waves arrive although costs cannot be wholly neglected. The useful length is limited by the fact that random fading occurs at distances as short as ten wave-lengths and it is doubtful if an antenna this long would realize the computed improvement. The cost per decibel gained is small for the initial steps, but it mounts very rapidly as the length of antenna increases. The height also is limited by cost Vs ^ r \ \ 1 a! ^^^^^n^ CURRENT NODE %. VOLTAGE ANTI-NODE t %» CURRENT NODE ^8 Fig. 10 — Diagram of simple directive receiving antenna. and by the necessity of allowing for considerable variation in the verti- cal angle of arrival as discussed in a previous paragraph. The antennas at Netcong are six wave-lengths long and the lowest conductors are about 10 ft. off of the ground. The gains over that of a half wave vertical antenna are in the order of 16 db (power ratio of 40). The average improvement in signal-to-noise ratio is of the same order. There are certain null points toward the sides and rear for which the ratio of directional discrimination is very large. The transmission lines are constructed of inner and outer copper tubes respectively 3/16 in. outside diameter and 5/8 in. inside diameter The tubes are held concentric by torroidal shaped insulators made of Isolantite, a ceramic product similar to porcelain and well adapted for high-frequency voltages. This same material is used for insulating purposes throughout the transmitting and receiving antennas. Trans- mission lines are supported a few inches above the ground and are connected to earth at short intervals. The lines vary in length from 288 BELL SYSTEM TECHNICAL JOURNAL 200 to 1500 ft. One of the interesting problems in connection with their design is the provision of means for allowing variation in length with temperature. Ordinary expansion joints introduce difficulties with electrical contacts and impedance irregularities. To avoid these the lines are made 10 per cent longer than otherwise necessary and they follow a sinuous course which permits the necessary bending. Sharp turns are not permissible because experiments have shown that they cause reflection disturbances. The measured loss in 1000 ft. of line at 20,000 kilocycles is 2 db. Receiving Station The radio problems encountered in the layout of the receiving sta- tion, in general, include most of those already mentioned in connection with the transmitting station, but their solution in some instances is quite different. In addition there are requirements imposed by sources of radio noise both within the station itself, and in the surrounding area which is beyond the control of the station. The number of antennas is determined, of course, by the frequency assignments of the distant transmitting station. Where two assign- ments are within 100 kilocycles it is possible to use the same antenna for both, but thus far, this has not been done at Netcong. The size of the antennas is not limited appreciably by the length of transmission lines because other factors make it necessary to separate them rather widely. On this account and also because the receiving apparatus and its power plant are small, comparatively inexpensive units, it is economical to place the receivers in small buildings centrally located with respect to the group of antennas for one channel. In this case the lengths of transmission lines are not controlling factors and the dimensions of antennas are governed primarily by the considerations previously outlined when describing the individual antenna. The small height of the antenna permits them to be placed in the line of reception of other antennas spaced ten wave-lengths or more away and of widely different frequencies such as those of one channel. Antennas adjusted for the same order of frequency are separated more than this. On the other hand, to avoid adverse reactions no two are placed adja- cent and end-to-end as at the transmitting station. The end-to-end separation at Netcong is in the order of four wave-lengths. The areas surrounding antennas are cleared of trees and kept free of all overhead wires or conducting structures to avoid reflection effects which disturb the directional characteristic of the antenna systems. The locations of antennas are also influenced materially by the neces- sity of avoiding interference from the ignition systems of internal TRANSOCEANIC TELEPHONE SERVICE 289 combustion engines. This imposes a requirement that the station site be isolated from air routes and roads carrying heavy traffic. The antennas are placed as far as possible from secondary roads which cross their line of reception. The layout at Netcong is shown in Fig. 11. There are thirteen antennas arranged in four groups with a receiver building for each group. A headquarters building located at the road entrance contains the wire terminating equipment, line repeaters, and voice frequency GREAT CIRCLE TO LONDON 42.9M CENTRAL TERMINAL BUILDING RECE>\;'ER 20.6 m 33.2M 16.0 M 303 M RECEIVER *2 / I4.2M GREAT CIRCLE TO BUENOS AIRES 207M Fig. 11 — Arrangement of receiving antennas at Netcong receiving station. testing apparatus. The power plant at each receiver and the entire central terminal apparatus at the headquarters building are placed in electrically shielded rooms to prevent radio noise disturbances emanat- ing from them and reaching the receivers directly or via the antennas. The radio stations described herein are pioneer commercial applica- tions in the development of short wave telephone transmission. Although progress has been rapid and far-reaching our knowledge of the behavior of short waves is by no means complete. It is reasonable, therefore, to expect that the future holds many improvements and that the information obtained by further fundamental investigations may materially alter both our views of the transmission phenomena and our ideas of what the apparatus and stations should be. The Words and Sounds of Telephone Conversations By NORMAN R. FRENCH, CHARLES W. CARTER, JR., and WALTER KOENIG, JR. This paper presents data concerning the vocabulary and the relative fre- quency of occurrence of the speech sounds of telephone conversation. Tables are given showing the most frequently used words, the syllabic struc- ture of the words, the relative occurrences of the sounds, and, for each vowel, the percentage distribution of the consonants which precede and follow it. Comparisons are made with the vocabulary and relative occurrence of speech sounds in written English. Introduction CONVERSATION resembles other forms of communication in its use of symbols, in themselves merely physical phenomena, but which combined in sequence are by convention endowed with meaning. The elementary symbols used in conversation are the acoustic dis- turbances called speech sounds. A language is characterized by the speech sounds which it uses and by the combinations of speech sounds which form syllables and words. The physical description of a lan- guage involves a statement of the characteristics of the individual sounds and also of the frequency of occurrence of each sound and combination of sounds. The latter or statistical aspect of conversation is treated in this paper. ^ Studies of the relative frequency of English speech sounds have been made previously, but they have been confined, so far as the writers have ascertained, to the analysis of written matter. Of these an extended investigation is that made by Godfrey Dewey. ^ For peda- gogical purposes in connection with difficulties in spelling and in developing methods of shorthand writing, which seem to have been the aims in the previous studies, written matter is the natural point of departure. There are obvious differences between English when read aloud from printed matter and English used as a medium of conversation, which might be expected to produce differences between analyses based on the two forms. Written matter is permanent and, to some degree, self- conscious; it receives qualification by dependent clauses and preposi- 1 Some of the results of this study were presented at the May, 1929, meeting of the Acoustical Society of America. See French and Koenig, JourjialA. S. of A., October, 1929, p. 110. ' " Relative Frequency of English Speech Sounds," Harvard Studies in Education, IV. Harvard University Press, 1923. 290 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 291 tional phrases and it makes use of synonyms and a vocabulary more or less ample according to the writer's fancy and ability. In conversation attention seems to be paid more to the thought than the form of ex- pression, with the exception, perhaps, that certain modes acceptable in writing may be considered as too formal for conversation. It is doubt- ful, however, that conversation should be described as more concise than written matter. The sentences are, indeed, likely to be shorter. They are often incomplete, in fact. But often in conversation even a single statement is completed only after a number of fumbling attempts, an oral manifestation of crystallizing thought, whereas in written matter the final expression alone would appear. In repetition of a thought, synonyms are less likely to be found in conversation than in written matter. Dependent clauses are less frequent than in written matter. Qualification and description often take the form of separate sentences, so that those words characteristic of involved construction tend to be less prominent in conversation, while the framework words, such as the auxiliary verbs and pronouns, are more intensively used. These differences, which tend to restrict the vocabulary, will be found reflected in the comparisons given later in this paper. The material for the present study was obtained from telephone conversations over typical toll circuits terminating in the city of New York. The process of noting the words of the conversations was carried out in the following manner: During one week the observer recorded nothing but the nouns used, during another week she re- corded only verbs, and during a third week only adjectives and adverbs. This routine was repeated until observations had been made on 500 conversations for nouns, 500 conversations for verbs, and 500 con- versations for adjectives and adverbs. Three other classes of words were recorded: prepositions and conjunctions, pronouns, and articles; but for these classes approximately 150 conversations in each case were judged to be sufficient. Certain classes of words were, for various reasons, omitted entirely. These are names, titles, exclamations, letters, numbers and the name- less sound which may be transliterated as "er" or "uh," so frequently punctuating a haltingly expressed sentence. A more comprehensive method, but based on a much smaller number of conversations, indi- cates that the ratio of the total number of occurrences of words in the omitted classes to the number of occurrences of the words discussed in this paper is about one to four. Within the omitted group the division is roughly as follows: proper names and titles, 20 per cent; exclamations and interjections, such as "yes," "no," "well," "yeah," "uh-huh," "oh," "all right," "hello," "good-by," laughter 292 BELL SYSTEM TECHNICAL JOURNAL and profanity, 40 per cent; letters and numbers, 25 per cent; and the sound "er," 15 per cent. The words which were obtained by the process of sampling con- versations for specific parts of speech are not, of course, identical with those which would have been obtained had the entire conversation been recorded. The representativeness of the most frequent words, which largely determine the relative frequency of the speech sounds, was investigated by a later test in which a different observer recorded the verbs from 250 conversations. These results will be discussed later, but it may be pointed out here that the word list obtained by the two observers corresponded so closely that it is felt that the samples of parts of speech were recorded with sufficient accuracy and were sufficiently large to justify taking the words obtained as a good repre- sentation of the main body of telephone conversation. The kinds of conversations encountered are shown in Table I. The great preponderance of business calls is reflected, as will be shown later, TABLE I Types of Telephone Calls on which Observations Were Made a. Material Business Calls 89.0 per cent All other Calls 11.0 per cent b. Speakers Two Men 86.5 per cent Two Women 10.4 per cent Man and Woman 3.1 per cent in the vocabulary. If a smaller percentage of the calls had been busi- ness in nature and if a larger percentage had been between women the vocabulary would probably have been different. Whether any marked change would have been found is open to some doubt when it is re- called that business may cover a wide range of topics and that in the 1,900 conversations from which samples were taken there may have been as many as 3,800 different speakers. Evidence will be given, however, which indicates that the relative frequency of the speech sounds would have been changed very little. Words The number of conversations on which observations were made was regulated to some extent by the ratio of the number of total words to the number of different words recorded in each class. In the early stages of observing many of the total words recorded were different, making this ratio low, but as the observations continued fewer and fewer new words were encountered. In Figure 1 curves are given THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 293 which show, for two classes of words, the way in which the number of different words in each class varied as the total number of words in that class increased. To take the nouns, of the first 200 about half were different, of the first 1,000 about a third were different, of the total of 11,660 nouns recorded about one tenth were different. An extrapolation of the curve indicates that the observations would need 2000 (/) Q CC O 5 I- z UJ a u u. o 1000 500 i 100 z 50 "'' v *^ ^'i> - ^^-' •aC >^^ fr .h ^ j>^ X fl b ^.^^f--^ ,/ r- ^ ^ r' f ^-^ ^ / y ^ ^ ^ ^ v' ^r .^ • y ^ ^ 100 1,000 10,000 TOTAL NUMBER OF WORDS IN EACH CLASS 100,000 Fig. 1 — The number of different words occurring in a given total number of words, for nouns and for verbs. to be increased tenfold from this point in order to double the number of different nouns. Approximately the same extension of the ob- servations would be required to double the number of different verbs. In neither case, however, could a material change in the relative oc- currence of the speech sounds be expected if the observations were so extended. This will be shown below. Table II shows the total number of words and the number of dif- ferent words for each part of speech separately. The verbs and aux- iliary verbs, which were recorded together, have been separated in the table. The numbers of total words for the other three minor classes have been found by multiplying the observed figures by the ratio of 500 to the actual number of conversations (about 150) on which observa- tions were made for these classes. The numbers of different words for these classes are not similarly increased since virtually all the possible different words were obtained in the observations. In finding the number of different words the various forms of the words, such as the plural form of the nouns, the different tenses of the verbs, and the comparative and superlative forms of the adjectives, have not been counted as separate words, although they were recorded and are 294 BELL SYSTEM TECHNICAL JOURNAL TABLE II Occurrence of Parts of Speech Parts of Speech Number of Words Ratio Total Total Different to Different Nouns Adjectives and Adverbs Verbs Auxiliary Verbs ... 11,660 9,880 12,550 9,450 17,900 12,400 5,550 1,029 634 456 37 45 36 3 11.3 15.6 27.5 255 Pronouns * Prepositions and Conjunctions * Articles * 398. 344. 1850. 79,390 2,240 35.4 * Derived from data on less than 500 conversations. treated separately in the analysis for speech sounds. An exception to this is that each form of the auxiliary verbs "be," "can," "may," etc., was counted as a separate word. It is of interest to find that of approximately 80,000 words so ob- tained, only 2,240, or less than 3 per cent, are different words. If each of the modifications of a word is counted as a different word the number of different words is increased to 2,822 ; but even on this basis less than 4 per cent of the total words are different words. Even among the nouns the number of different nouns is only a tenth of the total number of nouns. The five minor parts of speech shown in the last four lines of Table II form only 5 per cent of the different words and yet make up 57 per cent of the total words. The nouns, which constitute 46 per cent of the different words, contribute only 15 per cent of the total words. Such figures indicate clearly that conversation is based on a framework built up of a relatively small number of different words, arranged in many patterns, which supports the more variegated words which convey most of the meaning. A more detailed idea of this framework is given by Tables III-o and 1 1 1-6, which contain a list of the words which were observed in at least 1 per cent of the conversations. In Table Ill-a the words are arranged in order according to the total number of times they were recorded. This is approximately, but not quite, the same as the order of the num- ber of conversations in which they occurred as may be seen by exam- ining the numbers following each word. In Table 1 1 1-6 the same words are arranged alphabetically, for ease in reference. The list comprises 737 words out of the 2,240 different words recorded. The importance of the list lies in the fact, as will be shown later, that these words almost completely determine the relative frequency with which the elementary THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 295 Column A: Column B: TABLE Ill-a Word List — Numerical Order * Words which Occurred in One Per Cent or More of the 500 Telephone Conversations Analyzed Total number of times the word (or some form of it) was used. Number of conversations in which the word occurred. A B A B A B 3,990 I 467 336 what 193 170 anything 100 3,540 you 499 330 morning 191 170 my 97 3,110 the 496 326 an 178 2,060 a 487 321 just 211 168 night 107 2,046 on 458 317 over 208 159 call, n. 111 1,942 to 472 296 be 175 157 your 100 1,792 that 397 156 little 117 1,605 it 417 295 or 178 146 stuff, n. 92 1,506 is 419 295 take 207 146 won't 115 1,363 and 391 276 am 172 140 last, a. 106 274 come 168 140 she 50 1,360 get 393 274 make, v. 169 139 all 100 1,305 will, aux. 402 273 give 172 139 better 103 1,190 of 396 268 very 165 1,170 in 408 264 send 172 139 number, n. 80 1,115 he 297 262 as 125 138 out 90 1,100 we 294 259 right, a. 173 137 try 100 913 they 253 133 ask 101 887 see 328 247 order, n. 119 133 sell 81 883 have 367 243 good 149 131 not 96 823 for 330 241 minute 155 130 those 100 241 price 123 125 only 84 753 know 325 238 here 157 121 business 83 640 don't 301 234 car 88 120 office 83 638 do 302 230 had 151 618 are 293 229 time 165 118 late 94 599 want 297 228 can't 132 118 no, a. 77 597 go 280 226 much 160 117 all right 74 553 tell 264 115 pretty 92 518 with 263 224 there 144 115 shipment 80 496 me 283 222 week 120 113 back, a. 79 486 him 223 215 let 148 112 look, V. 85 214 letter 112 112 mean, v. 82 480 about 266 209 any 140 112 off 67 476 at 238 200 did 144 109 hear 85 474 think 232 199 more 134 473 this 240 195 didn't 142 108 ship, V. 68 458 day 251 193 talk, V. 131 108 way 81 418 thing 235 193 today 124 106 his 70 410 say 211 105 dollar 66 396 can, aux. 221 190 other 128 105 too 77 386 call, V. 200 186 company 111 105 wire, n. 78 379 would 207 186 fine, a. 122 104 haven't 83 184 could 124 104 then 88 370 them 170 183 same 127 103 how 78 358 was 194 179 put 114 103 who 74 339 now 216 178 wait, V. 135 338 from 196 176 has 114 98 buy 60 *In ambiguous cases the part of speech is denoted as follows: noun, n.; verb, v.; adjective or adverb, a.; auxiliary verb, aux.; preposition, prep. 296 BELL SYSTEM TECHNICAL JOURNAL TABLE Ill-a (Cont'd) A B A B A B 97 man 67 64 some 43 43 chance 37 97 wouldn't 79 63 been 53 43 coffee 8 96 before 79 63 but 54 43 every 36 96 first 71 63 contract 31 42 stock, n. 33 96 market 51 63 out of 25 42 than 33 93 something 67 63 sample, n. 37 92 month 70 63 these 57 41 feel 30 92 well, a. 71 40 different 30 89 case 47 61 few, a. 50 40 meet 28 61 ton 27 40 reason, n. 32 89 find, V. 72 61 train, n. 33 40 show, V. 30 88 by 75 60 best 52 40 which 33 88 probably 69 60 everything 47 40 yesterday 35 87 afternoon 62 60 may 50 39 pound, n. 21 87 line 60 60 thank 56 38 doing 34 87 name, n. 52 59 check, n. 35 38 keep 28 86 like, V. 71 58 along, prep. 33 86 sure 72 58 job 44 38 old 22 86 yet 67 37 awful 31 85 fellow 70 58 tonight 40 37 bag 24 58 up, prep. 42 37 certainly 28 85 pay, V. 55 57 home 35 37 difference 31 84 talk, n. 67 57 our 47 37 information 29 84 write 61 56 another 46 37 matter, n. 31 83 new 62 56 away 45 37 must 30 83 next 61 56 should 43 37 phone, n. 35 83 were 66 55 expect 49 37 seem 33 82 understand 63 54 around 54 82 when 69 54 copy, n. 36 36 boy 31 79 people 59 36 hand, n. 31 78 year 53 54 idea 38 36 hour 30 53 bad 46 36 house 27 77 us 63 53 couldn't 47 36 mind, n. 30 76 soon 63 52 bill, n. 39 35 early 26 75 place, n. 55 52 nice 38 35 figure, V. 24 73 money 56 52 tomorrow 36 35 oil, n. 18 72 guess, V. 63 52 word 45 35 question, n. 31 71 after 54 51 big 42 35 quite 32 71 hold, V. 60 51 where 42 71 through 46 51 whole 40 34 ahead 27 69 isn't 52 34 point, n. 27 69 leave 54 50 cent 31 34 wonder 28 49 figure, n. 32 33 offer, n. 20 68 coal 27 49 glad 39 33 speak 31 68 might, aux. 59 49 ship, n. 28 33 unless 33 68 work, V. 50 48 report, n. 29 32 bid, n. 24 67 again 62 47 suppose 41 32 deliver 24 67 her 30 46 into 42 32 less 23 67 its 67 45 boat 22 32 possible 24 67 so 53 45 couple 38 67 their 63 45 high 34 31 believe 25 66 long 55 31 check, V. 28 65 because 47 45 ought 37 31 low 22 45 trouble 33 31 situation 25 65 use, V. 50 44 barrel 20 31 touch, n. 29 65 work, n. 49 44 delivery 36 31 why 25 64 listen 55 43 anybody 40 30 basis 21 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION TABLE Ill-a (Cont'd) 297 A B A B A B 30 fix, V. 21 24 hope.^v. 19 18 steel 10 30 move 23 18 trip, n. 17 30 ready 25 24 near 23 18 wasn't 18 24 piece 16 17 above 13 30 receive 22 24 start, v. 22 17 accept 14 30 sorry 25 24 wrong 20 17 against 13 30 town 22 23 busy 17 17 amount 15 29 between 29 23 ever 22 17 appointment 14 29 does 27 23 foot 13 17 cable 10 29 dope, n. 24 23 lot 19 17 cover, v. 14 29 mail, n. 22 22 card 9 29 many 25 22 forget 18 17 definite 13 29 moment 26 17 goods 13 29 need, v. 22 22 friend 14 17 plant, n. 9 22 special 15 17 possibility 15 29 paper 17 22 wire, V. 18 17 size 12 29 telegram 19 21 balance, n. 15 17 somebody 17 29 telephone, n. 27 21 change, n. 15 17 still, a. 16 29 though 29 21 loan 5 17 story 12 28 able 26 21 mail, V. 16 17 ticket 9 28 customer 22 21 welcome, a. 21 17 within 17 28 instruction 20 20 account, n. 16 28 note, n. 24 20 agreement 8 16 handle, v. 14 28 ring, n. 23 16 like, a. 16 28 room 19 20 anyhow 17 16 part, n. 15 20 cut, V. 18 16 quote 14 28 sale 25 20 exactly 17 16 tank 7 27 arrange 23 20 happen 15 16 truck 13 27 bring 24 20 list, n. 13 15 along, a. 13 27 doesn't 23 20 niessage 11 15 also 11 27 done 23 20 most 15 15 answer, n. 15 27 maybe 26 20 record, n. 18 15 board 8 27 never 23 20 stop, V. 18 27 order, v. 24 20 terrible 13 15 cargo 8 27 really 25 15 clean, v. 14 27 share, n. 10 19 address, n. 15 15 clear, a. 14 19 department 16 15 cocoa 9 27 stay 23 19 far 15 15 cost, v. 15 27 wish, v. 22 19 hold, n. 17 15 date 14 26 book 17 19 load, V. 16 15 interest, n. 9 26 inch 7 19 meeting 9 15 item 10 26 machine 14 19 nearly 19 15 station 8 26 proposition 21 19 plan, n. 12 15 spend 11 26 railroad 19 19 position 15 26 run, V. 20 19 rate 11 15 worry, v. 14 26 short 21 14 already 14 25 bank 12 19 straight 15 14 arrangement 11 18 anyway 16 14 bid, V. 11 25 change, v. 22 18 cheap 13 14 club 7 25 city 18 18 even 17 14 extra 11 25 hasn't 25 18 imagine 17 14 fact 14 25 help, y. 19 18 lunch 18 14 finish, V. 10 25 material 14 18 pier 10 14 full 12 24 absolutely 21 18 possibly 14 14 help, n. 10 24 care, v. 24 18 quotation 13 24 down 20 18 small 17 14 hotel 11 24 hard 22 14 open, a. 12 298 BELL SYSTEM TECHNICAL JOURNAL TABLE Ill-a (Cont'd) A B A B A B 14 operator 10 12 real 11 10 sheet 8 14 particular 13 12 satisfactory' 11 10 street 8 14 perfectly 12 12 several 11 10 territory 5 14 profit 11 12 somewhere 12 10 together 8 14 read 11 12 steamer 10 14 report, v. 12 12 warehouse 8 10 transfer, n. 8 14 second, n. 12 10 warm 7 14 set, n. 8 11 afraid 11 10 whatever 10 11 almost 11 10 woman 5 14 sign, V. 12 11 arrive 10 10 yourself 10 14 stand, V. 14 11 both 10 9 build 7 14 surely 14 11 box, n. 7 9 care, n. 6 14 turn, V. 11 11 cold, n. 6 9 careful 9 13 across 13 11 complete, v. 8 9 certain 8 13 answer, v. 9 11 concern, n. 10 9 charge, v. 8 13 bond 8 11 confirm 7 13 building 11 11 definitely 10 9 color 8 13 charge, n. 8 9 complete, a. 8 13 condition 12 11 detail 10 9 conference 7 11 drawing 8 9 decide 9 13 connection 12 11 funny 11 9 end, n. 7 13 deal, n. 12 11 light, a. 7 9 express, n. 7 13 direct, a. 11 11 mile 8 9 game 8 13 drop, V. 12 11 motor 7 9 hospital 6 13 further 11 11 personally 8 9 immediately 7 13 general, a. 8 11 quality 10 9 large 8 13 himself 13 11 rather 11 13 insurance 11 11 use, n. 10 9 mention 7 13 interested 10 9 necessary 9 13 least 12 10 air 6 9 outside 9 10 awfully 10 9 personal 9 13 luck 12 10 bother, v. 9 9 remember 8 13 notify 6 10 carload 9 9 sit 8 13 offer, V. 12 10 cold, a. 7 9 sometime 9 13 party 12 10 crazy 8 9 statement 9 13 person 13 10 dinner 7 9 suggestion 8 13 quick 13 10 double 7 9 supply, v. 7 13 test, n. 8 10 easily 9 13 without 13 10 either 9 9 true 9 12 agree 11 9 up, a. 8 12 always 10 10 enough 10 9 weren't 7 10 everybody 10 9 willing 7 12 appreciate 11 10 explain 9 9 wise 7 12 bed 10 10 final 6 8 additional 8 12 brother 11 10 freight 8 8 advise 7 12 close, V. 11 10 having 10 8 agent 6 12 consider 9 10 head 9 8 agreeable 7 12 else 12 10 important 10 8 anxious 7 12 expense 10 10 kind, n. 9 12 fair 12 10 limit, n. 8 8 average, n. 7 12 great 11 8 beyond 8 12 loss 10 10 load, n. 8 8 carry 7 10 mark, n. 8 8 certificate 5 12 original 10 10 particularly 9 8 close, a. 8 12 per cent 8 10 positively 10 8 each 8 12 pick, V. 11 10 power 5 8 easy 7 12 policy 6 10 service 10 8 engineer 5 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 299 TABLE Ill-a (Cont'd) A B A B .4 B 8 except 8 7 locate 7 6 offhand 6 8 fill 8 7 lovely 6 6 operate 6 7 mind, v. 7 6 opportunity 6 8 firm, a. 5 7 mother 7 6 package 6 8 girl 6 7 once 5 6 practically 6 8 guarantee, n. 7 7 ours 7 6 promise 6 8 heavy 6 6 realize 5 8 look, n. 8 7 phone, V. 6 6 represent 6 8 middle, a. 7 7 proper 7 6 shall 6 8 mistake, n. 7 7 sake 6 6 simple 6 8 news 7 7 satisfied 7 8 ordinary 6 7 side 7 6 straighten 6 8 owe 6 7 state, n. 6 6 such 6 7 store, n. 5 6 thanks 6 8 plan, V. 8 7 supply, n. 7 6 touch, V. 6 8 push, y. 6 7 throat 5 6 unload 5 8 quantity 6 7 wonderful 6 5 advisable 5 8 reasonable 7 5 allow 5 8 regular 8 7 yard 5 5 approval 5 8 reply, n. 7 6 advice 6 5 catch 5 8 sail, V. 7 6 afford 5 5 conversation 5 8 second, a. 7 6 appear 5 8 settle 7 6 argument 6 5 correct 5 8 shape 8 6 begin 6 5 crowd 5 6 broker 5 5 difficulty 5 8 simply 8 6 bunch 5 5 disa pointed 5 8 single 7 6 cancel 5 5 discuss 5 8 suggest 8 6 claim, V. 5 5 doctor 5 8 sweet 7 5 estimate, v. 5 8 weather 5 6 clear, v. 5 5 grade 5 8 weight 5 6 collect 6 5 holiday 5 8 whether 7 6 competition 5 5 increase, v. 5 8 world 8 6 cost, n. 5 7 actual 5 6 dandy, a. 6 5 inform 5 7 ago 5 6 dealer 5 5 insist 5 6 delay, v. 6 5 instead 5 7 apparently 6 6 depend 6 5 intend 5 7 available 5 6 fairly 6 5 interesting 5 7 buyer 5 5 form, n. 5 5 mix 5 7 clean, a. 7 5 operation 5 7 cover, n. 7 6 impossible 5 5 pardon, n. 5 7 desk 7 6 indeed 6 5 payment 5 7 evening 7 6 inquiry 6 5 reach 5 7 event 7 6 issue, n. 5 7 evidently 7 6 lay 6 5 reduction 5 7 exact 7 6 lose 5 5 return 5 6 mark, v. 6 5 show, n. 5 7 favor 7 6 memorandum 6 5 sort, n. 5 7 follow 7 6 notice, n. 6 5 specification 5 7 indicate 6 6 notice, v. 6 5 surprised 5 7 life 7 5 until 5 300 BELL SYSTEM TECHNICAL JOURNAL Column A: Column B: TABLE Ul-b Word List — ^Alphabetical Order * Words Which Occurred in One Per Cent or More of the 500 Telephone Conversations Analyzed Total number of times the word (or some form of it) was used. Number of conversations in which the word occurred. A B A B A B A 18 anyway 16 10 bother, v. 9 7 apparently 6 11 box, n. 7 2,060 a 487 6 appear 5 36 boy 31 28 able 26 17 appointment 14 27 bring 24 480 about 266 12 appreciate 11 6 broker 5 17 above 13 5 approval 5 12 brother 11 24 absolutely 21 618 are 293 9 build 7 17 accept 14 6 argument 6 13 building 11 20 account, n. 16 54 around 54 6 bunch 5 13 across 13 27 arrange 23 121 business 83 7 actual 5 14 arrangement 11 23 busy 17 8 additional 8 11 arrive 10 63 but 54 19 address, n. 15 262 as 125 98 buy 60 6 advice 6 133 ask 101 7 buyer 5 5 advisable 5 476 at 238 88 by 75 8 advise 7 7 available 5 6 afford 5 8 average, n. 7 C 11 afraid 11 56 away 45 71 after 54 37 awful 31 17 cable 10 87 afternoon 62 10 awfully 10 386 call, v. 200 67 again 62 159 call, n. 111 17 against 13 B 396 can, aux. 221 8 agent 6 6 cancel 5 7 ago 5 113 back, a. 79 228 can't 132 12 agree 11 53 bad 46 234 car 88 8 agreeable 7 37 bag 24 22 card 9 20 agreement 8 21 balance, n. 15 24 care, v. 24 34 ahead 27 25 bank 12 9 care, n. 6 10 air 6 44 barrel 20 9 careful 9 139 all 100 30 basis 21 15 cargo 8 5 allow 5 296 be 175 10 carload 9 117 all right 74 65 because 47 8 carry 7 11 almost 11 12 bed 10 89 case 47 58 along, prep. 33 63 been 53 5 catch 5 15 along, a. 13 96 before 79 50 cent 31 14 already 14 6 begin 6 9 certain 8 15 also 11 31 believe 25 37 certainly 28 12 always 10 60 best 52 8 certificate 5 276 am 172 139 better 103 43 chance 37 17 amount 15 29 between 29 25 change, v. 22 326 an 178 8 beyond 8 21 change, n. 15 1,363 and 391 32 bid, n. 24 13 charge, n. 8 56 another 46 14 bid, V. 11 9 charge, v. 8 15 answer, n. 15 51 big 42 18 cheap 13 13 answer, v. 9 52 bill, n. 39 59 check, n. 35 8 anxious 7 15 board 8 31 check, V. 28 209 any 140 45 boat 22 25 city 18 43 anybody 40 13 bond 8 6 claim, V. 5 20 anyhow 17 26 book 17 15 clean, v. 14 170 anything 100 11 both 10 7 clean, a. 7 * In ambiguous cases the part of speech is denoted as follows: noun, n.; verb, v.; adjective or adverb, a.; auxiliary verb, aux.; preposition, prep. THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 301 TABLE Ul-b (Cont'd) .1 B .4 B A B 15 clear, a. 14 195 didn't 142 7 favor 7 6 clear, v. 5 37 difference 31 41 feel 30 12 close, V. 11 40 different 30 23 foot 13 8 close, a. 8 5 difficulty 5 85 fellow 70 14 club 7 10 dinner 7 61 few, a. 50 68 coal 27 13 direct, a. 11 49 figure, n. 32 15 cocoa Q 5 disappointed 5 35 figure, V. 24 43 coffee 8 5 discuss 5 8 fill 8 10 cold, a. 7 638 do 302 10 final 6 11 cold, n. 6 5 doctor 5 89 find, V. 72 6 collect 6 29 does 27 186 fine, a. 122 9 color 8 27 doesn't 23 14 finish, V. 10 274 come 168 38 doing 34 8 firm, a. 5 186 company 111 105 dollar 66 96 first 71 11 complete, v. 8 27 done 23 30 fix, V. 21 9 complete, a. 8 640 don't 301 7 follow 7 6 competition 5 29 dope, n. 24 823 for 330 11 concern, n. 10 10 double 7 22 forget 18 13 condition 12 24 down 20 6 form, n. 5 9 conference 7 11 drawing 8 10 freight 8 11 confirm 7 13 drop, V. 12 22 friend 14 13 connection 12 338 from 196 12 consider 9 E 14 full 12 63 contract 31 11 funny 11 5 con-\-ersation 5 8 each 8 13 further 11 54 copy, n. 36 35 early 26 5 correct 5 10 easily 9 G 15 cost, V. 15 8 easy 7 6 cost, n. 5 10 either 9 9 game 8 184 could 124 12 else 12 13 general, a. 8 53 couldn't 47 9 end, n. 7 1,360 get 393 45 couple 38 8 engineer 5 273 give 172 7 cover, n. 7 10 enough 10 8 girl 6 17 cover, V. 14 5 estimate, v. 5 49 glad 39 10 crazy 8 18 even 17 597 go 280 5 crowd 5 7 evening 7 243 good 149 28 customer 22 7 event 7 17 goods 13 20 cut, V. 18 23 ever 22 5 grade 5 43 e^-ery 36 12 great 11 D 10 e^'erybody 10 8 guarantee, n. 7 60 e^-eryth^ng 47 72 guess, V. 63 6 dandy, a. 6 7 evidently 7 15 date 14 7 exact 7 H 458 day 251 20 exactly 17 13 deal, n. 12 8 except 8 230 had 151 6 dealer 5 55 expect 49 36 hand, n. 31 9 decide 9 12 expense 10 16 handle, v. 14 17 definite 13 10 explain 9 20 happen 15 11 definitely 10 9 express, n. 7 24 hard 22 6 delay, v. 6 14 extra 11 176 has 114 32 deliver 24 25 hasn't 25 44 delivery 36 F 883 have 367 19 department 16 104 haven't 83 6 depend 6 14 fact 14 10 having 10 7 desk 7 12 fair 12 10 head 9 11 detail 10 6 fairly 6 109 hear 85 200 did 144 19 far 15 8 hea\-y 6 20 302 BELL SYSTEM TECHNICAL JOURNAL TABLE Ul-b (Cont'd) A B A B A B 1,115 he 297 K 112 mean, v. 82 25 help, V. 19 40 meet 28 14 help, n. 10 38 keep 28 19 meeting 9 67 her 30 10 kind 9 6 memorandum 6 238 here 157 753 know 325 9 mention 7 45 high 34 [20 niessage 11 486 him 223 L 8 middle, a. 7 13 himself 13 68 nu'ght, aux. 59 106 his 70 9 large 8 11 mile 8 71 hold, \'. 60 140 last, a. 106 36 mind, n. 30 19 hold, n. 17 118 late 94 7 mind, v. 7 5 holiday 5 6 lay 6 241 minute 155 57 home 35 13 least 12 8 mistake, n. 7 24 hope, V. 19 69 leave 54 5 mix 5 9 hospital 6 32 less 23 29 moment 26 14 hotel 11 215 let 148 73 money 56 36 hour 30 214 letter 112 92 month 70 36 house 27 7 life 7 199 more 134 103 how 78 11 light, a. 7 330 morning 191 86 like, V. 71 20 most 15 / 16 like, a. 16 7 mother 7 10 limit, n. 8 11 motor 7 3,990 54 18 9 10 6 1,170 26 5 ' '? 5 37 I idea 467 38 17 7 10 5 408 7 5 6 6 5 29 87 20 line, n. list, n. 60 13 30 226 move much 23 160 64 listen 55 37 must 30 imagine immediately important impossible in inch 156 10 19 21 7 66 little load, n. load, V. loan locate long 117 8 16 5 7 55 170 87 24 my N name, n. near 97 52 23 increase, v. indeed indicate inform information 112 look, V. 85 19 nearly 19 8 6 12 23 look, n. lose loss lot 8 5 10 19 9 29 27 83 necessary need, v. never new 9 22 23 62 6 5 5 inquiry insist instead 6 5 5 7 31 13 lovely low luck 6 22 12 8 83 52 news next nice 7 61 38 28 4 -y instruction 20 18 lunch 18 168 night 107 13 5 insurance intend 11 5 M 118 131 no, a. not 77 96 15 13 interest, n. interested, a. 9 10 26 machine 14 28 6 note, n. notice, v. 24 6 5 46 1,506 69 interesting, a. into is 5 42 419 52 29 21 274 mail, n. mail, V. make, v. 22 16 169 6 13 339 notice, n. notify now 6 6 216 isn t 97 man 67 139 number, n. 80 6 1,605 15 67 issue, n. it item its 5 417 10 67 29 10 6 96 many mark, n. mark, v. market 25 8 6 51 1,190 0 of 396 25 material 14 112 off 67 J 37 matter, n. 31 ii offer, n. 20 60 may 50 13 offer, V. 12 58 job 44 27 maybe 26 6 offhand 6 321 just 211 496 me 283 120 office 83 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 303 TABLE 1 1 1-6 (Cont'd) A B .4 B A B 35 oil, n. 18 18 possibly 14 183 same 127 38 old 22 39 pound, n. 21 63 sample, n. 37 2,046 on 458 10 power |5 12 satisfactory 11 7 once 5 6 practically '6 7 satisfied 7 125 only 84 115 pretty 92 410 say 211 14 open, a. 12 241 price 123 14 second, n. 12 6 operate 6 88 probably 69 8 second, a. 7 5 operation 5 14 profit 11 887 see 328 14 operator 10 6 promise 6 37 seem 33 6 opportunity 6 7 proper 7 133 sell 81 295 or 178 26 proposition 21 264 send 172 247 order, n. 119 8 push, V. 6 10 service 10 27 order, v. 24 179 put 114 14 set, n. 8 8 ordinary 6 8 settle 7 12 original 10 Q 12 several 11 190 other 128 6 shall 6 45 ought 37 11 quality 10 8 shape 8 57 our 47 8 quantity 6 27 share, n. 10 7 ours 7 35 question, n. 31 140 she 50 138 out 90 13 quick 13 10 sheet 8 63 out of 25 35 quite 32 108 ship, V. 68 9 outside 9 18 quotation 13 49 ship, n. 28 317 over 208 16 quote 14 115 shipment 80 8 owe P 6 R 26 50 40 short should show, V. 21 43 30 26 railroad 19 5 show, n. 5 6 package 6 19 rate 11 7 side 7 29 paper 17 11 rather 11 14 sign, V. 12 5 pardon, n. 5 5 reach 5 6 simple 6 16 part, n. 15 14 read 11 8 simply 8 14 particular 13 30 ready 25 8 single 7 10 particularly 9 12 real 11 9 sit 8 13 party 12 6 realize 5 31 situation 25 85 pay, V. 55 27 really 25 17 size 12 5 payment 5 40 reason, n. 32 18 small 17 79 people 59 8 reasonable 7 67 so 53 12 per cent 8 30 receive 22 64 some 43 14 perfectly 12 20 record, n. 18 17 somebody 17 13 person 13 5 reduction 5 93 something 67 9 personal 9 8 regular 8 9 sometime 9 11 personally 8 9 remember 8 12 somewhere 12 37 phone, n. 35 8 reply, n. 7 76 soon 63 7 phone, V. 6 48 report, n. 29 30 sorry 25 12 pick, V. 11 14 report, v. 12 5 sort, n. 5 24 piece 16 6 represent 6 33 speak 31 18 pier 10 5 return 5 22 special 15 75 place, n. 55 259 right, a. 173 5 specification 5 19 plan, n. 12 28 ring, n. 23 15 spend 11 8 plan, V. 8 28 room 19 14 stand, V. 14 17 plant, n. 9 26 run, v. 20 24 start, V. 22 34 point, n. 27 7 state, n. 6 12 policy 6 5 9 statement 9 19 position 15 15 station 8 10 positively 10 8 sail, v. 7 27 stay 23 17 possibility 15 7 sake 6 12 steamer 10 32 possible 24 28 sale 25 18 steel 10 304 BELL SYSTEM TECHNICAL JOURNAL TABLE III-& (Cont'd) A B .4 B A B 17 still, a. 16 71 through 46 222 week 120 42 stock, n. M 17 ticket 9 8 weight 5 20 stop, V. 18 229 time 165 21 welcome, a. 21 7 store, n. 5 1,942 to 472 92 well, a. 71 17 story 12 193 today 124 83 were 66 19 straight 15 10 together 8 9 weren't 7 6 straighten 6 52 tomorrow 36 336 what 193 10 street 8 60 ton 27 10 whatever 10 146 stuff, n. 92 58 to-night 40 82 when 69 6 such 6 105 too 77 51 where 42 8 suggest _ 8 31 touch, n. 29 9 whether 7 9 suggestion 8 6 touch, V. 6 40 which 33 9 supply, V. 7 30 town 22 103 who 74 7 supply, n. 7 61 train, n. 33 51 whole 40 47 suppose 41 10 transfer, n. 8 31 why 25 86 sure 72 18 trip, n. 17 1,305 will, aux. 402 14 surely 14 45 trouble 33 9 willing 7 5 surprised 5 16 truck 13 105 wire, n. 78 8 sweet 7 9 true 9 22 wire, V. 18 137 try 100 9 wise 7 T 14 turn, V. 11 27 518 wish, V. with 22 263 295 take 207 U 17 within 17 193 talk, V. 131 13 without 13 84 talk, n. 67 82 understand 63 10 woman 5 16 tank 7 2,i unless a 34 wonder 28 29 telegram 19 6 unload 5 7 wonderful 6 29 telephone, n. 27 5 until 5 146 won't 115 553 tell 264 58 up, prep. 42 52 word 45 20 terrible 13 9 up, a. 8 68 work, V. 50 10 territory 15 77 us 63 65 work, n. 49 13 test, n. 8 11 use, n. 10 8 world 8 42 than ii 65 use, V. 50 15 worry, v. 14 60 thank 56 379 would 207 6 thanks 6 V 97 wouldn't 79 1,792 that 397 84 write 61 3,110 the 496 268 very 165 24 wrong 20 67 their 63 370 them 170 W Y 104 then 88 224 there 144 178 wait, V. 135 7 yard 5 63 these 57 599 want 297 78 year 53 913 they 253 12 warehouse 8 40 yesterday 35 418 thing 235 10 warm 7 86 yet 67 474 think 232 358 was 194 3,540 you 499 473 this 240 18 wasn't 18 157 your 100 130 those 100 108 way 81 10 yourself 10 29 though 29 1,100 we 294 7 throat 5 8 weather 5 speech sounds occur. They form 96 per cent of the total occurrences of the words. It is to be noticed that no word was observed to occur in all the conversations. Of the 1,503 different words not shown on the list, 819 were ob- served only once and 320 only twice. It is quite likely that if the THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 305 observations were repeated this part of the list would be dupUcated very imperfectly, since these words, while in general well-known, tend to be technical or specific, hence dependent on particular types of subject matter. All except ten of the omitted words are nouns, verbs, adjectives or adverbs. 100 g80 o i^ 60 o 40 u a. uj 20 y y y' ■0' / y 4 A y ^^ Fig. 2- 1 5 10 50 100 500 1000 NUMBER OF DIFFERENT WORDS IN ORDER OF OCCURENCE -The cumulative curve obtained when the different words are arranged in order of occurrence. The importance of a relatively small number of different words which are used very frequently is shown graphically in Fig. 2. The curves shown are cumulative, giving the percentage of the total words contributed by the different words when arranged in the order of their occurrence. The curve labeled "Written" is based on the list given in the study by Dewey, cited above. The economy exercised in con- versation, or the poverty of conversational expression, according to the point of view, contrasts sharply with written English. In conver- sation 30 words account for half the total, in written English 69 words; in conversation 155 words form 80 per cent of the total, in written English 640. The 50 most common words in telephone conversation and in written English are shown in Table IV, arranged in their order of frequency of occurrence. These words form 60 per cent of the total in conversation and 46 per cent in written English. There are 29 words which are common to the two lists. The personal nature of telephone con- versation is shown in the two words which head the list. The most striking difference between the two is the large number of active verbs which occur in the list for conversation: "get," "see," "know, etc., 306 BELL SYSTEM TECHNICAL JOURNAL TABLE IV Fifty Commonest Words in Telephone Conversation Compared with Written English Telephone Written Telephone Written Conversation English Conversation English 1. I the 26. GO HIS 2. you of 27. TELL BUT 3. the and 28. with they 4. a to 29. me ALL 5. on a 30. HIM OR 6. to in 31. ABOUT WHICH 7. that that 32. at will 8. it it 2>i. THINK from 9. is is 34. this HAD 10. and I 35. DAY HAS 11. GET for 36. THING ONE 12. will be 37. SAY OUR 13. of was 38. CAN an 14. in AS 39. CALL BEEN 15. he you 40. would NO 16. we with 41. THEM THEIR 17. they he 42. was THERE 18. SEE on 43. NOW WERE 19. have have 44. from SO 20. for BY 45. what MY 21. KNOW NOT 46. MORNING IF 22. DON'T at 47. an me 23. DO this 48. JUST what 24. are are 49. OVER would 25. WANT we 50. be WHO The 21 words not common to both lists appear in capital letters. 12 in all. None of these appears among the 50 commonest words of written English. Three nouns, "day," "thing" and "morning," appear in the conversational list, none in the other. Only one con- junction is found in the conversational list, while five appear in the list for written English. When the first 100 words in telephone conversation are compared with the first 100 in written English two somewhat unexpected facts emerge. In telephone conversation 14 out of the first 100 are words of more than one syllable; in written English there are ten. Four two- syllable words appear among the first 50 telephone words; the first 59 of written English are monosyllables. A more striking difference concerns the origin of the words. Among the first 100 telephone words there are 11 which are derived through old French from the Latin; in written English there are only two from the Latin. Six of the 11 words occur in the first 65 telephone words, while the first word of Latin origin in written English is the 70th. The telephone words of Latin origin are, in order of occurrence: "just," "very," "order," "minute," THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 307 "price," "car," "letter," "fine," "company," "stuff," "number"; in written English these words are "people" and "very." The pre- dominance of business words in this list for telephone conversation suggests the influence of trade between England and France in the Middle Ages. More detailed comparisons may be drawn from Table V, which lists the first 25 nouns, the first 25 verbs and the first 25 adjectives and TABLE V Twenty-five Commonest Words by Parts of Speech Compared with Written English Nouns Verbs Adjectives and Adverbs Telephone Written Telephone Written Telephone Written Conversation English Conversation English Conversation English day- man get say now not thing time see make just all MORNING WAR know come very no ORDER PEOPLE want take RIGHT SO MINUTE day go know good WHEN PRICE YEAR tell go HERE any CAR thing think see MUCH more time way say get THERE now WEEK WORLD call give any UP LETTER COUNTRY take think more out COMPANY PART make LIKE TODAY other NIGHT business come tell other only CALL LIFE give USE FINE GREAT STUFF FACT SEND call SAME SOME NUMBER LINE LET want little HOW business GUN TALK GOVERN LAST very OFFICE case PUT STAND BETTER SUCH SHIPMENT HOME WAIT ask all FIRST way CENT TRY SEEM out good WIRE POWER ask SHOW not EVERY DOLLAR PRESENT SELL look onlv THEN man HOUSE look NEED LATE little MARKET LOSS MEAN SAVE no here month month HEAR WORK ALL RIGHT just case PEACE SHIP BELIEVE PRETTY WELL The words not common to both lists appear in capital letters. adverbs, for both telephone conversation and written matter. Among the nouns only eight are common to the two lists. The effects of business are apparent in the telephone list. On the other hand, the nouns of the written English list reflect the fact, pointed out by Dewey, that the list was obtained from a study made soon after the war. Among the verbs 15 words are common to the two lists and those which differ are concentrated at the end. Approximately half the adjectives and adverbs appear in both lists. The nouns from telephone conver- 308 BELL SYSTEM TECHNICAL JOURNAL sation shown in this table form 2.4 per cent of the different nouns and 40 per cent of the total nouns; the verbs form 5.5 per cent of the different verbs and 72 per cent of the total verbs; while the adjectives and adverbs form 3.9 per cent of the different adjectives and adverbs, but 48 per cent of the total. An examination of the origin of the words in Table V shows that the influence of Latin on the frequently used words is largely confined to nouns. Eleven of the first 25 nouns of telephone conversation, and eight of the first 25 nouns of written English come from the Latin. Among the first 25 telephone nouns, aside from the eight nouns men- tioned above among the first 100 words, there are: "office," "market" and "case"; among the first 25 nouns of written English the following are of Latin origin: "people," "country," "part," "fact," "cent," "power," "present" and "peace." Only one of the first 25 telephone verbs comes from Latin: "try," and three of those in written English: "use," "govern" and "save." Among the adjectives and adverbs there are found in the telephone list: "just," "very" and "fine," as above, and in the written English list the word "just" is added to "very," which was in the first 100 words. Referring once more to the small number of different words found it may be pointed out that this shows how difficult it would be to estimate the size of vocabularies by recording spoken words. The 80,000 words of this study are equivalent to a complete record of seven hours' conversation, taking a rate of 200 words per minute. As noted before, the number of different words was only 2,240, even though the conversations covered a wide range of topics by many different speakers. To increase this number notably, the curves of Figure 1 indicate that the observations would need to be very extended, since the rate at which new words appear has already become very low. For example, if the conversations were to go on continuously for a week at the above rate a total of 2,000,000 words might be expected. By extrapolating the curves of Fig. 1, and using a similar curve for adjectives and adverbs, which lies between the curves shown, it may be estimated that only about 5,000 of these words would be different words. Extrapolation is a rough tool, but even with its inaccuracies in mind, the conclusion seems safe that to measure a vocabulary by recording spoken words involves the risk of gross underestimation unless the observations are exceedingly prolonged. It is suggested that teachers of languages may find the 737 words in Tables lll-a and IH-^ to be of practical use in their profession. Pre- sumably the progress of a student in speaking a foreign language would be materially assisted by a thorough knowledge, early in his course, THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 309 of the words which are met with great frequency. The present methods of teaching the spoken language no doubt approximate to this, as a result of experience. It is suggested that the present word list, which contains the words used so frequently as to form 96 per cent of the total number observ^ed in this study, provides a guide for the selection of important words to be taught. Additions are needed to the list as it stands, in order to care for certain obvious situations not encountered in telephone conversation concerning, for example, hotels, restaurants and trains. With these points in mind, the list given has the advantage of being founded on a study of actual conversation. Syllables As a preliminary to analysis of the words into their component sounds the words were divided into syllables. With regard to the fact that the study concerned conversation the division was made on phonetic lines, which, as unabridged dictionaries show, differ from the ortho- graphical divisions. Likewise a few words such as "every," "prefer- ence," "average" and the like were divided into two syllables, accord- ing to the usual colloquial pronunciation. TABLE VI The Syllabic Structure of Coxversatioxal Vocabulary Parts of Speech Per Cent of Words Having Number of Syllables Sho\vn Average Number of 1 2 3 4 5 6 Syllables Nouns Verbs 53.3 81.9 57.8 94.8 82.0 33.8 15.0 30.7 4.7 13.8 9.7 2.8 8.0 0.6 3.2 2.7 0.3 2.8 0.1 0.86 0.47 0.66 0.15 0.03 0.02 0.01 1.63 1.21 Adjectives and Adverbs Minor All Words 1.58 1.06 1.23 In Table VI a summary is given of the syllabic structure of words, based on the total occurrence of the words. It may be noticed that words longer than two syllables make up only a trifle more than 4 per cent of the words observed. Nouns tend to be more polysyllabic than other classes, but even so the nouns having more than two sylla- bles occur so infrequently as to form only 13 per cent of all the nouns observed. The types of phonetic syllables which are found range in complexity from a single vowel through various combinations of consonants with a vowel. The relative number of the different types is shown in Table \'II. The letters V and C represent "vowel" and "consonant," 310 BELL SYSTEM TECHNICAL JOURNAL TABLE VII Types of Phonetic Syllables in telephone Conversation Relative Occurrence per Hundred Type Occurrence V 9.7 VC 20.3 CV 21.8 CVC 33.5 VCC 2.8 CCV 0.8 CVCC 7.8 CCVC 2.8 CCVCC 00.5 100.0 respectively, and the letters CC are used to denote a compound con- sonant form, that is, two or more consecutive consonants. It may be seen that the typical syllable is of the CVC type, closely followed in importance by the CV and VC types. The syllables having two or more consecutive consonants form about one seventh of the total. Speech Sounds The analysis of the words into their constituent sounds was at- tended by certain difficulties which should be borne in mind in consider- ing the tables which follow. It was not feasible to record the original words phonetically, just as they were pronounced by the telephone subscriber. Instead the words were recorded and their phonetic values assigned later. In so doing the dictionary was not adopted as an authority for the pronunciation since in the informality of conversa- tion, even among educated persons, there are elisions and changes of stress which cause departures from the dictionary standard. Certain very common words, for example, receive various treatments in con- versation, depending on their situation In the sentence, the emphasis desired and the speed of talking. The word "and" may be pronounced as spelled, but quite often it is reduced to " 'nd" or even " 'n'. " The prepositions "to" and "of" are similarly varied. Altogether about 40 common words were found, of this type, each of which seemed subject to several different pronunciations, even in speech which would not be regarded as unduly careless. These were all from the minor classes: auxiliary verbs, pronouns, prepositions and conjunctions. The modification, in general, is such as to give the vowel its unstressed value. In the analysis these different forms are included, the weight- ing for each modification necessarily being a matter of judgment. The remaining words were each assigned a single pronunciation, selecting that which we regarded as being the typical pronunciation THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 311 heard in reasonably enunciated conversation among educated persons in New York. The departures from dictionary standards are largely confined to the vowels. As a result the analysis is affected to some degree by the speech habits of the writers.^ It is regrettable that some arbitrariness should be introduced, but this seems to be a difficulty common to discussions of vowel sounds. Some of the difficulty is avoided by making separate classifications for vowels for which the pronunciation is indefinite, such as the vowels in unstressed positions. The articles "the," "a" and "an" were omitted entirely from the analysis on account of the large number of variant pronunciations to which they are subject. The results of the analysis into speech sounds are shown in Table VIII. Three divisions are given: vowels, initial consonants and final consonants, based on the division into phonetic syllables. The method followed was: first, to divide the words into phonetic syllables, second, to assign phonetic symbols to the sounds and third, to weight each sound by the total number of times the word was recorded. The sounds are identified in the table, where necessary, by key words. No difficulties were encountered in analysis of the consonants, but a few special points which arose in assigning the vowel qualities may be noted. The key word "pot" is used to denote a vowel sound which is pronounced differently by many natives of New England and those whose habits of speech were formed elsewhere.* With these New Englanders the sound tends toward the quality of the vowel in "pawn," although shorter in duration. The same New Englanders make a real distinction between the vowel of "pot" and the vowel of "palm." By many speakers elsewhere no such distinction is made and the two are lumped into a single intermediate sound which is neither the New Englander's "pot" nor "palm." To avoid confusion the class denoted by "pot" has been made to include "not" and many other monosylla- bles of the same ending, as well as "on," "job," "stock," etc., which grouping is believed to be homogeneous on either basis. The few words of the class of "palm" which were encountered have been in- cluded under "par." The class denoted by "par" may be subdivided into: "par," 1.24; "palm," 0.07. The class denoted by "palm" would be somewhat larger if the class which we may denote by " path," such as "can't," "last," "ask," etc., had not been classified under ^ For the benefit of phoneticians who may be interested it may be stated that the writers are residents of Greater New York of more than six years' standing, that their boyhoods were spent in Maine, Illinois and New Jersey, respectively, and their college years at Maine and Princeton, Harvard and Oxford, New York University and Harvard, respectively; this seems a background sufficiently varied to bring to light many of the principal variants of American speech. * Just what the geographical lines may be, the writers do not pretend to know. A phonetic map would be of interest. 312 BELL SYSTEM TECHNICAL JOURNAL TABLE VIII Relative Occurrence of Speech Sounds in Telephone Conversation All Words {Except Articles) Vowels pin . . pine . . pan. . pen. . peel . . pool. . pot. . pane, pole. . pawn . pun. . pull., pout . par. . pair . . purr, pew. . poise . 75.36 Unaccented Vowels possible 5.5^ about, differ. . . receive . notion . wanted . peop/e . . 5.33 4.56 3.78 2.65 1.83 .97 24.64 100.00 Total Number of Sounds 92,522 Initial Consofiants 10.27 W 7.58 T 6.89 TH" (then 6.60 Y 6.44 D 6.26 AI 5.21 H 4.78 K 4.74 S 4.15 N 4.14 B 2.96 G(gun; 1.69 L 1.31 F 1.09 R .80 P .26 TH' (thin .19 SH J CH Z ZH NG 9. 7. 6. 6. 6. 5. 5. 5. PR HW ST TR FR PL KW BL SP KL Others .38 .86 .72 .48 .21 .89 .75 .55 5.46 4.99 4.64 4.33 4.31 3.96 2.78 2.54 2.02 1.74 1.25 .83 .55 .34 .02 93.60 Compounds 1.06 .91 .87 .69 .62 .36 .28 .23 .19 .18 Final Consonants t . . . 14.30 r . . . 13.05 n . . . 12.52 1 . . . 8.40 z . . . 6.01 m . . . 5.48 d . . . 4.44 V . . . 4.23 ng . . . 3.57 s ... 3.13 k . . . 2.85 f ... 1.37 th" (with) 1.25 P 1.24 ch .53 b .42 g .38 sh .32 ] .14 th'(myth) .04 zh (azure) .01 h — ■ w — ■ y . . . — Compounds nt nd St ts nk Id rz ks kt rd 1.01 Others 6.40 83.68 4.40 2.56 1.18 1.11 .76 .75 .57 .47 .42 .37 3.73 16.32 100.00 64,043 100.00 65,544 "pan," such being the more common American pronunciation. Actually the occurrence of words in the class of "path" is not high; if they had been given a special class in Table VIII their relative occur- rence figure would have been 0.78, reducing the figure for "pan" to 6.11. Special categories are given to the vowel sounds in the classes denoted by "pair" and "purr" since there is often disagreement con- THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 313 cerning the quality of a vowel which precedes "r." Likewise it was found expedient to make a number of classifications of vowels in unaccented positions. Since the figures of Table VIII are likely to find application as weighting factors it is convenient to have them add exactly to 100 per cent, consequently they are given to two places of decimals. An estimate of the representativeness of these figures may be obtained from the data presented in Table IX, which were worked out from TABLE IX Comparison with Check Test Relative Occurrence < of Consonants in Verbs Sound First Observations Check Test Difference Sound First Observations Check Test Difference B 1.02 1.02 .00 S 10.31 9.67 -.64 D 4.46 4.83 + .37 T 16.97 17.39 + .42 F 1.73 2.18 + .45 V 2.36 2.20 -.16 G 11.15 9.40 -1.75 W 4.87 4.54 -.33 H 1.40 1.66 + .26 Y .55 .53 -.02 T .22 .23 + .01 Z 1.09 1.44 + .35 K 8.90 8.74 -.16 CH .32 .75 + .43 L 7.70 7.94 + .24 SH 1.35 1.17 -.18 M 4.45 3.96 -.49 TH' 2.53 2.85 + .32 N 6.87 7.10 + .23 TH" .05 .06 + .01 P 3.34 3.47 + .13 ZH .00 .00 .00 R 3.97 3.88 -.09 NG 4.39 5.00 + .61 100.00 100.00 observations mentioned before, conducted by a different observer at a different time, but on the same set of toll circuits. Records were made only of verbs, and for 250 instead of 500 conversations. The vocabu- lary collected in the check test resembled that of the first observations closely. Arranging the words in the order of occurrence, the first 17 words of the first observations are also the first 17 of the check test, although the order is not repeated exactly. In the first observations the first few words run: "get," "know," "see," "want," "go," "tell," "think" and "say"; in the check test the order is: "get," "see," "know," "want," "tell," "think," "go" and "say." Table IX shows the analysis of the words as to the simple consonants, lumping initial and final consonants together. Only one of the differences is greater than 1 per cent and all but three are less than 0.5 per cent. One check test is not sufficient for a final statement, but judging by these results the observing method and the samples taken seem to justify considering the figures of Table VIII as representative as far as the figures in the 314 BELL SYSTEM TECHNICAL JOURNAL digits position for most of the sounds and as to order of magnitude for the infrequent sounds. The effects of restricting the word Ust in various ways are shown in Figure 3. The first hne shows graphically the relative occurrence per ^ RELATIVE OCCURENCE OF THE INITIAL CONSONANTS IN TELEPHONE CONVERSATION P T K F TH' S SH CH M N NG L R W Y H Compound B D G V TH" Z ZH J n t-in ,-. _ „nn V^r^ n ■— -L-l 1— 5 n n .— . n rn^n i— i r~l — |r— 1 0_i__u..,^l^m 'i— i|_j 5 _ LJ Fig. 3 — The relative occurrence of initial consonants — effects of restricting the word list. Line I — Relative occurrence of initial consonants for all words. Line II — Differences resulting from omission of minor parts of speech (118 words). Line III — Differences resulting from omission of the 100 commonest words. Line IV- — Differences resulting from omission of the 1,500 least common words. hundred for initial consonants as in Table VIII. If the minor parts of speech are excluded before the analysis, which eliminates only 118 different words, but nearly half the total words, the resulting changes are shown on the second line. Notable decreases occur for "th," "w" and "y," which may largely be traced to the omission of "that," "they," "this," etc.; "will," "with," "would," etc.; and "you," respectively. These elisions enhance the relative contributions from "get," "see" and "know." When the 100 most common words are omitted there are also large changes, as shown by the third line. Since 50 of the 100 most common words are of the minor parts of speech the similarity of this line to the second is not surprising. The omission, on the other hand, of the 1,500 least common words, namely, those omitted from the vocabulary of Table III, changes the distribution by negligible amounts as shown in the fourth line. Since, then, the 737 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 315 commonest words seem to determine the relative frequency of the sounds of conversation, the writers are encouraged to believe that if this study were repeated on telephone calls of which a greater propor- tion were social rather than business in nature the analysis into sounds would be changed very little. Likewise the conclusion is drawn that if the study were prolonged tenfold so as to double the number of dif- ferent words no material change in the relative frequency of the sounds would be found. TABLE X Relative Occurrence of Speech Sounds in Telephone Conversations Nouns, Verbs, Adjectives and Adverbs VoiL'els Initial Consonants Final Consonants 10.63 7.56 7.19 6.38 5.96 5.78 5.40 5.25 4.59 4.01 2.49 1.83 1.58 1.53 1.39 1.33 .49 .35 73.74 Unaccented Vowels differ 5.79 receive 5.73 possible 4.82 obout 3.96 wanted 2.54 peop/f- 1.76 notion 1.66 pen. . pin. . . pane, pole . . pawn . peel . . pine, pun. . pot . . pan. . pull., pout . pair. . par. . pool . . purr. . pew . . poise . Total number. 26.26 100.00 50,161 S N T M G K L D W B R P F SH TH' H J Y V TH" CH Z ZH NG Compounds PR . ST . TR . PL . HVV . KW . BL . SP . KL . GR . Others. 8.34 7.94 7.55 7.40 6.87 6.70 6.65 5.25 4.86 4.38 4.11 4.06 3.88 2.42 2.41 2.38 1.33 1.25 1.21 .97 .87 .55 .03 91.86 1.69 1.39 1.11 .58 .49 .44 .37 .30 .29 .27 1.21 8.14 k m s d z P V f th" ch b g sh J th' zh h w Compounds nt St nd . nk . Id rz ks . kt . rd . ns Others. 14.64 13.53 9.99 8.62 5.10 4.96 4.50 3.89 3.66 2.42 2.00 1.94 1.28 .82 .81 .73 .66 .57 .24 .07 .02 80.45 3.37 2.07 1.66 1.32 1.31 .98 .82 .73 .64 .54 6.11 19.55 100.00 40.107 100.00 37,493 316 BELL SYSTEM TECHNICAL JOURNAL For some purposes weighting lists based on the words of speech which carry the meaning are appropriate. This is approximated to by the figures given in Table X, in which the sounds are analyzed for nouns, verbs, adjectives and adverbs only. The outstanding changes in the weighting of initial consonants have just been commented on in con- TABLE XI Relative Occurrenxe of Speech Sounds in Telephone Conversations Conversational Weighting Note: The sounds of each word are weighted b}' the number of conversations in which the word is used, instead of by the total occurrences of the word. Vowels Initial Consonants Final Consonants pin 11.22 W 8.26 r 13.87 pen 7.90 T 7.09 t 11.98 pan 6.40 M 6.69 n 10.92 peel 6.21 D 6.52 1 8.13 pine 5.95 K 5.90 m 5.43 pane 5.60 S 5.90 d 5.20 pole 5.18 L 5.44 z 5.13 pun 4.66 B 5.32 ng 4.05 pawn 4.64 H 5.31 v 3.72 pot 4.08 N 5.09 s 3.64 pool 3.40 TH" 5.01 k 3.41 pull 3.24 F 4.10 p 1.55 pout 1.89 G 4.00 f 1.41 par 1.33 R 3.53 th" 1.18 pair 1.31 Y 3.17 ch 65 purr 1.11 P 3.09 b 52 pew 38 SH 2.09 g 49 poise 24 TH' 2.06 sh 45 V 1.43 j 17 74.74 J 94 th' 06 CH 74 zh 02 Unaccented Vowels Z .47 h • — ZH 03 w — about 5.39 XG — y — differ 5.35 receive 4.85 92.18 81.98 possible 3.57 notion 2.69 Compounds Compounds wanted 2.16 people 1.25 PR 1.27 nt 4.68 ST 1.06 nd 2.09 25.26 HW 1.03 st 1.43 TR 82 ts 1.14 FR 62 Id 89 100.00 PL 49 nk 71 Total Xumber. . . 54,656 KW 40 rz 63 BL 30 ks 61 SP 26 k-t 56 KL 26 rd 51 Others 131 Others 4.77 7.82 18.02 100.00 100.00 39,924 40,993 THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 3l7 nection with Fig. 3, line 2. The vowel weighting reflects the enhanced importance of the "e" in "pen" from the verbs "get," "tell," "send" and shows a considerable reduction in the vowel in "pool," largely from the loss of "you" and "to." The unstressed vowels, especially as in "about" are also diminished. Among the final consonants the largest change is a reduction in "z," which results from the elimination of "is," "was," "as," etc. Comparing the distributions of Table VIII and Table X as a whole, however, both show about the same degree of non-uniformity; the maximum and minimum weightings do not differ greatly. One more type of analysis is given in Table XI. In this case the sounds of each word are weighted by the number of conversations in which the word occurred, instead of by the total number of times the word was used. This seems to be a somewhat radical change in method, involving, as it does, a considerable reduction in the weighting of the words at the head of the list. When the effect of eliminating the first 100 words entirely, shown in Fig. 3, line 3, is recalled, large changes might be expected. Actually the result is remarkably similar to the figures of Table VIII. The relatively diminished importance of "you" and "to" is seen in the vowel list, of "you" again among the initial consonants, and of "it," "that" and "get" in the list of final consonants. The range covered by the relative weightings is still much the same as in Tables VIII and X, Comparisons with Written English Some of the differences between the vocabularies of telephone conversation and written English have been pointed out. The effects of these differences may be seen in the relative occurrence of the sounds as shown by Figures 4, 5 and 6 for vowels, initial consonants and final consonants, respectively, using the analysis based on all the words (except articles). The upper line in each case is a graphical representa- tion of the corresponding data of Table VIII, after certain changes have been made to put them on the same basis as the tables given by Dewey for written English. In the case of the consonants the only change needed was omission of the compound consonants. In the case of the vowels it was necessary to combine some of our classifications, since Dewey made but 17 distinctions among the vowels. We believe the combinations made are those followed by Dewey himself, as ascer- tained from examples given by him in his text. The phonetic symbols given in Figure 4 are those used by him. The combinations made were as follows: "pen" and "wanted"; "pane" and "pair"; "pin," "pos- sible" and "receive"; "pun" and "purr"; "about," "differ," "peop/g" 21 318 BELL SYSTEM TECHNICAL JOURNAL and "notion." The comparisons are made with Table XVI of Dewey's book, which does not include "the," and from which we have sub- tracted the article "a." The outstanding differences between the vowel frequencies in telephone conversation and written matter are the excess in conversa- tion of "about," "pine," "pool" and the deficiencies in "pan," "pin" % 20 15 10 5 0 RELATIVE OCCURENCE OF THE VOWEL SOUNDS IN TELEPHONE CONVERSATION i i M m ^ == ^ n i I <^y/A Y/A I I i 5 0 -5 Fig. A IX ty ©■ AT vr ju, JUL 3q: 3 u oi -j -j -i o 0. Q.I- <^ °-5|UJ U OL Q. 5 0.3 OU- ;:Idl' o d o 8S' a Q. Q. U U I- § Z (O 3 u D- O ? O- TT n -rnn n TJ PERCENTAGES IN EXCESS OF THOSE FOR WRITTEN MATTER 4 — Comparison with written English — relative occurrence of the vowels. and "pot." The greater occurrences of "pine" and "pool" are almost entirely accounted for by the greater use of the words " I " and "you." The deficiencies mentioned do not, on analysis, seem to depend on one or two words, but rather on the whole vocabulary, except that of the increase in the unstressed vowel denoted by "about" nearly 1.7 per cent comes from the vowels of words which in the study of written English were classified under "pan." Among the initial consonants (Fig. 5) the greatest change is in the occurrence of "y," which is much more frequent in conversation. This again is largely caused by the pronoun "you." Much of the increase in "g" may be traced to the greater use of "get" and "go." The sounds "w" and "t" are the most frequent sounds in written English, as well as in conversation. Figure 6 shows that in the case of the final consonants the sounds "t" and "1" are notably more frequent in conversation than in written THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 319 matter. The increase in "t" arises almost entirely from "that," "it" and "get" which combined have a contribution about 4.9 per cent larger in conversation than in written matter. About half the increase RELATIVE OCCURRENCE OF THE INITIAL CONSONANTS IN TELEPHONE CONVERSATION (compound CONSONANTS NOT INCLUDEd) fS MNNGLRWYH nr-.n. ,^nn, ,^^ TIT "ODD izr PERCENTAGES IN EXCESS OF THOSE FOR WRITTEN MATTER Fig. 5 — Comparison with written English — relative occurrence of the initial consonants. RELATIVE OCCURRENCE OF THE FINAL CONSONANTS IN TELEPHONE CONVERSATION (compound CONSONANTS NOT INCLUDED) ''b "^ D *" G ^ v"'''tH^ Z^^Zh'^'^jMNNGL RWHY IT 'UU^ n _ U PERCENTAGES IN EXCESS OF THOSE FOR WRITTEN MATTER Fig. 6 — Comparison with written English — relative occurrence of the final consonants. in "1" is attributable to the words "will" and "tell." Some of the deficiency in " v" may be traced to the word "of" which has a contribu- tion 1.8 per cent greater in written matter. On the other hand the 320 BELL SYSTEM TECHNICAL JOURNAL words "have" and "give" together contribute 1.1 per cent more to conversation, so that the net difference in "v" is to be traced to small accretions from the whole vocabulary rather than a few specific words. Relative Occurrence of Combinations of Sounds A more elaborate analysis of the phonetic syllable is given in Table XII, which shows, for each vowel, the frequency of occurrence of the consonants preceding the vowel and also of the consonants which follow the vowel. The complete word list (except articles) was used as a basis. The cases in which no consonant occurs in front of the vowel are included, as well as the cases in which there is no following con- sonant. These figures are shown as a double column under the key word denoting the vowel sound. In each double column the figures on the left apply to initial consonants and on the right to final con- sonants. The figures are given in per cent, so that each column adds to 100. The consonants are grouped by phonetic classes. The table is to be read as follows: of the syllables in which the vowel sound is that in "pan," 28 per cent begin with "th" (as in "that"), 26 per cent have no initial consonant, 16 per cent begin with "h," 7 per cent with "k," 6 per cent with compound consonants, 5 per cent with "b," etc.; while 29 per cent end with compound consonants, 27 per cent with "t," etc. Where no figure is entered the occurrences were less than 0.5 per cent ; where a dash is shown no combinations of the kind indicated were observed. If the figures are taken by rows instead of columns no meaning can be attached to them before they are multiplied by the relative occurrence of the different vowels. In studying this table it is to be remembered that because the dif- ferent vowels have very different frequencies of occurrence the sub- divided data shown in different columns cannot be considered as equally representative. Syllables having the vowel as in "pin," for example, were present, as shown in Table VIII, to the number of 0.1027 X 92,522, or 9,500. The syllables in this class which begin with "t" are shown in Table XII to be 1 per cent, representing 95 occurrences. In the class having the vowel of "poise," however, there were only 176 examples, so that the 37 per cent of these syllables beginning with "p" result from only 65 occurrences. It is to be seen that only one vowel, "pool," is preceded by a par- ticular sound more than 50 per cent of the time, this sound naturally being "y." Six vowels are preceded by particular sounds more than 25 per cent of the time. The sounds of "pair," "purr," "par" and "differ" must be followed by "r," a blank, or a compound consonant beginning with "r," as a result of the way in which the analysis was THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 321 Q. -P -i£ Q. 1- X -Q "D CP to Q U <;- r in X r V <0 0 I II U. 1- l/l w> (J > X N X -1 +) N I I > 1- N N -) £ C U) ^ L 5 J 2 z z _) tr s >- H BLANK BLANK COMPOUND COMPOUND oil 1 1 1 .n 1 1 1 1 — I ^ C: >n 1 1 1 -co - 1 1 1 1 1 1 1 rvi - 1 OJ - r 1 1 't 2 1 1 1 1 1 1 - ?^ ru 1 1 f 1 1 - 1 (M 1 - 1 1 c\j 1 ■ 'J - 1 01 gai 1— S <:i 1 ^ ' 1 lO (V - 1 1 1 r). p 1 ^ 1 r I 1 11-11 - 1 t») 1 - 1 1 r~ 1 m CM - 1 1 1 1 1 1 - 1 CM in 5 1 - ^ 1 - 1 — 1 1 11-11 - CM 1 1 1 II 5 CO 1 .n 1 ^ (3> « en , , 1 1^ 1 oj 5 1 1 1 1 1 1 1 1 CO 1 ^n CM 1 1 1 * - 1 ^ - , <*> 1 in COUJ OCD. cx:lu o UJ z Q. (— 2 = C\J r) oj OJ 1 1 1 O 1 C\J 1 - 1 fl) fM 1 I - r Si 1 . 1 ^ 1 1 in ( — 1 1 ro 1 1 — 1 II 1 1 >n 1 1 1 1 1 1 1 - 1 — 1 1 1 1 1 1 1 1 1 1 1 - 1 1 1 1 1 i~- 1 1 (0 1 1 1 (0 <0 \f 1 Oi 1 1 1 ^ ^ 1 (o r^ CM I 1 -^ 1 1 1 1 1 1 1 cy 1 1 1 1 CO 1 — 1 1 1 'J " - S <° i ^ 2 - CO _ , rvj ro „CM ^ 1 o ' CM - 2 LLI Q. 1 - 1 - ro (0 r- Kl -- 1 t 1 1 — 1 (\l - lO - r- 01 1 — 1 OJ 1 11-11 III 1 ^ 1 1 1 1 CM 1 1 1 1 1 1 1 1 1 - — CO 1 ro 1 1 1 CM CM 1 \ \ — 1 — ^ O -III = It 1 - - 'J . 1 1 1 f;^ - 1 1 > ^ > ^ S -^ 1 1 1 1 - ■r> - 1 in - ^ _ ffl 1 1 1 t in — - - CM - n - ■^ 1 If) (g r~ CO II (0 1 ij - 2 1 - CO 1 ^ 2. 1 1 £ N ^ - I I > 1- N N -) E C CP -- L S 31 5 z z -I cr § >• •Ci:x:Qc3 <<5d -l_lOO I (CcDaCL Q o ►J w > a w o ? o J J o l-H Q 1=1 Z Xj < «< u f-H W c/) o z o o b O o H H CT! 322 BELL SYSTEM TECHNICAL JOURNAL made, similarly for the "1" of "peop/e" and the "n" of "notion." Aside from these no single consonant occurs as often as 50 per cent of the time after a particular vowel. With five vowels a particular consonant ends the syllable more than 25 per cent of the time. In nearly every case the most frequent combinations can be traced directly to the first 50 words of the vocabulary. Five vowels are preceded by blanks more than 50 per cent of the time and eight are followed by blanks in more than 50 per cent of the cases. The combinations of different vowels with compound consonants vary considerably in importance, ranging in the final position from practically none with the vowel of "pew" up to the vowel of "pun," which is terminated by a compound consonant 63 per cent of the time. Conclusion In concluding, a brief review is presented of the main points of interest. The paper has for its basis a Hst of 80,000 words obtained from telephone conversations. This list has been studied with respect to the number of different words contained in it, the relative occurrence of the different speech sounds and the combinations of sounds which form syllables. In so far as the authors know this is the first study of this type based on conversations as contrasted with written matter. Perhaps the most striking aspect of the word list is the small number of different words contained in it, only 2,240 out of the total of 80,000. Of these 2,240 words 819 occur only once. The balance, or 1,421 words, constitute practically 99 per cent of the total words recorded ; of these the 121 different words which constitute the minor parts of speech form 45,000 of the total occurrences. The pronouns "I" and "you" together occur over 7,500 times. This intensiveness with which a small number of words is used in conversation is considerably greater than in the written English ana- lyzed by Dewey. In conversation the 155 most frequently used words make up 80 per cent of the total occurrences; to reach the same per- centage in the written English analyzed by Dewey 640 words must be included. The frequently used words of conversation are character- ized, as compared with written English, by the greater prominence of certain active verbs, such as "get," "see," "know," etc., 12 of which occur in the first 50 words of conversation, while there are none in the first 50 words of written English. The most frequent words of con- versation differ from written English also in the greater number of words of Latin origin which appear frequently in conversation: 11 from the first 100 of the list for conversation, as compared with two from the first 100 of written English. THE WORDS AND SOUNDS OF TELEPHONE CONVERSATION 323 The word list is characterized by a large percentage of monosyllables. Over four fifths of the 80,000 occurrences are of this type, a result largely brought about by the frequent repetition of the minor parts of speech, among which 95 per cent are monosyllables. When the words are analyzed into phonetic syllables about one fifth are found to be of the type vowel-consonant, about one fifth consonant- vowel, and a third of the type consonant-vowel-consonant. The relative occurrences of the different speech sounds were obtained by assigning phonetic values to the sounds of the phonetic syllables and weighting each by the total number of times it was used. Twenty- five categories are used for the vowels. Seven of these are for vowels in unaccented positions, which make up, altogether, about 25 per cent of the vowels. The relative occurrences of the individual sounds differ greatly for different vowels. The range extends from about 10 per cent for the vowel of "pin," and about 8 per cent for the vowel of "pine," down to 0.3 per cent for "pew" and 0.2 per cent for "poise." Among the initial consonants 94 per cent are single sounds, and the remaining are compounds of two or more successive consonants. The range extends from about 9 per cent for "w," and about 8 per cent for "t" down to about 0.3 per cent for "z" and the slightest trace, .02 per cent for "zh." The most frequent compound initial consonant is "pr," with an occurrence of 1 per cent. Among the final consonants the compounds are somewhat more prominent, forming 16 per cent. The most frequent final consonant is "t," 14 per cent, the next is "r," 13 per cent, the range extending down to 0.1 per cent for "zh." The most frequent compound final consonant is "nt," 4.4 per cent, and the next is "nd," 2.6 per cent. Considering the marked differences between the word lists for conversa4;ion and for written English, a comparison of the relative frequency of the speech sounds in the two cases is perhaps more remark- able for the likenesses than the difTerences. About the same range of percentages is covered in both cases. Certain sounds do show marked difTerences. Among the vowels the unaccented vowel denoted by "about" is more frequent in conversation and the vowel of "pan" less frequent. The initial "y" and the final "t" are also more frequent in conversation. Many of the differences can be traced directly to one or two words which in their frequent use are typical of conversation. In considering the occurrence of speech sounds in telephone conversa- tions from the point of view of their contribution to the ease or diflfi- culty of carrying on conversations it seemed of interest to determine how the occurrence of the speech sounds was affected by changing the list in certain ways. Omission of the minor parts of speech changes the 324 BELL SYSTEM TECHNICAL JOURNAL relative occurrence of a number of the sounds materially, although the general range of percentages covered is changed very little. Omission of the 1,500 least common words has a negligible effect. When the words are weighted by the number of conversations in which they occurred, out of 500, instead of by their total occurrence, the effect on the distribution of sounds is surprisingly small, considering the radical change in method. While the analysis into speech sounds for purposes connected with the design of telephone circuits was the real goal of this study, it is hoped that the information concerning both words and sounds will be of service also to those working in the fields of phonetics and philology. The Reciprocal Energy Theorem By JOHN R. CARSON This paper gives a simple theorem determining relative transmission efficiencies in a two-way transducer, and showing that the conditions for equal efficiencies of transmission in the two directions are simply those for maximum output and maximum reception of energy. The theorem is then applied to radio communication and a second theorem stated and proved by which the ratio of the transmitting efficiences of any two antenna systems is expressed in terms of their receiving efficiences. The paper closes with a mathematical note on a generalization of Rayleigh's Reciprocal Theorem. THE Reciprocal Theorem, originally enunciated by Rayleigh, which has proved so useful to communication engineers, may be stated, with sufficient generality for engineering purposes, as follows: Let an e.m.f. E] , inserted in any branch, designated as No. 1, of a transducer,^ produce a current I2 in any other branch No. 2; correspond- ingly let an e.m.f. Ei" inserted in branch No. 2 produce a current I\' in branch No. 1; then I\ E\ = 1-2 El ■'1 and when £/ = Eo" the currents in the two branches are equal. The engineer, however, is primarily interested in energy rather than current relations, whereas the theorem says nothing explicitly regarding energy relations and relative efficiencies in two-way transmission. It is, however, a simple matter to deduce from it quite general and useful formulas relating to relative transmission efficiencies. In the present paper there will be formulated and proved a reciprocal energy theorem for the general transducer, after which it will be applied to the question of antenna transmission efficiency in radio communication. Consider a transducer having two sets of accessible terminals 1,1 and 2,2. With terminals 2,2 closed by an impedance S2 = ^2 + ixt, let the driving point impedance, as measured from terminals 1,1 be denoted by Zn = Rn + iXn; similarly with terminals 1,1 closed by an impedance Zi = r^ -\- ixi, let the driving point impedance, as measured from terminals 2,2 be denoted by Z22 = -R22 + iX22. Now with the terminals closed by the impedances z-i and 22, let an e.m.f. £1 be inserted in series with the terminal impedance Zi; then the current In, delivered to the transducer at the sending terminals 1,1 is * A transducer is defined as a complete transmission system which may or may not include a radio link, which has two accessible branches, either of which may act as the transmitting branch while the other acts as the receiving branch. These branches may be designated as operating branches. 325 326 BELL SYSTEM TECHNICAL JOURNAL ^"=F^ (1) and the current /12, received by the terminal or load impedance, 22, is given by /12 = ^ ' (2) Here Z12 is the transfer impedance of the transducer for the specified terminations. The power Pu" developed by the generator of e.m.f. Ex is The power Pji delivered to the transducer is P„ = K„|/„l= = ^;^£,= (4) and the power P12 delivered to the load impedance z-i is P,2 = r,\Ii2? =y^,E:~. (5) Now reverse the direction of transmission; that is insert an e.m.f. E2 in series with the terminal impedance z^; corresponding to equa- tions (3)-(5) we have then ^'' -iz22 + 22r^' ^^ I Z22 + Z2 r P21=-^£2^ (8) As a consequence of the Reciprocal Theorem the transfer impedances are equal ; that is Z'ii = Z12. (") From the preceding we get at once the following expressions for the ratios of the powers delivered to the load impedances; Pi2^r2/£iY P2X rAEi) THE RECIPROCAL ENERGY THEOREM 327 rj r 2 \ / JV22 i/Un Ri2 + r^X Zu + zi + n ^22 I 22 Pu" 22 i? 22 Ru Zn + 2i Z22 + Z2 P22 (10) (11) From (10) it follows that for equal total generated powers, the relative transmission efficiency in the two directions is given by 12 P21 R22 + ^2 Ru + r, Zn + Z22 + S2 (12) while on the basis of equal powers delivered to the transducer, the relative transmission efficiency is, by (11) V = 12 P21 R22 Ru Zu + Zi Z22 + Z2 (13) Now in correctly designed communication transmission systems, the terminal impedances are so proportioned with reference to the characteristics of the transducer itself as to secure maximum output and maximum transfer of power from generator to load; the required condition is that the terminal impedances 2) and 22 be the 'conjugate image impedances' of the transducer; analytically stated Zi = Ru — iX u and Z2 = R22 — iX 22- Introducing these relations into (12) and (13), we have ^^ = r/ = 1 (14) and the relative transmission efficiencies are the same in the two directions. We thus have the following propositions: — If a transducer is terminated in its conjugate image impedances — the condition for maximum output and maximum transfer of power — the efficiency of transmission is the same in the two directions. We shall now apply the preceding to the derivation of a simple formula which enables us to determine the relative transmission efficiencies of any two long wave radio antennas.- Consider any antenna, designated as No. 1, and let it be acting as * As pointed out in the paper on "Reciprocal Tlieorems in Radio Transmission " Proc. I. R. E., the Reciprocal Theorem does not hold rigorously in radio transmission if the earth's magnetic field plays an appreciable part in the transmission phenomena. Consequently the formula and proposition which follow apply rigorously only to' long wave transmission; they are probably, however, approximately correct for short wave transmission except in the neighborhood of the critical wave-length 214 meters. See a paper by Nichols and Shelling, "Propagation of Electric Waves over the Earth '' B. S. T. J., April 1925. 328 BELL SYSTEM TECHNICAL JOURNAL a transmitter to a reference antenna, designated as No. 3, which is located at any desired point 3. Let £13 denote the intensity of the (vertical) electric field produced at point 3 by antenna No. 1. Then the current induced in the receiving branch of No. 3 will be azEu, the parameter as being the receiving sensitivity of antenna No. 3. The power Pn transferred from 1 to 3 is then Pi3 = rzaz^Eu^, where r^ is the equivalent resistance of the receiving branch of antenna No. 3. Now reverse the direction of transmission; we have Pzi = riarEzi^. We now suppose that the terminal impedances are adjusted for maxi- mum output and maximum transfer of power and that the power Pu developed by No. 1 when transmitting is equal to the power P33 developed by No. 3 when transmitting. Then it follows at once from the reciprocal energy theorem, that Pu = Pzu and ExzY ^i«i! . 2 £31 / ''30:3' Now replace antenna No. 1 by any other antenna, designated as No. 2 ; we then have from the foregoing E,zY _r,a.^ _ Ez2 / rzdz^ By virtue of the terminal impedances specified, r^ = Ri and ^2 = R2 where Ri and R2 are the resistances of the two antennas as measured from their operating terminals. Consequently, since £32 = Ezi, we have EuV Riar i?l/^l- where hi and h-z are the equivalent heights of the two antennas. The ratio rjn will be termed the 'relative transmission figure of merit' of the two antennas No. 1 and No. 2 with respect to trans- mission between any two specified points. For directional antennas, the parameters cci and a^ will depend on the direction of transmission; that is, the location of the receiving with respect to the transmitting point. THE RECIPROCAL ENERGY THEOREM 329 The foregoing may be summed up in the following proposition. The relative transmission figure of merit oj any two antennas with respect to transmission from a given transmitting point to a given re- ceiving point is equal to the ratio of their resistances as measured from their operating branches, multiplied by the square of the ratio of their receiving sensitivities with respect to transmission from the receiving point to the transmitting point. This theorem has a considerable field of practical utility. For example it enables us to deduce the relative transmitting properties and efficiency of any antenna system from its receiving efficiency. It has already been so applied in one actual case of large importance. Note on the Reciprocal Theorem The proof of the Reciprocal Theorem, as given originally by Lord Rayleigh, was applicable only to 'quasi-stationary' transducers, that is transducers which obey the simple laws of electric circuit theory. In the July 1924 issue of the Bell System Technical Journal the writer stated and proved a generalized theorem subject, however, to the restriction that the permeability ix of the medium shall be everywhere unity. The theorem referred to is Let a distribution of impressed periodic electric intensity F' = F'{x, y, z) produce a corresponding distribution of current intensity u' = u'{x, y, z), and let a second distribution of equi-periodic impressed electric intensity F" = F"{x, y, z) produce a second distribution of current intensity u" = u"{x, y, z), then f{F'-u")dv = f(F"-u')dv, the volume integration being extended over all conducting and dielectric media. F and u are vectors and the expression (F-u) denotes the scalar product of the two vectors. Later Pleijel ^ stated the theorem for unrestricted values of n. In discussing reciprocal theorems in the June 1929 issue of the Proc. I. R. E. the writer expressed some doubt as to the validity of Pleijel's proof (which is entirely different from my own). Subsequent study, however, has convinced me that except for minor and easily remedied errors, the proof is entirely sound. Later the writer discovered that the restriction n = 1 can easily be removed from his own original proof as will now be shown.^ '"Two Reciprocal Theorems in Electricity," Ingeniors V'etenskaps Akademien Nr. 68, 1927. * Another and somewhat different extension of the proof has been derived by my associate Dr. W. H. Wise. 330 BELL SYSTEM TECHNICAL JOURNAL If ju 5^ 1 everywhere and if we write u; = u + curl M = \E + curl M (V) equation (8) of my paper becomes ^ 1 , too r W I iwr\ , ^ , 1 , njr _^ + _J_exp^--jJ. = G+^curlM (2') X and correspondingly equation (9) becomes f{w'-G") - {w"-G')]dv + I ^ {u;'-curl M") - (w;"-curl M')]dv = 0. (3') If now in (3') we replace u; by w + curl M and note that u/X = E, (3') reduces to f{iu'-G") - {u"-G')]dv - f{{G'- curl M") - (G" • curl M') } dv (4') + f{E' -cml M") - (E"-cur\ M')]dv = 0. Finally since E - G = -—A, (4') reduces to f{{u'-G") - {u"-G')}dv ^l^fUA'- curl M") - {A" ■ curl M') ]dv ^ {). (5') c But /(A' -curl M")dv = /(M"-curl A')dv = T- f- ^(5"-curl^0^i; 47r J II = -T- f ^^-^ (curl A" ■ curl A')dv, 47r J n SO that the second integral of (5') vanishes and f{{u'-G") - {u"-G')}dv = 0, (60 which is equation (9) of the original paper. The rest of the proof of the theorem is now simply that of the original paper. It will be observed the theorem is stated for the current u = \E; that is the conduction (plus polarization) current. Ballantine ^ in ^ The paper itself must be consulted for the significance of the symbols and the method of attack and proof. 6 June 1929 issue of Proc. I. R. E. THE RECIPROCAL ENERGY THEOREM 331 discussing this subject states that the theorem holds for the current w = \E -\- curl M. This cannot be true in general, however, because from the foregoing in order that the theorem should hold for the current w, it is clearly necessary that f{{F'-cm\ M") - (f"-curl M')]dv = 0. This is only true in the exceptional cases where the impressed force is derivable from a potential; that is, curl Z' = 0, or else f = 0 where M ?^ 0. The Approximate Networks of Acoustic Filters By W. P. MASON The approximate equivalent electrical networks of acoustic filters are developed in this paper, from the lumped-constant approximation networks for electric lines. In terms of this network, design formulae have been developed for all single band pass filters. It is possible, from these formulae, to determine the physical dimensions of an acoustic filter necessary to have a given attenuation and impedance characteristic. THE original theory of acoustic filters given by Stewart ^ is based upon the representation of such filters by means of lumped constants in the form of a 7" network. More recently, the writer ^ has presented a theory of acoustic filters, showing that they are equivalent to a combination of electric lines. Lines, as an approxi- mation, can be represented by networks with lumped constants, and hence an acoustic filter has a lumped-constant approximation network, which should represent the filter well at low frequencies. It is here shown that the network proposed by Stewart is a first approximation to the network of electric lines given in the former paper.^-^ This first approximation represents the low pass filter well at low fre- quencies, but does not very adequately represent the band-pass filters. Accordingly, a second approximation is developed. All of the single band-pass filters have been analyzed and design formulae are given for them in terms of the second approximation network. The Approximate Lumped-Constant Networks of Acoustic Filters An acoustic filter, as developed so far, consists of a main conducting tube, and a side branch. In a symmetrical filter, the side branch is connected to the main conducting tube half-way between the two ends, as shown on Fig. 1 . The type of filter obtained depends primarily on SIDE BRANCH / ^MAIN CONDUCTING TUBE Fig. 1 1 Stewart, Phys. Rev., 20, pp. 528-551, 1922. Phys. Rev., 25, pp. 90-98, 1925. * Mason, Bell System Technical Journal, 6, pp. 258-294, 1927. ^ This fact has also been pointed out by Stewart, Journal of the Optical Society, July 1929, and by Lindsay, Phys. Rev., 25, pp. 652-655, 1929. 332 APPROXIMATE NETWORKS OF ACOUSTIC FILTERS ?,?>?> what type of side branch is used, the resonances of the latter deter- mining the frequencies of maximum suppression. The equivalent electrical circuit for an acoustic filter, was shown in a previous paper ^ to be two lines shunted by the impedance of the side branch. This representation is shown on Fig. 2. To obtain a lumped- Fig. 2 constant representation for this network, it is necessary first to con- sider the lumped-constant representation of a line, which is discussed below. A. Lumped- Constant Representation of a Line In a previous paper - it was shown that the propagation constant of a tube is given by the equation p-. _ -^[(•-tVS)-^V£^]. ^'^ while the characteristic impedance is given by the expression Z = pc'P (2) In these equations co is 27r times the frequency, c the velocity of sound, Po the perimeter of the tube, 5 its area, p the density of the medium and 7'-, a constant 1 elated to the viscosity, which for air has the value 4.25 X 10-* in c.g.s. units. A tube is the analogue of an electric line with distributed resistance, inductance, and capacity. No quantity corresponding to leakance is present. To determine the values of these quantities, use is made of the well known equations for a line Z = R -f jcoL G + jcoC ' P = V(i^-f jcoL)(G-f jcQ, (3) where R, L, G and C are respectively the distributed resistance, induc- tance, leakance, and capacity of the line per unit length. Comparing - Loc. cit. 22 334 BELL SYSTEM TECHNICAL JOURNAL (3) with (1) and (2), it is found that R L C G P (4) pc 0, 2 » neglecting small correction terms. These are the equivalent distrib- uted constants per unit length of the pipe expressed in acoustic impedance units. The representation of lines with distributed constants by means of networks containing lumped constants has received considerable atten- tion.^ With three impedances, either the T or it network representa- tion shown on Fig. 3, can be used. /OL 2S Ppi-Ml'^P^ X SL PC' 2 S X X PoL\/r^fpuj _ 25^ V 2 Fig. 3 The impedances of short or open circuited lines can be represented approximately by fewer elements than three. The first approximation for a short circuited line is an inductance and resistance equal to the sum of the distributed inductances and resistances of a line, while the first approximation for an open circuited line will be a capacity equal to the distributed capacities of the line. These approximations hold for very low frequencies. The second approximation for open and short circuited lines can be obtained with three impedances, as shown 4 A. E. Kennelly "Artificial Electric Lines, 1917." K. S. Johnson "Transmission Circuits for Telephone Communication, 1925," page 151. APPROXIMATE NETWORKS OF ACOUSTIC FILTERS 335 on Fig. 4. These representations follow directly from the T or ir 2S2V 2 — wwv — PL 2S SL -OC2 Fig. 4 yOL s2 V 2 rVWW — O^MXIS SL 20C2 network representation shown on Fig. 3, by open or short circuiting the T and tt networks, respectively. B. Lumped-Constant Representation oj an Acoustic Filter In his theory of acoustic filters, Stewart has represented an acoustic filter by the network shown on Fig. 5, where Z^ is the impedance of the /PL s, PL Fig. 5 side branch. Stewart has represented the side branch impedance, by either one or two elements, depending on the side branch, and the main branch by a single inductance, equal to the sum of the distributed inductances of the tube. This corresponds to the first approximation of the representation of a line by lumped constants. This repre- sentation gives good results for the low pass filter, but does not repre- sent, very adequately, the band-pass filters. The best second approximation for an acoustic filter, employing two elements to represent the main conducting tube, is shown on Fig. 6. PL S| ■ S|L PC2 ;z2 ■ S|L "PC2 Fig. 6 336 BELL SYSTEM TECHNICAL JOURNAL The main conducting tube is represented by an L network containing the total distributed capacity of the tube in the shunt arm, and the total distributed inductance of the tube in the series arm. The side branch impedance shunts the two L networks at their center. The propagation constant and characteristic impedance of this structure are given by the expressions cosh P = 1 2ix)'-L'- jicpL I (jo'-L'~ /^2^ 1 1 + joopL 2Zo.S', J PC-' 1 1 - •) f •' \ -1 wLSi{2Z2) L- 1 - CO '-IJ {^) where S\ is the area of the main branch. If these equations are compared with those given in the former paper,- it is seen that they are approximately those obtained by taking the first two terms of the expansions of the trigonometrical functions. The characteristics of the filter are not very readily seen from equation (5), but can be readily found by transforming the network shown on Fig. 6, into the much more general lattice network shown in Fig. 7. APPROXIMATE NETWORKS OF ACOUSTIC FILTERS 337 That the network shown on Fig. 7 is the equivalent in characteristic impedance and propagation constant of that shown on Fig. 6, can readily be verified by substituting the impedances of the lattice net- work into the formulae for a lattice network Z = ^ZaZb; cosh P Zn + Za Zb ~ Za (^') where Za is the impedance of one of the series arms, and Zb that of one of (he lattice arms. A lattice network has a pass band when (he reac- SERIES y/ ARM ~~y^ / FREQUENCY / LATTICE /*— ARM r' Fig. 8 tance of the series arm is of opposite sign to that of the lattice arm. When the reactances of the two arms have the same sign, an attenua- tion band results, while when the reactances of the two arms are equal, an infinite attenuation constant results, since here the lattice will be a balanced Wheatstone bridge. For example, suppose that a side branch impedance, equivalent to an inductance and capacity in series, is used. The impedance of the lattice arm has two zero impedance points — one of which is at an infi- nite frequency — and two infinite impedance points- — one of which is at zero frequency — as shown on Fig. 8. The impedance of the series arm 3.^8 BELL SYSTEM TECHNICAL JOURNAL Is that of an anti-resonant circuit, as shown on Fig. 8. There are two possible impedance characteristics for the series arm, in relation to the lattice arm, which will give a single band filter. One of these is obtained by letting the series arm have an infinite impedance when the lattice arm has a zero impedance, which results in a low pass filter. The second relation — which is that shown on Fig. 8 — is obtained by letting the series arm have an infinite impedance when the lattice arm has an infinite impedance. The pass band is between zero frequency, and the frequency at which the lattice arm resonates. In a similar manner, the other types of acoustic filters can be ana- lyzed. C. Side Branch Impedances The possible types of side branches can be divided into two classes, those which are entirely enclosed, and those which are open to the air. The first kind are characterized by a series capacity, while the second kind always have a shunt inductance. One of the simplest side branch impedances is a short tube open on the end. The first approximation to this side branch is an inductance, as shown on Table I, No. 1, equal to the total distributed inductance of the tube. This approximation holds well if the product of the tube length by the frequency, is not too large. A longer tube, open on the end, can be represented by an inductance and capacity in parallel as discussed in Section A and shown on Table I, No. 2. A tube closed on the end can be represented by an inductance and capacity in series as shown on Table I, No. 4. When these tubes are used as side branches, an additional factor comes in — an end correction. That is, the side branch must be con- sidered as extending into the main branch for a distance proportional to the radius, because a motion of air in the direction of the side branch, occurs in the main branch. The value of this effect has been investigated by Rayleigh, who found that this effect can be calculated by increasing the length of the tube by a length equal to .785 times the radius. Another correction applies to an open ended tube, which has been determined experimentally as .57 times the radius. Hence the length of an open ended tube must be considered as /' = / + (.785 + .57)r. A straight tube can give all the combinations of side branch imped- ances, but one of its dimensions is necessarily limited, namely the area. For the area cannot become larger than the area of the main tube, since otherwise it could not be connected to the main tube. By APPROXIMATE NETWORKS OF ACOUSTIC FILTERS 339 ELEMENT -^IKJCOM^- STRUCTURE NO. I L = P\' VALUES OF CONSTANTS V = 1 -+- l.355r S = TTr2 ELEMENT L c — L = PX Xi S ^ = 2^ > C2 STRUCTURE N0.2 /J VALUES OF CONSTANTS V = 1 + 1.355 r S = TTr2 ELEMENT p\- I'S STRUCTURE NO. 4 I 1' I '1 r VALUES OF CONSTANTS V= X + 0.785 r S = TTr2 ELEMENT L=^ C=^ 2S /0C2 STRUCTURE NO. 5 -^t t-- 2/-h— '^ VALUES OF CONSTANTS Log m). 0.46 t r2t (r2-r2)T.lX (^i-f) T + - vft S = n Log m)^°-^^ ELEMENT STRUCTURE NO. 3 -\2^ - o (T 8 NOIiVONjliV O 8 N0llVnN3iiV 8 N0IXVnN3iIV 8 NOIlVnN31iV o tyw C4- u t= (\J ^1?f ^8 I2i (M wp ■^c^ <.8 1? c~ tC CJ 1 1 ur Q. ^ o N u^ 1 (v_8 wp 0J_ 1 Q. T 't o fsi 8 N0IIVnN3ilV o ^ t f ' Q. c/i 8 N0llVnN3IlV '^ri^-^ Ci u) + lOlftI T- 8 N0llvnN3iiV Qk/l ^ tf» 1= «-l^ 8 NOiivnNiii* ° al' "iLSi^T, ~|cv|8 nMvi~8 ~^|(vi8 i-iLS ~rt° Ol - •3 8 N0livnN311VO ^ is; t- < < X <£5 Contemporary Advances in Physics, XX Ionization of Gases by Light By KARL K. DARROW The subject of this article, the ionization of gases by ultraviolet light, is a narrow but singularly inviting department of modern physics. The obstacles to experiment are so great that they are only now being overcome by the latest improvements in laboratory technique; nearly all the valu- able data are of recent acquisition, and the period of discoveries is not yet past. Some of the results afford excellent confirmation of present atomic theory; others are still obscure and challenging. THE subject of this article is more narrowly restricted than those of many others of the series. It is narrower even than the title might imply; for by "ionization" I mean for the present only the detachment of the most loosely bound electron of a molecule or an atom, and by "light" only the waves of the visible spectrum and that adjoining range of wavelengths to which the name of ultraviolet is customarily confined. Either of these limitations is implicit in the other; for though most molecules are fashioned with electrons bound with varying degrees of tightness, and the removal of any one thereof is an act of ionization, it is beyond the power of such light-waves to abstract any except the loosest. Perhaps it will be found instructive if for so definite and circumscribed a problem I relate the methods of experiment, the data of the experiments, the simple theory, and the artifices which have been conceived to reconcile the theory and the data, sometimes with success and at other times in vain. Most of the really valuable data are of recent acquisition, for there are difficulties hampering the attack upon the problem, which the progress of laboratory technique is only gradually clearing away. Consider, for instance, the question of providing the light. It is de- sirable to be able to illuminate the gas with monochromatic light of any wavelength, photons of any energy. When ionization by electrons is being studied, one varies the energy and the wavelength at will by varying the voltage impressed on the electrons. With light, this is not within our power; one has to take the quanta as they are supplied by a luminous source. If the spectrum of the source consists of bright lines widely separated, the ionization which any of them alone produces may be measured, and the energy of the quanta is very narrowly defined. On the curve of ionization vs. frequency, then, every spectrum line supplies an 341 23 Contemporary Advances in Physics, XX Ionization of Gases by Light By KARL K. DARROW The subject of this article, the ionization of gases by ultraviolet light, is a narrow but singularly inviting department of modern physics. The obstacles to experiment are so great that they are only now being overcome by the latest improvements in laboratory technique; nearly all the valu- able data are of recent acquisition, and the period of discoveries is not yet past. Some of the results afford excellent confirmation of present atomic theory; others are still obscure and challenging. THE subject of this article is more narrowly restricted than those of many others of the series. It is narrower even than the title might imply; for by "ionization" I mean for the present only the detachment of the most loosely bound electron of a molecule or an atom, and by "light" only the waves of the visible spectrum and that adjoining range of wavelengths to which the name of ultraviolet is customarily confined. Either of these limitations is implicit in the other; for though most molecules are fashioned with electrons bound with varying degrees of tightness, and the removal of any one thereof is an act of ionization, it is beyond the power of such light- waves to abstract any except the loosest. Perhaps it will be found instructive if for so definite and circumscribed a problem I relate the methods of experiment, the data of the experiments, the simple theory, and the artifices which have been conceived to reconcile the theory and the data, sometimes with success and at other times in vain. Most of the really valuable data are of recent acquisition, for there are difficulties hampering the attack upon the problem, which the progress of laboratory technique is only gradually clearing away. Consider, for instance, the question of providing the light. It is de- sirable to be able to illuminate the gas with monochromatic light of any wavelength, photons of any energy. When ionization by electrons is being studied, one varies the energy and the wavelength at will by varying the voltage impressed on the electrons. With light, this is not within our power ; one has to take the quanta as they are supplied by a luminous source. If the spectrum of the source consists of bright lines widely separated, the ionization which any of them alone produces may be measured, and the energy of the quanta is very narrowly defined. On the curve of ionization vs. frequency, then, every spectrum line supplies an 341 23 342 BELL SYSTEM TECHNICAL JOURNAL experimental point of which at least the abscissa Is certain. But at the frequencies between the lines, there is no way of getting information; and many a published curve is traced by guesswork right across the regions of major importance where data are essential, simply because Nature left those regions vacant of lines in the spectrum of the mercury arc! If on the other hand the source has a continuous spectrum or one crowded with bright lines, the device for resolving or filtering the light will transmit to the ionizable gas photons not of a single wavelength, but of a range of many Angstroms — dozens or scores of Angstroms, perhaps even a hundred. A single measurement may be, and usually is, plotted as if it belonged to the central wavelength of the transmitted band; but actually the ionization results from waves of all the wide range, and not even from a uniform distribution of energy across the range, but from a distribution affected by the qualities both of the source and of the resolving apparatus. With chemical filters, i.e. with coloured absorbing liquids, bands of transmitted light may be formed in various parts of the spectrum. They are likely to be broad and hazily bounded, and limited in number; but the liquid filters are inexpensive and easy to handle, and in tracing backward the sequence of observations on any one substance, one often finds that the very earliest were made with filters. Monochromators— which is to say, spectroscopes — form bands of which the central wavelength and the width may be varied at will. This sounds ideal; but in practice, of course, the narrower the transmitted range of wavelengths, the scantier the transmitted energy; and one must compromise as best one may between a band too narrow to produce a measurable degree of ionization, and one so broad that it is hard to apportion the credit for the effect which it produces among the frequencies which make it up. It will be evident that the ideal curve, drawn through experi- mental points scattered thickly all through the spectrum and each corresponding to a single wavelength, is difficult of attainment and even of approach ! As the atmosphere of the earth prevents us from observing the spectra of the stars at shorter wavelengths than some 280w/i, so the opacity of all terrestrial solids prevents us from projecting quanta of smaller wavelength than \2Smix, into an enclosure. Indeed, it is only from fluorite and only from occasional samples of fiuorite that one can make windows which are transparent so far out; the next best and much the commoner substance, quartz, ceases to transmit at about 145m/z. We are thus almost entirely debarred from observations on the noble gases and on the common diatomic gases, which is deplorable. CONTEMPORARY ADVANCES IN PHYSICS 343 The desired photons being successfully fired into the gas, the next problem is that of distinguishing the ionization they may cause in it from whatever other liberation of charge they may effect in striking walls, electrodes, or any of the other furniture within the tube. Light of sufficient frequency to ionize a gas will usually be able to produce an outflow of electrons from almost any metal. One takes of course the elementary precaution of designing one's tube in such a way, that the beam of light traverses it from entrance-window to exit-window without touching any electrode; but the disparity of the effects is nevertheless so great, that a modicum of scattered light may evoke more electrons from the metal than the primary beam in all its strength detaches from atoms of the gas. Some experimenters use alternatively two electrodes of very unequal size, expecting that a current due to ionization of the gas will be the same in magnitude whichever they use as cathode, while a current due to light falling on the electrodes will be greater when the larger is the cathode. Some vary the density of the gas, assuming that if the current is proportional to density it must be due to the effect which they seek; but it would also vary as the density, if instead it consisted of electrons expelled from the electrodes by light proceeding from excited atoms of the gas. Some finally have so designed their apparatus that they perceive positive ions only; this seems to be the safest way. Like Hughes,* I will divide the data according to the character of the gases to which they refer: the common or "permanent" (chiefly dia- tomic) gases first, then mercury, finally the alkali metals. The permanent gases can be disposed of in short order, for our knowledge in this field is scanty, though surprising. Measurements of ionization by electron-impacts, and what little has yet been deduced from spectra, agree in indicating that for all of them (with the slight exception of nitric oxide NO) the ionizing-potential is greater than 10 volts. Translating this figure into wavelengths of light, we infer that only photons of smaller wavelength than 125w7/x can ionize such a molecule in a single impact. Therefore light which is able to traverse any window of solid substance should be unable to ionize any perma- nent gas (except NO) ; in other words, any such gas enclosed in a tube should be immune to ionization by any radiation entering from outside. Yet there is unimpeachable evidence that air, and oxygen and nitrogen separately, and possibly hydrogen and iodine, are ionized by light which has penetrated fluorite. The threshold for these gases must * A. L. Hughes, Ionization of Gases and Vapors by Light (Washington University Studies, 1929). 1 have benefited much by this article, and also by that of F. L. iVIohler, Recombination and Photoio7iization (Reviews of Modern Physics 1, 216-227 (1929)). 344 BELL SYSTEM TECHNICAL JOURNAL therefore lie on the long- wave side of 125m/x. For air it is presumed to lie between this and 145m/x, since a plate of quartz holds back the ioniz- ing rays. The discrepancy between these and the expected thresholds may not seem large, but it is important. Naturally one has recourse to the idea of a two-stage ionization, occurring when two quanta in succession are absorbed by the same molecule— an idea which, we shall see, is frequently invoked in other cases. If this is valid, the ionization should increase as the square of the intensity of the light. There seem to be no data bearing on this point. To quote from Hughes, "further investigations in this field are badly needed."^ With mercury the situation is much more definite, but for those who like to have simple theories verified completely it is no more satisfac- tory. The spectrum of the mercury atom is well mapped and well inter- preted, and the ionizing-potential for electron-impacts has been deter- mined over and over again. From both of these it follows that ioniza- tion by single photons should be possible only at wavelengths smaller than 1188A. However it is certain that the light of the famous res- onance-line of mercury, 253 7A, is able to ionize the vapor of the ele- ment whence it proceeds.- This seems the natural equivalent of the well-known fact that when mercury atoms are bombarded by a sufficiently dense electron-stream, ionization begins at the resonance-potential. The quanta of the wavelength 2537A have 4.9 equivalent volts of energy. Such a photon strikes an atom, and excites it transferring if from the normal into a certain excited state, denoted by the symbol 2^Pi; a second comes along and likewise is absorbed, bringing the energy of excitation of the atom up to twice 4.9 equivalent volts; this amount falls short of the ionizing potential by only 0.6, and a third photon more than sup- plies what is required. So runs the simple interpretation; but we shall see that only the first of these steps is confirmed by further experiment, and that the rest of the process must happen in some other way, though the way is far from clear. 1 For references to the literature I refer to Hughes {I.e.) Among the latest papers are those of A. L. Hughes {Proc. Camb. Phil. Soc, 15, pp. 483-491 (1910)); F. Palmer (Phys. Rev. 32, pp. 1-22 (1911)); E. B. Ludlam {Phil. Mag. (6) 23, pp. 757-772 (1912)); W. West, E. B. Ludlam {Proc. Roy. Soc. Edinb., 45, pp. 34-41 (1925)). Some of the early work on air was confused by what appears to have been a photoelec- tric effect of particles of colloid size ("nuclei" or Kerne) produced by the action of ultraviolet light on impurities in the air — one of the once-popular and now forgotten problems of physics. 2 Literature: G. F. Rouse and G. W. Giddings, Proc. Nat. Acad. Sci., 11, pp. 514-517 (1925); 12, pp. 447-448 (1926). F. G. Houtermans, ZS. f. Phys., 32, pp. 619-635 (1925). Twenty years ago W. Steubing {Phys. ZS., 10, pp. 787-793 (1909)) observed that light coming from a mercury arc and passing through quartz was able to produce a current in a tube containing mercury vapor; but his result has been impugned. CONTEMPORARY ADVANCES IN PHYSICS 345 It was the beautiful experiment of Rouse and Giddings which con- firmed that the first of the steps is the entry of the atom into the 2'P] state; for they showed that ionization of the gas occurs only when the impinging quanta have just the energy required for that transition, not when they have a little less or even a little more. This they were able to show because of the phenomenon of "self-reversal." When a luminous gas becomes dense and hot, the lines of its emission-spectrum broaden ; for the atoms perturb one another, the energy- values of their stationary states are changed by various amounts, and the frequencies of many of the quanta which emerge are appreciably shifted upwards or downwards from the original or "standard" values appropriate to isolated atoms. If in addition the region of density and heat is sur- rounded by another where the gas is cooler and more rarefied, the atoms in this outer zone, being relatively unperturbed, will be able to absorb the quanta having the standard frequencies, but not those others of which the frequencies are shifted. In technical language, the "core" of the line is absorbed; only the "wings" pass through; the Hne exhibits "self -reversal." In the spectrum of the ordinary mercury- vapor lamp, the line 2537 is notably self -reversed. Cooling the lamp with flowing water or an air-blast, however, abolishes the effect; the line shrinks to its normal narrowness, the wings disappear, but the photons of the core are able to escape from the tube. Any action there- fore which is performed by the light of a cooled mercury-vapor lamp, but ceases when the cooling is suspended, must be due to quanta having energies adjusted exactly to the values which are able to excite isolated atoms of mercury. Rouse and Giddings found that ionization of gaseous mercury is precisely such an action.^ We cannot so readily conclude that the second step in the ionization- process is the absorption of a second 4.9-volt quantum. It would be rather of an odd coincidence, if there were an excited state of the mercury atom differing in energy from the 2^P] state by just so much as this latter differs from the normal state — not, however, an impossible coincidence. Another and a stronger argument is furnished by the fact that when quanta of various wavelengths shorter than 2537 — in- cluding some which could transfer the atom from the 2^Pi into other known excited states — are projected into the gas along with 2537, the rate of ionization is not augmented. If none of these can help the electron to escape, it is not so likely that a second quantum of precisely the wavelength 2537 can achieve it. Moreover the duration of the ' There was still a residual current in the irradiated tube when the cooling of the lamp was discountinued; but it depended on the size of the cathode in such a way as to suggest that it was due to light falling on that electrode (cf. page 343). Houter- mans later verified this result. 346 BELL SYSTEM TECHNICAL JOURNAL 2^Pi state is known to be so short (of the order of 10"'^ second) that under the actual conditions of some of the experiments an atom would not often meet two quanta in such quick succession that at the advent of the second it would still be in the 2^P] state into which the first had put it. In other words, the number of 2^P] atoms in the gas at any moment is too small. This last would be a serious obstacle to any theory, but for the fact that the mercury atom possesses another stationary state slightly below the 2^Pi, the which is metastable. This is the 2^Pq state, of 4.7 equivalent volts; its mean duration may amount to something like a hundredth of a second. Now collisions of mercury atoms in the 2^Pi state with atoms of certain other kinds, argon notably, may cause the former to pass over into 2^Po- This is an instance of "collisions of the second kind." When mercury-vapor is mixed with a much larger quantity of argon and is illuminated with 2537 light, the number of 2^Po atoms is at any moment much greater than the number of 2^Pi atoms would be, if the argon were absent ; further, it is proportional to the amount of argon. Now F. G. Houtermans found that the rate of ionization, in mercury mixed with argon and irradiated by 2537, is proportional to the amount of argon. Therefore, in one stage of its progress from a normal atom to an ion, the mercury atom must be in the 2^Po state. It enters this state from the 2^Pi because of a collision with an argon atom. How does it leave? by absorption of a second 4.9-volt quantum? Two of the considerations of the last paragraph but one speak against this idea, and Houtermans thinks that the 2^Po atom collides with another which is in the 2^Pi state, and there is an interaction — this would be another sort of "collision of the second kind" — in which one of them adds to its store of energy all or most of what the other possesses. So it arrives within an equivalent volt or so of the state of ionization; if one were to take over all the energy of excitation of the other, it would have 4.7 + 4.9 = 9.6 equivalent volts, out of the 10.4 required. Still a third step seems to be essential. The reader may have wondered that I have as yet said nothing about the dependence of ionization on intensity of light, for evidently the former should increase as the cube of the latter if the process is a three-stage one as I have sketched. The matter has been tested by experiment; the answer was unexpected, for the ionization varies as the square of the light — as though the process were of two stages.^ We * This simple result was obtained only over certain ranges of temperature and pressure of the vapor, but these were precisely the ranges where both are low, and we should expect the result to be most reliable and least subject to confusion by secondary effects. As the pressure rises so does the exponent n in the relation ionization = {intensity)'^. CONTEMPORARY ADVANCES IN PHYSICS 347 cannot suppose that the atom after its second gulp of energy picks up the remaining 0.8 volt in a collision with a fast-moving ordinary atom, for at normal temperatures such fast-moving atoms would be exces- sively rare. Houtermans suggests that when a 2^i'o and a 2^Pi atom collide with one another, they unite to form an ionized molecule Hg2+. This is far from being the only case in which a molecule is invoked as the deus ex machina to help out with an otherwise untenable theory. We turn now to the alkali metals, or rather to the three heavier among them, caesium, rubidium, and potassium. With these it is more nearly possible to get a full view of the situation. The phenom- ena are not confined to spectrum ranges in or beyond the remotest attainable fringes of the ultra violet. Indeed, in these four cases, even the wavelength where ionization by single impact should begin is well within reach, being in the nearer ultra violet; 241 2A for Na, 2856 for K, 2968 for Rb and 3183 for Cs. Ionization currents are provoked by light at even greater wavelengths ; this resembles the case of mercury irradiated by 2537, and is equally perplexing, indeed more so. They are however much greater, near or beyond the limiting wave- length for one-stage ionization; and there, we seem to be witnessing the simplest process of all. With caesium, rubidium and sodium, the data in this range conform to simple theories in a gratifying way. I will consider these first, and then the most mysterious case of all, that of potassium. The vapor pressures of the alkali metals increase with atomic num- ber, and for rubidium and caesium are great enough to permit the methods employed with the gases mentioned above: which is to say, that stationary vapor of known density may be illuminated by light of known intensity, and the amount of ionization be measured absolutely by drawing off all the ions. This I denote as the "absolute" method. There is another, the "method of space-charge annulment." The tube containing the vapor contains also a hot-filament cathode and some form of anode, and the filament is kept so hot, the P.D. between it and the anode kept so low, that the electron-borne current between cathode and anode is limited by its own space-charge. W^hen positive ions are formed in the vapor, as in these experiments they are by light, a fraction of the negative space-charge is annulled, and the current increases. The change in the current is a measure of the num- ber of positive ions formed. Nothing of the sort results if light falls on solid objects in the tube and ejects electrons, an insensitiveness which is a great advantage of the method. For positive ions it is a very sensitive method; one finds such statements as "each positive ion formed causes a million extra electrons to flow from cathode to anode," 348 BELL SYSTEM TECHNICAL JOURNAL and Foote and Mohler, who were the first to apply this method to ionization by Hght, perceived the effect at pressures of mercury vapor as low as .002 mm. Hg. It does not permit of absolute measurements; but one may use it to make accurate measurements of the relative ionizing-power of light of any number of wavelengths, and then stand- ardize them en bloc by a determination at a single wavelength with the absolute method. 3200 3000 2800 2600 WAVE LENGTH 2400 2200 Fig. 1 — Ionization by light plotted as function of wavelength for caesium (Critical wavelength: 3184A). (F. L. Mohler, C. Boeckner). I now reproduce two of the most recently published curves of ioniza- tion vs. wavelength: Fig. 1 for caesium, from F. L. Mohler and C. Boeckner;^ Fig. 2 for rubidium, from E. O. Lawrence and N. E. Edlefsen.^ It is the downward trend of these curves from the limiting wavelength towards shorter waves which interests us now. Ionization by light of a given intensity is most abundant when the quanta have just the energy required to detach the electron, and no more. The more the energy of the photon exceeds the strictly necessary value, the less it is likely to be captured and have its energy spent for ioni- zation. The various theories, except for one, predict a steady downward trend; one in particular, that of R. Becker, supplies the broken curve 5 Bur. Stand. Journ. Res., 3, pp. 303-314 (1929). ^Phys. Rev., (2) 34, pp. 233-242 (1929). CONTEMPORARY ADVANCES IN PHYSICS 349 of Lawrence and Edlefsen's figure (relative ordinates have no signifi- cance, it is only the trends of the curves which should be compared). Mohler and Boeckner also measured the actual number of ions produced by light of known intensity in a known quantity of gas, using of course the absolute method, and expressing their results in the following way. Suppose a thin stratum of gas, of thickness dx and area A. Denote by N the number of atoms per unit volume of the 2400 2600 2800 3000 3200 ANGSTROMS 3400 3600 3800 Fig. 2 — Ionization by light plotted as function of wavelength for rubidium (Criti- cal wavelength at 2968A). Circles and crosses correspond to different densities. (E. O. Lawrence, N. Edlefsen.) gas; then NAdx will stand for the number in the stratum. Denote by Q the total number of photons striking the stratum in unit time; suppose that they fall upon it perpendicularly, and are evenly dis- tributed over its area. The number of ions formed in the stratum in unit time, / will be proportional to NAdx and to Q/A. Write: / = kNQdx the coefficient k is the quantity of which the experiments are designed to reveal the value. (We should not be entitled to expect this to be constant, if more than a small fraction of the quanta were spent in ionization; but in the practical cases we may.) The values which they give are (2.3 ± 0.2) • 10~^^ for caesium and 1.1 • 10~^^ for rubidium, 350 BELL SYSTEM TECHNICAL JOURNAL at the limiting frequency in each case. Earlier E. M. Little ' had got a value two orders of magnitude lower for caesium; this difference has not been reconciled. These values will later be compared with those to which the theories lead. The upturn in the curve of Fig. 1 on the shortwave side of 2600A may serve ^ as an introduction to the case of potassium. Adjourning therefore the discussion of the righthand part of the curve of Fig. 2, I take up next this strange and singular case. The first who plotted an ionization-vs-wavelength curve for potas- sium was E. O. Lawrence.^ The vapor-pressure of potassium being low, he so designed his tube that the beam of light passed across the vertically-rising jet of gas distilling from a pool of highly-heated metal. This expedient was used by all the other physicists who worked upon potassium, and was at one time held responsible for the curious results, until finally Mohler and Boeckner confirmed the previous data by measurements on stagnant vapor. The ionization-current was col- lected by electrodes placed on either side of the jet and away from the light; so the method is fit to give the relative ionizing-powers of light of various wavelengths, though not an absolute measurement, the density in the jet being unknown. Lawrence's monochromator provided beams of light extending over some 80A of the spectrum. Few data can have been more unexpected, indeed more positively unwelcome, than those which he obtained; for what they intimated was, that ionization begins, or at least the sharp increase of ionization occurs, at a wavelength definitely too small. It seems as though a photon could not ionize a potassium atom without having definitely more than the necessary energy; a conclusion which would be in disaccord with fundamental theory, and with the (subsequent) experiments upon rubidium and casium. New experiments upon potassium gave comfort to the theory, but also demonstrated the anomaly which Lawrence had discovered.'" The 7 Phys. Rev., (2) 30, pp. 109-118, pp. 963-964 (1927). * However it does not appear in the corresponding curve obtained by Lawrence and Ediefsen. ^Phil. Mag., 50, pp. 345-359 (1925). There had been four precursors: S. H. Anderson, L. A. Gilbieath, R. C. Williamson, R. Samuel (for the references, see Hughes, I.e.). The earliest two reported ionization at wavelengths where it now seems unlikely that true ionization of the vapor would have been perceptible; the others used chemical filters and so were unable to plot a curve, but seem to have observed the weak ionization produced between 2800 and 3100A. " Such a proof would relieve us from one of the greater difficulties of the " molecule " hypothesis — the necessity of assuming that ionization of a K2 molecule by light is an event thousands of times as probable as that of a K atom, for in potassium vapor under the actual conditions free atoms are believed to be a thousand times more abun- dant than molecules, and yet the ions which we are ascribing to the latter are much more plentiful. R. W. Ditchburn and F. L. Arnot {Proc. Roy. Soc. 123, pp. 516-536 (1929)) found nothing but K+ ions in the ionized vapor, thus disposing of the notion that the process might consist in the detachment of an electron from a thenceforward stable K2 particle. CONTEMPORARY ADVANCES IN PHYSICS 351 curve which I display as Fig. 3 is taken from the latest paper,' ^ but the marked points comprise those of Lawrence's first article (large circles) and those obtained in the interim by R. C. Williamson. '- The monochromators used in these late researches gave narrower wave- POTASSIUM 2200 2400 2600 ANGSTROMS 2800 3000 Fig. 3 — Ionization by light plotted as function of wavelength for potassium (Critical wavelength: 2856A). Circlets, crosses and large circles correspond to different sets of observations by Lawrence & Edlefsen, Williamson & Lawrence. (E. O. Lawrence, N. Edlefsen.) length-bands than those used formerly, and so revealed the small peak at the proper limiting-frequency which had eluded Lawrence at the outset. The much more prominent peak at shorter waves remains outstanding. The data, be it mentioned, are here reduced to equal intensities of light for the various wavelengths. 11 Lawrence & Edlefsen, Phys. Rev., (2) 34, pp. 1056-1060 (1929). i2Proc. Nat. Acad. Set., 14, pp. 793-799 (1928). 352 BELL SYSTEM TECHNICAL JOURNAL The molecule was invoked at once as the deus ex machina; the ioniza- tion beginning beyond the proper wavelengths was supposed to be ionization of molecules, with or without dissociation. So long as the threshold was thought to be near 2600 or 2550, this idea was fortified by the following calculation. Suppose that a photon of wavelength 2555A has just the energy required to split a K2 molecule into a K atom, a K+ ion, and a free electron; and that a photon of 2856A has just the energy required to split a K atom into a K+ ion and a free electron. One easily sees that then the difference between the energies of these two photons would be just the energy required to split a K molecule into two neutral K atoms. The difference amounts to 0.5 equivalent volt. This figure agrees ^^ with independent estimates of the value of the latter quantity, which is the heat of dissociation of K2. The force of this agreement has just been weakened by the curve of Fig. 3, showing as it does that the ionization in question begins near 2700A — weakened, but not destroyed, for the ions pro- duced by waves shorter than 2555 might be explained in a way which the reader will easily imagine after the next two paragraphs. The other alternative is, to hope that quantum mechanics will presently prove that the ionization-vs-frequency curve for the potassium atom ought to display both the maxima which are found. Return now to the curve of Fig. 2 for rubidium. On the long-wave side of the limiting-frequency there is a series of peaks ; they lie at the frequencies of the various members of the principal series of lines in the Rb spectrum. Even more striking peaks of this sort were earlier obtained with caesium by Foote, Mohler, and R. L. Chenault;^^ the relevant part of one of their curves is shown as Fig. 4. Palpably these are phenomena of the same sort as one meets when mercury is irradiated by 2537; and they signify an ionizing-process of two or more stages, the first of which is excitation by the absorption of a photon. There is probably no need to suppose more than two stages; the energy received by the atom from the photon is always much more than half of what is required to ionize. It is supposed by those who have obtained the data that the process is completed by an impact of fast-moving atom, one of those which by virtue of Maxwell's distribu- tion have the necessary excess of energy over the relatively modest mean value corresponding to the actual temperature. The relative heights of the peaks would then be determined partly by the relative abundance of atoms having the necessary energies, and partly by the relative probabilities of the corresponding types of excitation, which " R. W. Ditchburn, Proc. Camb. Phil. Soc, 24, pp. 320-327 (1928). ^^Phys. Rev., (2) 26, pp. 195-207 (1925); 27, pp. 37-50 (1926). CONTEMPORARY ADVANCES IN PHYSICS 353 are quantities with which the theories deal. According to Foote, Mohler and Chenault, these relative heights are in fair accordance with the theories. The actual heights, however, depend on the mean duration of the excited states. I do not know whether it has been proved that these last long enough to permit the explanation. 36 32 28- IS-4p Ui >- t > 24 20 z u i 12 _L 162°C - .28 - 24 - .20 .16 .12 08 .04 .00 3900 3800 3700 3600 3500 \ IN A.U. 3400 3300 3200 Fig. 4 — Ionization of caesium vapor by light, at wavelengths greater than the critical (3184A). (Aiohler, Foote & Chenault.) Since the quanta spent in ionization vanish from the light, the transmitted beam when spread into a spectrum reveals absorption at their frequencies. These absorption-spectra supply all that is known as yet about the process of ionization by light in sodium and in atomic hydrogen and valuable additions to the data for the three heavier alkali metals. It will be remembered that the lines of a line-series in an absorption- spectrum occur because the photons of the corresponding frequencies can be absorbed by atoms in a particular initial state (normal or excited) which thereupon pass over into higher states of excitation; that as the lines converge upon the limit of the series, the corresponding terminal states approach that of ionization; that the limiting or 354 BELL SYSTEM TECHNICAL JOURNAL convergence-frequency itself, multiplied by //, give the energy required to ionize an atom from that initial state which is common to the entire series. Thus photons having the convergence-frequency of any series are just able to detach an electron from an atom in the corresponding state. Consequently photons having any greater frequency have energy sufficient to detach an electron, and give it some kinetic energy in addition. Now we are not aware of any "quantum" limitations on the amount of energy which a freed electron may receive. We thus infer that light of any frequency superior to a convergence-frequency will be able to ionize atoms and to be absorbed in doing so, and that there will be a continuous region of absorption in the spectrum extend- ing upwards from the limit of each series. For such a region I will use the terms continuous band and continuum. Bohr drew this inference in the first of his epoch-making papers on the interpretation of spectra. He was able then to point to only one example; a continuum beyond the limit of the principal series of sodium, observed by R. W. Wood.^^ Afterwards J. Hartmann ^^ searched the spectrograms of the stars, and in those of the so-called "hydrogen stars" he found a continuous band beyond the limit of the Balmer series. This, be it noted, is the sign of ionization of hydrogen atoms initially not in the normal, but in a certain excited state. The continua beyond the principal series of the alkali metals, however, are due to ionization of normal atoms. Those of sodium and potassium were studied by Holtsmark; ^"^ those of caesium and rubid- ium have been discerned (Harrison, I.e. infra); and the former two were measured, that is to say the variation of absorption-coefficient with frequency was measured, for sodium by G. R. Harrison ^^ and B. Trumpy,^^ and for potassium by R. W. Ditchburn.^^ Obviously if the fundamental theory is correct, absorption is pro- portional to ionization, and the curves representing the two as functions of wavelength should coincide everywhere if scaled to coincide at any one point; and measurements of either should make the other nugatory. Unfortunately it is difficult to measure the absorption properly, perhaps impossible to do it with anything like the precision feasible with the other measurement.'-^ Harrison managed to get smooth absorption- '^Phil. Mag., (6) 18, pp. 530-534 (1909). ^^Phys. ZS., 18, pp. 429-432 (1917). ^"^ Phys. Rev., 20, pp. 88-92 (1919). ^^Phys. Rev., (2) 24, pp. 466-477 (1924). 13 Z5./. Phys., 47, pp. 804-813 (1928). 2° Proc. Roy. Soc, 117, pp. 486-508 (1928). ^1 Mohler and his colleagues state that with an amount of ionization tenfold greater than that which is observed with caesium at the series-limit, a stratum of the gas at 230° would have to be nitie metres deep to give a 50 per cent absorption. CONTEMPORARY ADVANCES IN PHYSICS 355 curves (obtained of course by applying the densitometer to the spec- trogram) with sodium. On the other hand, the experiences of Ditch- burn with potassium are not encouraging. Not only did he have to shoot a jet of rapidly-distilling vapor across the beam of light, but he was obliged to swamp it in a vast excess of nitrogen — partly to keep the metal from boiling away in a rush, partly it seems to prevent the vapor from attacking the quartz windows. The curves are very crinkly, and it is difficult to tell what share of the absorption should be credited to molecules and what to atoms. Nevertheless Ditchburn was able to deduce a value of the coefficient k having the same order of magnitude — 10~^^ — as those which Mohler and Boeckner had obtained with caesium and with rubidium when they were measuring, not the disappearance of photons from the beam, but the advent of ions in the gas. Mohler and Boeckner themselves observed the absorption of light in caesium, and they found for k the value 4-10~^^, — a good agreement, but they qualify it with the words "subject to great uncertainty because of the low value of the total absorption." Let it be pointed out in closing, that agreements such as these are proof that in this region of the spectrum, photons ionize when they are absorbed, and absorption is due to ionization. To physicists familiar with the new atomic theories, this seems self- evident, and scarcely worth the proving; but it is not self-evident, and there was a time, not many years ago, when such a proof would have been a sensational event. Motion of Telephone Wires in Wind By D. A. QUARLES This paper deals with the position of equilibrium of a loop of wire in a steady transverse wind and with the swinging of such a loop in one or more gusts of wind. In the first part, the loop is assumed to be inelastic and to swing as a rigid body. Under these conditions, nomograms are given from which may be read the deflection of loops of wire .104" or .165" in diameter as a function of steady wind velocity. The maximum additional swing of such a loop with a single gust and with a succession of gusts of given peak velocities may also be read from the nomograms. A chart is also included giving the effect of wind velocity on the sag of .104" and .165" hard drawn copper wires at tensions and span lengths common in the telephone plant. UNTIL recent years, most of the important open wire toll circuits of the Bell System had the two wires of a pair spaced 12 inches apart. This wide spacing, with the consequent high mutual induct- ance between the several pairs on a pole line, limited the use of the lines for multiplex transmission with high frequency or "carrier" currents. A reduction in the separation of the wires of a pair with the retention of the present center-to-center spacing of the pairs was one of the measures which offered the opportunity of increasing the message carrying capacity of a pole line. The controlling factor in limiting such a reduction in spacing was the hazard of the wires of a pair swinging together in the wind thus interrupting or impairing the transmitted messages. About two years ago the 12-inch spacing was reduced to 8 inches in some cases. This was considered to be as great a change as could be safely taken from a mechanical point of view, based on the available data. These data consisted in part of experiments made on an experimental line and in part of an analysis of the performance of certain working wires in the telephone plant which, for various reasons, had been installed on a close spaced basis. It was realized that if the wires of pairs could be placed even closer together, materially lower crosstalk between the circuits would result, thus increasing the circuit capacity of open wire lines, and therefore effecting economies. Accordingly, a comprehensive investigation of the wire spacing problem was begun. As some of the factors involved in a theoretical determination of the chance of two parallel wires swinging together in the wind were rather obscure and difficult of evaluation, it was decided to attack the problem experimentally. A field site was selected some distance from New York, where the terrain and weather conditions were suitable for such an investigation, and an experimental station was constructed and appropriately equipped. 356 MOTION OF TELEPHONE WIRES IN WIND 2,hl Some time will be required, however, before definite conclusions can be drawn from the experimental work of this new laboratory. As an aid in the interpretation of the experimental results, certain theoretical work has been done on the dynamics of a wire loop swinging in the wind. It is this phase of the problem that is dealt with in this article. In the first part of this discussion, the wire loop is treated as an inelastic, rigid body.^ As it was later found that under the conditions applying in our problem there was a considerable increase in the sag of the wire due to the wind, an investigation was made of the magnitude of the correction required when the elasticity of the wire is taken into account, the results of which are given in the latter part of this article. Fig. 1. Consider an element of the wire, shown in Fig. 1 in cross-section, swinging about axis 0, at a radius y. The wind is assumed horizontal and transverse to the axis. The sag a is also assumed small compared with the span length so that to a sufficient approximation the length of the wire is equal to the length of the span and the surface of the wire opposing the wind is independent of the angle of deflection {a) of the wire in the wind. The velocity of the element of wire relative to axes fixed with respect to the earth is ya. The wind velocity relative to the same coordinate ^ An article entitled "The Behavior of Overhead Transmission Lines in High Winds " by Professor E. H. Lamb, which appeared in the October 1928 Journal of the Institu- tion of Electrical Engineers, gives an analysis of the inelastic, rigid loop problem which has been followed in general outline in the present treatment. There is disagreement, however, with one of the fundamental assumptions upon which Professor Lamb's analysis is based and our formulae are therefore generally at variance with those derived in his article. Mr. R. L. Peek, Jr. of Bell Telephone Laboratories, working independently, ar- rived at results in agreement with those given in the present article. 24 358 BELL SYSTEM TECHNICAL JOURNAL system is V and the wind velocity relative to the wire at any instant is therefore the vector difference V — ya which has the magnitude VF" + {yaY — IVya COS a. It is assumed that the wind pressure against this element is propor- tional to the square of this vector and acts along its direction. The moment about the axis of the wind pressure on the element ds is therefore given by: ^[F- + (3'a)" — ly'aV COS a\y cos ^ds, where U is the ratio of wind pressure per unit length to square of veloc- ity. Evaluating cos ^ and noting that ya is small compared with F, this reduces to: kdsy V^ cos a 1 — 4t ( cos a -\ i • L F \ cos a / J Putting y = and ds = dx and integrating, the total moment of wind pressure is 4 16 -kV^aC cos a — -r-EkVa^Ca (cos^ a -{- 1). If the line through the supports is inclined to the horizontal by angle 7 this expression becomes: 4 16 -kV^aC cos a cos y — ■r-rkVa'^Ca (cos- a -\- 1) cos^ y The dynamic equation for the motion of the loop then becomes: a H (1 + COS" a)a -\- -r- s\n a = -; , m 4 a 4 ma cos y where m is the mass of unit length of wire. Static equilibrium is then given by: tan a = mg cos 7 Proceeding with the analysis, an equation is found for small motions MOTION OF TELEPHONE WIRES IN WIND 359 about any position of equilibrium (deflection a) of the form '(p -\- 2eip + n-ip = 0, where _ (t + cos^c^) kV ^ ~ 2m and 7 £, « W 4o cos a For cases of practical interest in this investigation n^ > e^ and the motion about equilibrium is periodic and of period rp _ ^TT _ kfl cos a where a is the sag in feet and g the acceleration of gravity in feet per second per second. The ratio of the period of small oscillations about equilibrium to the period when a is zero is given by T/Tq = Vcos a. The damping as measured by the ratio of successive half swings, X, is given by log, X = , = - € ■\n^ — e^ n If a wire, held at a deflection a by a steady wind F, is subjected to a gust of wind having maximum velocity V\, the additional throw of the wire will depend on the duration of the gust and may in general be either greater than or less than the increase in steady deflection which Vi, if sustained, would produce. The maximum throw will be given by a gust of most favorable duration and /x,„g has been defined as the ratio of this maximum throw to the increase in deflection that would result if the peak velocity were sustained. Similarly, for a periodic succession of gusts, there is a most favorable timing which in general will produce displacements greater than would a wind which sustained the velocity of the gust peaks. The ratio of the throws produced by a most favorably timed succession of gusts to the increase in deflection which would result if the peak velocity of the gusts were sustained, has been defined as ^imp- The formulae derived above have been applied to the practical conditions of the telephone line problem,- where our interest is centered in hard drawn copper wire, commonly of .104" or .165" diameter, with spans ordinarily from 90 to 200 feet and sags commonly from 7" to 20" 2 This work was carried out in the Bell Telephone Laboratories by Mr. V. Nekrassoff. 360 BELL SYSTEM TECHNICAL JOURNAL though occasionally considerably greater. The method, which will be described in more detail elsewhere, was to reduce the expressions for wind pressure per unit length of wire, F, angular displacement a, periods of small oscillations, T and Tq, damping constant X, and the effects of single and periodic gusts, /i,„s and ^mp, to explicit functions of the wind velocity in miles per hour, the diameter of the wire in inches, sag of the wire in inches and trigonometric functions of the deflection of the loop a and inclination of the loop 7. The factor k does not appear directly in the equations, having been replaced by fractional powers of wind velocity and wire diameter derived from the experimental results of Relf.3 The following nomograms have been constructed by this method. Nomogram No. 1 (Fig. 2) gives the steady deflection « of a span of wire inclined to the horizontal at an angle 7 and the force in pounds per linear foot of wire for a normal wind of velocity V. It also gives the ratio of the period of small oscillations about the equilibrium position to the natural period about the vertical position, this ratio depending only on a. The actual value of the period in seconds may be read on nomogram No. 2 (Fig. 3). By the use of nomogram No. 3 (Fig. 4), the damping constant X, and the gust ratios n,ns and ju,„p may be computed from the sag a, the wind velocity V and the diameter D. These nomograms in short give the numerical solution for our prob- lem for wires of the two diameters assumed, namely .104" and .165". Two major assumptions should be noted, first, that the wire loop swings in a plane and second, that the wire is inelastic. The first assumption has a certain justification in that each element of wire if independent of adjacent elements would be in equilibrium in the same deflected angle a as is found for the loop as a whole. Expressing this in another way — if it be assumed that the wind is uniform along the span there would be no forces, considering only first order effects, to distort the loop out of a plane. The second assumption is not so readily justified, in fact the sag of the wire may be greatly affected by the wind pressure. The equili- brium deflection a is, however, independent of the sag of the wire and is found to be the same when the elasticity of the wire is taken into ac- count as that derived for an inelastic wire. Considering only the case where the line through the supports is horizontal (7 = 0), we define 2r as the unstressed length of wire in the loop and note that this may be either greater or less than the span length 2c depending upon the tension at which the wire is suspended. ^ British Advisory Committee for Aeronautics — Report Xo. 102. o 6 hK .n(ni-o)c«)o»o otftjui 1^ (\K^nr^r^a)(r rt i m ■nuitnininin>nininin«) ioio e cibcidododood 666 6 «0 (fl r- » « ffi (o «e (0 6 6 6 n 00 - (\j d d d n (^ 6 o CO d o o CO d o a> 6 'J to 6 n 6 CD 6 6 (0 • 6 Ok 6 o» o> o o» 6 as. ooiOinoom^r- no on CD in ti>— (T>^nco>rt(Mnr\j (Mn — oo — 0000^*^^*0 coaj h- p^ h- 0) — rrroooooo ooo Q ooodobo ocidd odd o >n o in n CO ggg coo •n oj tn m o O d d g o d Q o t o d in 3 6 n o> a> t^ vi in ^ tn <\i — 10 (£1 to O) tO tn (A 1. 1 ■A ,1 L 1 ,1 t 1, 1 J s 1 , n 1 n J II M 1 J? 1 tn d L. « :?^?SSp£$S'8£|s"SSS 5 I S "^ I 1 1 1 1 1 1 ■to 1 r n 1 r 1 1 1 1 ^1 or-^oo-ooo-no inooinotfi'O "O .n to n ©^••oooo^o-noi to « o> w iM n ij «o © - ••r-*Jt^TO(M-o (J> (0 r^ « to lO ■^ m i\i - — S g eO (0 CO O d (0 o d I- o d s o d n o d Si o 6 o d in n o d tn o d lA tn o 6 o 8 o 6 doooodododd d d d d d d 6 6 d 6 d d d l-lH" ^-r-<»«*-««o>oo> oJiOh-onf^ n d) ■^ n n,^rnirft«>*t^«» n^tntocDo. - ty v to .n>i d nj (\1 fvi rvi r\j (\j (\1 (*) (O p) o n o o O w a> 01 o — Ai n ^ n fi -vi -4 •* -ir r tW •* ■« V lo ri' I ' I' I ' I ' I ' OO O O O O Q >no o o o o o r-i^ to t/X ^ m r\^ AJnjni N rg n] r\j 01 ____;:!— _^_ w cjf^^ I I'l 'l I I |ml)llMinulllll|ll^Mlll|lll>ll^M/ll>ll^lW|Hll'nlllV^l^lllll|lll'li^li'N'll^il'''ll/l'l''l'''VlV|'l^^Vl'\" o ,>i o f< .2 M>>< „ <3 o > II II II — — :— ~ ^ — fi td OfVKOOiOO r^ o -< 00> ^ O t~ IB omorf) — t^ w «) ©"louj — t» (\J ■ I'l'i'i't'J^tvti OO0 --<\I(\J IM CJ rvi o o> to « — — - — — — — <\i oj -< >ONoio>in^ w ■jJj.T.T.T.Jv, (OfOioo> 'i — r- N «> -< o(Mtf>p o Ai K) o <0 ^ 40 ^0(0^ ^ U) 01 V --fUNNN ni N --0 — mietooi-— ntnr^fft— ^^p^ «j ^ 4 W 4 VW ^ to mS "i o mS |C o J - _ s asa?s?s8sssg ^ ^ ^ O^ ^ r- a A o - o o o — t * * ^ i^r^^gs 4 O r> O fu ni — o O'O'OiB*)* f«. h- ^. * "O « >« — — — — oppopoooooo oo p o o o o a — V r^ 0 — f~ i00<0^ ^ « t r> o fO 'H o>«Ai«)i/ (^ (>J CV — O 99i90a . - - OOOOC . . . lO (M O OOOOOOOO O O O O O O N 0> (\J04WN- - - — SO(MU>0>Qr^^ON^tO«»0— N4«0 — (*) "> »* o^ ^ to n*^«o • *^ lOON — t*) 0«ltO'')NOo>OOtOOOO M«^«-« r--«>«No-o«to — rtiot-9— ^h-o« rt f 1 / 1 1 7 20 / / 1 f/' / / / 1 4 ^ / f // / y / 1 1 /', / _ * -^ X 1 / 1 / 15 1 /, 'oV / J 1 "^h 1 / / 1 1 7 / / / ;^ / 1 10 / 1 1 / ^^_ ^ y 1 1 — f 1 J x' 1 f / 5 ^ ^ X 10 1 1 1 1 1 1 1 1 1 -0.104" DIAMETER WIRE — -- ^ I 65 1 1 + POINT WHERE FO OF WIND EQUA RTF 1 LS 1 WEIGHT OF WIRE 60 / 1 r 1 1 1 ',/ / f S5 1 1 1 f 1 1 1 ' 1 // 1 1 // 1 50 1 It 1 -f — / 1 1 ' oV 1 1 1 1 oV f // 1 *v / 45 1 / / -i 1 / / / / 1 f / / / 1 / / / 1/ 4 ' +> f ' / 40 / / / 1 y ' / / ^y y f- 1 o^l' / / ^/l 1 ."/' 1 35 i r, 1 1 / / 1 / 1 oV 1 1 /- ^ f // '■7 30 / 1 ' / / / / / / / / / 1 / / / / r t f / / / / / / ^/ / / / / / / i^ . 1 PO / / / t f '2 + t 15 7 1 / J. y 10 0 20 40 60 80 100 WIND VELOCITY IN MILES PER HOUR 0 20 40 60 80 100 WIND VELOCITY IN MILES PER HOUR SPAN OF 130 FEET Fig. 5-.4. SPAN OF 200 FEET Fig. 5-5. MOTION OF TELEPHONE WIRES IN WIND 361 50 0.104" DIAMETER WIRE 0.165" DIAMETER WIRE 45 + POINT WHERE FORCE OF WIND EQUALS WEIGHT OF WIRE 40 35 1 1 1 30 1 1 / / 10 1 1 I o t 1 / z 1 1 i f z '^/l o < 2S 1 1 o° I 1 t in'-^ 1 f A?/ 1 1 f / / / / 1 J f / / 1 J 20 / / 1 9' f / 1 1 '^A 1 f ,+ / 1 1 / '/ / I n / ^ ^> y^ y / / 1 / 15 / /< %'/ / / ' / '^/ 1 / / / / 7 / ,/ f , / f / 1 / 10 / 1 1 / ^ ^ ^y' y / 1 f -i 1 / / A / s ^^^ 1 1 1 1 1 1 1 1 -\JA\J'\ UIMMtP LK WIHL j 65 I 1 + POINT WHERE FORCE OF WIND EQUALS WEIGHT OF WIRE 1 1 1 60 1 —h 1 1 1 ' / / / 55 1 / 1 f — 1 1 // / 1 // / 50 1 1 1 f- 1 t 1 1 1 1 "7 ■f — / 1 1 o7 f 1 1 /y i / 45 1 / 1 1 / 1 / 1 1 f / / 1 1 / / 1 1/ 4 f r / 40 / / / / J ' / r / f / ,y y / o'^y / / - /' / / ."/' / 35 i r, J 1 y 1 1 / ' "V 1 1 /- r f // H 30 1 i ' 1 ^ I 1 1 / 1 1 / 1 1 1 / 1 1 i 25 t /+ 1 / / / 1 / ^/ 1 / / / 1 / __^ i^ , 1 ?n / / / / / t 'J +' t 1 s 7 1 .^ ^' y 10 0 20 40 60 80 100 WIND VELOCITY IN MILES PER HOUR 0 20 40 60 80 100 WIND VELOCITY IN MILES PER HOUR SPAN OF 130 FEET Fig. 5-^. SPAN OF 200 FEET Fig. 5-B. 362 BELL SYSTEM TECHNICAL JOURNAL If E is the modulus of elasticity in pounds per square inch cross-section, D the diameter in inches, a the sag in feet and m the weight of wire per linear foot, the approximate relationship^ is: a' + — (c - r)a = ttD^E As only horizontal winds normal to the line of supports are being considered, the wind pressure when the loop is in equilibrium is horizontal. The weight of the wire being vertical the two forces add at right angles, their resultant being the square root of the sum of their squares. This resultant lies of course in the plane of equilibrium of Fig. 6 — Test House and Line. the loop. The wind pressure component is about equal to the gravity component for a velocity of 38 m.p.h. in the case of .104" wire and about 47 m.p.h. in the case of .165" wire. The effective weight of the wire under these conditions would be greater by a factor of V2 than the true weight. In general, ni in the above formula is the effective weight of the wire per unit length. A wire having a sag of 5" in a 130' span with a temperature of — 10° F. would have a sag of about 9" at 50° F. and about 16" at 100° F. due to thermal expansion. The sag of such a wire would be increased by wind pressure as shown in Fig. 5-A , the wind being given in true normal velocity. The figure shows the increase to be most marked for low temperatures and small diameters as would be expected. Similar ^ Due to Mr. J. A. Carr of Bell Telephone Laboratories. MOTION OF TELEPHONE WIRES IN WIND 363 results are shown in Fig. S-B for a span of 200'. Both indicate the marked increase of sag under not uncommon wind conditions. While the above formula and charts give a fairly definite picture of the effect of elasticity on the solution of the problem of static equili- brium, the much more complex problem of the motion of an elastic loop in a varying wind has not been attacked.^ The necessity for such additional refinements can probably not be determined until the field experiments above referred to have progressed to the point where fairly comprehensive data are available for analysis and for a check of the theoretical conclusions arrived at in this paper. * An article by Karl Wolf in Zeitschrift f iir Angewandte Mathematik und Mechanik of April 1927 treats certain aspects of the dynamics of an elastic loop, with particular reference, however, to power lines. As yet, no attempt has been made to apply the results of this work to our particular problems. Economic Quality Control of Manufactured Product^ By W. A. SHEWHART That we cannot make all pieces of a given kind of product identically alike is accepted as a general truth. It follows that the qualities of jjieces of the same kind of product differ among themselves, or, in other words, the quality of product must be expected to vary. The causes of this variability are, in general, unknown. The present paper presents a scientific basis for determining when we have gone as far as it is economically feasible to go in eliminating these un- known or chance causes of variability in the quality of a product. When this state has been reached, the product is said to be controlled because it is then possible to set up limits within which the quality may be expected to remain in the future. By securing control, we attain the five economic ad- vantages discussed in Part III. I Introduction 1. What is the Problem of Control? WHAT is the problem involved in the control of quality of manu- factured product? To answer this question, let us put our- selves in the position of a manufacturer turning out millions of the same kind of thing every year. Whether it be lead pencils, chewing gum, bars of soap, telephones or automobiles, the problem is much the same. He sets up a standard for the quality of his product and then tries to make all pieces of product conform with this standard. Here his troubles begin. For him standard quality is a bull's-eye, but like a marksman shooting at such a target, he often misses. As is the case in everything we do, unknown or chance causes exert their influence. The problem then is: how much may the quality of a product vary and yet be controlled? In other words, how much variation should we leave to chance? To make a thing the way we want to make it is one popular concep- tion of control. We have been trying to do this for a good many years and we see the fruition of this effort in the marvelous industrial develop- ment around us. We have accepted the idea of applying scientific principles but now a change is coming about in the principles them- selves which necessitates a new concept of control. A few years ago we were inclined to look forward to the time when a manufacturer would be able to do just what he wanted to do. We shared the enthusiasm of Pope when he said "All chance is but direction thou canst not see," and we looked forward to the time when we would see that direction. In other words, emphasis was laid on the exactness ^ Paper presented before A. A. A. S. on December 28, 1929, at Des Moines, Iowa. 364 ECONOMIC QUALITY CONTROL OF PRODUCT 365 of physical laws. Today, however, the emphasis is placed elsewhere as is indicated by the following quotation from a recent issue, July, 1927, of the journal Engineering: "Today the mathematical physicist seems more and more inclined to the opinion that each of the so-called laws of nature is essentially statistical, and that all our equations and theories can do, is to provide us with a series of orbits of varying probabilities." The breakdown of the old orthodox scientific theory which formed the basis of applied science in the past necessitates the introduction of certain new concepts into industrial development. Along with this change must come a revision in our ideas of such things as a controlled product, an economic standard of quality and the method of detecting lack of control or those variations which should not be left to chance. Realizing, then, the statistical nature of modern science, it is but logical for the manufacturer to turn his attention to the consideration of available ways and means of handling statistical problems. The necessity for doing this is pointed out in the recent book on the "Ap- plication of Statistics in Mass Production," by Becker, Plant and Runge. They say: "It is therefore important to every technician who is dealing with problems of manufacturing control to know the laws of statistics and to be able to apply them correctly to his problems." Another German writer, K. H. Daeves, writing on somewhat the same subject says: "Statistical research is a logical method for the control of operations, for the re- search engineer, the plant superintendent, and the production executive." This statement is of particular interest because its author has for several years been associated with the application of statistical methods in the steel industry. The problem of control viewed from this angle is a comparatively new one. In fact, very little has been written on the subject. Progress in modifying our concept of control has been and will be comparatively slow. In the first place, it requires the application of certain modern physical concepts and in the second place, it requires the application of statistical methods which up to the present time have been for the most part left undisturbed in the journals in which they appeared. This situation is admirably summed up by the magazine Nature of January, 1926, as follows: "A large amount of work has been done in developing statistical methods on the scientific side, and it is natural for any one interested in science to hope that all this work may be utilized in commerce and industry. There are signs that such a move- ment has started, and it would be unfortunate indeed if those responsible in practical affairs fail to take advantage of the improved statistical machinery now available." 366 BELL SYSTEM TECHNICAL JOURNAL 2. Object The object of this paper is the presentation of a scientific basis for interpreting the significance of chance variations in quality of product and for eliminating causes of variability which need not be left to chance, making possible more uniform quality and thereby effecting certain economies. 3. Nature of Control Let us consider a very simple example of our inability to do exactly what we want to do and thereby illustrate two characteristics of a controlled product. Write the letter a on a piece of paper. Now make another a just like the first one; then another and another until you have a series of a's, a, a, a, a, . . . . You try to make all the a's alike but you don't; you can't. You are willing to accept this as an empirically established fact. But what of it? Let us see just what this means in respect to control. Why can we not do a simple thing like making all the a's just alike? Your answer leads to a generalization which all of us are per- haps willing to accept. It is that there are many causes of variability among the a's: the paper was not smooth, the lead in the pencil was not uniform and the unavoidable variability in your external surroundings reacted upon you to introduce variations in the a's. But are these the only causes of variability in the a's? Probably not. We accept our human limitations and say that likely there are many other factors. If we could but name all the reasons why we cannot make the a's alike, we would most assuredly have a better understand- ing of a certain part of nature than we now have. Of course this conception of what it means to be able to do what we want to do is not new; it does not belong exclusively to any one field of human thought; it is a commonly accepted conception. The point to be made in this simple illustration is that we are limited in doing what we want to do ; that to do what we set out to do, even in so simple a thing as making a's that are alike requires almost infinite knowledge compared with that which we now possess. It follows, therefore, since we are thus willing to accept as axiomatic that we cannot do what we want to do and that we cannot hope to understand why we cannot, that we must also accept as axiomatic that a controlled quality will not be a constant quality. Instead a controlled quality must be a variable quality. This is the first characteristic. But let us go back to the results of the experiment on the a's and we shall find out something more about control. Your a's are different from my a's; there is something about your a's which makes them yours ECONOMIC QUALITY CONTROL OF PRODUCT 367 and something about my a's that makes them mine. True, not all of your a's are alike. Neither are all of my a's alike. Each group of a's varies within a certain range and yet each group is distinguishable from the others. This distinguishable and, as it were, constant variability within limits is the second characteristic of control. 4. Definition of Control For our present purpose a phenomenon will be said to be controlled when, through the use of past experience, we can predict, at least within limits, how the phenomenon will be expected to vary in the future. Here it is understood that prediction within limits means that we can state, at least approximately, the probability that the observed phenomenon will fall within the given limits. In this sense the time of the eclipse of the sun is a predictable phenomenon. So also is the distance covered in successive intervals of time by a freely falling body. In fact, the prediction in such cases is extremely precise. It is an entirely different matter, however, to predict the expected length of life of an individual at a given age; the velocity of a molecule at a given instant of time ; the breaking strength of a steel wire of known cross section; or numerous other phenomena of like character. In fact, a prediction of the type illustrated by fore- casting the time of an eclipse of the sun is almost the exception rather than the rule in scientific and industrial work. In all forms of prediction an element of chance enters. The specific problem which concerns us at the present moment is the formulation of a scientific basis for prediction, taking into account the element of chance, where, for the purpose of our discussion, any unknown cause of a phenomenon will be termed a chance cause. II. Scientific Basis for Control 1. Three Important Postulates What can we say about the future behavior of a phenomenon act- ing under the influence of unknown or chance causes? I doubt that, in general, we can say anything. For example, let me ask: "What will be the price of your favorite stock thirty years from today?" Are you willing to gamble much on your powers of prediction in such a case? Probably not. However, if I ask: "Suppose you were to toss a penny one hundred times, thirty years from today, what proportion of heads would you expect to find?," your willingness to gamble on your powers of prediction would be of an entirely difterent order than in the previous case. 368 BELL SYSTEM TECHNICAL JOURNAL The recognized difference between these two situations leads us to make the following simple postulate: Postulate 1. All chance systems of causes are not alike in the sense that they enable us to predict the future in terms of the past. Hence, if we are to be able to predict the quality of product at least within limits, we must find some criterion to apply to observed vari- ability in quality to determine whether or not the cause system pro- ducing it is such as to make possible future predictions. Perhaps the natural course to follow is to glean what we can about the workings of unknown chance causes which are generally acknowl- edged to be controlled in the sense that they permit of prediction within limits. Perhaps no better examples could be considered than those which influence length of human life and molecular motion, for it often appears that nothing is more uncertain than life itself, unless perhaps it be molecular motion. Yet there is something certain about these uncertainties. In the assumed laws of mortality and distribution of molecular displacement, we find some of the essential characteristics of control within limits. A . Law of Mortality The date of death always has seemed to be fixed by chance even though great human effort has been expended in trying to rob chance UJ O < UJ > «>- t^ < CD ^m ±o z o I- u < a. 45 50 AGE IN YEARS Fig. 1— Law of mortality — ^law of fluctuations controlled within limits. of this prerogative. We come into this world and from that very in- stant on are surrounded by causes of death seeking our life. Who knows whether or not death will overtake us within the next year? ECONOMIC QUALITY CONTROL OF PRODUCT 369 If SO, what will be the cause? These questions we cannot answer. Some of us are to fall at one time from one cause, others at another time from another cause. In this fight for life we see then the element of uncertainty and the interplay of numerous unknown or chance causes. However, when we study the effect of these chance causes in produc- ing deaths in large groups of individuals, we find some indication of a controlled condition. We find that this hidden host of causes produce deaths at an average rate which does not differ much over long periods of time. From such observations we are led to believe that, as we approach the condition of homogeneity of population and surroundings, we approach what is customarily termed a "Law of mortality" such as indicated schematically in Fig. 1. In other words, we believe that in the limiting case of homogeneity the causes of death function so as to make the probability, let us call it dy, of dying within given age limits, such as forty-five to fifty, constant: That is, we believe these causes are controlled. In other words, we assume the existence of a kind of statistical equilibrium among the effects of such an unknown system of chance causes expressable in the assumption that the probability of dying within a given age limit, under the assumed conditions, is an objective and constant reality. B. Molecular Motion Just about a century ago, in 1827 to be exact, an English botanist. Brown, saw something through his microscope that caught his interest. It was motion going on among the suspended particles almost as though they were alive. In a way it resembled the dance of dust particles in sunlight, so familiar to us, but this dance differed from that of the dust particles in important respects — for example, adjacent particles seen under the microscope did not necessarily move in even approximately the same direction, as do adjacent dust particles suspended in the air. Watch such motion for several minutes. So long as the temperature remains constant, there is no change. Watch it for hours, the motion remains characteristically the same. Watch it for days, we see no difference. Even particles suspended in liquids enclosed in quartz crystals for thousands of years show exactly the same kind of motion. Therefore, to the best of our knowledge there is remarkable permanence to this motion. Its characteristics remain constant. Here we cer- tainly find a remarkable degree of constancy exhibited by a chance system of causes. Suppose we follow the motion of one particle to get a better picture of this constancy. This has been done for us by several investigators, 370 BELL SYSTEM TECHNICAL JOURNAL notably Perrin. In such an experiment he noted the position of a particle at the end of equal intervals of time, Fig. 2. He found that Fig. 2 — ^A close-up of molecular motion appearing absolutely irregular, yet controlled within limits. the direction of this motion observed in one interval differed, in general, from that in the next succeeding interval. He found that the direction of the motion presents what we instinctively call absolute irregularity. Let us ask ourselves certain questions about this motion. Suppose we fix our attention on the particle at the point A. What made it move to B in the next interval of time? Of course we answer ECONOMIC QUALITY CONTROL OF PRODUCT 371 by saying that a particle moves at a given instant in a given direction, say AB, because the resultant force of the molecules hitting it in a plane perpendicular to this direction from the side away from B is greater than that on the side toward B ; but at any given instant of time there is no way of telling what molecules are engaged in giving it such mo- tion. We do not even know how many molecules are taking part. Do what we will, so long as the temperature is kept constant, we can- not change this motion in a given system. It cannot be said, for ex- ample, when the particle is at the point B that during the next interval of time it will move to C. We can do nothing to control the motion in the matter of displacement or in the matter of the direction of this displacement. Let us consider either the x or y components of the segments of the paths. Within recent years we find abundant evidence indicating that these displacements appear to be distributed about zero in accord with what is called the normal law. That is to say, if x represents the deviation from the mean displacement, zero in this case, the probability dy of X lying within the range x to x -\- dx is given by dy = ^^e-^^'l^'^'Hx, a) where a is the root mean square deviation. Such evidence as that provided by the law of mortality and the law of distribution of molecular displacements leads us to assume that there exist in nature phenomena controlled by systems of chance causes such that the probability dy of the magnitude X of a characteristic of some such phenomenon falling within the interval X to X + dX is express- able as a function / of the quantity X and certain parameters repre- sented symbolically in the equation dy=f(X,K\,, ■■■,\J,dX, (2) where the X's denote the parameters. Such a system of causes we shall term constant because the probability dy is independent of time. W^e shall take as our second postulate: Postulate Z — Constant systems of chance causes do exist in nature. To say that such systems of causes exist in nature, however, is one thing; to say that such systems of causes exist in a production process is quite another thing. Less than ten years ago it seemed reasonable to assume that such systems of causes existed in the production of telephone equipment. Today we have abundant evidence of their 372 BELL SYSTEM TECHNICAL JOURNAL existence. The practical situation, however, is that in the majority of cases there are unknown causes of variability in the quality of a product which do not belong to a constant system. This fact was dis- covered very early in the development of control methods, and these causes were called assignable. The question naturally arose as to whether it was possible, in general, to find and eliminate causes of variability which did not form a part of a constant system. Less than ten years ago it seemed reasonable to assume that this could be done. Today we have abundant evidence to justify this assumption. We shall, therefore, adopt as our third postulate: Postulate 3 — Assignable causes of variation may be found and eliminated. Hence, to secure control, the manufacturer must seek to find and eliminate assignable causes. In practice, however, he has the difiiculty of judging from an observed set of data, whether or not assignable causes are present. A simple illustration will make this point clear. 2. When Do Fluctuations Indicate Trouble? In many instances the quality of the product is measured by the fraction non-conforming to engineering specifications or as we say the fraction defective. Table 1 gives for a period of 12 months the ob- TABLE 1 Apparatus Type A Apparatus Type B Month n No. Insp. Hi No. Def. p = mln Fraction Def. Month n No. Insp. No. Def. p = n\ln Fraction Def. Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec 527 610 428 400 498 500 395 393 625 465 446 510 4 5 5 2 15 3 3 2 3 13 5 3 .0076 .0082 .0017 .0050 .0301 .0060 .0076 .0051 .0058 .0280 .0112 .0059 Jan Feb Mar Apr May June July Aug Sept Oct Nov Dec 169 99 208 196 132 89 167 200 171 122 107 132 1 3 1 1 1 1 1 1 2 1 3 1 .0059 .0303 .0048 .0051 .0076 .0112 .0060 .0050 .0117 .0082 .0280 .0076 Average .... 483.08 5.25 .0109 149.33 1.42 .0095 served fluctuations in this fraction for two kinds of product designated here as Type A and Type B. For each month we have the sample size n, the number defective Wi and the fraction p = ni/n. We can ECONOMIC QUALITY CONTROL OF PRODUCT 373 better visualize the extent of these fluctuations in fraction defective by plotting the data as in Fig. 3-a and Fig. 3-b. 0.04 0.03 - UJ > o UJ ^. 0.02 z o H o < DC 0.01 I t • • • . • • • 0.041- a 0.03 > H- O UJ u. UJ 0.02 Q Z o 1- o < 0.01 M M J J A MONTHS O J L M M J J A S MONTHS N J L Fig. 3 — Should these variations be left to chance? a. Apparatus Type A. b. Apparatus Type B. What we need is some yardstick to detect in such variations any evidence of the presence of assignable causes. Can we find such a 25 374 BELL SYSTEM TECHNICAL JOURNAL yardstick? Experience of the kind soon to be considered indicates an affirmative answer. It leads us to conclude that it is feasible to es- tablish criteria useful in detecting the presence of assignable causes of 0-04r £;0.03 UJ > u UJ LI. ^ 0.02 z o \- o 0.04 r a 0.03 UJ > I- o liJ lii 0.02 o z o I- o < a. u. 0.01 J L « t J I I I I L J FMAMJ JASOND MONTHS M M J J A MONTHS Fig. 4 — Should these variations be left to chance? a. Xo. b. Yes. variation or, in other words, criteria which when applied to a set of observed values will indicate whether or not it is reasonable to believe ECONOMIC QUALITY CONTROL OF PRODUCT 375 that the causes of variabiHty should be left to chance. Such criteria are basic to any method of securing control within limits. Let us, therefore, consider them critically. It is too much to expect that the criteria will be infallible. We are amply rewarded if they appear to to work in the majority of cases. Generally speaking, the criteria are' of the nature of limits derived from past experience showing within what range the fluctuations in quality should remain, provided they are to be left to chance. For example, when such limits are placed on the fluctuations in the qualities shown in Fig. 3, we find (see Fig. 4) that in one case two points fall out- side the limits and in the other case no point falls outside the limits. Upon the basis of the use of such limits, we look for trouble in the form V H _J < a O z o I- o z D u. Ul 1 O in • • • ONLY SUCH VARIATIONS •SHOULD BE « LEFT* TO , CHANCE AS TIME GOES ON Fig. 5 — Art plus modern statistical machinery makes possible the establishment of such limits. of assignable causes in one case but not in the other. However, to be of really practical interest, we should be able to answer the following question: Can we expect to be able to find and eliminate causes of variability only when deviations fall outside the limits? First, let us see what statistical theory has to say in answer to this question. Upon the basis of postulate 3, it follows that we can find and remove causes of variability until the remaining system of causes is constant or until we reach that state where the probability that the deviations in quality remain within any two fixed limits (Fig. 5) is constant. How- ever, this assumption alone does not tell us that there are certain limits within which all observed values of quality should remain provided the causes cannot be found and eliminated. In fact so long as the limits are set so that the probability of falling within the limits is less than 376 BELL SYSTEM TECHNICAL JOURNAL unity, we may alweiys expect a certain percentage of observations to fall outside the limits even though the system of causes be constant. In other words, the acceptance of this assumption gives us a right to believe that there is an objective state of control within limits but in itself it does not furnish the practical criterion for determining when variations in quality, such as given in Fig. 3, should be left to chance. Furthermore, we may say that mathematical statistics as such does not give us the desired criterion. What does this situation mean in plain every day engineering English? Simply this: such criteria, if they exist, cannot be shown to exist by any theorizing alone, no matter how well equipped the theorist is in respect to probability or statistical theory. We see in this situation the long recognized dividing line 4.0 3.5 UJ 3.0 > 1- O UJ 2.5 a. UJ Q 2.0 1- Z o 1.5 a. Q. 1.0 0.5 AVERAGE AVERAGE AVERAGE 1923 - 1924 1925 YEAR 1926 (9 months) Fig. 6 — Evidence of improvement in quality with approach to control. between theory and practice. The available statistical machinery referred to by the magazine Nature is, as we might expect, not an end in itself but merely a means to an end. In other words, the fact that the criterion which we happen to use has a fine ancestry of high- brow statistical theorems does not justify its use. Such justification must come from empirical evidence that it works. As the practical engineer might say, the proof of the pudding is in the eating. Let us therefore look for the proof. 3. Evidence that Criteria Exist for Detecting Assignable Causes A . Fig. 6 shows the results of one of the first large scale experiments to determine whether or not indications given by such a criterion applied to quality measured in terms of fraction defective were justified by experi- ECONOMIC QUALITY CONTROL OF PRODUCT 377 ence. About thirty typical items used in the telephone plant and pro- duced in lots running into the millions per year were made the basis for this study. As shown in this figure during 1923-24, these items showed 68 per cent control about a relatively low average of 1.4 per cent defec- tive.^ However, as the assignable causes indicated by deviations in the observed monthly fraction defective falling outside of control limits were found and eliminated, the quality of product approached the state of control as indicated by an increase of from 68 per cent to 84 per cent control by the latter part of 1926. At the same time the quality improved; in 1923-24 the average per cent defective was 1.4 per cent whereas by 1926 this had been reduced to .8 per cent. Here we get some typical evidence that, in general, as the assignable causes are removed, the variations tend to fall more nearly within the limits as indicated by an increase from 68 per cent to 84 per cent. Such evi- dence is, of course, one sided. It shows that when points fall outside the limits, experience indicates that we can find assignable causes, but it does not indicate that when points fall within such limits, we cannot find causes of variability. However, this kind of evidence is provided by the following two typical illustrations. TABLE 2 Electrical Resistance of Insulations in Megohms, Should Such Variations be Left to Chance? 5045 4635 4700 4650 4640 3940 4570 4560 4450 4500 5075 4500 4350 5100 4600 4170 4335 3700 4570 3075 4450 4770 4925 4850 4350 5450 4110 4255 5000 3650 4855 2965 4850 5150 5075 4930 3975 4635 4410 4170 4615 4445 4160 4080 4450 4850 4925 4700 4290 4720 4180 4375 4215 4000 4325 4080 3635 4700 5250 4890 4430 4810 4790 4175 4275 4845 4125 4425 3635 5000 4915 4625 4485 4565 4790 4550 4275 5000 4100 4300 3635 5000 5600 4425 4285 4410 4340 4450 5000 4560 4340 4430 3900 5000 5075 4135 3980 4065 4895 2855 4615 4700 4575 4840 4340 4700 4450 4190 3925 4565 5750 2920 4735 4310 3875 4840 4340 4500 4215 4080 3645 4190 4740 4375 4215 4310 4050 4310 3665 4840 4325 3690 3760 4725 5000 4375 4700 5000 4050 4185 3775 5075 4665 5050 3300 4640 4895 4355 4700 4575 4685 4570 5000 5000 4615 4625 3685 4640 4255 4090 4700 4700 4685 4700 4850 4770 4615 5150 3463 4895 4170 5000 4700 4430 4430 4440 4775 4570 4500 5250 5200 4790 3850 4335 4095 4850 4300 4850 4500 4925 4765 5000 5100 4845 4445 5000 4095 4850 4690 4125 4770 4775 4500 5000 B. In the production of a certain kind of equipment, considerable cost was involved in securing the necessary electrical insulation by means of materials previously used for that purpose. A research pro- gram was started to secure a cheaper material. After a long series of preliminary experiments, a tentative substitute was chosen and an 1 Jones, R. L., "Quality of Telephone Materials," Bell Telephone Quarterly, June, 1927. 378 BELL SYSTEM TECHNICAL JOURNAL extensive series of tests of insulation resistance were made on this material, care being taken to eliminate all known causes of variability. Table 2 gives the results of 204 observations of resistance in megohms taken on as many samples of the proposed substitute material. Reading from top to bottom beginning at the left column and con- tinuing throughout the table gives the order in which the observations were made. The question is: "Should such variations be left to chance?" No a priori reason existed for believing that the measurements form- ing one portion of this series should be different from those in any other portion. In other words, there was no rational basis for dividing the SHOULD THESE VARIATIONS BE LEFT TO CHANCE? 52 10 50 5 I O O 48 UJ 2 46 UJ O 44 Z 34 •• • •-• -•-s? • • 52 ■ (0 50 ? I g48 - UJ 5 46 . Z •• UJ ^ > • •« • z ^42 in Ul a. 40 1- UJ o < 3fl — a. UJ > < 36 ~ 34 1 1 1 1 10 20 30 SAMPLE ANSWER: "NO" 40 50 10 SAMPLE ANSWER; "YES" 20 Fig. 7. total set of data into groups of a given number of observations except that it was reasonable to believe that the system of causes might have changed from day to day as a result of changes in such things as atmos- pheric conditions, observers, and materials. In general, if such a change is to take place, we may readily detect its effect provided we divide the total number of observations into comparatively small sub- groups. In this particular instance, the size of the sub-group was taken as four and the black dots in Fig. 1-a show the successive averages of four observations in the order in which they were taken. The dotted lines are the limits within which experience has shown that these observations should fall, taking into account the size of the sam- ECONOMIC QUALITY CONTROL OF PRODUCT 379 pie, provided the variability should be left to chance. Several of the observed values lie outside these limits. This was taken as an indica- tion of the existence of causes of variability which could be found and eliminated. Further research was instituted at this point to find these causes of variability. Several were found and after these had been eliminated, another series of observed values gave the results indicated in Fig. 1-b. Here we see that all of the points lie within the limits. We assumed, therefore, upon the basis of this test, that it was not feasible for research to go much further in eliminating causes of variability. Because of the importance of this particular experiment, however, 48.0 o X a> a. o 47.8 O O < cr tu > < 47.6 L 0 4 8 12 16 SAMPLE Fig. 8 — \'ariations that should be left to chance. Does the criterion work? "Yes." considerably more work was done, but it failed to reveal causes of variability. Here then is a typical case where the criterion indicates when variability should be left to chance. C. Suppose now that we take another illustration where it is reason- able to believe that almost everything humanly possible has been done to remove the assignable causes of variation in a set of data. Perhaps the outstanding series of observations of this type is that given by Millikan in his famous measurement of the charge on an electron. Treating his data in a manner similar to that indicated above, we get the results shown in Fig. 8. All of the points are within the dotted limits. Hence the indication of the test is consistent with the ac- cepted conclusion that those factors which need not be left to chance had been eliminated before this particular set of data were taken. 380 BELL SYSTEM TECHNICAL JOURNAL 4. Role Played by Statistical Theory It may appear thus far that mathematical statistics plays a relatively minor role in laying a basis for economic control of quality. Such, however, is not the case. In fact, a central concept in engineering work of today is that almost every physical property is a statistical distribu- tion. In other words, an observed set of data constitutes a sample of the effects of unknown chance causes. 1 1 is at once apparent, therefore, that sampling theory should prove a valuable tool in testing engineering 60 r 50 60 70 80 90 100 MODULUS OF RUPTURE IN 100 POUNDS PER SQUARE INCH Fig. 9 — Variability in modulus of rupture of clear specimens of green sitka spruce typical of the statistical nature of physical properties hypotheses. Here it is that much of the most recent mathematical theory becomes of value particularly in analysis involving the use of comparatively small numbers of observations. Let us consider, for example, some property such as the tensile strength of a material. Provided our previous assumptions are justi- fied, it follows that after we have done everything we can to eliminate assignable causes of variation, there will still remain a certain amount of variability exhibiting the state of control. Let us consider an ex- tensive series of data recently published by a member of the Forest Products Laboratories'- (Fig. 9). Here we have the results of tests for tensile strength on L304 small test specimens of sitka spruce, the kind - Xewlin, J. A., Proceedings of the American Societv of Civil Engineers, September, 1926, pp. 1436-1443. ECONOMIC QUALITY CONTROL OF PRODUCT 381 of material used in aeroplane propellers during the war. The wide variability is certainly striking. The smooth solid curve is an approx- imation to the distribution function for this particular property repre- senting at least approximately a state of control. The importance of going from the sample to the smooth distribution is at once apparent and in this case a comparatively small amount of refinement in statisti- cal machinery is required. Suppose, however, that instead of more than a thousand measure- ments we had only a very small number, such as is so often the case in engineering work. Our estimation of the variability of the distribution function, representing the state of control, upon the basis of the inform- ation given by the sample would necessarily be quite different from that ordinarily used by engineers (see Fig. 10). This is true even though i.Sr Z 1.6 o 1- < > LU Q gl.4 < Z < I- (/) ti. O 1.2 UJ I- < \- LJ LOh INEW OLD i- 1 ' 1 1 1 ^ L_^ 1 1 \ I I I 1 I I I I 0 3 5 10 20 50 100 n Fig. lO^Correction factors made possible by modern statistical theorj' are often large. — ^Typical Illustration. we make the same kind of assumption to begin with as engineers have been accustomed to do in the past. This we may take as a typical example of the fact that the production engineer finds it to his advan- tage to keep abreast of the developments in statistical theory. Here we use new in the sense that much of modern statistical machinery is new to most engineers. 382 BELL SYSTEM TECHNICAL JOURNAL 5 . Conclusion Based upon evidence such as already presented, it appears to be practicable to set up criteria by which to determine when assignable causes of variations in quality have been eliminated so that the product may then be considered to be controlled within limits. This state of control appears to be, in general, a kind of limit to which we may expect to go economically in finding and removing causes of variability without changing a major portion of the manufacturing process as, for example, would be involved in the substitution of new materials or designs. III. Advantages Secured through Control 1. Reduction in the Cost of Inspection If we can be assured that something we use is produced under con- trolled conditions, we do not feel the need for inspecting it as much as fi,6 A S O N 1927 M J J / 1928 MONTHS S O N D J F MAM 1929 0.6 <-0,2 _1_ _!_ JASON 1927 _1_ _1_ J_ _1_ _1_ _1_ MAMJ JASON 1926 MONTHS F M A M 1929 Fig. 11 — Approach to stable equilibrium or control as assignable causes are weeded out, thus reducing the need for inspection. we would if we did not have this assurance. For example, we do not waste our money on doctors' bills so long as we are willing to attribute the variability in our health to the effects of what in our present termin- ology corresponds to a constant system of chance causes. In the early stages of production there are usually causes of varia- bility which must be weeded out through the process of inspection. As ECONOMIC QUALITY CONTROL OF PRODUCT 383 we proceed to eliminate assignable causes, the quality of product usu- ally approaches a state of stable equilibrium somewhat after the man- ner of the two specific illustrations presented in Fig. 11. In both instances, the record goes back for more than two years and the process of elimination in each case covers a period of more than a year. It is evident that as the quality approaches what appears to be a comparatively stable state, the need for inspection is reduced. 2. Reduction in the Cost of Rejections That we may better visualize the economic significance of control, we shall now view the production process as a whole. We take as a specific illustration the manufacture of telephone equipment. Picture, if you will, the twenty or more raw materials such as gold, platinum, silver, copper, tin, lead, wool, rubber, silk, and so forth, literally col- lected from the four corners of the earth and poured into the manu- facturing process. The telephone instrument as it emerges at the end of the production process is not so simple as it looks. In it there are 201 parts, and in the line and equipment making possible the connec- tion of one telephone to another, there are approximately 110,000 more parts. The annual production of most of these parts runs into the millions so that the total annual production of parts runs into the billions. How shall the production process for such a complicated mechanism be engineered so as to secure the economies of quantity production and at the same time a finished product with quality characteristics lying within specified tolerances? One such scheme is illustrated in Fig. 12. Here the manufacturing process is indicated schematically as a funnel, at the small end of which we have the 100 per cent inspection screen to protect the consumer by assuring that the quality of the finished product is satisfactory. Obviously, however, it is often more econom- ical to throw out defective material at some of the initial stages in production rather than to let it pass on to the final stage where it would likely cause the rejection of a finished unit of product. For example, we see to the right of the funnel, piles of defectives, which must be junked or reclaimed at considerable cost. It may be shown theoretically that, by eliminating assignable causes of variability, we arrive at a limit to which it is feasible to go in reducing the fraction defective. It must sufiice here to call attention to the kind of evidence indicating that this limiting situation is actually approached in practice as we remove the assignable causes of variability. Let us refer to the information given in Fig. 6 which is particularly significant because it represents the results of a large scale experiment 384 BELL SYSTEM TECHNICAL JOURNAL carried on under commercial conditions. As the black sectors in the pie charts decrease in size, indicating progress in the removal of as- signable causes, we find simultaneously a decrease in the average frac- RAW MATERIAL INSPECTION TO REDUCE COST OF PRODUCTION PARTS PARTS too t- O UJ u. LU Q 6.20 I- Z UJ o a. UJ °- 11.70 • • ,•• • . • • • • ^ • • * *»• • • •• •••• • • ' • • ■ • -— — w-r*- — «>«' — •^-tt"* «-• ^. — " vx= • • • BAKERY I BAKERY 2 BAKERY 3 BAKERY 4 BAKERY 5 BAKERY 6 • • ••-. ♦. "• 9 • •» »-^» = BAKERY 7 • • 3.78 4.90 4.82 •• • • ••• ^•^« • . • • • •^ =F5^ • • • •• • • • - _- . • • • •• • • BAKERY 8 BAKERY 9 •• ■,*>■ • • • • = BAKERY 10 I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I I 0 5 10 15 20 25 30 35 40 WEEKS Fig. 13 — Results showing how control effects a reduction in the cost of rejections. Some time ago it became possible to secure the weekly record of return of stale bread for ten different bakeries operating in a certain metropolitan district. These observed results are shown graphically in Fig. 13. At once we see that there is a definite lack of control on the 386 BELL SYSTEM TECHNICAL JOURNAL part of each bakery. The important thing for us to note, however, is that the bakery having the h^west percentage return, 1.99 per cent, also shows better control than the other bakeries as judged by the number of points falling outside the control limits in the period of 36 weeks. 3. Attainment of Maximum Benefits from Quantity Production The quality of the finished product depends upon the qualities of raw materials, piece parts and the assembling process. It follows from simple theory that so long as such quality characteristics are controlled, the quality of the finished unit will be controlled, and will therefore exhibit minimum variability. Other advantages also result. For example, by gaining control, it is as we have already seen, possible to establish standard statistical distributions for the many quality char- acteristics involved in design. Very briefly, let us see just how these statistical distributions, representing states of control, become useful in securing an economic design and production scheme. Suppose we consider a simple problem in which we assume that the quality characteristic Y in the finished product is a function / of m different quality characteristics, Xi, X^, • • • , X^, representable symbolically by Equation (3). Y =f{X„ X,, ■■■, XJ. (3) For example, one of the X's might be a modulus of rupture, another a diameter of cross section, and Y a breaking load. Engineering re- quirements generally place certain tolerances on the variability in the resultant quality characteristic F, which variability is in turn a func- tion of the variabilities in each of the m different quality characteristics. As already stated, the quality characteristic Y will be controlled provided the m independent characteristics are controlled. Knowing the distribution functions for each of the m different independent variables, it is possible to approximate very closely the per cent of the finished product which may be expected to have a quality characteristic Y within the specified tolerances. If it is desirable to minimize the variability in the resultant quality Y by proper choice of materials, for example, and, if standard distribution functions for the given quality characteristics are available for each of several materials, it is possible to choose that particular material which will minimize the variability of the resultant quality at a minimum of cost. 4. Attainment of Uniform Quality Even TJiough Inspection Test Is Destructive So often the quality of a material of the greatest importance to the individual is one which cannot be measured directly without destroying ECONOMIC QUALITY CONTROL OF PRODUCT 387 the material itself. So it is with the fuse that protects your home; with the steering rod on your car; with the rails that hold the locomotive in its course; with the propeller of an aeroplane, and so on indefinitely. How are we to know that a product which cannot be tested in respect to a given quality is satisfactory in respect to this same quality? How are we to know that the fuse will blow at a given current; that the steer- ing rod of your car will not break under maximum load placed upon it? To answer such questions, we must rely upon previous experience. In such a case, causes of variation in quality are unknown and yet we are concerned in assuring ourselves that the quality is satisfactory. Enough has been said to show that here is one of the very important applications of the theory of control. By weeding out assignable causes of variability, the manufacturer goes to the feasible limit in assuring uniform quality. 5. Reduction in Tolerance Limits By securing control and by making use of modern statistical tools, the manufacturer not only is able to assure quality, even though it cannot be measured directly, but is also often able to reduce the tolerance limits in that quality as one very simple illustration will serve to indicate. Let us again consider tensile strength of material. Here the measure of either hardness or density is often used to indicate tensile strength. In such cases, it is customary practice to use calibration curves based upon the concept of functional relationship between such characteris- tics. If instead of basing our use of these tests upon the concept of functional relationship, we base it upon the concept of statistical rela- tionship, we can make use of planes and surfaces of regression as a means of calibration, thus in general making possible a reduction in the error of measurement of the tensile strength and hence the establish- ment of closer tolerances. It follows that this is true because, when quality can be measured directly and accurately, w^e can separate those samples of a material for which the quality lies within given tolerance limits from all others. Now, when the method of measurement is indirect and also subject to error, this separation can only be carried on in the probability sense assuming the errors of measurement are controlled by a constant system of chance causes. It is obvious that, corresponding to a given probability, the tolerance limits may be re- duced as we reduce the error of measurement. Fig. 14 gives a simple illustration. Here the comparative magni- tudes of the standard deviations of strength about the two lines of regression and the plane ^ of regression are shown schematically by the ^ For definition of these terms see any elementary text book on statistics. 388 BELL SYSTEM TECHNICAL JOURNAL a i/i 40,000 I I- o g 30,000 q: I- <0 to Z 20,000 L t; 40 60 60 100 2.4 2.6 2.8 3.0 ROCKWELL HARDNESS (E) DENSITY IN GRAMS PER CENTIMETER CUBE A B Z = TENSILE STRENGTH IN LBS./SQ. IN. Y=ROCKWELL HARDNESS (E) X = DENSITY IN GRAMS PER CENTIMETER CUBE Fig. 14 — How control makes possible improved quality through reduction in tolerance limits. ECONOMIC QUALITY CONTROL OF PRODUCT 389 lines in Fig. 14-J. The lengths of these are proportional to the allow- able tolerance limits corresponding to a given probability. Customary practice is to use the line of regression between tensile strength and hardness. Note the improvement effected by using the plane of regres- sion. By using the hardness and density together as a measure of tensile strength in this case, the tolerance limits on tensile strength corresponding to a given probability can be reduced to approximately one-half what they would be if either of these measures were used alone. IV. Conclusion It seems reasonable to believe that there is an objective state of control, making possible the prediction of quality within limits even though the causes of variability are unknown. Evidence has been given to indi- cate that through the use of statistical machinery in the hands of an engineer artful in making the right kind of hypothesis, it appears possible to establish criteria which indicate when the state of control has been reached. It has been shown that by securing this state of control, we can secure the following advantages: 1. Reduction in the cost of inspection. 2. Reduction in the cost of rejections. 3. Attainment of maximum benefits from quantity production. 4. Attainment of uniform quality even though inspection test is destructive. 5. Reduction in tolerance limits where quality measurement is indirect. 26 optimum Reverberation Time for Auditoriums By WALTER A. MAC N AIR i The suggestion is made that the sound damping material in an auditorium should be such that the loudness of tones will decay at the same rate for all frequencies. To attain this the reverberation time at 80 cycles must be twice what it is at 1000 cycles. The change of optimum reverberation time with volume is shown to be derivable from a single hypothesis. I. Reverberation Time vs. Frequency THERE is very little published data in regard to the change in reverberation time with frequency in auditoriums which are considered near ideal. It is often mentioned by engineers and physi- cists that to secure the best acoustical results, the reverberation time should be the same for all frequencies in any one room. This specifies that the sensation level shall decay at the same rate for all frequencies of interest. It seems more reasonable, however, to specify that the loudness of all pure tones shall decay at the same rate for all frequencies since it is the loudness of a tone which takes into consideration not only the energy level but also its ultimate effect upon one's brain. In Fig. 1 - are plotted data which show the relation between the loudness as judged by a considerable number of observers and the sensation level. It will be seen that for frequencies between 700 and 4000 cycles per second these two quantities are equal to each other so that the two points of view mentioned above demand identical conditions through- out this frequency band. Outside of this band, however, any change in the sensation level gives a greater change in the loudness, as may be seen. The maximum loudness in which we are interested at present is about 73.^ In the figure the curves may be replaced by straight lines which represent fair approximations to the observed data up to this loudness. This family of straight lines may be represented by the expression Lt = AfSu (1) where Aj \s the slope of the line adopted to fit the data for the fre- 1 Presented before Acoustical Soc. of Amer., Dec, 1929. Jour. Acou. Soc. Amer., Jan., 1930. -This is Fig. 108 from "Speech and Hearing" by H. Fletcher. ^ This is the loudness that the source chosen in Part II of this paper will produce in a room of 1000 cubic feet having a reverberation time of 0.8 seconds. 390 OPTIMUM REVERBERATION TIME FOR AUDITORIUMS 391 quency /. The values of A; chosen from this figure are given by the next, Fig. 2. This approximation simpHfies our calculations very much and introduces errors which are not intolerable. Referring back to Fig. 1, if we wish to adjust the absorption of the room so that the loudness of all pure tones will decay at the same rate, say for the moment 60 units per second, it is seen that the sensation level must drop 60 db per second for frequencies between 700 and «, V / y y <^ /> y « ^ / / / /, <>, y / ^y /' / 'a '/ AD / / ;> 1/ 'A / A ^ / A ° / / / f / / / A? fA An ,f '+ A f A f k / / UJ Z r J / A / / a f / / / y %' / -J f / / / / V f ^k y / / / / V. / **n l\ ^ 0 V L "a y 1 — pn /^ y ^ J in i A » f 0 / 10 20 30 40 50 60 70 SENSATION LEVEL 80 90 100 Fig. 1. — Loudness of pure tones. 4000 cycles and for other frequencies the sensation level must drop 60/.(4/ db per second; or in other words, the reverberation time for frequencies between 700 and 4000 cycles should be one second and outside of this band it should be Aj seconds. Fig. 2, then, which is a plot oi A} vs. frequency now becomes also an illustration of the shape of the reverberation time vs. frequency curve which a room should have in order that the loudness of pure tones of all frequencies shall decay at the same rate. 392 BELL SYSTEM TECHNICAL JOURNAL According to Sabine's well known formula the reverberation time is inversely proportional to the number of absorption units in the room so that, if we assume this, we may immediately infer the shape of the curve which represents the number of absorption units necessary at any frequency, referred to the amount required at 1000 cycles, to obtain our required condition. These values are plotted in Fig. 3. If it should happen that the greater part of the sound absorption in a room is caused by one particular kind of surface, then the curve in Fig. 3 is the shape of the absorption curve that this material should have. A pertinent observation on which every one seems to agree is that if 2.4 2 34568 2 34568 10 100 1,000 10,000 FREQUENCY IN CYCLES PER SECOND Fig. 2. — \ 'allies of Af vs. frequency. an auditorium has an unusually long reverberation time and conse- quently is of little use, when empty, it attains excellent acoustic conditions when filled with a large audience. In these cases a very large part of the absorption is caused by the audience. The absorp- tion of an average audience has been measured by W. C. Sabine ^ and his results are also plotted in Fig. 3. The close agreement between this curve and the one we have obtained from our hypothesis gives con- siderable confidence in our general viewpoint. II. Reverberation Time vs. Volume It is generally accepted that the best acoustical conditions in a room are obtained when the reverberation time is adjusted to a definite value known as the optimum reverberation time. Observations reported in literature agree that the value of the optimum reverberation time in- ^ "Collected Papers on Acoustics," page 86. OPTIMUM REVERBERATION TIME FOR AUDITORIUMS 393 creases with the size of the room in the way shown in Fig. 4 where the curves are the choices reported by Watson,^ Lifschitz,^ and Sabine.*' These experimental results have served as the basis of successful adjustment and design of many auditoriums. One naturally seeks the factor which determines a choice of reverberation time of two seconds for a million cubic feet theatre and on the other hand a choice of near one second for a 10,000 cubic foot music room. It is our pur- pose now to point out the factor which apparently does this. We will set down a condition which we believe to be this factor and then will show that the requirements demanded by it agree quite 100 90 /• ? > ^ ^ \ s y y Y /" s 70 / 60 50 y 1 y 40 y i 1 )EAL SPECIFIED BY THE PRESENT WORK 30 20 in PRODUCED BY AN AUDIENCE DATA OF W.C.SABINE VALUE 100 ASSIGNED FOR 1000 CYCLES 0 10 4568 2 34568 100 1,000 FREQUENCY IN CYCLES PER SECOND 4 5 6 8 10,000 Fig. 3. — Relative number of absorption units vs. frequency. closely with the empirical results illustrated in Fig. 4 and mentioned above. The condition is Ltdt = - K, (2) in which ^o is the time a sustained source of sound E is cut off, /i the time the sound becomes inaudible, Lt the loudness of the sound at any instant t, and K a constant. As shown in Fig. 1, the loudness of a one thousand cycle note is equal to the sensation level, that is, Lt = Si for 1000 cycles. ■•Watson, Architecture, May, 1927. * Lifschitz, Phys. Rev., 27, 618, 1926. « Sabine, Trans, of S.M.P.E., XII, 35, 1928. 394 BELL SYSTEM TECHNICAL JOURNAL Since, during the time of decay, Si decreases uniformly with time, and therefore Lt also, then for a thousand cycle note, evaluating our integral we have Lt,T, = 2K (3) or St^Ti = 2K, where T, = t,- to. This last expression is practically in the form in which this condition was first stated by Lifschitz.'' In (3) there are three unknowns and a fourth is implied, namely, the power of the source, E. We now turn our attention to finding the relation between the volume of a room and the reverberation time dictated by the stated condition. Following P. E. Sabine let us take the rate of emission of the source, E to be 10^° cubic meters (35.3 X 10^" cubic feet) of sound of threshold density per second. Now ^ 4V, 4 X 35.3 X 10'" Ti = — loge , ca c • a where F is the volume of the room in cubic feet. c is the velocity of sound, 1120 feet per second. a is the number of absorption units in sq. feet and '^ T c ini 4 X 35.3 X 10'" Li, = St, = 10 logio ~ If we should substitute these values in (3) we would obtain a relation between V, a, and K which must be satisfied when condition (2) is satisfied. In other words, this relation would specify the amount of absorption, for a one thousand cycle note, a room should have if it complies with (2). If we assume Sabine's well known formula, namely, ^ _ .057 where T is the reverberation time in seconds we may express this relation in terms of V, T, and K with the result (2KYI- 10.40 + log To, - log V = ^ ^g3 j.^ ,., , (4) ' See Crandall "Theory of Vibrating Systems and Sounds," page 211. * See Crandall "Theory of Vibrating Systems and Sounds," page 210, and the definition of sensation level. OPTIMUM REVERBERATION TIME FOR AUDITORIUMS 395 where Top is the value of T imposed by our condition (2) for a thousand cycle tone. Referring to Fig. 4 it will be seen that all three observers agree rather closely that the reverberation time for an auditorium of 1,000,000 cubic feet should be 2.0 seconds. This value refers to a tone of 512 cycles, the customary frequency used for experimental obser- vation. It has been shown above that the reverberation time for 8 10,000 100,000 VOLUME IN CUBIC FEET 4 5 6 8 l,000,000' Fig. 4. — Optimum reverberation time \s. volume in cubic feet for 512 cycles. 1000 cycles should be 92.5 per cent of the reverberation time for a 512 cycle tone, so that the 2.0 seconds above corresponds to 1.85 seconds for 1000 cycles. We can evaluate K in (4) by adapting this latter value of Top for a volume of 1,000,000 cubic feet. This gives K = 32.6. Substituting this value in (4) we obtain 6.35 log V = 10.40 + log Top - T 1/2 ■* op (5) 1.8 1.6 §1.2 o O 1.0 tiJ - ^ ^ ir^ -H - - -"'^ . — - — 0.8 z " 0.6 0.4 0.2 0 1,000 4568 2 34568 10,000 100,000 VOLUME IN CUBIC FEET 4 5 6 8 1,000,000 Fig. 5. — Optimum reverberation time vs. volume in cubic feet for 1000 cycles. From (5) we may obtain Top for 1000 cycles for any volume. See Fig. 5. As mentioned above these values of Top are 92.5% of Top 396 BELL SYSTEM TECHNICAL JOURNAL for 512 cycles so that these latter may be easily deduced for com- parative purposes. These values are plotted to give curve number 4 in Fig. 4. It is seen that this curve agrees very well with those showing the choices of competent judges. III. The Mork General Hypothesis Equation (2) may be written as follows, since we have assigned a value to K: Ltdt = - 32.6 (6) L and it will be remembered that we have considered Li^ to be the loud- ness set up by a certain standard source. Allowing V to vary with / constant (1000 cycles) we have obtained a relation between the opti- mum reverberation time and volume of rooms for 1000 cycles. We wish to point out now that exactly this same condition (6) with V constant and / variable, will give the same results that we have ob- tained in Part I of this paper with the only further requirement that for other frequencies than 1000 cycles the strength of the source E shall be such that the loudness L/^ set up in the room at the frequency considered shall be exactly the same as the loudness which our standard source would set up at 1000 cycles. In Part I of this paper our stated condition was that the loudness of all pure tones shall decay at the same rate for all frequencies. Since we have specified that the loudness at the time /o shall be the same for all test frequencies and also that the loudness at the time ti shall be zero for all frequencies, it is quite evident that the above integral can have the same value at all these frequencies only when the loudness decays at the same rate for all frequencies concerned. In other words, this condition stated as an integral specifies exactly the same require- ment on the decay of loudness that we expressed in our statement early in Part I of this paper. IV. Conclusions To recapitulate, we have set down an equation, together with a specification of the strength of the virtual source in each case, from which we obtain the value the reverberation time for any frequency tone should have in any sized room according to the condition which apparently controls the choice of observers. One naturally turns to see what meaning may be attached to this significant expression, namely, the integral of the loudness taken OPTIMUM REVERBERATION TIME FOR AUDITORIUMS 397 throughout the time of decay to inaudibility. Since this integral has the same value for all auditoriums which are considered ideal, it implies that one's brain is a ballistic instrument which is concerned with not only the maximum value of loudness but also with the effect of loudness integrated throughout a considerable interval of time. Abstracts of Technical Articles from Bell System Sources Phenomena in Oxide Coated Filaments.^ Joseph A. Becker. A theory of the changes in activity in oxide coated filaments is proposed. From a comparison of the behavior of these filaments and filaments with composite surfaces such as thorium on tungsten, caesium on tungsten, and casium on oxygen on tungsten it appears probable that oxide coated filaments owe their high activity to adsorbed metallic barium. The changes in emission from a coated filament produced by changes in plate potential and by currents sent into or drawn from it, are ascribed to electrolysis of the oxide. When electrons are sent into a coated filament barium is deposited on the surface and the activity increases until an optimum is reached beyond which the activity decreases. When current is drawn from the oxide, oxygen is deposited on the surface. If the oxygen is beneath the adsorbed barium, it increases the activity; if it is above the barium, it decreases the activity. Both barium and oxygen diffuse readily from the sur- face into the oxide and vice versa. This theory is tested, confirmed, and extended by numerous experiments. An experimental technique is employed by which relative rates of evaporation of small amounts of electropositive and electronegative materials can be determined with considerable precision. The same technique might be useful in a number of similar investigations. Metallic barium or oxygen which evaporate from a coated filament are allowed to deposit on one side of a flat tungsten ribbon whose thermionic activity is followed. When the plausible assumption is made that an optimum activity is obtained when the tungsten is covered with a single layer of electropositive material, the relative rates of evaporation can be converted to absolute rates. This tech- nique is also employed to determine the factors which control the evaporation of oxygen from a coated filament. Estimation of the Volatile Wood Acids Corrosive to Lead Cable Sheath.^ R. M. Burns and B. L. Clarke. The detection of volatile acids in the air drawn from creosoted wood conduit corrosive to lead cable sheath made desirable the development of a suitable method for the extraction and estimation of volatile wood acids. Such a method consists in the condensation of the volatile constituents of wood sawdust removed under reduced pressure and titration of the conden- 1 ThePhys. Rev. Nov. 1929. ' Jndust. and Eng. Chem., Jan. 1930. 398 ABSTRACTS OF TECHNICAL ARTICLES 399 sate using a modified differential potentiometric electrode. Acidity data have been obtained for Douglas fir, western hemlock, southern yellow pine, western pine, spruce, redwood, cedar, and oak, and a correlation is attempted between these acidities and the observed corrosive character of the woods. Electron Waves.^ C. J. Davisson. This paper is a brief review of the experiments made on the diffraction of electrons by crystal during the first two years following the discovery of this phenomenon, and an indication of the paths along which future experimentation may be expected to proceed. Television in Colors by a Beam Scanning Method.'^ Herbert E. Ives and A. L. Johxsrud. It has been recognized ever since the practical achievement of television, and indeed before, that television might be achieved in colors by utilizing the principles used in three-color photography. The requirements in the two cases are very closely parallel. Three-color photography had to wait for its practical achievement, on photographic materials sensitive to all colors of the visible spectrum. The parallel requirement in the case of television is for photoelectric cells similarly color sensitive. The requirements of television as to primary colors to be used for the synthesis of the colored image are relatively more difficult of fulfillment than in the case of color photography because in television we need not merely colored light sources, but light sources which shall be capable of following the variations of the television signal current with high speed. If, how- ever, these two requirements, namely color sensitive photoelectric cells and high speed-colored lights, are met, television in color could conceivably be realized by utilizing any one of a number of devices for analyzing and recombining images which have been successfully applied in three-color photography. Air Transport Communication.'" R. L. Jones and F. M. Ryan. The successful operation of an air transportation system depends in no small degree on the communication facilities at its command. Rapid and dependable communication between transport planes in flight and the ground is essential. Two-way radio telephony provides this necessary plane-to-ground contact. The design of a radio telephone system for this service requires quantitative knowledge of the transmission conditions encountered in ^ Jour. The Franklin Inst., Nov. 1929. * Jour. Opt. Soc. of Amer., Jan. 1930. 'Jour. A. /.£.£., Jan. 1930. 400 BELL SYSTEM TECHNICAL JOURNAL plane-to-ground communication. An experimental investigation of these conditions over the available frequency range has been carried out and the results are described. A complete aircraft radio telephone system designed for the use of air transport lines and an airplane radio receiver designed for reception of government radio aids to air navigation are also described. A Study of Noise in Vacuum Tubes and Attached Circuits. '^ F. B. Llewellyn. The noises originating in vacuum tubes and the attached circuits are investigated theoretically and experimentally under three headings: (1) shot effect with space charge, (2) thermal agitation of electricity in conductors, (3) noise from ions and secondary electrons produced within the tube. A theoretical explanation of the shot effect in the presence of space charge is given which agrees with experiment insofar as a direct deter- mination is possible. It is shown that the tubes used should be capable of operating at full temperature saturation of the filament in order to reduce the shot effect. In the computation of the thermal noise originating on the plate side of a vacuum tube, the internal plate resistance of the tube is to be regarded as having the same temperature as the filament. Noise produced by ions within the tube increases as the grid is made more negative. With tubes properly designed to operate at temperature saturation it is possible to reduce the noise on the plate side to such an extent that the high impedance circuits employed on the grid side of the first tube of a high gain receiving system contribute practically all of the noise by virtue of the thermal agitation phenomenon. On the Nature of ''Active'' Carbon? H. H. Lowry. Practically all investigators have used for their measure of "activity" the adsorp- tive capacity of the carbon (charcoal) under certain arbitrary condi- tions. In several previous papers data have been given which indicate that the adsorptive capacity of carbon is increased by any process which increases either the total surface per unit weight or the degree of unsaturation of the surface atoms, or both. No exceptions to this generalization have been encountered. Since the adsorptive capacity is dependent on two factors which may be independently varied, it seems hardly logical to continue its use as a measure of the activity of carbon. Since it is generally recognized that the forces effective in adsorption processes are a result of the unsaturation of the surface * Proc. The Inst. Radio Engiiieers, Feb. 1930. ''Jour, of Fhys. Cliem., Jan. 1930. ABSTRACTS OF TECHNICAL ARTICLES 401 atoms, the ratio of the adsorptive capacity to the total adsorbing sur- face would appear to be much more satisfactory for a measure of the activity. The data shown graphically in this paper show that starting with a given raw material, i.e., an anthracite coal, an increase in the tempera- ture to which the material is heated above 1000° decreases the adsorp- tive capacity per unit pore volume. It is pointed out that the pore volume may be considered a measure of the extent of adsorbing sur- face and that the activity of an adsorbent carbon (charcoal) should be measured by the amount of gas adsorbed per unit area of its surface. The data, therefore, indicate that the activity of a charcoal is indepen- dent of the atmosphere in which it is prepared and dependent only on the maximum temperature to which it is heated. At any temperature between 900 and 1300° an increase in the adsorptive capacity is most probably accompanied by a proportional increase in the extent of the adsorbing surface. For example, although the adsorptive capacity of the samples prepared at 1100° ranged from 1.8 to 23.1 c.c. carbon dioxide per gram at 0° and atmospheric pressure, the actually meas- ured values of activity ranged from 0.201 to 0.295, while the weighted average for all the samples prepared at the same temperature was 0.27 : the variations observed are believed to be due to the limitations, which have been discussed, of the measure of the surface area rather than to a real difference in the activity. The Operation of Modulators from a Physical Viewpoint.^ E. Peterson and F. B. Llewellyn. The mathematical expressions which occur in the treatment of non-linear devices as circuit elements are interpreted in terms of a graphical physical picture of the processes involved. This picture suggests, in turn, several useful ways of apply- ing the equations in cases where the driving forces are so large that the ordinary power series treatment becomes prohibitively cumbersome. In particular, the application has been made in detail to the calculation of the intermediate-frequency output to be expected from a heterodyne detector having an incoming radio signal and locally generated beating oscillator voltage applied on its grid and a circuit of finite impedance to the intermediate frequency attached to its plate. A Study of the Output Power Obtained from Vacuum Tubes of Different Types} H. A. Pidgeon and J. O. McNally. Economical operation of the large number of tubes involved in the Bell System makes nec- essary the adoption of common supply voltages. This requires that * Proc. The Inst. Radio Engineers, Jan. 1930. ^ Proc. The Inst. Radio Engineers, Feb. 1930. 402 BELL SYSTEM TECHNICAL JOURNAL repeater tubes of various types be designed to operate at a fixed plate voltage. For this reason the design of ampHfier tubes to give as large a power output as possible at the operating plate voltage is of considerable importance. In the case of three-electrode tubes it is possible from theoretical considerations to compute, approximately, the electrical parameters a tube must have in order to give the maximum output power of a given quality obtainable under fixed operating conditions. The electrical characteristics and output of fundamental, second, and third harmonics of two of the more common telephone repeater tubes are given. It is of considerable interest to determine whether greater power output of comparable quality can be obtained from tubes containing more than one grid. Since no sufficiently exact theoretical analysis of multi-grid tubes is yet available to permit the determination of the parameters of optimum tubes, a comparative experimental inves- tigation of a number of such structures has been undertaken. The electrical characteristics and output of fundamental, second, and third harmonics of several such experimental tubes are given. The power output of multi-grid tubes and of three-element tubes is compared. The reasons for the comparatively large power output of certain types of multi-grid tubes are discussed. Effect of Small Quantities of Third Elements on the Aging of Lead- Antimony Alloys}^ Earle E. Schumacher, G. M. Bouton, and Lawrence Ferguson. The data presented in this paper definitely show that small quantities of certain elements when added to lead — 1 per cent antimony alloys have a very marked effect on the rate at which antimony is precipitated from supersaturated solid solution. Some suggestions of the mechanism of this change can be had from a consider- ation of the experimental findings in conjunction with the pertinent equilibrium diagrams. Although the literature shows that the third elements studied are insoluble in lead in the solid phase, no results have been reported on alloys containing these elements in very low concentrations. Even though they should be insoluble in lead, antimony may so change the lead lattice that they become soluble in lead-antimony. Furthermore, since these elements form either compounds or solid solutions with anti- mony, there are forces of attraction between them which may be strong enough to carry small quantities of the third elements, along with the antimony, into solid solution in the lead. The resulting ter- ^" Indust. and Eng. Cliem., Nov. 1929. ABSTRACTS OF TECHNICAL ARTICLES 403 nary solutions, by their different energy relations, may cause the ob- served effects on the rate of precipitation of antimony. The Tube Method of Measuring Sound Absorption Coefficients.^^ E. C. Wente. The general principles underlying the tube method of measuring sound absorption can be derived conveniently from the analogous equations for the electrical transmission line. These equa- tions show that the actual method of measurement is capable of many modifications, some of which have already been adopted by various experimenters. However, if reliable results are to be obtained, it is important that the apparatus be so designed that the propagation along the tube be rectilinear and the attenuation small, and that the tone be kept free from harmonics. In the tube method the absorption is measured at perpendicular incidence, whereas in the reverberation method it is measured at ran- dom incidence. A theoretical study of the absorption of sound by por- ous materials as a function of the angle of incidence shows that in some cases there may be a considerable discrepancy between the values obtained by the two methods. The tube method may also give im- practicable results for materials which are to be used in the form of large panels and absorb sound largely by virtue of inelastic bending rather than because of their porosity. " Jour, of the Acoust. Sac. of Amer., Oct. 1929. Contributors to this Issue Ralph Bown, M.E., 1913, M.M.E., 1915, Ph.D., 1917, Cornell University, Captain Signal Corps, U. S. Army, 1917-19; Department of Development and Research, American Telephone and Telegraph Company, 1919-. Mr. Bown has been in charge of work relating to radio transmission development problems. He is a Past President of the Institute of Radio Engineers. John R. Carson, B.S., Princeton, 1907; E.E., 1909; M.S., 1912; American Telephone and Telegraph Company, 1914-. Mr. Carson is well known through his theoretical transmission studies and has published extensively on electric circuit theory and electric wave propagation. Karl K. Darrow, B.S., University of Chicago, 1911; University of Paris, 1911-12; University of Berlin, 1912; Ph.D., University of Chicago, 1917; Western Electric Company, 1917-25; Bell Telephone Laboratories, 1925-. Dr. Darrow has been engaged largely in writing on various fields of physics and the allied sciences. Some of his earlier articles on Contemporary Physics form the nucleus of a recently pub- lished book entitled "Introduction to Contemporary Physics" (D. Van Nostrand Company). A recent article has been translated and published in Germany under the title "Einleitung in die Wellen- mechanik." William Fondiller, B.S., College of the City of New York, 1903; E.E., Columbia University, 1909; M.A., Columbia University, 1913; Engineering Department, Western Electric Co., Inc., 1909-25; Bell Telephone Laboratories, Inc., 1925-. Mr. Fondiller's work has re- lated to the development of transmission apparatus, such as loading coils, filters, transformers, etc. and is now Assistant Director of Ap- paratus Development of Bell Telephone Laboratories, Inc. In this capacity he is responsible for the design of telephone apparatus and investigations of materials. Norman R. French, A.B., University of Maine, 191-i; A.M., 1916; Instructor, Physics Department, University of Maine, 1914-16; In- structor, Princeton University, 1916-17; General Staff, A.E.F., 1917- 18; Commanding Officer, Flash and Sound Ranging Sections, Army Engineers' School, A.E.F., 1918; American Telephone and Telegraph Company, Department of Development and Research, 1919-. Mr. French's work has related chiefly to loading, submarine cables and transmission quality. 404 CONTRIBUTORS TO THIS ISSUE 405 Charles W. Carter, Jr., A.B., Harvard, 1920;B.Sc., Oxford, 1923; American Telephone and Telegraph Company, Department of Devel- opment and Research, 1923- Mr. Carter's work has had to do with the theory of electrical networks and with problems of telephone quality. Walter Koenig, Jr., A.B., Harvard, 1923; Instructor and Re- search Assistant, Harvard, 1923-24; American Telephone and Tele- graph Company, Department of Development and Research, 1924-. Mr. Koenig has been engaged chiefly in studies relating to trans- mission quality. W. A. MacNair, B.Sc, Colgate Univ., 1920; Ph.D., Johns Hopkins Univ. 1925; National Research Fellow in Physics, 1925-27; Bell Telephone Laboratories, 1929-. W. P. Mason, B.S., University of Kansas, 1921; M.A., Columbia, 1924; Ph.D., Columbia, 1928. Engineering Department, Western Electric Company, 1921-25; Bell Telephone Laboratories, 1925-. Mr. Mason's work has been largely in transmission studies. A. A. Oswald, B.S., Armour Institute of Technology, 1916; E.E., 1927. Western Electric Company, Engineering Department, 1916-24; Bell Telephone Laboratories, Inc., 1925-. Mr. Oswald's work has been concerned with the development of long and short wave trans- atlantic and ship-to-shore radio-telephone systems; and, during the War, of systems for airplane radio-communication and radio-control. D. A. Quarles, A.B., Yale University, 1916; U. S. Army, 1917- 19; Engineering Department, Western Electric Company, 1919- 25; Bell Telephone Laboratories. 1925-. Mr. Quarles was earlier engaged in transmission studies of circuits and networks. More recently he was in charge of inspection engineering on apparatus products. As Assistant Director of Apparatus Development, he is now engaged in development work on Outside Plant products. Walter A. Shewhart, A.B., University of Illinois, 1913; A.M., 1914; Ph.D., University of California, 1917; Engineering Department, Western Electric Company and Bell Telephone Laboratories, 191 8-. Mr. Shewhart is making a special study of the application of probability theories to inspection engineering. The Bell System Technical Journal July, 1930 Radio Telephone Service to Ships at Sea * By WILLIAM WILSON and LLOYD ESPENSCHIED The ])aper discusses the American end of the ship-to-shore radio telephone system and the connecting equipment on board the Leviathan. The most suitable wavelengths for this service are in the short-wave range, but the use of these wavelengths complicates the problem, since different wavelengths are required according to the distance of the ship from shore, the time of day, season of year, etc. The problem on shipboard is further complicated by the fact that the transmitting and receiving systems are necessarily near together and special precautions are necessary to take care of interference from the radio telephone transmitter and the radio telegraphic services. In addition to interference from these sources, there is a background of interference in the ships' electrical equipment, all of which necessitates a much more powerful land station than is necessary on shipboard. In the present system, the shore transmitter has a power rating of 15 kw. and the ship transmitter of 500 watts. The shore transmitting station is lo- cated at Ocean Gate, N. J., and the receiving station at Forked River, N. J. At both of these locations, directive antennas are employed which cover the ships' lanes. The stations are connected by wire to the Long Lines toll office in New York, and the o\'er-all control of the circuit is carried out from this ])oint. Both the ship and shore transmitters are crystal controlled. The ship's receiver is highly selective and is of the double-detection type. Com- munication between the ship and the shore is carried out by use of a pair of frequencies, one for transmission in each of the two directions, separated from each other by about three per cent. Ships of a number of nations are being equipped with wireless telephone apparatus and as the service expands, it will undoubtedly be necessary to formulate a plan in which inter- national agreement is reached on the allocations of frequencies for ship-to- shore telephony and telegraphy, in order that undue interference within the ser\'ices themselves or between the two services shall not ensue. IN view of the developments which have recently taken place in the field of ship-to-shore radio telephony, it would appear appropriate to review the state of the science and to discuss the problems which have arisen, the facilities which have been installed, and the general results obtained. The ship-to-shore radio telephone system, which is here described, was opened for public service between the Leviathan and the United States on December 8, 1929. This was the first extension of the public telephone service to a ship at sea and enabled calls to be made between the vessel and any Bell System subscriber. The system as set up is intended primarily for giving telephone service to the larger passenger- carrying vessels as an extension to the wire network, and should be distinguished from the more simple uses which have been made of radio * Presented at the North Eastern District Meeting of the A. I. E. E., Springfield, Mass., May 1930. 407 27 408 BELL SYSTEM TECHNICAL JOURNAL telephony in the marine field, such as that of enabling a coastal station operator to talk with coast guard vessels, fishing trawlers, etc. This paper is concerned with the developments which have been carried out in the United States, including the establishment of trans- mitting and receiving stations on the New Jersey coast, the equipping of the Leviathan and the establishment of service to that ship. It is significant of the wide-spread interest in this type of service that developments have also gone forward rapidly on the European side where the British, Germans, and French are preparing coastal stations and equipping some of the larger ships for public telephone service. The British have already initiated service to two of the White Star Liners, the Olympic and the Majestic, and before the summer is over it is likely that half a dozen of the larger transatlantic vessels will be undertaking this service, connecting with both the American and the European networks.^ Early Developments Attempts to apply telephony in the maritime field date back to the pioneer work on radio telephony itself, but it was not until the applica- tion of the vacuum tube were developed that radio telephony for any service became finally practicable. Following the long distance, point-to-point radio telephone experi- ments of 1915, there was carried out in the following year what is believed to have been the first trial of two-way radio telephony from the wire telephone system to a vessel at sea. This trial was conducted by Bell System engineers in cooperation with the Navy Department. On that occasion the Secretary of the Navy, in his office in Washington, carried on two-way conversations with the captain of the U. S. S. New Hampshire oflf Hampton Roads. Following the further development of radio telephony during the War, there was undertaken, in the years 1920-1922, an extensive devel- opment of ship-to-shore radio telephony, looking toward the linking of ships at sea with the land line telephone network.- At that time there was built a coastal radio telephone station at Deal Beach, N. J., and several ships were equipped on a trial basis. Extensive engineering tests were made and a number of demonstrations carried out which proved the physical feasibility of establishing such connections. W^hile the trials were successful from the technical standpoint, the development was not carried into commercial use because the adverse economic conditions existing in the post-W^ar period did not appear to ^ Ship-to-shore telephone service is now given (July, 1930) from both U. S. and British shores to'the Leviathan, Olympic, Majestic and Homeric. -"Radio Extension"" of the Telephone System to Ships at Sea," by H. W. Nichols and Lloyd Espenschied, /. R. E. Proceedings, Vol. 11, 1923. RADIO TELEPHONE SERVICE TO SHIPS AT SEA 409 justify the initiation of the new service at that time. Furthermore, the waves in the range of 300-500 meters, which had been used in these early trials, were soon thereafter assigned for broadcasting. In the last few years the whole outlook has changed considerably. The development of short-wave radio systems has greatly increased the message carrying capacity of the radio spectrum and has made it feasi- ble to maintain communication over a greater range of distances than was previously practicable for ships. Transoceanic radio telephone services have been inaugurated, and with the large increase in steam- ship travel there has arisen a renewed interest in the extension of telephone service to ships at sea. When it became evident that short-wave transmission might be CC Ul ^- UJ a. ui a. If) b o > o a. o o z u a: y- in a ui o < 10,000 1,000 100 10 ^ = ^ ^ ^ ^ ^ ^ s s s B S Nf = = ^ — / 1 TRANSMISSION FROM 2XG TO FT. ST. GEORGE ENROUTE TO BERMUDA ISLES 4 OCT. 21-24, 1925 _| 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 - 11 =: ^ :^ = — . — — — — — — — — — — zc *— J i / \ "N \y / y \ \ t' A ^- — — — — 3 — ■— — =^ — — -*- E — :^ =:: / 1 / \ / / ^ ' ' I ^ = _ 1 ^ — — — ' 3 EE E 1- HA Kill 1 T HM ^ — — — (n ~ — — — — — ' — — 1 — — ^ HARBOR f- — — — ^ 1 ; ^^ ^^ ^^ /I ^^ 1 1 1 20 Ml 250 Ml. ^ 8f ),r ^1. 680 Ml. 7« >o| Ml 76 5N !!l 765MI, 1 1 7'p 5N .11. 12 M 12 MID M 12 M 12 MID N 12 M 12 MID M Fig. 1 — Received Fields, New York-Bermuda Run 1925. desirable for ship-to-shore telephone service, there was undertaken a program involving the measurement of the strength of the electric fields received aboard ship from a shore transmitter. This work was part of a general program intended to obtain fundamental data upon short-wave transmission, for purposes of point-to-point, as well as for ship-to-shore telephone services. The tests were first made in 1925 on vessels running between New York and Bermuda. Further measure- ments were made on other ships in subsequent years. Fig. 1 is an example of the result of these earlier tests. Transmission was from Deal Beach on 4.5 megacycles (66 meters). The curve shows the relatively weak field which was received as the vessel left dock, due to the considerable stretch of land which intervened in the transmission 410 BELL SYSTEM TECHNICAL JOURNAL path, the rise of the field to high values as the ship passed out of the harbor, and the gradual diminution as the vessel continued on her course. It will he oljserved that transmission on this frequency was effective at night all the way to Bermuda, but that during the daytime the transmission failed for distances greater than a few hundred miles. Corresponding measurements showed that daylight transmission could be secured by means of a higher frequency, such as 9 megacycles (v33 meters). Measurements of this kind, supplemented by data obtained for a wide range of distances over land, and for transatlantic distances, have built up a fairly complete set of quantitative data on short-wave transmission over different distances and for various times of the day and year. Along with this study of transmission conditions, there was carried on the development of short-wave apparatus technicjue for telephony. The first application was in the field of point-to-point transatlantic operation and the considerable art built up there, including the design of transmitters, receivers, directive antennas, and the working out of two-way operating methods, served as a very useful basis from which to develop the coastal and ship stations for the maritime system. With this background of development, preparations were made to set up a two-way, short-wave radio telephone system for commercial service. This service was centered upon New York because of the large concentration of ocean-going trafftc at that port. The Technical Problem One of the most important problems to be solved in the design of a short-wave system is that of determining the frequencies necessary for giving the service involved. The frequencies which are best suited to the different distances, time of day, and season of the year for trans- mission over the North Atlantic are indicated in the curves of Fig. 2. The curves for the greater distances refer to the transmission which appears to take place in the upper regions of the earth's atmosphere and is usually referred to as sky-wave transmission. Each of the sky- wave curves traces the optimum frequency-distance relation for the time of day and season of the year indicated. The curves merely give a general picture of the frequency relation and do not take account of other effects, such as fading, magnetic storms, etc. The figure brings out very clearly the necessity for using a variety of wavelengths if the ship lanes are to be adequately covered. Fortu- nately, there is a considerable band on each side of the curves shown, in which good transmission can be obtained, and this enables one to choose a small number of frequencies in the short-wave range which are ade- RADIO TELEPHONE SERVICE TO SHIPS AT SEA 411 quate to cover the conditions. Actually, it is found that a set of about four frequencies will suffice to cover the North Atlantic. For distances greater than a few hundred miles this characteristic obtains irrespective of whether the transmission is over water or over land, by reason of the fact that the transmission appears to take place in the upper regions of the earth's atmosphere. Closer in to the transmitting station, however, there is the so-called surface component, the attenuation of which is much less over sea water than over land. It will be seen that the surface wave may be relied upon for distances of the order of 200-300 miles, for frequencies of about 4 megacycles. The transmission of this component is much more stable and reliable than is the transmission of the sky wave. It O < U' Z 3 I >-l 0 O z UJ D 8 O OA 5^ I , ""^ ' ^ ■ 0 'r .^■ r' 2' / y SllM /E.p;_ , ■ / / N(l^ ,/ / ^ ( \ X , N GHT -WIN TER '^^ < '-i C £000 DISTANCE. - STATUTE UILE.S Fig. 2. — Distance-frequency characteristic. seemed important, therefore, to utilize the surface wave to the maxi- mum extent possible. W'ith this in mind, a series of transmission measurements was made over a stretch of water between New Jersey, Long Island, and Nan- tucket for the purpose of more accurately evaluating the effectiveness of the surface wave component, particularly in so far as it bears upon the question of how close to the water front the coastal station need be placed. Transportable transmitting and receiving stations were used in these tests. It was found that as the transmitting or the receiving station was moved away from the water front, the attenuation in- creased materially. For example, moving either terminal a mile back from the coast line increases the attenuation some 8 decibels at 4.5 megacycles. On the other hand, a narrow stretch of land, such as a sand bar, out a few miles from the coast, introduces relatively little loss. 412 BELL SYSTEM TECHNICAL JOURNAL These results indicate that if full advantage is to be obtained from the more reliable surface-wave component, the coastal station should be immediately upon the seacoast or a salt-water bay. An important factor in connection with radio reception on ship- board is that of electrical interference. The modern steamship re- quires for its operation and its service a large amount of electrical machinery. In addition to this, it is equipped with various radio telegraphic services. The operation of all of this electrical equipment produces interference in a receiver which is much in excess of that normally encountered in a shore receiving station which can be so located as to be reasonably free from electrical disturbances. Further- more, there is on the ship another source of disturbance which is due to Fig. 3 — U. S. coastal station, circuit between New York and ship. charging and discharging of various parts of the rigging in the strong electromagnetic fields of the various radio transmitters. These various sources of disturbance were found in the earlier shipboard experiments and the high noise levels are, in general, the predominant factor in limiting the communication range. These factors made it desirable to employ at the shore end as powerful a transmitter as was available and to use whatever benefit could be obtained from antennas designed to be roughly directive along the transatlantic ship lanes. A trans- mitting set of the type used in transatlantic communication, but adapted for the ship-to-shore wavelengths, was therefore employed. Since the shore receiver can be located in a comparativ^ely quiet situation and since use can also be made of roughly directive receiving antennas, there is no advantage in transmitting as large an amount of RADIO TELEPHONE SERVICE TO SHIPS AT SEA 413 0 0 0 0 0 1 o m 0 0 0 vD ■4^ t rO rO >^ ^^--^^ \ \ ^"^^'^^'^ \ 0 SC 2 -1 xQ \ >v ^ \ 0 in O In 8^ -\ ^_,^---^ \ ^^^-\^'^^ 0 mi / U.' >- A o< JX-- \ z^j^lY 0 J.^^^V^^ (D / ' \ " -\ \ O V^ Ae ^^^T; BAY" BISC ^ \ _l \j \ :^^=^^ 0 0 0 ^^>=^s^^^ — V \ \ ^^li---" \ ■ in '^.c^" \\ ^ ^^X^r" \ 0 in 0 ^ — ^ — \ " "'-. *\ — \ o — - — *""' v\ \ ^_— -Pt- '0' \ 0 \__- — -" '^ 1 0 0 \ m — (M 0 \ _____ \j\: O -4 \ \ w \ 1 ^ 0 0 U^ in C\J Si ' 1ST 5T 8. ST 1ST \\^\ ^-watt screen-grid tube which can be used either as an amplifier or a frequency doubler. The output of this tube, except in the case of the higher frequencies, is applied directly to the grid of a 50-watt screen-grid amplifier. For the higher frequencies a second frequency doubler can be switched into the circuit. The output of the 50-watt tube is coupled through a balanced trans- former to the final amplifier stage. The amplifier or frequency doubler RADIO TELEPHONE SERVICE TO SHIPS AT SEA 421 stages are separately shielded and radio-frequency filters are provided in all power supply leads. The power amplifier consists of an air-cooled, three-element, one- kilowatt tube. Neutralization is accomplished by the familiar balanc- ing arrangement shown in the figure. The output circuit of this stage consists of a parallel resonant circuit with provision for tapping in the connection to the antenna. Fig. 12— Leviathan transmitter. Modulation takes place in the plate circuit of the final amplifier stage, the plate current supply being fed through a special transformer, the secondary of which is connected to two 250-watt tubes connected push- pull and fed by a 50-watt amplifier. The power supply is obtained from motor-generator sets operated from the 110-volt, d-c. ship supply. Protection of the operators and 422 BELL SYSTEM TECHNICAL JOURNAL apparatus is provided by means of relays and contactors in the high- voltage supply circuits which prevent the high voltages from being applied if the filament or grid circuits are not closed or if the doors of the transmitter are open. An illustration of the ship's transmitter is shown in Fig. 12. The picture is somewhat out of perspective owing to difficulty in taking the photograph in the limited space available on shipboard. The receiving problem on shipboard is complicated by a number of factors. The transmitting and receiving frequencies must be within a few per cent of each other, if the best transmission conditions for the time and place are to be utilized and if the frequencies are to remain in the bands assigned internationally to the mobile services. This re- quirement, as well as the noivSe conditions on shipboard, calls for a receiver of high selectivity, which is obtained, in the present instance, by the use of a double-detection set. The over-all selectivity is accom- plished both by having a number of highly selective circuits ahead of the first detector and by using tuned circuits in the intermediate fre- quency stages, the high-frequency selectivity being used primarily to prevent overloading of the first tube and the intermediate frequency circuits being used to obtain the final selectivity required. A reduction of the disturbances due to stay noises and better dis- crimination against the transmitted carrier is obtained if the trans- mitting and receiving antennas are widely spaced. On the other hand, for operating reasons, it is desirable to have the transmitter and receiver located in the same room. In the case of the Leviathan installation, the transmitting antenna is located directly above the radio room, between the second and third stacks, and the receiving antenna is placed as far as possible behind the third stack. The receiving antenna is connected through a suitable step-down circuit to a shielded trans- mission line, the other end of which is connected to the receiver, the receiver itself being very thoroughly shielded to avoid direct interfer- ence from the transmitter. On account of limited space, only two antennas are provided to handle the four frequencies, each antenna representing a compromise between the most efficient antennas which could be put up to handle the separate wavelengths. As stated above, the receiver itself is of the double-detection type, using heater type tubes throughout. Screen-grid tubes are used for the first detector and intermediate frequency amplifiers and three- element tubes in the remaining positions. A photograph of the re- ceiver and associated voice-frecjuency equipment, as it is installed on the Leviathan, is shown in Fig. 13, and a diagrammatic representation of the receiver is shown in Fig. 14. The high-frequency selective sys- RADIO TELEPHONE SERVICE TO SHIPS AT SEA 423 tern consists of four separately shielded tuned circuits, coupled by small capacities. The use of a screen-grid tube in the detector circuit gives a two fold advantage over the use of a three-element tube in that a higher input impedance is maintained at the higher frequencies and the necessity for neutralizing against the reaction of the beating oscilla- tor on the input circuit is eliminated. The beating voltage is made of the order of 75 to 100 volts for the purpose of reducing the effective tube noise in the detector plate circuit. With this arrangement no d-c. plate voltage is ordinarily required. The screen voltage is 22 Fig. 13 — Leviathan receiver. volts. The output circuit is tuned to the intermediate frequency of 300,000 cycles and connection with the first intermediate amplifier is effected by means of a low impedance transmission line. The inter- mediate frequency amplifier stages are coupled by means of doubled tuned circuits. The use of properly designed circuits of this type makes it possible to obtain a high degree of selectivity against unde- sired frequencies while obtaining sufficient band width to maintain ease of tuning and to pass the desired frequencies. The second de- tector is of the conventional grid bias type. Automatic gain control 28 424 BELL SYSTEM TECHNICAL JOURNAL is provided in which a certain amount of the carrier is taken at the end of the intermediate frequency stages, ampHfied and rectified. The resulting d-c. current produces a voltage drop across a resistance, which is applied to thegridof the first detector in such a manner that an increase in the intermediate frequency output brings about a reduction in the total set gain and vice versa. Manual gain control for following wide changes in the received fields is accomplished by variation of the voltages applied to the grid and the screen of the first detector. The voice-frequency equipment, in addition to the desk telephone H F FILTCR i^'' OET. ANO BEATING OSC. Fig. 14 — Ship receiver schematic diagram. set located in the subscriber's booth, comprises a technical operator's position located adjacent to the ship's receiver, and an attendant's desk located on a lower deck in a room adjacent to the subscriber's booth. The control equipment consists of repeaters, volume control devices, and volume indicators, by means of which the levels of the incoming and outgoing signals can be properly adjusted. Keys are provided which enable the technical operator to talk either over the radio circuit or to the ship subscriber. The booth attendant has facilities by which he can talk either to the subscriber or to the control operator and has a connection with the ship's telephone system for the purpose of locating persons on the ship and calling them to the radio telephone booth. RADIO TELEPHONE SERVICE TO SHIPS AT SEA 425 The subscriber's booth is provided with a desk telephone set having a high-grade transmitter. The outgoing and incoming circuits are shielded from each other and brought separately to the transmitter and receiver of the subscriber's set. An illustration of the subscriber's booth on the Leviathan is shown in Fig. 15. Fig. 15 — Subscriber's booth on Leviathan. The Wavelength Situation and Simultaneous Telephone AND Telegraph Operation Communication between ship and shore is carried out by the use of a pair of frequencies, one for transmission in each of the two directions, separated from each other by about 3 per cent. The specific frequen- cies which were first assigned by the Federal Radio Commission to the shore station and the Leviathan were necessarily chosen on more or less 426 BELL SYSTEM TECHNICAL JOURNAL of a makeshift basis, in the absence of any comprehensive wavelength plan for this new service. The Commission has recently had under study the setting up of more adequate provisions for ship-to-shore telephone channels, whereby it is hoped a series of frequencies may be designated for telephone service exclusively and whereby there may be established the relation between the telephone and the telegraph frequencies necessary for the avoidance of interference between the two services. Especially is coordination of the two sets of frequencies necessary on the larger vessels, in order that simultaneous telegraph and telephone service may be given without mutual interference. On the larger liners simultaneous use of the radio telephone and radio telegraph service must be provided for. This means that the trans- mitters of both services must keep accurately on their frequencies and be free of spurious components, and that the receivers must be highly selective. It further entails that the transmitting and receiving frequencies in each of the two cases be so coordinated that the trans- mission frequency of one service does not lie too near the receiving frequency of the other, and bespeaks a considerable amount of mutual cooperation between the operating agencies involved. Difficulties of fitting in the two services were encountered in the early work on the Leviathan and, although the problem has not been worked out to final solution, sufficient progress has been made, in cooperation with the engineers of the Radio Corporation of America, to enable the telegraph and telephone services to be conducted simultaneously without undue interference. In view of the fact that ships of a number of nations are already pre- paring to give radio telephone service on the transatlantic routes and with the probability of this service also extending to other parts of the world, it would appear to be a matter of importance that the whole question of marine frequency allocations be worked out in the near future not merely on a national but also on an international basis. Transmission Results The transmission results which have been obtained with the Levia- than on her first trip of commercial service are summarized in Fig. 16. It will be noted that practically continuous 24-hour communication was maintained for distances within 1000 miles of the shore, correspond- ing to two days out. The service at greater distances was more inter- mittent. This was largely due to the fact that during this first trip the effort was concentrated on covering reliably the more important nearer- in distances, and the ship was not prepared to transmit on frequen- cies above 8 megacycles. The service proved to be much in demand RADIO TELEPHONE SERVICE TO SHIPS AT SEA 427 EASTBOUND DEC. 1929 TIME OF DAY — EST AM NOON PM 6 8 10 12 2 4 6 8 10 NO, OF tALLS -r "T" n r NOON DIST. AMBROSE SUN. N.Y.-SHIP 8 SHIP- NY LEFT N.Y 12 PM ^^r^ 10 MON. NY -SHIP 9 SHIP- NY 13 TUE. N.Y-SHIP 10 SHIP-NY. I 8 _ iHIGH NOISE 13 HEAVY CRACKLES' WED. NY-SHIP II SHIP-N.Y FIGURES bENOTE APPROXIMATE FREQUENCY IN MEGACYCLES NOISE NOISE THUR. N.Y-SHIP 12 SHIP-NY LOW FIELDS, NOISE ?I?i?L FRI. 13 LOW FIELDS, NOISE --^-r"^— 8 ' 4 ARRIVED CHERBOURG 170 7 20 1270 1820 2390 WESTBOUND TUE. 17 WED. N.Y-SHIP 18 SHIP-N.Y THUR, N.Y-SHIP 19 SHIP-N.Y FRI, N.Y-SHIP 20 SHIP- NY SAT N.Y-SHIP 21 SHIP-N.Y SUN. N.Y-SHIP 22 SHIP-NY MON. N.Y-SHIP 2 3 SHIP-NY TUE. N.Y-SHIP 24 SHIP-NY. TOTAL ENTIRE N>: -SHIP TRIP SHIP-NY LEFT CHERBOURG ® LOW FIELDS, ship's MOTOR NOISE --5--I-?. — LOW FIELDS — T--r— - 4 ' 8 CRACKLES^"^®^-- HIGH NOISE-- QRM ship's SET i^ LOW FIELDS 8 ' 4 l4iai 4 NOISE 8 W 8 -„„^...„. ^^, QRM0N4MCn QRM SHIPS SET 13 I 4 iSiM i8i4 .CRACKLES LOW-FllilDSa CRACKLES ^ '« ' ^ 13 xl*_A_l. QRM SHIPS SET^ ± i L 8 |4\ 8_ ARRIVED AMBROSE LIGHT _L J_ a_ _L _L _l_ 246 8 10 12 2 46 8 AM NOON PM TIME OF DAY —EST 10 18 28 32 46 151 2640 2100 1530 1110 680 195 DOCK Fig. 16 — Transmission results between 5. 5. Leviathan and Xew York. 428 BELL SYSTEM TECHNICAL JOURNAL by the passengers, as is indicated by the number of calls completed each day, particularly on the return trip. A similar number of test and demonstration calls was made during the voyage. The calls were completed without undue delay, there being only one ship in- volved, and a fairly high grade of communication was obtained. In conclusion it will be realized that the solution of the techincal problem of ship-to-shore telephony is now well in hand and has been carried to the point of having proved the practicability of giving this service. Further problems are naturally arising in carrying the devel- opment into more general effect, particularly operating problems and those concerned with the international coordination of the service. The indications are that the larger transoceanic ships will be rather generally equipped for telephony and that the service will become one of permanent value in the maritime field. A General Switching Plan for Telephone Toll Service By H. S. OSBORNE * This paper outlines a comprehensive plan for improved switching of long haul toll telephone traffic in the United States and Eastern Canada. A brief discussion is given of the methods of designing the toll plant to give ade- quate transmission efficiency for all connections established in accordance with this plan. This includes a new method of providing amplification at intermediate switching points replacing the cord circuit repeatei method. ON January 25, 1915, telephone service was, with due ceremony, inaugurated between the Atlantic and Pacific Coasts of this country. This occasion marked a great step forward both technically and commercially. Before that time, the limit of practicable telephone transmission had been about 1,500 miles. The transcontinental service was made possible by the completion of numerous important developments and particularly by the perfection of telephone repeaters and of means for applying them to long wire circuits. Until then the Pacific States and their neighboring states had been isolated telephonically from the eastern and midwestern parts of the country. The demonstration of commercially practicable telephone circuits across the continent gave a great impetus to the idea of universal service, that is the provision of a telephone plant such that telephone service could be given at commercially attractive rates between any two telephones in the country. In the fifteen years since the opening of the first transcontinental circuits, the ideal of universal service has to a large extent been realized. Practically all the telephones of the United States and a large part of Canada now have provision for connection with the countrywide toll telephone network, more than 99 per cent being included. To achieve universal service, however, involves a great deal more. Circuits must be provided in such numbers and so arranged that connections between any two telephones can be estab- lished quickly and without too many intermediate switching points. Also the telephone plant must be designed for such standards of transmission that these connections, when established, permit satis- factory conversation. In general, the technical advances which have been made during the last fifteen years to achieve the present standards of toll service have been described from time to time before the American Institute of Electrical Engineers, and it is not within the scope of this paper to review them. * Presented at Convention A. I. E. E., Toronto, June 1930. 429 430 BELL SYSTEM TECHNICAL JOURNAL Associated Avith this development of the telephone plant has been a very rapid increase in trat^c. Fig. 1 indicates this increase in the United States and Canada since 1915. A striking characteristic of this growth is that the increase has been much more rapid for the longest lengths of haul than for the shorter lengths of haul. For example, during the last five years in which the messages on lengths of haul up to 250 miles approximately doubled, the messages on hauls from 250 to 1,000 miles increased five times and those over 1,000 miles increased more than ten times. This characteristic is also 1 1000 800 600 400 200 1915 1920 1925 1930 ( EST.) Fig. 1 — Total toll messages in millions per year — Bell system. illustrated in Figs. 2, 3, and 4 which show respectively the growth in the number of circuits between Toronto and Detroit 240 miles in length, Buffalo and Chicago 550 miles in length, and direct circuits from New York and Chicago to the Pacific Coast, averaging about 2,500 miles in length. This particularly rapid growth in very long haul trafftc has made it practicable to establish a considerable number of long haul circuit groups and has greatly assisted in the problem of handling satisfactorily calls between widely separated points. It has led to the condition today in which 74 per cent of the long distance (Long Lines) messages are handled over direct circuits and 20 per cent with one intermediate switch. SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 431 The part of the business on which it is most difficult to give a high grade of service is naturally the scattering business between widely separated points. In these cases each item of traffic, that is the business between two specific points, is relatively small but the number of items of traffic is great. The number of messages involved in each item of traffic does not justify direct circuits and in very large LI 15 10 1915 1920 1925 1930 (EST.) Fig. 2 — Growtli in number of toll circuits — Toronto to Detroit. numbers of cases it is necessary, in order to provide a connection, to make more than one intermediate switch. This applies at present to six per cent of the long distance telephone business of the country. All measures of the quality of service — speed, accuracy and trans- mission— show that the difficulty of satisfactorily handling the service increases rapidly with the number of intermediate switches involved. The development of the toll business has led to a great increase in the amount of business between large numbers of widely separated points. There has also been an extensive trend toward concentration of the plant used in handling the business in important toll offices and along important routes. The long haul toll business is now handled at about 2,500 "toll centers" out of approximately 6,400 432 BELL SYSTEM TECHNICAL JOURNAL central offices in the United States and eastern Canada. Furthermore, the technical developments in toll circuits have led to great increases in the numbers of circuits along a given route. The extension of the use of carrier telephone has increased the capacity of a 40-wire pole line from 30 circuits to 70 circuits. On the heaviest toll routes, moreover, circuits are now provided by means of toll cable construction, a single cable carrying 200 or 300 circuits. During the past year 14 12 10 1915 1920 1925 1930 (EST.) Fig. 3 — Growth in number of toll circuits — -Buffalo to Chicago. or two the growth has been so rapid as to stimulate a very large amount of construction of underground toll conduit routes, providing in many cases for several thousands of telephone circuits on a single route. General Toll Switching Plan The conditions outlined above form the background which has made it both possible and important to adopt a new fundamental arrange- ment for the layout of toll plant and the routing of toll messages. This is called the "General Toll Switching Plan." The purpose of this plan is to provide systematically a basic plant layout designed for the highest practicable standards of service consistent with economy. SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 433 including speed, accuracy and directness of routing between any two points in the country and suitable transmission standards. This involves the layout of the plant in such a manner as to limit as much as practicable the number of switches required for providing a con- nection between any two telephones and the establishment of standards of design and construction providing satisfactory transmission over any route thus established. The plan is, therefore, of particular 50 40 30 20 10 1915 1920 1925 1930 (EST.) Fig. 4 — Growth in number of toll circuits — New York and Chicago to San Francisco, Los Angeles and Seattle. value in improving the service conditions of switched toll traffic, that is, traffic requiring the connection of two or more toll circuits. The general features of the plan will be understood by reference to Figs. 5 and 6. Figure 5 shows the application of the plan to a limited operating area such, for example, as a State. Within the area are selected a small number of important switching points designated as "primary outlets." Each toll center is connected directly to at least one of these outlets and all primary outlets within the area are directly interconnected. This makes possible the interconnection of any two toll centers within the area with a maximum of two switches and within the part of the area served by one primary outlet, with a maximum of one intermediate switch. 434 BELL SYSTEM TECHNICAL JOURNAL The primary outlets were selected after a careful study of the present switching and operating conditions and the probable develop- ment of toll traffic within the various areas with a view to obtaining the minimum number of primary outlets capable of handling the traffic economically. The routings provided by the plan are supple- mented by direct circuits, or by other routings where the amount of business justifies such additional circuits as indicated by the dashed lines in Fig. 5. In general the requirement is made that these supple- mentarv routes shall be at least as satisfactory, both as regards SOLID LINES - FUNDAMENTAL ROUTES OF GENERAL PLAN DASHED LINES - SUPPLEMENTARY DIRECT CIRCUIT GROUPS O PRIMARY OUTLET • TOLL CENTER Fig. 5 — General toll switching plan — application in local company area. number of switches and transmission, as the routes provided by the fundamental switching plan. However, when the supplementary routes are used only as alternates to a primary routing they may be somewhat less satisfactory in these respects. The tentative selection of primary outlets is shown in Fig. 7. It is interesting to note that it is found practicable to take care of switching for the 2,500 toll centers of the United States and eastern Canada by the establishment of approximately 150 of these as primary outlets. For handling the business throughout the country eight of the primary outlets are designated as regional centers, which are indicated SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 435 in Fig. 7. The method of routing calls is indicated by Fig. 6. Each primary outlet is connected with at least one regional center and with as many more as practicable. Each regional center is directly con- nected to every other regional center in the country. By this means, any one of the primary outlets, which are the 150 most important switching centers in the country, can be connected to any other primary outlet in the country with a maximum of two switches and within the area served by a regional center with a maximum of one intermediate switch. As an illustration of the concentration of switching which results, New York serves as regional center for the entire northeastern section of the United States and eastern Canada. SOLID LINES - FUNDAMENTAL ROUTES OF # REGIONAL CENTER GENERAL PLAN Q PRIMARY OUTLET DASHED LINES- SUPPLEMENTARY DIRECT • TOLL CENTER CIRCUIT GROUPS Fig. 6 — General toll switching plan — illustration of interconnection of important switching offices throughout Bell system. The extent to which intermediate switching is limited by the application of this plan is indicated by Fig. 8, which shows the maxi- mum number of switches recjuired under the plan between different types of toll centers. It is estimated that the percentage of long haul messages requiring more than one intermediate switch will, by means of this plan, be reduced by more than 50 per cent. As an example of the benefit resulting from the adoption of this plan between two remote points, consider a connection which was requested between Pembroke, Ontario and St. Anthony, Idaho. Under the old routing instructions such a call required intermediate 436 BELL SYSTEM TECHNICAL JOURNAL 'c 3 O V--'- 1» !.--# 1^ ■|.° !• « ^ o I. i2 ^o 'I* I ^,_ to a: ~3 C H D O c < z o < OJ o C t') o III ft PC a. 0) • • u tn ■a ^""""^ C i rt tn ,' C -1-1 £ 2» ,/'■ / .c/5 o c .9 03 O O nJ .S O ■^ i3 c o be SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 437 switches at Ottawa, Toronto, Chicago, Denver, Salt Lake City, Pocatello and Idaho Falls, a total of seven. The chance of establishing such a connection within satisfactory limits of time was, of course, relatively small and the resulting circuit, when established, did not permit the conversation to be held. Under the general toll switching plan, this call will be routed with switches at Ottawa, New York, Denver and Pocatello, a reduction of three switches. Furthermore, the circuits involved in this connection will be designed with such transmission standards as to give satisfactory conversation. To— From Same Regional Area Another Regional Area Re- gional Center Pri- mary Outlet Toll Center Di- rectly Con- nected to Re- gional Center Toll Center Di- rectly Con- nected to Pri- mary Outlet Re- gional Center Pri- mary Outlet Toll Center Di- rectly Con- nected to Re- gional Center Toll Center Di- rectly Con- nected to Pri- mary Outlet Regional Center Primary Outlet Toll Center (directly connected to Regional Center) 0 0 1 0 1 1 2 0 1 1 2 1 2 2 3 0 1 1 2 1 2 2 3 1 2 2 3 2 3 3 4 Toll Center (directly connected to Primary Outlet) Fig. 8 — Maximum number of switches under general toll switching plan. The routes provided by the plan for countrywide service are also supplemented by more direct routes of equivalent or better service characteristics in cases where the amount of business is sufficient to make this economical. Furthermore, the routes to regional centers are, in some cases, supplemented by alternate routes through what are called "secondary outlets." These are distinguished from the primary outlets in that they do not necessarily have direct circuit connections to all toll centers in their areas but serve a useful purpose as an alternate route for the toll centers connected to them. The essential features of the general toll switching plan from the standpoint of the interconnection of the switching offices may be summarized as follows: Regional Centers Regional centers are switching offices strategically located to cover the various parts of the country and completely interconnected with direct circuits, thus forming the basis of a countrywide toll network. 438 BELL SYSTliM TKCIINICAL JOURNAL Primary OtUlets Primary outlets are switching offices having direct circuits to one or more regional centers and each having direct circuits to all toll centers in the area for which it is the primary outlet. Also, each primary outlet is connected to every other outlet within as large an area as practicable, usually within a State. Supplementary Offices Secondary Outlets Secondary outlets are switching offices having direct circuits to one or more regional centers and are intended primarily to furnish alternate routes for toll centers for reaching the regional centers, thus providing a greater degree of flexibility in the plant. Secondary Switching Points Secondary switching points are additional switching offices intended to provide routes which are more direct thus reducing back haul for intra-area business. Transmission Considerations of General Toll Switching Plan An important part of the development of the plan was the determi- nation of proper transmission requirements such that any toll con- nection established in accordance with the plan would have satisfactory transmission efficiency. Before the perfection of telephone repeaters, the provision of satis- factory transmission efficiency depended largely upon limiting the total attenuation loss of the complete circuit. At the present time the perfection of repeaters has practically removed that limitation. For example, the attenuation in a New York-Chicago circuit in cable is such that without the use of repeaters the ratio of input power to output power for speech currents transmitted over the circuit would be 10^^, while by the use of repeaters at the terminals and at 17 inter- mediate points the ratio actually is 10. The removal of the limitation formerly set by circuit attenuation makes possible the increase of the efficiency of circuits to the limit determined by some other characteristic of the circuit. There are various things which under different conditions may determine this limit. One is the effect on transmission of echoes, namely, portions of the speech currents reflected back from the distant end of the circuit or from intermediate points. Another is the distortion due to the building up of greater transmission gain at certain frequencies than at others, which effect may result if repeaters introduce too SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 439 great an amplification into the circuit. As an extreme case, this might result in a sustained oscillation or singing on the circuit. Other effects which may be important are those of crosstalk between tele- phone circuits, or of noise induced in the telephone circuits from outside sources, both of which are increased by increasing repeater gains. On the longer connections, echoes are almost always the controlling factor, whereas on the shorter connections, such effects as crosstalk, singing and noise generally are limiting. A reduction in any of these effects generally involves more expensive types of con- struction. The difference between the attenuation loss of the circuit and the total transmission gain introduced into the circuit by repeaters is spoken of as the net equivalent. For long telephone circuits it is generally economical to provide sufficient repeater gain so that the circuit can be operated at the minimum net equivalent permissible, this minimum equivalent being determined by the transmission factors just mentioned. Therefore, in establishing satisfactory trans- mission efficiencies for the overall toll connections in accordance with the toll switching plan, each link must be designed on the basis of the minimum working net equivalent which it will contribute to an overall connection made up of several circuits switched together. The establishment of satisfactory and economical transmission requirements for the toll circuits laid out in accordance with the plan involves the following steps: a. The establishment of satisfactory overall net transmission equiva- lents. h. The coordinated design of all classes of toll circuits, and of the subscribers' circuits, toll switching trunks and tributary trunks connected to them, in such a way that the desired overall transmission standards will be given at a minimum total cost when suitable transmission gains are provided by repeaters in the toll circuits and at toll switching points. c. The economical and satisfactory distribution of transmission gain, permitting all toll circuits to be operated at their minimum net equivalents when this is desirable. The overall transmission equivalents to be given under the plan are based on standards which have heretofore been used for a large part of the toll business but which it has been impracticable to meet in many cases between widely separated points. With the means now available for operating circuits at their minimum working net equivalents, it was found that satisfactory overall transmission 29 440 BELL SYSTEM TECHNICAL JOURNAL equivalents could be provided under the plan even for the maximum number of switches using standards for the construction of toll circuits very comparable with those already applied to new circuits. Ex- pressed in terms of the transmission reference standard, the plan set up gives a maximum of 25 db overall equivalent within one inter- connected area (two intermediate switches) and a maximum of 31 db between any two telephones of the United States and eastern Canada. In order to determine the most economical distribution of these overall equivalents, a study was made based upon the estimated total number of toll circuits of each class in 1932 and their distribution by length. It is also necessary to include the corresponding estimates for the plant between the toll office and the subscriber, the loss in this part of the plant being on the average about half of the overall net equivalent of the connection. Based upon these estimates, it was possible to determine, by an economic study, the distribution of the overall minimum net equivalent between these various parts of the circuit which would give minimum total expenses. The toll terminal losses and the minimum net equiva- lents for toll circuits established in this way are shown in Fig. 9. Classification of Toll Circuit Involved Minimum Working Net Loss of Toll Circuit — db Maximum Via Operating Equivalent — db Transmission Margin — db Toll Center to Primary Outlet Toll Center to Regional Center Primary Outlet to Regional Center Regional Center to Regional Center Primary Outlet to Primary Outlet Toll Center to Toll Center 3.0 3.5 3.5 4.0 4.0 6.0 9.0 7.0 4.0 4.0 3.0 3.0 3.0 6.0 + 1.0 + 0.5 - 0.5 - 1.0* ■ - 1.0 Direct Toll Circuit (for terminal use only) Toll Terminal Loss * Circuits equipped with echo suppressors may be designed with greater negative margins. Fig. 9 — Transmission design data of general toll switching plan. In addition to the circuits involved in multi-switch business, the studies connected with this plan necessarily include circuits used for terminal business only, and others for which switching is limited to a single intermediate switch at points where transmission gain is not required. These circuits are associated with the plan because the portions of the circuit between the toll center and the subscriber are common for these circuits and for circuits directly involved in the general plan. Design standards for these classes of circuits are also shown in I'ig. 9. SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 441 Provision of Transmission Gain at Intermediate Switching Points The third step mentioned previously is the determination of the best distribution of repeater gains to permit the individual circuit to be operated by itself or in conjunction with other toll circuits at approximately the minimum net equivalent as determined by the several effects mentioned previously. In so far as the gain of repeaters permanently inserted at intermediate points in a toll circuit is con- cerned, this is a matter of economical design of the circuit and has been adequately covered in other papers. We are interested here, however, in considering the provision of gain at the intermediate switching points when two toll circuits are connected together. As indicated previously, echo effects are usually controlling on the longer connections, whereas crosstalk, singing and noise will usually control on the shorter connections. This is due to the fact that for the great majority of toll circuits the echo effects on individual circuits increase more rapidly with length than do crosstalk and noise. Singing tendencies also increase at a rapid rate with increase in length on two-wire circuits but tend to be independent of length on four-wire cable and carrier telephone circuits which are used to a large extent to provide the circuits between the primary outlets and regional centers and between the regional centers. Furthermore, when two or more toll circuits are connected together, the echo effects of the indi- vidual circuits add together almost directly, whereas the effects of crosstalk, singing and noise increase at a much less rapid rate. The result of these general considerations is that when a toll circuit is switched to another toll circuit, the overall combination can, in general, be operated at a lower net equivalent as determined by echo effects than the sum of the two circuits when operated individually in which case the minimum equivalent is determined by the crosstalk, singing and noise effects. Therefore, it is necessary in the case of connections built up by connecting together a number of toll circuits to introduce repeater gain at the intermediate switching points. If gain were not introduced at intermediate points, it would be neces- sary in order to obtain the same overall results on connections involving more than one toll circuit to design and build a considerably more expensive type of toll circuit plant in which the crosstalk, singing and noise effects would be greatly reduced. In the past, gain was inserted at intermediate switching points by the use of cord circuit repeaters. These familiar devices consisted of telephone repeaters inserted in the cord circuits and associated by means of double plugs with the toll circuits and with individual 442 BELL SYSTEM TECHNICAL JOURNAL balancing networks designed for each toll circuit. By this means intermediate gains of from 4 to 10 db were inserted at the switching points when connection was made between two toll circuits. The use of cord circuit repeaters has been an outstanding element in the provision of improved transmission on switched connections. It has, however, some disadvantages which have increased in impor- tance with the increase in transmission efficiency of circuits and with the rapid development of toll business. The routine for inserting the cord circuit repeaters when needed is necessarily somewhat cumber- some, involving considerable expense for operators' labor and for increased use of the toll circuits by operators. Furthermore, under practical conditions it was found to be not possible to insure that the cord circuit repeaters would always be used when required by the routing instructions. Recent developments in the types of toll circuit have greatly increased the numbers of toll circuits provided with repeaters at their terminals as a part of the most economical design of a circuit. When such repeaters are available, the desired switching gain can be obtained by making use of the gain available in these repeaters. The great increase in the number of terminal repeaters required for other reasons, important reductions in the cost of repeaters and the savings of operators' labor and circuit time have made it practicable to adopt a plan of providing, at certain points, terminal repeaters for every circuit, thus doing away entirely with cord circuit repeaters at these points. With the terminal repeater arrangement, the insertion of transmission gain on switched connections is done automatically by taking out of each circuit on such connections a section of artificial line. This is, of course, the equivalent of increasing the gain of the terminal repeater. Satisfactory transmission results for all connections under the general toll switching plan involve the insertion of repeater gain on all connections switched at important switching points. This will be carried out by the terminal repeater plan just described. The artificial lines or pads which are cut out of the circuit on switched connections have losses of from 1 to 4 db, depending upon the circumstances of each case. This means that when two toll circuits are switched together, from 2 to 8 db is automatically subtracted from the con- nection at each switching point. The arrangement is indicated schematically in Fig. 10. The design of each circuit must, of course, be such that when either end of the circuit is connected to a subscribers' station, the repeater gain at that end will not be greater than that permissible under the terminating condition, but that when two or SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 443 more of such circuits are connected together for a long built-up toll connection, the complete circuit will operate at as nearly as practicable its minimum working net equivalent. While under these conditions the permissible values of the pads associated with the terminal re- peaters naturally vary in individual cases, it has been found possible to work out for general use a series of values which should give satis- factory results. These are indicated in Fig. 11. It will be noted that these values are such that a circuit switched at both ends to other toll circuits is operated at either .5 db or 1 db less than its minimum working net equivalent, this deficit being made up by a corresponding margin at the ends of the circuit. For example, by reference to Fig. 11, it will be noted that whereas the design values of the three intermediate links of a five-link connection equate to 11 db, these links will contribute a total loss of only 9 db. On the other F1G.1 3 o-vwv-ixi- SWITCHING OFFICE 3 3 -{XHvw\r-o o-vwHXH SWITCHING OFFICE 3 3 -N-^/w\^-o o^vwHXh 3 MXHww-o FIG.2 o-^wWKl- 3 3 KXhVWV-O CMAA/V rM- 3 3 -IXhVWV-O O-AAftA, rM- 3 -[Xpvwv-o FIG. 3 o- 3 3 -tXl-]V\AAr-0 o-^wv rN- Fig. 10 — Illustration of typical transmission data of terminal repeater-;-switching pad method of operation. Fig. 1^ — -Circuits between switching pad offices in terminal condition. Fig. 2- — Circuits of Fig. 1 interconnected at switching pad offices. Fig. 3 — Connection between non-pad offices switched at pad office. hand, the end links will contribute a total of 8 db, whereas their design values equate to only 6 db. The 2 db marginal deficiency in the intermediate links is compensated for by the 2 db marginal surplus in the end links. When intermediate links are used as end links in built-up connections, the switching pads at the terminating ends restore the necessary positive margins. The design of the very long intermediate circuits, such as some of those connecting two regional centers, requires special consideration and treatment to meet the transmission requirements specified. By making use of a fundamental feature of four-wire circuits equipped with echo suppressors and by employing circuits with the highest velocities of propagation for this purpose, these circuits may be designed in practically all cases to contribute not more than the desired 444 BELL SYSTEM TECHNICAL JOURNAL operating equivalent for an intermediate link. Four-wire circuits equipped with echo suppressors are unique in that at the longer circuit lengths the increase in minimum net equivalent with further increase in length becomes very slight. Two general arrangements for removing the switching pads from and restoring them to the toll line circuits are available depending upon the type of switchboard facilities involved. Either arrangement PO 3 3* is> iVVV^Wp QNGC ■ TC «v SAME AREA DISTANT AREA DISTANT AREA Wr| DISTANT AREA PO INTRA AREA TC (3)[-0 SAME AREA ■"PO TC=TOLL CENTER P0= PRIMARY OUTLET RC= REGIONAL CENTER NGC = NON-GAIN SWITCHING CENTER. ( ) MINIMUM WORKING NET LOSS MAXIMUM TOLL CIRCUIT EQUIVALENT I 7 db O OPERATING VIA EQUIVALENT MAXIMUM OVERALL CONNECTIONS 3ldb [] TRANSMISSION MARGIN ASSUMED LIMITING TOLL TERMINAL LOS5--7db *VALUE or PAD IN TERMINAL LINKS DEPENDENT ON NOISE AND CROSS-TALK CONDITIONS Fig. 11 — Diagrammatic representation of transmission data for handling switclied toll traffic under general toil switcliing plan. requires the modification of both the toll line circuit and the switch- board circuits. One method controls the switching pad by a marginal relay in the sleeve of the toll line circuit. In the other arrangement, the pad is under the control of relays operated by battery supplied from a simple.x bridge in the connecting circuits. With the general toll switching plan the number of places in which switching gain is required is greatly limited, being, as pointed out above, a total of about 150 out of 2,500 toll centers. This number will be somewhat increased by secondary switching points in which it is found economical to insert switching gain in order to save the back-haul involved in following the routing provided by the plan. However, the net result is that under the toll switching plan the number of points at which switching gain is provided will be materially limited, with corresponding economies. SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 445 Programming the Establishment of the General Toll Switching Plan The full application of the general toll switching plan involves a large number of individual rearrangements of plant layout, the establishment of certain new circuit groups and the rerouting of a considerable amount of switched business, the conversion of the switching offices to the terminal repeater arrangement, and the modification of the transmission requirements of certain of the circuits. The date at which these rearrangements will be completed is naturally different for different sections of the country and is determined by the regular program of plant additions and rearrangements to take care of increasing business and of needed service improvement. The existence of a comprehensive plan of this sort insures that the program of rearrangements as carried out will be along the lines of greatest economy and maximum improvement in service. The present plans of the telephone companies in the United States and Canada indicate that the plan as now established will be very closely approximated by the actual plant in the course of about five years. Future View Such a plan as has just been discussed is naturally not a static thing but is subject to continual modification to bring it into corre- spondence with changed conditions. In connection with such changes it is of interest to consider briefly the probable long time trend of the development of the plan. One possible ultimate development would be the increasing con- nection of primary outlets to a single regional center so that ultimately only one regional center would be necessary. If this were to take place, the regional center would undoubtedly be Chicago. Fig. 12 is interesting as showing the extent to which the primary outlets already are connected directly with Chicago, over one half of them having such direct connection. If Chicago ultimately became the only regional center, it would reduce the maximum number of switches to three. It seems evident, however, that such a plan would have many disadvantages. It seems clear that with such an arrangement, numerous secondary regional centers would be necessary to avoid uneconomical back-haul of large amounts of traffic, and the economies of such an arrangement do not look promising. Furthermore, it would lead to a tremendous con- gestion of through switching at one point, this congestion going far beyond the limits of economical concentration and leading to serious operating difficulties. 446 BELL SYSTEM TECHNICAL JOURNAL > J3 ■^s 71 ^1 c 0) ujq: o cca. >— I u a 3 O u be 3 O o bo c d O 'So c 3 O 3 bo In (J '% CO 03 i-i 0) c O be ■ «-« SWITCHING PLAN FOR TELEPHONE TOLL SERVICE 447 A second, and it is believed more promising general trend would result from the gradual increase in the number of regional centers as the continued development of business makes this economical. With this growth would come also a continued increase in the number of toll centers connected directly to a regional center. By this process there would be a continued growth in the number of toll centers which can be interconnected with a maximum of two intermediate switches, and it is possible that ultimately the primary outlets can drop out of the picture completely, giving a maximum of two inter- mediate switches for the entire country. While any such outcome is evidently many years away, it seems probable that it is along these lines the growth in development of the plan should be directed. Although this direction of development avoids the congestion which would be brought about by the single regional center plan, even under this plan the rapidly growing amount of toll switching to be done in large metropolitan centers offers a very important problem for the future. Toll switching at these points is rapidly outgrowing the capacity of a single manual switchboard, as the switching of local calls did long ago. Equipment changes are being made which increase this capacity, but they can be but a temporary relief. Looking to the future, an increasing amount of the outgoing traffic will be handled by operators in the local central offices, reaching the toll line over toll tandem trunks. It is evident, however, that the ultimate solution of the problem will involve the use of machine methods for the selection of the toll line by the operators, as is now done in certain segregated toll tandem systems. The entire trend of recent years is thus to decrease the differences between the handling of exchange messages and of toll messages. At the present time more than 95 per cent of the toll messages are completed while the subscriber remains at the telephone, with speeds of completion only slightly slower than those of exchange messages. Transmission standards, while naturally somewhat better for the shorter distances involved in exchange messages, are, nevertheless, rapidly becoming very comparable. The present view of trends for the future is for continuation of this process, perhaps even to the use of similar types of machine equipment at central offices for switching the various classes of messages. The author gratefully acknowledges the assistance of many of his associates in the preparation of this paper, and particularly of Mr. J. V. Dunn. Image Transmission System for Two-Way Television* By HERBERT E. IVES, FRANK GRAY and M. W. BALDWIN A two-way television system, in combination with a telephone circuit, has been developed and demonstrated. With this system two people can both see and talk to each other. It consists in principle of two television systems of the sort described before the June, 1927, Convention of the American Institute of Electrical Engineers. Scanning is by the beam method, using discs containing 72 holes, in place of 50 as heretofore. Blue light, to which the photoelectric cells are quite sensitive, is used for scanning, with a resultant minimizing of glare to the eyes. Water-cooled neon lamps are employed to give an image bright enough to be seen without interference from the scanning beam. A frequency band of 40,000 cycles width is required for each of the two television circuits. Synchronization is effected by transmission of a 1275 cycle alternating current controlling special synchronous motors rotating 18 times per second. Speech trans- mission is by microphone and loud speaker concealed in the television booth so that no telephone instrument interferes with the view of the face. DITRING the past few years, since the physical possibihty of television has been established, the chief problems which have received attention have been those of one-way transmission. In particular, the experimental work in radio television has had for its principal goal the broadcasting of television images, which is inher- ently transmission in one direction. At the time of the initial de- monstration of television at Bell Telephone Laboratories in 1927,^ one part of the demonstration consisted of the transmission to New York of the image of a speaker in Washington simultaneously with the carrying on of a two-way telephone conversation. At that time it was stated that two-way television as a complete adjunct to a two-way telephone conversation was a later possibility. It is the purpose of this paper to describe a two-way television system now set up and in operation between the main offices of the American Telephone and Telegraph Company at 195 Broadway and the Bell Telephone Laboratories at 463 West Street, New York. It con- sists in principle of two complete television transmitting and re- ceiving sets of the sort used in the 1927 one-way television demonstra- tion. In realizing this duplication of apparatus, however, a number of characteristic special problems arise, and the paper deals chiefly with matters peculiar to two-way as contrasted with one-way tele- vision. * Presented at June, 1930, meeting of A.I.E.E., Toronto, Canada. ' Bell System Technical Joitnial, October, 1927, ]>]>. 551-652. 448 IMAGE TRANSMISSION SYSTEM 449 Physical Arrangement and Operation The detailed description of the optical and electrical elements of the two-way television system will be more readily grasped if it is preceded by an account of the general arrangement of the parts and of the method of operation of the system from the standpoint of the user. The physical arrangement of the two-way television system is shown by the pictorial sketch Fig. 1, and in the photographs Fig. 2 and Fig. 3. The terminal apparatus is largely concentrated into a booth, — the television booth — similar in many respects to the familiar telephone booth, and a pair of cabinets, which contain the scanning discs and light sources. As in the 1927 demonstration, scanning is performed by the beam method, the scanning beam being derived from an arc lamp whose light passes through a disc furnished with a spiral of holes and thence through a lens on the level of the eyes of the person being scanned. The light reflected from the person's face is picked up by a group of photoelectric cells for subsequent ampli- fication and transmission to the distant point. The signals received from the distant point are translated into an image by means of a neon glow lamp directly behind a second disc driven by a second motor placed below the first and inclined at a slight angle to it. The two discs, which are shown in the center cabinet of Fig. 2, are of slightly different sizes; the upper one 21" in diameter and the lower one 30". They difl'er from the discs used in the earlier demonstration in that in place of the 50 spirally arranged holes formerly used, they carry 72 holes whereby the amount of image detail is doubled. While with the earlier "50 line" picture recognizable images of a face were obtainable, the aim in this new development was to reproduce the face so clearly that there would never be any doubt of recognizability, and so that in- dividual traits and expressions would be unmistakably transmitted. This doubled number of image elements necessarily requires, for the same image repetition frequency (18 per second) twice the transmission band, or approximately 40,000 cycles as against 20,000 for the 1927 image. The only part of the television apparatus visible to the user is the array of photoelectric cells which are in the television booth behind plates of diffusing glass. In addition to the photoelectric cells and their immediately associated amplifiers, the booth contains a con- cealed microphone and loud speaker. By means of these, the voice is transmitted to the distant station and received therefrom without the interposition of any visible telephone instrument which could obscure the face. 450 BELL SYSTEM TECHNICAL JOURNAL in •■'ri:!'^ .» ,•'--. ■-;.'--.^'-^-;n'--'-l'~.':^.'.'.?-W^ a. ^ i a. J~> O > CD * < 0) _J z Id 1 Z O .•t; I (/) Q. ■« LJJ in -J UJ ^ H (0 _i -+ _i UJ CD c .2 '33 'E ■(-' % ■i-> o •g 1 in !2 'C o r, d CU u 1 X r-( Q. ^ >s u -I-" LxJ _1 O LU ^ h- 1- o ^ z ^ < v LU z z O ^ X % a. UJ _j LU p h- CD z m < o> u c: LU 2 < IMAGE TRANSMISSION SYSTEM 451 From the standpoint of the user, the operation of the combined television and telephone system is reduced to great simplicity. He enters the booth, closes the door, seats himself in a revolving chair, swings around to face a frame through which the scanning beam reaches his face, and upon seeeing the distant person, he talks in a natural tone of voice, and hears the image speak. Conversation is carried on as though across a table. Fig. 2 — The three major cabinets of the television-telephone apparatus. Optical Problems Some of the more special of the problems encountered in two-way television are primarily optical in character. The principal one is that of regulating the intensity of the scanning light and of the image which is viewed so that the eye is not annoyed by the scanning beam or the neon lamp image rendered difficult of observ^ation. It has been necessary for the solution of this problem to reduce the visible in- tensity of the scanning beam considerably below the value formerly used and to considerably increase the brightness of the neon lamp. The means adopted consists first, in the use of a scanning light of a color to which the eye is relativ^ely insensitive but to which photoelectric cells can be made highly sensitive. For this purpose blue light has been used, obtained by interposing a blue filter in the 452 BELL SYSTEM TECHNICAL JOURNAL path of the arc light beam, and potassium photoelectric cells specially sensitized to blue light and more sensitive than those previously used have been developed. The number of these cells and their area has also been increased over those used in the earlier television apparatus so that the necessary intensity of the scanning beam is decreased. The second half of the problem, namely that of securing a max- Fig. 3 — Interior of the television booth. imum intensity of the neon lamp, has been attained by the develop- ment of water-cooled lamps capable of carrying a high current. The net result of the use of the blue light for scanning, of more sensitive photoelectric cells, and of the high efficiency neon lamps is that the user of the apparatus is subjected only to a relatively mild blue IMAGE TRANSMISSION SYSTEM 453 light sweeping across his face, which he perceives merely as a blue spot of light lying above the incoming image. Figure 3 shows the in- terior of the television booth with the frame through which the observer sees the image of the distant person. A second optical problem is the arrangement of the photoelectric cells required in order to obtain proper virtual illumination of the observer's face. As we have previously pointed out in discussing the beam scanning method,^ the photoelectric cells act as virtual light sources and may be manipulated both as to their size and position like the lights used by a portrait photographer in illum- inating the face. In the present case, it is desired to have the whole face illuminated and accordingly photoelectric cells are provided to either side and above. One practical difficulty which is encountered is that eyeglasses, which often cause annoying reflections in photog- raphy are similarly operative here. For this reason, it is important that the photoelectric cells be placed as far to either side or above as possible. The banks of photoelectric cells shown in Fig. 3 are accordingly much farther removed from the axis of the booth than were the three cells used in the first demonstration. In the position which has been chosen for the cells, reflections from eyeglasses are not annoying unless the user turns his face considerably to one side or the other. The number of cells has been so chosen as to secure a good bal- ance of effective illumination from the three sides and it has been found desirable to partly cover the cells on one side in order to aid in the modelling of the face by the production of slight shadows in one direction. Another optical problem is the illumination of the interior of the booth. There must, of course, be sufficient illumination for the user to locate himself, and it is also desirable that the incoming image and the scanning spot be not seen against an absolutely black background. The illumination of the booth is by orange light, to which the cells are practically insensitive, and so arranged that the walls and floor are well illuminated. In addition to the wall and floor illumination, a small light is provided on the shelf bar in front of the observer so as to cast orange light on the front wall surrounding the viewing frame. This light contributes materially to reducing the glaring effect of the scanning beam, and to the easy visibility of the incoming image. In addition to the optical features which are visible to the person sitting in the booth, there are very necessary optical elements which ^Jul. optical Soc. of America, Mardi, 1928, p. 177. 454 BELL SYSTEM TECHNICAL JOURNAL have to do with the positioning of the outgoing and incoming images. A practical problem which is encountered when customers of various heights use the apparatus is that the scanning beam, if fixed in its posi- tion, would strike too high or too low upon many faces. In order to direct the beam up or down as is required, a variable angle prism, consisting of two prisms arranged to rotate in opposite directions, Fig. 4 — Optical means for controlling heights of scanning and viewing beams. is interposed in the path of the scanning beam. This prism, which lies directly in front of the projection lens used with the upper disc, is shown in Fig. 4 at P. Its rotation is controlled by a knob with a numbered dial. The exact position is determined by the operator by reference to a monitoring image which will be described below. Another optical element which serves two purposes, is a large con- vex lens lying between the receiving disc and the observing frame, IMAGE TRANSMISSION SYSTEM 455 shown at L, Fig. 4. This lens is used to magnify the incoming image to such a size that the image structure is just on the verge of visibiUty, under which condition the face of the distant person appears as though he were approximately eight feet away. In addition to acting as a magnifier, this lens serves to position the incoming image to fit the height of the user. For, by raising or lowering it by means of a knob, the operator, using the information as to the observer's height obtained from the position of the scanning beam, locates the lens so that the virtual image appears in the proper position. Photoelectric Cells and Associated Circuits The photoelectric cells used in this apparatus are similar in shape to those used in the first demonstration, but somewhat larger. Each 100 80 I- z u Cj 60 40 20 ^ /c '\ ' 1 \ \ VIOLET VGREEN RED 0.3 0.4 0.5 0.6 WAVE LENGTH IN p. 0.7 o.a Fig. 5 — .4. Relative optical transmission of the blue filter through which the scanning beam passes. B. Relative sensitivity of the photoelectric cells to various parts of the spectrum. C. Relative sensitivity of the eye to A-arious parts of the spectrum. cell is twenty inches long and four inches in diameter, giving it an area of approximately eighty square inches for collecting light. The anode is made in the form of a hollow glass rod wound with wire. This construction prevents the electrical oscillations that would otherwise result from mechanical vibrations of the anode. The sen- sitive cathode consists of a coating, covering the rear wall of the tube, of potassium sensitized with sulphur.'^ This kind of cell is consider- ably more sensitive than the older type of potassium-hydride cell 3 A. R. Olpin, Phys. Rev., 33, 1081 (1929). 30 456 BELL SYSTEM TECHNICAL JOURNAL while still having most of the sensitiveness in the blue region of the spectrum. Figure 5 shows the response of the photoelectric cells used to the various parts of the spectrum together with the trans- mission of the blue filter and the brightness of the various parts of the spectrum as evaluated by the human eye. The very great efficiency of the photoelectric cells and the inefficiency of the eye to the light used are apparent. To amplify the photoelectric current, the cells are filled with argon at a low pressure. Photoelectrons passing from the sensitive film to the anode ionize the gas atoms along their paths and thus cause a greater flow of current. The ionization of the gas does not, however, instantaneously follow sudden variations of the true photoelectric emission from the sensitive film, that is, there is a time lag in the If) _i ffl u UJ Q + 4 + 2 0 10 (0 O lU > i< C- EQUALIZED ,,- --^OUTPUT ^~^^^^^^^r"^ \ "^"\ ^^^^ A- SINGLE ^ ^^ CELL ''^ "v. \B- GROUP OF \ CELLS \ 100 10,000 FREQUENCY IN CYCLES PER SECOND 1,000 100,000 Fig. 6 — Loss in response of photoelectric cells at high frequencies. ionization of the gas and in the disappearance of ionization. This lag results in a relative loss and phase shift of the high frequency components of a television signal with respect to the low frequency components which become serious in the wider frequency range utilized in the 72 line image. The relative loss in output from a single large photoelectric cell at high frequencies is shown in decibles by curve A of Fig. 6. In the television booth, the twelve large cells mounted in the walls of the booth present an area of approximately seven square feet to collect light reflected from a subject. To secure the desired effective illumination, the cells are mounted in three groups, comprising a group of five cells in each of the two side walls of the booth and a group of two cells in the sloping front wall above the subject. The twelve cells are enclosed in a large sheet copper box, provided with IMAGE TRANSMISSION SYSTEM 457 doors to each group. The cells of each group are connected in par- allel through the input resistance of a two stage resistance-capacity coupled amplifier similar to those previously used. This raises the level of the signal to such a point that the output of the three ampli- fiers may be carried through shielded leads and connected in parallel to a common amplifier. The metal anodes and lead wires of the cells in parallel in any one group give an appreciable capacity to ground, which results in a further loss in amplitude and phase shift of the high frequency com- ponents of the signal. The combined loss introduced by ionization of the gas in the cells and by capacity to ground is shown by curve B of Fig. 7 — Schematic of interstage amplifier coupling to equalize for the high frequency losses in the photoelectric cells. Fig. 6. This combined loss is equalized by an interstage amplifier coupling, Fig. 7. The equalized output from the photoelectric cells is shown by curve C, Fig. 6. Two-Way Image Signal Amplifiers The vacuum tube system used to amplify the photoelectric cell currents in two-way television is patterned closely after that used previously in one-way television, and the description here will be con- fined chiefly to novel features. These new features are necessary to take care of the doubled frequency band which results when the scanning is done with a 72-hole disc rather than with a 50-hole one, and to provide sufficient power to operate the high intensity neon lamp which is essential to two-way television. Certain other new features have been introduced in order to simplify the apparatus and to reduce the maintenance required to keep it in good working con- dition. The vacuum tubes which operate at low energy levels are the so-called "peanut" type, chosen because of their freedom from microphonic action and their low interelectrode capacities. Protection against mechanical and acoustical interference is secured by mounting these tubes in balsa wood cylinders which are loaded with lead rings 458 BELL SYSTEM TECHNICAL JOURNAL and cushioned in sponge rubber. The tubes are electrically connected in cascade by means of resistance-capacity coupling, so that the whole amplifier system is stable over long periods of time and is also uniformly efficient over the required frequency band. Grid bias for the small tubes is supplied by the potential drop across a resistance in the filament circuit; the power requirements for the low level stages of the amplifier are filled by 6-volt filament batteries and 135- volt plate batteries, all located externally where they can be checked and replaced conveniently. The amplifier system is divided into units of convenient size as shown in Fig. 8. Associated with each of the three banks of photo- o a. O lUiO U _J Oo O X 0. CC UJ o_i CLU llJ 2-1 (£0. "is O I- 111 t£ WtL 10 -^ 10-4 10"^ Fig. 8 — Schematic diagram of the complete television channel and the relative voltage levels of the signal along the channel. electric cells is a two-stage unit known as the photoelectric cell amp- lifier; the combined output of these three units is carried to a four- stage unit known as the intermediate amplifier whose output is of sufficiently high level to be carried outside the copper cell cabinet to the three-stage transmitting power amplifier on the relay rack. A four-stage unit known as the receiving power amplifier is also on the rack, and serves to amplify the signal from the other station to a level which will yield an image of satisfactory contrast when it is translated into a light variation by means of the neon lamp. The final stage of this amplifier consists of two special 250-watt tubes in IMAGE TRANSMISSION SYSTEM 459 parallel. These large tubes are used because their plate impedance is of the same order of magnitude as the impedance of the neon lamp, and because they will supply the necessary direct current to the neon lamp without overheating. Figure 8 also shows what may be termed a voltage level diagram for the whole system. Ordinates on this diagram represent voltage amplitudes at the junctions between units of the system, and by themselves tell nothing at all about the power conditions in the system, since the impedances are not specified. It is interesting to observe that the signal voltage produced by the three banks of photoelectric cells has an effective value of about 50 microvolts across the 50,000 ohm input resistance; the transmitting amplifier delivers about 1 volt to the 125-ohm cable circuit, and the receiving amplifier delivers about 100 volts to the 1,000 ohm neon lamp circuit. The signal cur- rent through the neon lamp has an effective value about a thous- and million times greater than that of the current variation in one of the photoelectric cells. The most outstanding contribution to the development of tele- vision amplifiers is the combination of output and input transformers whose transmission characteristics are shown in Fig. 9, A, and whose impedance characteristics are shown in Fig. 9, B, and C. The ex- ceptionally wide frequency range, corresponding to a ratio of. limiting frequencies of 5,000 to 1, transmitted by these transformers is due largely to the use of chrome permalloy, a recently developed core material having very high permeability. The improved char- acteristics are also the result of refinements in design which involve the use of adjusted capacities and resistances to control the character- istics at the higher frequencies. Due to the fact that each terminated transformer looks like a resistance of 125 ohms over practically the entire frequency range of the image signal, it makes no difference in the form of the overall voltage amplification characteristic of the cir- cuit whether the transformers are connected together directly or by means of the equalized cable circuit whose characteristic in shown in Fig. 10. Advantage of this circumstance is taken in providing switch- ing means whereby each transmitting amplifier may be connected through a resistance pad to its local receiving amplifier, enabling a person to see his own image in the television booth, which is a conven- ience in making apparatus adjustments. Transformers of this type must be carefully protected against magnetizing forces which might cause polarization of the core material. In order to keep the plate current of the final tube of the transmitting power amplifier from flowing through the winding of the output 460 BELL SYSTEM TECHNICAL JOURNAL +12 +11 +10 o +9 _i ill (D UJ Q -18 -19 -20 -21 10 INPUT _, ^,_ — - — ' ■ ■' -^ OUTPUT ~~ — ' — ^^ ^ ■*v. 100 FREQUENCY 1,000 10,000 IN CYCLES PER SECOND 100.000 175 150 125 100 ^75 ^50 25 0 -25- -50 RESISTANCE ^^ ■ REACTANCE ^ B 100 FREQUENCY 1,000 10,000 IN CYCLES PER SECOND 100,000 175 150 125 100 -75 I O 50 25 0 -25 -50 RESISTANCE ^ — ^~^ .^^ " REACTANCE ■ ■ 10 100 FREQUENCY 1,000 10,000 IN CYCLES PER SECOND 100,000 Fig. 9 — A. Voltage ratio characteristics of VV-TSTQ input transformer and \V-7880 output transformers, each connected between its rated impedance. B. Impedance characteristic of W-7880 output transformer with 1765 ohm resistance load. C. Impedance characteristic of \V-7879 input transformer with 20 mmf. capacity load. IMAGE TRANSMISSION SYSTEM 461 transformer, the transformer winding is shunted by a battery and a resistance in series. The resistance is made high, so that the trans- mission loss due to bridging it across the circuit is small; the voltage of the battery is made equal to the potential drop across the resistance due to the plate current of the tube, so that the average voltage across both the battery and the resistance, and hence across the trans- former winding, is zero. A vacuum thermocouple is connected in series with the line winding of the output transformer, serving as level indicator for the trans- mitting amplifier. The level indicator for the receiving amplifier is 12 II 10 - U u WITH EQUALIZER^ \ — / / / WITHOUT EQUALIZER^ ^ y i ^ — 10 100 1,000 10,000 FREQUENCY IN CYCLES PER SECOND 100,000 Fig. 10 — Insertion loss characteristic of cable cricuits which transmit the image signal, measured between 125 ohm resistances. a vacuum thermocouple in series with the grid resistance of the two 250-watt tubes. The electrical control panels associated with one terminal of the television apparatus are shown in Fig. 11. Transmission Circuits Two special requirements for the two-way television transmission circuits are to be emphasized. The first, which has already been referred to, is the wide frequency transmission band, from 18 cycles to 40,000 cycles, which must have a high degree of uniformity of trans- mission efficiency and freedom from phase distortion. The second is the necessity for two circuits for the television images. This arises from the fact that the two parties to the conversation must both see and be seen at all times. There can be no interruption of one face by the other, comparable with the alternation of the role of speaker and listener in telephony which permits the use of a single circuit for ordinary speech communication. The terminal stations of the two-way television system are con- 462 BELL SYSTEM TECHNICAL JOURNAL nected by eight underground circuits, each consisting of 13,032 feet of No. 19 gauge and 390 feet of No. 22 gauge non-loaded cable. Two circuits are used for transmitting the image signals, two for the accom- panying speech, one for the synchronizing current, two are used as Fig. 11 — Control apparatus panels associated with one terminal of the television apparatus. order wires, and one is kept as a spare. All of the circuits have identical transmission characteristics, but equalization is necessary only on the two which carry the image signals. Figure 10 shows the insertion loss characteristic of each circuit, and also shows the insertion loss charac- IMAGE TRANSMISSION SYSTEM 463 teristic of the image circuits when the image line equalizers are in- cluded. Although the distance between the stations is small the require- ments of the television system from the standpoint of freedom from noise and other interference require that considerable care be given to the selection of the cable circuits used. All terminal connections are made through balanced repeating coils or transformers so that all of the circuits are balanced to ground. Also, in order to insure that the crosstalk between the various channels be unnoticeable the terminal equipment is so adjusted that approximately the same amount of power is transmitted by each circuit. Neon Lamps and Associated Circuits After amplification, the received television signal is impressed on the grids of two power tubes in parallel to furnish current for a neon receiving lamp. The terminal lamp circuit is shown in Fig. 12. » _ X Fig. 12 — Schematic of neon lamp circuit. The grid bias of the two power tubes is varied by the operator to control the DC plate current, which replaces the original DC signal component suppressed at the sending end. The quality of the re- produced image is determined by the operator's control over the relative levels of the incoming AC signal and the restored DC current. The television current from the power tubes is translated back into light by a water-cooled neon lamp designed to operate on a large current. The structural details of the lamp are shown in Fig. 13. Heavy metal bands attach the rectangular cathode to a hollow glass stem occupying the central portion of the tube. Water from a small circulating pump flows continuously through the glass stem and cools the cathode by thermal conduction through the metal bands. To reduce sputtering of the cathode and consequent blackening of the 464 BELL SYSTEM TECHNICAL JOURNAL glass walls, the front surface of the cathode is coated with beryllium. This metal resists the disintegrating action of the glow discharge very satisfactorily and gives the lamp a prolonged life. Other metal sur- faces in the tube are shielded from the discharge by mica plates; and * Fig. 13 — Water-cooled neon lamp. the discharge passes from the frame-like anode to the front surface of the cathode, covering it with a brilliant layer of uniform cathode glow. Pure neon in a plate type of lamp gives a very inferior reproduction of an image. The impedance of the lamp is relatively high and com- prises both a resistance and a reactance which vary with frequency. The variation in the impedance causes a relative loss in the frequency components of the signal and also introduces spurious phase shifts. IMAGE TRANSMISSION SYSTEAI 465 In addition, pure neon has an after-glow; the gas continues to glow for an appreciable time after current ceases to flow. This after-glow casts spurious bands of illumination out to one side of the brighter image details. A small amount of hydrogen in the neon prevents such an afterglow; and at the same time improves the circuit characteristics of the lamp. The total impedance of the lamp is lower, making it a less influential part of the lamp circuit; and the resistance and reactance vary in such a manner that the phase shift is more nearly proportional to frequency (a phase shift proportional to frequency causes no distortion in the reproduction of an image). Other active gasses may be used with the neon to improve the operation of a television lamp, but one or two per cent of hydrogen is most satisfactory. Since hydrogen is absorbed by the electrodes in a glow discharge, it slowly disappears from the neon during operation of the lamp. For this reason the lamp is provided with a small side reservoir of hydrogen. The lamp and the reservoir carry porous plugs immersed in a pool of mercury; and a flexible rubber connection permits the two plugs to be brought into contact at will. Minute quantities of hydrogen may be introduced into the lamp by simply bringing the two plugs into con- tact for a short time. Even with this improvement the circuit characteristics of a lamp are not ideal. W'ith power tubes it is usually desirable to include a fixed resistance in series with the lamp to prevent semi-arcing con- ditions. Such a resistance also makes the lamp a less influential fraction of the total circuit impedance. Optical Monitoring System In order to insure that the incoming and outgoing images are prop- erly positioned, no matter what the stature of the person sitting in the booth, and that the images shall be of proper quality, it is essential to have some means for the operator to observe and adjust these images. The optical monitoring system provided consists of an outgoing mon- itor and means for adjusting the scanning beam, and an incoming monitor and means for adjusting the position of the viewing lens to suit the height of the sitter. The outgoing monitoring system is the same as that used in the one- way television apparatus which has already been described. A small neon lamp (Fig. 14, at bottom of top disc) is placed behind the sending disc but displaced several frames from the aperture through which the arc lamp beam passes. By continuing the spiral of holes part way around it is possible to see the complete image from the auxiliary neon 466 BELL SYSTEM TECHNICAL JOURNAL lamp, to which the outgoing signals are also supplied. In order to see this monitoring lamp from the operator's position, a right-angle prism and a magnifying lens are placed in front of the disc and the image is observed through an opening in the side of the motor cabinet. The Fig. 14 — Sending and receiving discs, with neon lamps and optical arrangements for image monitoring. task of the operator is to direct the scanning beam up or down by means of the variable angle prism until the face of the person in the booth is centrally located. This adjustment is facilitated by a wire IMAGE TRANSMISSION SYSTEM 467 which passes across the image and is placed at the height at which the user's eyes should appear. The height of the observer's eyes is an indication of the position which should be taken by the large magnifying lens L, and the operator, after having properly placed the scanning beam, reads the scale on the variable angle prism dial, and then sets the magnifying lens by turning its controlling knob to the same number. When both adjustments are complete, the person in the booth will not only be properly scanned but will be in the best position to see the image. In order to monitor the incoming image, an optical arrangement is adopted by means of which light from the water-cooled neon lamp is taken off at the side and reflected through the disc and thence reflected again, as shown in Fig. 14 (top of bottom disc), through a second, lower, observing hole on the side of the motor cabinet. Because of the small area of the side view of the neon lamp, a lens system is inserted which focusses the image of the lamp at the place to be occupied by the pupil of the operator's eye. When the eye is properly placed, the whole of the lens area is seen filled with light and exhibits the incoming image. In addition to the monitoring means just described, an additional view of the incoming image is provided by means of a 45° mirror which is carried on the back of a movable shutter which is shown at S in Fig, 4. This shutter carries an illuminated sign on the side turned to the user with the inscription "Watch this space for television image." The shutter with its sign covers the image until the adjustments just described are made, when it is dropped out of sight. While it is in place, the operator is provided with an additional monitoring image reflected from the 45" mirror. This view is, of course, in every respect identical with that which the user sees. The function of the incoming monitoring system is primarily to enable the operator to set the electrical controls to give the proper quality of image. He also has another task which is that of properly framing the image. This he can do by turning the framing handle, which is described elsewhere, while watching the image from the 45° mirror. This framing operation is preferably performed not on a person sitting in the booth but upon some suitable object such as a mirror located upon the rear door of the booth. In order to make this framing adjustment, the operators at the two terminals set their scanning beam dials to predetermined positions such that the scan- ning beams place the framing mirrors at the lower edge of the scanning rectangles, the phases of the incoming discs are then shifted until the images of the mirrors are seen properly located in the incoming mon- itors. 468 BELL SYSTEM TECHNICAL JOURNAL Signalling System In order to coordinate operations at the two terminal stations, an order wire system is provided. There are four telephone sets at each station; one on the attendant's desk in the ante-room, one concealed inside the television booth, one in the control room, and one at the control panels for the technical operator, who operates the small switchboard which is part of the system. Two of the underground cable circuits connect the two switchboards, so that there may be not more than two separate conversations between stations at one time. Ringing is accomplished by means of standard 20-cycle ringing current furnished by the Telephone Company. During a demonstration, the attendants' telephones are connected permanently over one of the cable circuits. To relieve the operators of the duty of ringing each time the attendants wish to communicate, a push button and buzzer are provided at each attendant's desk, operated by the standard ringing currents simplexed on the synchro- nizing circuit. This arrangement leaves the operators free to manip- ulate the television apparatus. The two order wire circuits are each simplexed to provide two additional circuits which operate signal lamps indicating to both operators when either chair in the television booths is occupied and turned in position. Discussion The primary objects in developing and installing the two-way television system have been two. The first was to obtain information on the value of the addition of sight to sound in person to per- son communication over the telephone. The second was to learn the nature of the apparatus and operating problems which are involved in a complete television-telephone service. While the in- stallation is entirely experimental, it is being maintained in practically continuous operation for demonstration to employees and guests of the Telephone Company, and interesting data are being gathered on all aspects of the problem. It may be said without fear of contradiction that the pleasure and satisfaction of a telephone conversation are enhanced by the ability ot the participants to see each other. This is, of course, more evident where there is a strong emotional factor, as in the case of close friends or members of the same family, particularly if these hav^e not been seen for some time. Were the television apparatus and required line facilities of extreme simplicity and cheapness it would be safe to predict a demand for its IMAGE TRANSMISSION SYSTEM 469 early use. At the present time, however, the terminal apparatus is complex and bulky, and requires the services of trained engineers to maintain and operate it. In addition to the cost of the terminal apparatus there is the unescapable item of a many-fold greater trans- mission channel cost. Because of the wide transmission bands re- quired for the television images, the inherent necessity for a television channel in each direction, and the extra channels for synchronizing and signalling, the total transmission facilities used in this demonstra- tion are those which could, according to current practice, carry about fifteen ordinary telephone conversations. It is to be expected, of course, that development work will result in some increase in the efficiency of the transmitting channels and in simplifications of the terminal apparatus. It is conceivable, therefore, that our present conception of the cost of the whole system may ultimately be materially changed. Synchronization System for Two-Way Television * By H. M. STOLLER In a previous paper presented before the June, 1927 Convention of the American Institute of Electrical Engineers, the method of securing syn- chronization of television signals was described as employed in the Bell System Television demonstration of April, 1927. The present paper de- scribes the development of a new control circuit which is in use in the new two-way television system between the Bell Telephone Laboratories at 463 West Street and the American Telephone and Telegraph Company build- ing at 195 Broadway, New York. TELEVISION transmission requires not only synchronization of the transmitting and receiving equipment but such synchronization must be held to a narrow phase angle so that the scanning discs at the transmitting and receiving end will never depart more than a small frac- tion of a picture frame width from the desired position.^ In the 1927 demonstration, 2125 cycle synchronous motors were employed with supplementary D.C. motors to facilitate starting. This plan required the use of vacuum tube amplifiers of large size in order to supply suf- ficient power to the synchronous motors. Such high frequency synchronous motors, however, are inefficient and expensive, so that when designing the new system, it was desired to solve the problem of synchronization with simpler and cheaper equipment and in a manner which would require less attention in starting. It was particularly desired to employ a motor which could be operated directly from the 110 volt lighting circuit without any auxiliary "A", "B" or "C" batteries for the control equipment. Description of Motor Figure 1 shows a photograph of the new television motor and its associated control equipment. The motor is a four pole compound wound D.C. motor with the following special features added. 1. An auxiliary regulating field, the current through which is controlled by the vacuum tube regulator. 2. A damping winding on the face of the field poles to prevent the field fiux from shifting (Fig. 3). * Presented at June, 1930, meeting of A.I.E.E., Toronto, Canada. 1 These requirements are more fully discussed in a previous paper. {Journal of the A. I. E. E., Vol. 46, page 940, 1927.) 470 5 YNCHRONIZA TION S YSTEM 471 Three slip rings are provided at points 120 electrical degrees apart for furnishing three phase power to supply plate and filament voltage for the regulating circuit. A pilot generator of the inductor type is built into the motor frame and delivers a frequency proportional to the motor speed for actuating the control circuit. Fig. 1 — New television motor and vacuum tube control circuit. 5. A hydraulically damped coupling is provided between the motor shaft and the scanning disc. (Fig. 4.) The motor frame was made from a standard 36 tooth stator punching by cutting out three teeth per pole, thus forming four polar areas of six teeth each. The shunt, series and regulating fields enclose the 31 472 BELL SYSTEM TECHNICAL JOURNAL entire polar areas. The damping winding consists of insulated closed turns of heavy copper wire distributed over the pole faces in the slots as shown in Fig. 3. It will be noted that this damping winding has no effect upon the flux through the poles as long as the flux density over the polar surface does not shift. In other words, the damping winding permits the total flux of the motor to increase or decrease as required by the regulating circuit but will oppose any tendency of the flux to shift back and forth across the pole face. As will be explained SCANNING DISK Fig. 2 — Schematic diagram of control circuit. later on, this feature is essential in order to prevent hunting or insta- bility of the image. The hydraulically damped coupling between the motor shaft and the scanning disc is also essential in order to avoid hunting. It employs flexible metal bellows filled with oil and connected by a small pipe equipped with a needle valve for adjusting the amount of damping. Figure 4 shows its construction. The scanning disc itself is centered on a ball bearing which allows the disc to rotate with respect to the shaft within approximately ± 5 degrees mechanical movement. 5 YNCHRONIZA TION S YSTEM 473 Control Circuit Figure 2 shows a schematic diagram of the control circuit. When the motor is operating at full speed the pilot generator delivers approx- imately 1 watt of power at 300 volts, 1275 cycles to the plates of a pair of push-pull detector tubes. The grids of these tubes are supplied with an e.m.f. of the same frequency from an oscillator or other source of power having a sufficiently constant frequency. The amount of power required for this grid circuit is only a few thousandths of a watt. The detector tubes rectify the plate voltage producing a po- tential drop across the coupling resistance R]. If the plate and grid voltages are in phase, so that the grids of the tubes are positive at the SLOT INSULATION Fig. 3 — Damping winding preventing shifting of field flux. same instant that the plates are positive, the plate current will be a maximum. If the grid voltage is negative when the plate voltage is positive the plate current is practically zero, so that the magnitude of this current is a function of the phase relationship between the grid and plate voltages as shown in Fig. 5. The voltage drop across the coupling resistance i?i is applied to the grid circuits of three regulator tubes. These tubes derive their plate voltage supply from a three phase transformer fed with power from the three slip rings provided on the motor. These tubes act as a rectifier whose output is controlled by the potential impressed upon the grids from the coupling resistance R\. The current of the regulator tubes is passed through the regulating field provided on the motor. This field is in a direction to aid the shunt field and series fields of the motor. 474 BELL SYSTEM TECHNICAL JOURNAL The operation of the circuit is as follows: In starting switch ^i is closed which applies direct current to the shunt field and armature circuits of the motor. The motor accelerates as an ordinary compound wound motor. Switch S2 is then closed applying three phase power from the slip rings of the motor to the transformer. As the speed of the motor approaches the operating point, the beat frequency between the pilot generator and the oscillator will cause beats in the current through the regulating field which are visible on the meter Mi. Let us assume that the field rheostat has been previously adjusted so that with SCANNING DISK-22"DIA. BELLOWS m ss ss Fig. 4— Hydraulically dampedcoupling'to prevent hunting of motor. the shunt field alone the motor will tend to run slightly over the desired operating speed. When the exact operating speed is obtained, the beat frequency in the regulating field will be zero and as the motor tends to accelerate, the phase relationship between the pilot generator and the oscillator will reach a point tending to give maximum strength to the regulating field. When this point is reached, the acceleration of the motor will be checked by the increased field and the speed will tend to fall until the phase of the pilot generator with respect to the oscillator has shifted sufficiently so that the regulating field current is 5 YNCHRONIZA TION S YSTEM 475 reduced to an equilibrium value, after which the motor continues to run at constant speed in accordance with the frequency of the os- cillator. Operating tests on the circuit show that the motor will hold in step over line voltage ranges from 100 to 125 volts and will be self-synchron- izing over somewhat narrower voltage limits. Thus, under normal conditions all that is necessary from an operating standpoint is to close the switch and wait for the motor to pull into step. Control Oscillator The control oscillator is a standard type of vacuum tube oscillator having a frequency precision of the order of 1 part in 1000, when PLATE CURRENT /effective I VALUE 90° PHASE ANGLE BETWEEN PLATE AND GRID VOLTAGES Fig. 5 — Phase detector tube characteristic. 180' delivering the negligible output of .005 watts to the grid circuit of the detector tubes. This frequency is delivered directly to the motor circuits at one end of the line and is transmitted over a separate cable pair to the control circuits at the other end of the line. It was found that the detector tubes would operate successfully over a considerable variation in power level, provided the minimum oscillator output was sufficient. An interesting alternative method was developed in which the synchronizing channel between stations may be omitted entirely, but this method was not used in the present system as the additional cost was not justified. The method, however, is described as it may prove of value if television transmission over long distances is con- sidered. 476 BELL SYSTEM TECHNICAL JOURNAL Mr. W. A. Marrison in his paper "A High Precision Standard of Frequency," Proceedings I. R. E. July, 1929, described a crystal controlled oscillator which would maintain a precision as to frequency of 1 part in 10,000,000. This oscillator employs a quartz crystal as its primary means of control and by means of secondary circuits the natural period of the crystal, which is approximately 100,000 cycles, may be stepped down to lower frequencies which are more convenient for such purposes as motor control. By this means, a frequency of the desired value may be obtained with a precision so great that the speed of the scanning discs under control of the above described circuit will be so nearly perfect that no synchronization channel at all is required. For example, if the period of observation of the television image is 5 minutes, the scanning disc will make 5300 revolutions when operating at a speed of 1060 r. p.m. Assuming a precision of control of 1 part in 10,000,000, the maximum error during the 5 minute interval will be 5300 divided by 10,000,000 or about 1/2000 of 1 revolution. Expressed in degrees on the periphery of the disc, this is equivalent to approximately 1/6 of 1 degree or since the width of the television image with 72 holes in the scanning disc is 5 degrees, the image will drift 1/30 of a frame width during the 5 minute interval. If the speed of the scanning disc at the other end drifts an equal amount in the opposite direction, the displacement of the television image will be only 1/15 of a frame width, which is a tolerable amount of drift. From a practical viewpoint, however, it is apparent that the addi- tional cost of very precise independent oscillators would be greater than the cost of providing the synchronization channel, except possibly for transmission over long distances. Framing Referring to Fig. 2, it will be noted that a phase shifter is provided between the oscillator and the input terminals to the control circuit. This phase shifter is designed with a split phase primary member producing a rotating magnetic field. The secondary member is sin- gle phase and is mounted on a shaft provided with a handle. By rotating the handle of the phase shifter in the desired direction, the frequency delivered from the phase shifter will be the algebraic sum of the frequencies of the oscillator plus the frequency of rotation of the armature of the phase shifter. It is, therefore, a simple matter for the operator at the receiving end to momentarily increase or decrease the control frequency and thus bring the picture into frame. SYNCHRONIZATION SYSTEM 477 Discussion During the development of the control system, one of the first difficulties encountered was hunting of the controlled motor. The problem of hunting, of course, becomes more difficult of solution the greater the precision of speed regulation desired and the greater the moment of inertia of the load connected to the motor, the latter state- ment applying only to controlled systems of the synchronous type. Since the moment of inertia of the scanning disc is large relative to that of the motor armature, it is seen that the conditions for securing stable rotation would be unfavorable in both the above mentioned respects if the scanning disc were mounted directly on the motor shaft. The hydraulically damped type of coupling above described was, therefore, inserted between the motor shaft and the scanning disc. It was found, however, that hunting still occurred. A further analysis of the problem showed that the axis of the field flux of the motor was shifting back and forth across the pole faces. The damping winding shown in Fig. 4 was then added with a marked improvement. It was also observed that a strong series field on the motor assisted in secur- ing stability and it was, in fact, necessary to employ all three ex- pedients to secure satisfactory performance. In the system as finally developed the television image, if disturbed by a momentary load such as the pressure of the hand against the disc, would come back to rest within approximately one second, there being two oscillations during this interval. In actual operation, it was found that the normal fluctuations in line voltage occurring on the commercial power supply produced no transients of sufficient magnitude to cause any objectional instability in the received image. In conclusion, it should be pointed out that this type of control system could be equally well employed with larger motors for other applications requiring precise speed regulation. While the circuit described is applicable only to a direct current motor, a similar system may be applied to an alternating current motor substituting a saturat- ing reactor in place of the regulating field winding in the manner described by the author in his paper ^ presented before the Society of Motion Picture Engineers, September, 1928. 2 S. M. P. E. Transactions, Vol. 12, No. 35, page 696. Sound Transmission System for Two- Way Television* By D. G. BLATTNER and L. G. BOSTWICK In this paper is described the speech transmission part of the two-way tele\'ision system described in companion papers. The system is designed to produce the best possible illusion of face-to-face communication between speakers located at a distance. Some of the novel features of the system described include the use of distant pick-up transmitters and loud speakers concealed in the wall of the booth, also the use of heavy glass windows through which the scanning beam and the reproduced image are projected as a means of preventing the admission of noise into the booth. IN the design of a sound transmission system to be correlated with a visual system, the requirements as to perfection of results desired are no more stringent than for other high grade sound rep- producing systems ^ that have been described in the literature from time to time. Rather in this case the peculiarities of the system are largely those incidental to the adaptation of old technique to meet new conditions. The principal limitation of the sound system imposed by the visual system is that the user be relieved of all necessity of holding a telephone in close proximity to the head. Such a limitation is highly desirable in order to secure the most natural pose of the features and the most satisfactory scanning. Obviously, the best way of meeting this lim- itation is by the use of telephone instruments of the type adapted for picking up and reproducing sounds at a distance. The use of such instruments has the further advantage that they can be located near the vision screen and so reproduce any peculiarities in tone quality that would result if the speaker were actually located at the position of the image. Of, course, the sharpness of this perspective effect obtained is influenced by the loudness of both the original and the reproduced sounds but the matter of location of instruments is also very important. It would thus seem that the use of distant pick-up and distant pro- jecting instruments offers certain rather fundamental advantages but it is also true that it presents certain other disadvantages. One of the disadvantages is that the distant pick-up microphone gives less output than a close-up device because of the reduced sound pressure on the diaphragm; also a sound producing device to give suitable reception * Presented at June 1930 meeting of A. I. E. E., Toronto, Canada. 1 "Public Address Systems" by J. P- Maxfield and I. W. Green in A. I. E. E., Feb. 14, 1923. Also "High Quality Recording and Reproducing of Music and Speech" by J. P. Maxfield and H. C. Harrison in A. I. E. E., Feb. 1926. 478 SOUND TRANSMISSION SYSTEM 479 at a distance must be supplied with a higher transmission level than would a close-up instrument. It thus becomes necessary to provide for greater gain in transmission and greater electrical power capacity than would be required were the instruments held close to the head. The use of the more elaborate transmission facilities is in itself dis- advantageous but it also tends to increase the feed-back from the loud speaker to the microphone; also the effect of any noise at the micro- phone position or at the listening position tends to interfere more seri- ously with the successful conduct of conversation. In the design of the two-way television system recently installed between the Bell Telephone Laboratories at 463 West Street and the American Telephone & Telegraph Co. at 195 Broadway in New York City, it was felt that it would be possible to overcome these technical objections to the distant type instruments and that the advantages mentioned would justify any measures necessary to do so. The question of instruments was solved by the use of the Western Electric 394 condenser type transmitter - and a dynamic direct radiator loud speaker. The transmitter is one of the type generally used for phonograph and sound picture recording and for other purposes where good quality, high stability and quietness of operation are essential. The direct radiator type of loud speaker was used instead of the usual horn type because of the limited mounting space available. It con- sists of a dynamic structure with a rigid duralumin diaphragm about 3" in diameter flexibly supported at the edge and radiating directly into free air. \\'hile such a structure is not highly efficient and permits of only a small sound power output these considerations are of second- ary importance in this case. The instruments were located in the front wall of the booth about 2' from the position of the user and ad- jacent to the viewing screen in order to enhance the perspective as described above, the microphone being above and the loud speaker below as shown in Fig. 1. These instruments were (in this particular case) connected into a four-wire circuit although in certain cases it might be desirable to use a 2-wire circuit. Such a change would of course be en- tirely feasible. The remainder of the apparatus used consisted of ampli- fiers located at the transmitting end of each channel and an attenuator at the receiving end, the two ends being connected by means of a loop of approximately 3 miles of non-loaded non-equalized cable. The amplifiers and the attenuators were each readily adjustable so that the sounds of different speakers could be reproduced at the optimum loud- ness. Observation of the performance of the system was made possible in each of the control rooms by means of a monitoring head ^ E. C. Wentc in Physical Review of May 1922. 480 BELL SYSTEM TECHNICAL JOURNAL type receiver bridged across the mid-point of an attenuator tying the two channels together. The attenuation used in the monitoring cir- cuit was such as to give no audible feed-back in either booth. The results obtained with this set-up were considered satisfactory from the standpoint of both volume and quality. Ready recognition of familiar Fig- 1 — Microphone and loud speaker in position above and below television scanning and viewing aperture. voices and the association of the source of the reproduced sounds with the image were the usual occurrence. Figure 2 shows in block form the complete circuit set-up and Figure 3 shows the combined response frequency characteristic of the microphone, amplifier and loud speaker. SOUND TRANSMISSION SYSTEM 481 The ordinates of this curve represent variations in sound pressure from the loud speaker for constant pressure on the transmitter diaphragm. These data were obtained with the loud speaker located in a heavily damped room. The measurements were made on the sound axis at a distance of 2', representing the relative position of the observer under conditions of actual use. In setting up such a system the chief consideration is in regard to the acoustic feed-back from the loud speaker to the microphone and in this connection the design of the booth is an important factor. The booth must necessarily be so shaped that the user, looking at the view- ing screen, can be satisfactorily scanned and the light reflected from 463 WEST ST. I CONTROL ROOM CONDENSER MICRO- I- PMONE CONDENSER MICRO- PHONE AMPLIFIER LOUD SPEAKER VOLUME INDICATOR MAIN AMPLIFIER MONITORING RECEIVER MONITORING ATTENUATOR 195 BROADWAY CONTROL ROOM RECEIVING ATTENUATOR MONITORING ATTENUATOR MONITORING RECEIVER LOUD SPEAKER MAIN AMPLIFIER VOLUME INDICATOR CONDENSER MICRO- PHONE AMPLIFIER CONDENSER MICRO- PHONE Fig. 2 — Circuit diagram for sound transmission system for two-way television. the scanned areas will strike the banks of photoelectric cells required for the reproduction of the visual likeness. This requires that the person scanned be located in close proximity to the scanning disc and to the photoelectric cells as well as to the microphone and loud speaker. Such an arrangement is objectionable from an acoustic standpoint in that in the present state of development the cells are necessarily large and poor absorbers of sound. They thus tend to cause part of the sound output from the loud speaker to reflect back into the microphone. If the sound so reflected or fed back is equal or greater in magnitude than the original sound picked up and is of the proper phase relation, the system will "sing" and the sound system become of no practical use. A further eff"ect of the design of the booth is that as a closed cavity, it tends to cause sounds of a certain pitch range to be accen- 482 BELL SYSTEM TECHNICAL JOURNAL tuated. To reduce these effects as far as possible, the television booths were made as large as other considerations would permit and all surfaces were covered where possible with acoustic absorbing material. They have a floor area of about 35 sq. ft. and are about 8 ft. high. Because of the increased transmission required for the proper interpretation of sounds in the presence of noise, the booths were made of heavy masonry material to insulate the user and the microphone from the noise incidental to the rotating parts of the television apparatus. It was thus necessary to project the scanning beam and to view the illuminated image through a window located in the front wall. The microphone and the loud speaker were fitted into this wall, which was then covered over with a thin screen to improve the appearance as (/I ■I- 10 u +5 UJ O z o 1/1 lA 2 - in 10 ~- - V •s y y - - ■ HIGH PASS FILTERS •s s \ / "^OW PASS FILTERS / \ \ \ \ / / / / / 2 34568 2 34566 100 1,000 10,000 CUTOFF FREQUENCY OF FILTERS IN CYCLES PER SECOND Fig. 1 — Syllable articulation of circuit similar to master reference system at optimum received volume under quiet conditions. of growth is relatively slower above 2,500 cycles than below, and that the total gain in going from this point to infinity is relatively small. Figure 2 shows on a somewhat different basis the upper part of this curve and also, for comparison, corresponding data for circuits having commercial terminal apparatus of the types used in the Bell System. The ordinates for these curves are the ratios of the increase in articu- lation in going from an upper cutoff frequency of 2,000 cycles to some higher point, to the total change in articulation in going from 2,000 cycles to infinity. For example, referring to the curve for the effect of upper cutoff frequency on articulation of the Master Reference System, it is seen that the articulation for the 2,000-cycle point is 70 TRANSMITTER FREQUENCY RANGE 485 per cent, for the 3,000-cycIe point is 87 per cent and for infinity is 97 per cent. Increasing the cutoff from 2,000 to 3,000 cycles gives a growth in articulation which is 17/27, or .63, of the total increase in articulation which would be obtained in going to a cutoff of infinity. The values for the other curves of Figure 2 are obtained in a corre- sponding manner, it being appreciated that the articulation values with commercial instruments are lower than those for the Master Reference Circuit. This method of plotting the results has the advantage of showing the rate of growth of articulation for the three kinds of circuits on a comparable basis. 1.0 09 OB 07 0.6 Ob 0.4 0.3 02 0.1 2000 2400 2800 3200 3600 4000 '.UTOFF FREQUENCY OF LOW PASS FILTER IN CYCLES PER SECOND DESKSTAND SETS^ .^ -'liANDSETS .^ y ^ / ^ /* ^ l-^IRCUIT SIMILIAR TO ^/(ASTER REFERENCE SYSTEM f // 7 / f / / Ordinate is: Af - A: 2000 Am - Aa Where Af = the syllable ar- ticulation with a low pass filter of cutoff frequency f. A2000 = the syllable articula- tion with the 2000 cycle low pass filter. Am = the syllable articulation obtained with no filters. Fig. 2 — Syllable articulation of telephone systems at optimum received volume under quiet conditions. It is seen from the curves of Figure 2 that raising the upper fre- quency limit from 2,000 to 2,500 cycles gives about one-half of the total increase which would be obtained in going to an infinite cutoff and raising to 2,750 cycles gives for the commercial instruments about two-thirds of the increase in articulation which would be obtained in going to an infinite cutoff. These curves do not indicate any particular cutoff frequency as a stopping point for commercial circuits but it is considered that going as far as about 2,750 cycles is justified. While there is some articulation advantage in going further, observations of the number of repetitions occurring in conversations over circuits having different cutoff frequencies have indicated but little reduction in repetitions by going beyond about 2,750 cycles with commercial types of terminal sets. For the lower end of the range, the lower cutoff frequency curve of Figure 1 shows little effect on articulation of cutoffs below 400 cycles. 486 BELL SYSTEM TECHNICAL JOURNAL The selection of the 250-cycle point for the specified frequency range is on the basis of maintaining reasonable naturalness. It has been found that with present commercial station sets little is gained either in intelligibility or naturalness by extensions of the trans- mitted frequency range beyond the limits which have been set. This range, moreover, permits effective utilization, particularly from the standpoint of intelligibility, of the capabilities of much better station instruments even if this improved apparatus should approach the ideal in performance. With such terminal apparatus, major extensions beyond the upper frequency limits give improvements from the stand- point of naturalness largely as the result of better reproduction of the fricative consonants and of some of the incidental sounds which accom- pany speech. The^extension necessary to effect a material improve- ment in this respect is a matter of a thousand cycles or more, rather than hundreds of cycles. It has been considered that such an extension for message circuits is not now justified. Further in this connection, it must be borne in mind that an exten- sion of the transmission range will in general increase the amount of noise on the circuit and magnify the crosstalk problem. For trans- mission systems such as carrier and radio where the noise may be assumed to be uniformly distributed over the sideband range, the added noise may be particularly important. Also a widening of the range increases the difficulties of securing proper impedance balances and of equalizing amplitude and phase distortion. On the basis of these considerations, it has been decided that new designs of telephone message circuits for the Bell System should have an effective transmission band width of at least 2,500 cycles, extending from about 250 to 2,750 cycles. Furthermore, this band width will be made greater in those cases where this can be accomplished without material increase in costs. Some Recent Developments in Long Distance Cables in the United States of America By A. B. CLARK THE transmission history of long distance circuits, and particularly long distance cable circuits, has been one of continually improving standards. It has also been one of continual reduction of circuit costs. These have resulted largely from new developments to which have been added economies resulting from heavy growth and improved engineering. To put it another way, present-day circuits are capable of trans- mitting a kilocycle of frequency range more cheaply than those of earlier days. As the cost per kilocycle of band width has been reduced, part of the cost reduction has naturally been used in furnishing tele- phone customers wider-band and generally better telephone circuits. The accompanying chart is of interest in comparing the transmission frequency characteristics of what were considered good telephone cir- cuits some time ago with what are considered good telephone circuits today and what are proposed for the future. At the left of the chart are shown various types of circuits which have been in use or proposed for New York-San Francisco service, a distance of a little over 3,000 miles. The original loaded transcontinental line, which remained in service from January 25, 1915, until April, 1920, when it was unloaded, gave a band width of only about 900 cycles.* The non-loaded circuit was better, giving about 1,800 cycles. The modern carrier telephone circuit is better still, giving about 2,700 cycles. The extra-light loaded type of cable circuit (H-44, which has been the standard loading system for long distance use for some time) will give a band even wider, extending up to at least 3,000 cycles. At the right of the chart are shown typical characteristics for New York-Washington (about 225 miles) two-wire cable circuits with various loadings. The now obsolete heavy-loaded system, H-245, gave an effective range of 1,400 cycles, the medium-heavy loaded or H-174 gave 1,900 cycles while a new system which is being considered, called B-88, will give about 2,700 cycles. (At the present time H-174 two- * The limiting frequencies are taken as those at which the loss is 10 db greater than the loss at 1,000 cycles. 32 487 488 BELL SYSTEM TECHNICAL JOURNAL wire circuits are restricted to shorter lengths, the curve being given simply for comparative purposes.) In addition to this matter of frequency band width, there has been improvement in the 1,000-cycle efficiency of long distance circuits and also improvement with respect to noise and crosstalk. The original loaded transcontinental circuit, for example, gave, during good weather. NEW YORK TO SAN FRANCISCO 165- MIL I OPEN-WIRE NON-LOADED CARRIER 2 -WIRE LOADED CABLE — 19 GA. NEW YORK-WASHINGTON 10 -I400CYCLES- 1 H-245 i H-174 / B-86 1 f 1900 CYCLES 07nn r*V("i re / J J / 7 / / T 5 , / \ \ 1 1 \ i 1 1 1 1 1 0 \ ^ J. J / • / -5 1000 2000 CYCLES PER SECOND 3000 1000 2000 3000 CYCLES PER SECOND Fig. 1 — Transmission-frequency characteiistics of representative types of telephone circuits. a 1,000-cycle transmission loss of about 20 db with a variation from this of at least 10 db during bad weather. The non-loaded circuit gave about 12 db during good weather with smaller variations. With both of these circuits the noise was somewhat in excess of 1,000 noise units. The carrier and cable systems compare very favorably with non-loaded voice- frequency circuits in the matter of transmission loss and are much quieter. With the two-wire cable circuits shown, the H-245 circuit gave about 12 db loss at 1,000 cycles, the H-174 circuit 10 db loss and it is ex- pected that the B-88 circuit will give about 9 db loss at 1,000 cycles. All of the cable circuits are strikingly quiet as compared with older type voice-frequency open-wire circuits. The cable circuits are also considerably better from the standpoint of crosstalk. It is of interest to consider the effect on service of the change in standards of toll circuits as illustrated by the characteristics of the RECENT DEVELOPMENTS IN LONG DISTANCE CABLES 489 above circuits. One way of indicating this is by the repetitions occur- ring per unit time in commercial conversations. Assuming present commercial telephone instruments, typical terminal circuit and room noise conditions, following are some estimates on this basis: Repetitions per Circuit 100 Seconds Loaded New York-San Francisco circuit 3 Non-loaded New York-San Francisco circuit 2 Carrier circuit. New York-San Francisco 1 H-44 cable circuit. New York-San Francisco 1 H-245 cable circuit. New York- Washington lyi H-174 cable circuit, New York-Washington \% B-88 cable circuit, New York- Washington 1 Short Cable Circuits Consideration is now being given to giving up the H- 172-63 two-wire circuits in favor of B-88-50 and H-88-50 two-wire circuits. H-1 72-63 four-wire circuits were given up some time ago. With the new two- wire circuits the important line constants and circuit characteristics are given in the following table. H-88-50 loading is being considered particularly for those repeater sections less than about 40 miles in length while B-88-50 loading is being considered particularly for repeater sections whose lengths are greater than about 45 miles. For intermediate repeater section lengths the choice of loading will be dictated by various considerations appli- cable to the particular circuit layout involved. With either of the above two-wire circuits, the following transmission results are anticipated : Circuits for terminal business up to about 250 miles in length to have a working net loss at 1,000 cycles of about 9 ± 2 db. The frequency range to extend from about 250 cycles to some frequency between 2,750 cycles and 3,000 cycles. Crosstalk between circuits to exceed 1,000 units in only about 1 per cent of the combinations. Noise measured at the receiving end of the circuit, including "babble," * less than 200 units. Circuits for "via" business to be limited to lengths in the neighborhood of 100 miles so that adding a circuit link of this type to a built-up connection will not, in general, add more than about 2 or 3 db to the overall loss. * Babble is the name given to the effect produced by a number of different circuits crosstalking into a particular circuit at a given time and producing an unintelligible murmur. 490 BELL SYSTEM TECHNICAL JOURNAL Long Cable Circuits Present plans are to retain H-44-25 four-wire circuits for the inter- mediate and longer distances. This type of circuit is well known so it is unnecessary to go into its characteristics. With the idea of trans- mitting frequencies up to about 3,000 cycles with very little evidence of phase distortion, phase correctors are being considered for very long circuits of this type, say, circuits exceeding 1,000 miles in length. Characteristic 77-88-50 5-88-50 Conductor gauge No. 19 B. & S. No. 19 B. & S. Side circuit cable capaci- tance per mile 0.062 microfarad 0.062 microfarad Phantom circuit cable ca- pacitance per mile 0.1 microfarad 0.1 microfarad Inductance of loading coils on side circuits 88 milhenries 88 milhenries Inductance of loading coils on phantom circuits 50 milhenries 50 milhenries Spacing of loading coils 6,000 feet 3,000 feet Nominal cutoff frequency of side circuit 4,000 cycles 5,600 cycles Nominal cutoff frequency of phantom circuit 4,200 cycles 5,900 cycles Nominal velocity of prop- agation of side circuit 14,000 mi. per sec. 10,000 mi. per sec. Nominal velocity of prop- agation of phantom cir- cuit 15,000 mi. per sec. 10,600 mi. per sec. Nominal impedance of side circuit 1,110 ohms 1,560 ohms Nominal impedance of phantom circuit 670 ohms 940 ohms 1000-Cycle attenuation of side circuit at 55° F. 0.36 db per mile 0.28 db per mile 1000-Cycle attenuation of phantom circuit at 55^ F. 0.30 db per mile 0.24 db per mile RECENT DEVELOPMENTS IN LONG DISTANCE CABLES 491 Importance of High Velocity Circuits in Cable Echo suppressors have proven quite effective in reducing the echo effects on long distance circuits. For very long distance cable circuits, however, echo is still a matter for concern, particularly with losses held down to figures such as those given in the paper by Mr. H. S. Osborne entitled "A General Switching Plan for Telephone Toll Service." When cable circuit lengths become very great the actual delay suf- fered by the speech waves in traveling from end to end of the circuit becomes important quite apart from echo. We must look forward to the time when a subscriber in San Francisco will talk by cable across the United States to New York, then by cable and open wire to New- foundland, by submarine cable to England and then by a long cable circuit, let us say, to Constantinople; in other words, a 10,000-mile circuit length. The highest velocity long distance cable circuits in use today will give, for conversations over such a circuit, about >^-second delay in going from one end to the other so that when one subscriber speaks the other's reply cannot possibly reach him in less time than one second. This is quite a long time interval. By utilizing speaking tube delay circuits, connections have been set up involving delays as great'as this. Conversations are possible over circuits with such delays but the delay is a serious interference particularly when voice-operated devices are added which tend to lock out portions of the conversations. It is thus evidently important to seek higher velocity circuits. Telephone Carrier in Cable In seeking ways for obtaining high velocity circuits in cable in an economical manner, consideration has been given to the proposition of applying telephone carrier to long distance cables. For large groups of long distance circuits it appears likely that a carrier-frequency range can be advantageously used in cables, as wide or wider than the fre- quency range which has been exploited on open-wire lines. Experi- mental work on systems of this kind is actively under way at the present time. The higher frequencies involved together with the accompanying attenuations and increased coupling between circuits introduce some very interesting and unusual noise and crosstalk problems, as well as problems of equalization and maintenance of transmission stability. Also, there are interesting economic problems of conductor size and type, loading versus no loading, repeater spacing, etc. It is interesting to note that if non-loaded circuits are utilized, a velocity of transmission of about 100,000 miles per second would be 492 BELL SYSTEM TECHNICAL JOURNAL obtained while with loaded circuits a velocity perhaps half as great. The non-loaded setup in particular would, therefore, provide circuits whose velocity, like open-wire circuits, would be great enough to leave no question of obtaining satisfactory conversations over any world- wide telephone network. It is too early, however, to predict just what the outcome of this development may be. Phase Distortion in Telephone Apparatus * By C. E. LANE This paper shows that if, over its transmitting range, the phase shift, B, in radians, of a four-terminal network may be written S = Oo + Oico (w = 2-wx frequency in cycles per second), there is no phase distortion if ao = N, N being any integer. However, there is a delay, for any signal, given by dB/dw = fli (seconds). If N is not an integer, there is a delay, Oi, and in addition a distortion, which distortion, generally for speech and music, may be neg- lected. Typical phase characteristics for lines, filters and all-pass networks are shown. In general over their transmitting range, such phase charac- teristics which usually are curved, may be regarded as the sum of two characteristics, a straight line having a slope corresponding to the minimum slope of the original and which introduces a delay without distortion and a curved portion to which all of the distortion of the signal may be ascribed. Oscillograms are given showing the distortion for a loaded line and for band filters for a signal which is of the form y sin (wo^ + 0) between ^ = 0 and t = T and zero for all other time. A description is given of the means em- ployed for reducing the amount of phase distortion in telephone cable and in low-pass filters in circuits used for program transmission and regular tele- phone service. Also, phase distortion in repeaters and transformers is described. Brief reference is made to the problem of phase distortion in telegraph, picture transmission, and television circuits. '^"^HE effects of amplitude distortion in the transmission of signals -L has been taken into consideration in the design of telephone systems for some time. Recently ^ increasing attention is being given to the phase changes which waves undergo in the process of their transmission. The necessity of this is, on the one hand, due chiefly to the use of long distance telephone systems involving greater lengths of loaded cable and numerous filters and repeaters in tandem, and, on the other hand, due to the demand for improved performance. One place where better quality has become particularly desirable is in circuits for interconnecting broadcasting stations. This paper will present some general considerations of the relation between the phase characteristics ^ of telephone apparatus and signal distortion,'' show the types of phase characteristics that most fre- quently require consideration and discuss the manner in which the amount of phase distortion is controlled. Brief reference will also be * Presented at New York Section, A. I. E. E., May 1930. 1 At the end of this paper, a bibliography is given containing references to previous publications on this subject. ^ For a definition of phase characteristic see Appendix I. The "insertion " phase characteristic and "image transfer" phase characteristic are defined there. ^ A companion paper by J. C. Steinberg deals specifically with the effects of phase distortion on the quality of speech and music. Another companion paper by H. Nyquist and S. Brand treats of the measurement of phase distortion. 493 494 BELL SYSTEM TECHNICAL JOURNAL made to phase distortion in systems for transmitting other than tele- phone signals. Interpretation of Phase Distortion Telephone systems must be so designed that the received signal approximates in wave form the sent signal within limits found by exper- ience to be tolerable. We are here concerned primarily with the de- parture of the received signal wave from the sent signal * which may be attributed to the phase characteristic * of such networks as lines, filters, repeaters, etc. which go to make up the complete system. Such distortion is called phase or delay distortion. The reason for the term, delay, will appear later. We shall summarize here some of the more general conclusions of the effect on signals of certain phase characteris- tics and discuss the validity of these conclusions in Appendix II. If the phase characteristic of any network is of the type shown by any of the dotted lines in Fig. 1 the received wave will be an exact copy of the sent or reference wave (assuming no amplitude distortion). In the case of Fig. 2 the received wave differs from the reference wave only in that it is reversed in sign which is equivalent simply to reversing the terminals of the load. In both cases the received wave is delayed ivith respect to the reference wave by a time interval that is given by the slope of the phase characteristic or dBjdw. If B is in radians, and co = 2irf, where/ is the frequency in cycles per second, the delay will be in seconds. There is no distinction in effect between the phase char- acteristics for any of the dotted lines and furthermore they are identical with any such solid broken line as that shown. Since this is true we may completely represent any of the phase characteristics of Fig. 1 if we choose by the single line passing through zero in which case the delay is 5/co. If the phase characteristics are straight lines and intersect the verti- cal axis at odd multiples of 7r/2 the received wave may be obtained from the reference wave first by delaying it by dB/dco and then shifting the phases of all its steady state sinusoidal components as obtained by Fourier Integral or Series analysis by 7r/2. If the straight line phase characteristics intersect the ordinate at intermediate values the received wave may be said to be the sum of * See Appendix I. If one desires to be specific the sent or reference wave in the case of considering image transfer phase shift may look upon as the current entering the network and in the case of insertion phase shift the current through the load with the network omitted; and the received wave the current through the load in either case with the network in place. * In actual apparatus certain general types of phase characteristics are associated with certain attenuation characteristics. For this reason it is difficult to separate to the extent one might sometimes desire the effects of the two types of distortion, attenuation and phase. PHASE DISTORTION IN TELEPHONE APPARATUS 495 two parts as follows: The first part is an exact copy of the original wave modified in amplitude by a factor cos a^ and delayed by dB/do:. The second part is obtained from the original wave by shifting all of the (720O) (540°) (360») a: (I80">J O Z (0°} bJ < (-180°) a. (-360°) (-540°) (-720°) 4TT I O^ - ' ^ ' b^"'' ,/- 3lT - ^ -- r""^^ c ,--- --- ^- ^•' 2tT .^--' ^' "'' d-- ^-^ -^'" TT - ^^ ^ ^^ -' " e ,- 0 ^'-^ , ,^^^^ , -^ ^^ i-"" f= ^^,^^ -IT - ^ ^ ' ^ --' 2tt '- ^^^ "^ 3tt - -- "^ 4TT ,-^ j = 2irf Fig. 1 — ^Phase characteristics which introduce no distortion. (720°; 4 IT - (-540°J -3tt fV20") -4tT - (.i=2TT-r Fig. 2 — Phase characteristics which introduce no distortion equivalent to those of Fig. 1 with connections to the load reversed. components of the original by 7r/2, multiplying the result by sin ao and then delaying by dBldoi. ao is the value of the angle at the point of intersection of the vertical axis. 496 BELL SYSTEM TECHNICAL JOURNAL An important point to note here is that if a given signal has all its important frequency components falling in a region between /i and /o either because of the nature of the signal or as a result of attenuation in the system we are only interested in the phase characteristic in that region. Thus for such a signal a sufficient condition for negligible phase distortion is that the phase characteristics be like those in Fig. 1 and Fig. 2 between /i and fx only. The phase characteristics actually found in telephone apparatus that frequently must be considered may for convenience be classified as follows: (1) those typical of the low pass filter and the loaded line, 180 160 - 0.002 < _j UJ o 0.001 0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 FREQUENCY IN CYCLES PER SECOND Fig. 13 — Insertion delay characteristics corresponding to phase characteristics of Fig. 11. TOTAL DELAY , :^ L-^ STH OF CAB _E F OR = B0( iBA^ t C^F DEL AY OF 5 0-M ILE LEN ^Ric :tor DEL AY C ^F PHASP /-^ ■~~- -^ single phase corrector consists of eleven sections, seven of one kind, three of another, and one of a third. Each section contains one two- terminal inductance coil and one three-terminal coil with mutual between the two windings and also two condensers. Fig. 13 gives '^ This cable consists of No. 19 gauge copper wire with 44 millihenry loading coils spaced every 6000 feet. 506 BELL SYSTEM TECHNICAL JOURNAL dBjdoi for the cable, the phase corrector, and the two combined. A 50 mile length of corrected cable gives a delay of .00375 seconds. A 3000 mile length " of this cable such as would extend from coast to coast would give a delay of .225 seconds, with a difference between the 36,000 32,000 28,000 24,000 (0 lU Ul a. o Q 20,000 I W 16,000 tu < I Q. 12,000 8;000 4,000 // / / k /y // w K- /'J w i- // 9 ^ >p ' 4 v^ •v^ :> /y ''/. // % w J A r / / A/ A >" pHA' iS-^ -i\rr_ OF „ — y~-r/-iD GORP f- ": ■ ' ^^ 400 800 1,200 1,600 2,000 2,400 FREQUENCY IN CYCLES PER SECOND 2,800 3,200 Hg. 14 — Insertion phase characteristics of a 500 mile length of cable for telephone message service. minimum and maximum value of .006 seconds in the corrected range. Without correction this difference would have been .055 seconds. Figs. 14, 15 and 16 correspond respectively to the previous three but are for a 500 mile length of the cable for regular telephone message service. The total delay for 500 mile length of this cable after correc- '' In designing apparatus going in long distance circuits in general the parts are so designed that if a circuit 3 or 4 thousand miles long is used the total accumulated distortion due to either amplitude or phase will be within tolerable limits. PHASE DISTORTION IN TELEPHONE APPARATUS 507 tion is .029 seconds. A 3000 mile length would give .174 seconds delay. The difference after correction for 3000 miles between minimum and maximum value of dB/dco is .007 seconds and before correction .035 seconds. This phase corrector (for 500 miles) consists of 12 sections, 8 of one kind and 4 of another. Each section contains four condensers and two four-terminal inductance coils with mutual between windings. Both this phase corrector and the previous one are formed by connect- ing together such all-pass network sections as to give the phase char- acteristic desired. In the first bridge T-sections are used and in the second lattice type.^~* The former are more economical when unbal- anced apparatus may be used, though similar phase characteristics may generally be obtained with either. Fig. 15 — Schematic of the phase corrector for 500 mile length of cable for telephone message service. 2. Filters. — The following factors influence the phase distortion in filters: (1) The width of the frequency band transmitted, (2) the amount of discrimination between transmitted and attenuated regions (cor- responds to number of filter sections), (3) the rate at which the atten- uation rises at the edges of the transmitting band, (4) the types of filter sections used, (5) the number of filters in tandem, (6) the amount of reflection due to impedance mis-match near the edges of the trans- mitting bands, and (7) the amount of dissipation in the filter elements. The insertion phase characteristics of Fig. 17 and the insertion delay characteristics of Figs. 18 and 19 are for two low pass filters ^^ of the usual type. As will be seen from their attenuation characteristics (insertion loss) each gives a discrimination of about 35 db, although the second requires an additional section in order to provide the rapidly '^ Nyquist, U. S. patents Nos. 1,675,460 and 1,735,052 and Zobel patent Xo. 1,701,552; Maximum Output Network for Telephone Substation and Repeater Cir- cuits, by G. A. Campbell and R. M. Foster, Trans. A. I. E. E., Vol. 39, pp. 231-280. ^^ This note explains symbols used in these three figures and also the following three. Zj is the image impedance. Zo for a low pass filter is the value of Zj at zero frequency and for a high pass filter at infinite frequency. Q is the ratio of the coil reactance to its effective resistance. Dissipation in the condensers is considered negligible. For a filter section having an attenuation peak at frequency, /«,, and a cut-off at, fc, "a " is the ratio /„//c for a low pass filter and/e//„ for a high pass filter. 508 BELL SYSTEM TECHNICAL JOURNAL rising attenuation at the edge of the band. These curves show that the delay distortion in the transmitting band is increased by increasing the slope of the attenuation curve at the cut-off although the minimum value — i.e. the delay which applies to the signal as a whole, does not increase appreciably. The effective band width transmitted depends upon both the delay and attenuation characteristics since especially 0.032 0.030 0.028 0.026 0.024 0.022 ^ 0.020 o z O 0.018 o UJ 0.016 ^ 0.014 _I UJ ° 0.012 0.010 0.006 0.006 0.004 0.002 TOTAL DELAY rE b^ ^OM^ ^''■ ,p.g£ ,-1^11-' : LEN GTH C 3F CABUE f OB'f EUEP^ De .AY C ^.0. ^ ^y 0 P PH> i£2o/ =? 400 800 1,200 1,600 2,000 2,400 FREQUENCY IN CYCLES PER SECOND 2,800 3,200 Fig. 16 — Insertion delay characteristics corresponding to the phase characteristics of Fig. 14. for a number of filters like these in tandem the delay of the frequency components of the wave near the cut-off may be so great that these components contribute little to articulation. Therefore in the design of filters a proper balance must be determined between the rate of attenuation and the delay distortion. A more complete discussion of the relation between delay, attenuation, and the effective cut-oft' is given in the paper by J. C. Steinberg.-" When low pass filters are to be designed with sharper cut-offs from the standpoint of both delay and attenuation, there are two ways in which this is usually done, one 20 Loc. cit. PHASE DISTORTION IN TELEPHONE APPARATUS 509 600 550 500 450 in (ij iiJ 400 cc o UJ Q 350 I- 300 I ^ 250 UJ 200 150 100 50 0 r L| 3M1 L2 cm ^ 1-3 L| 00( n Q = 200 O- l.25f- F^ ■^000 H - I,25f^ — 1 (X 0.= NO DISSIPATION 1 SCHEMATIC (A) CONSISTING OF ONE 1^ (X SECTION, ONE a = l.25 SECTION AND ONE a = 1.05 SECTION ,/^ \ >-l L| / /- ^^ =^5 -- // 7 / / SCHEMATIC (B) SAME AS (A) WITH 1.05 SECTION OMITTED / > / TERMINATED IN Zi ^ ^ o/^ / -- _A vN^ l >^ ^ ^ "^^K a^^ 0.2 0.4 0.6 0.8 1.0 FREQUENCY (-fj 1.2 1.4 1.6 F"ig. 17 — -Insertion phase characteristics and schematics of low-pass filters. (/I) For a filter consisting of 3 full sections, one section having no attenuation peak, one with a peak at 1.25/c and another at 1.05/c. (5) Same as (/I) with the 1.05 section omitted. 10 7 - «i|3 >- < _i UJ Q fl J 1 \ r200 -80 ) DISSIPATION r / \ ■s^ NC V / 1 \, \ \ '' z o 1- ~ < D 3- re DEC. \ UJ h- y 1 1 o£^ ■^ y I ' J 0.2 0.4 06 0.8 FREQUENCY (-^) 1.0 1.2 50 45 40 Z 30 - Z o < D 20 Z UJ \- 10 1.6 Fig. IS — Insertion delay and attenuation characteristics for filter .1 of Fig. 17 510 BELL SYSTEM TECHNICAL JOURNAL is the use of a few filter sections of a type different from the usual types above having phase characteristics with a negative second derivative rather than positive so that the combination will postpone the occurrence of phase distortion until very near the theoretical cut- off and the other is the addition of all pass network sections which accomplish the same general result. O ^ = 200 ) DISSIPATION / S ^, NC / / / / / o ' ~ i ' I / 2 * V / ' 1 DEL£1J„- ^ / 0.2 0.4 0.6 0.8 FREQUE 1.0 1.2 1.4 ncy(^) 50 45 40 35 -° 30? 25 Z o I- < 20 i I- 15 5 10 5 0 1.6 Fig. 19 — Insertion delay and attenuation characteristics for filter B of Fig. 17. One other point should be noted here. The shape of the insertion phase curve as shown at the cut-off frequency owes its departure from the image transfer phase shift without dissipation shown by the dotted line more because of the reflection than dissipation. The value of the Q so long as it is within the usual range makes little difference. Figs. 20, 21 and 22 correspond to those of 17, 18 and 19 but are for high pass filters. High pass filters introduce no initial delay to signals as a whole. The distortion of the signal is dependent upon sharpness of cut-off, Q etc. just as for low pass filters. Band pass filters give an initial delay defined by the shape of the phase curve. Other factors remaining the same this delay as well as the amount of phase distortion is inversely proportional to /o — /i the band width in cycles and is independent of the position of the band on the frequency scale. The effect of reflection, dissipation, sharpness of cut-off, etc., are about the same at the lower cut-off as for a high pass filter and at the upper cut-off as for a low pass. As already noted one distinguishing feature of the phase characteristic of a band pass filter PHASE DISTORTION IN TELEPHONE APPARATUS 511 FREQUENCY 1.2 1.4 Fig. 20 — Insertion phase characteristics and schematics for high-pass filters. {A) For a filter consisting of 3 full sections, one section having no attenuation peak, one with a peak at 1/1.25/c and another at 1/1.05/c. (S) Same as {A) with the 1/1.05/c section omitted. 10 <3. 3 6 ■01 T> u -^ 5 >- < -I 4 0.6 0.8 / \ ■* X Y \ ' \ t Q = 200 a = 8o NO DISSIPATION 1 /- 5 1 -I m z -c - > \ \ V \ \ \ •^^ W 'o^. — 1.2 1.4 FREQUENCY (^c) 1.6 1.8 2.0 50 45 40 35 30 25 20 15 10 2.2 Fig. 21 — Insertion delay and attenuation characteristic for filter .4 of Fig. 20. 512 BELL SYSTEM TECHNICAL JOURNAL is that the straight portion of the phase curve may if extended intersect the vertical axis at any point and does not like the low pass filter pass always through Ntv. As an example of a condition where it has been found necessary to take phase into consideration in designing low pass filters let us con- sider the case of line filters used in open wire circuits transmitting simultaneously both programs for broadcasting and carrier telephony. Here as many as 40 or 50 filters may be used in tandem. Fig. 23 shows the measured delay and attenuation characteristics i 3 "olx >- < u 9 8 7 / \ ^ / \>, = aoo 3 DISSIPATION 6 5 4 3 2 ' N( \ ft O V V \ ^ \ \ % \ ^^vL^ () 50 45 40 35 30 25 20 15 10 5 0 z o 1- < D Z UJ I- I- < 0.6 0.8 1.0 1.2 1.4 FREQUENCY 1^^) 1.6 1.8 2.0 2.2 Fig. 22 — Insertion delay and attenuation characteristic for filter B of Fig. 20. for 25 of the present 5000 cycle quality line filters now in use, these filters being connected in tandem. The circuit is so designed that it practically equalizes up to 5000 cycles for the attenuation distortion. When a number of these filters are used in tandem as in the longest "hook ups" the phase distortion of these filters is somewhat noticeable but not seriously so and their effective cut-off is appreciably lowered because of this phase distortion.'^ Fig. 24 gives the calculated delay and attenuation of twenty- five 8000 cycle low pass line filters connected in tandem. This filter is being considered for use in place of the 5000 cycle line filter of Fig. 23. The delay for 25 of these filters in tandem is ap- proximately constant up to 7500 cycles (within .001 seconds). The attenuation is also constant up to this frequency. The attenuation -' J. C. Steinberg, loc. cit. PHASE DISTORTION IN TELEPHONE APPARATUS 513 distortion between 7500 and 8000 cycles is purposely left uncorrected so that if these filters are used when a number of them are connected in tandem on a line, the effective cut-off will be lowered therefore eliminating the effects of any accumulating delay over this range. Thus for short distances 8000 cycle quality will be realized, and for very long circuits 7500 cycle quality, and for intermediate lengths something in between. In any case the effects of delay distortion will be negligible. Special filter sections are used in order to meet the unusual delay and attenuation requirements. The same results could L| I -2 SCHEMATIC 1-3 OF FILTER ■-4 L5 R| lO— ^005 Cu ■ vv c l_ - §1-6 |l7 C2^ C3^ C4: §1-6 k^i T V V V V ^~> OLq ?- g^S 125 Ll 1-2 "-3 L4 L5 ■i— ^AA/v-04 R| 1 100 / / 75 / / / 50 ^ / / 1 0 y' / - " / 25 1 ATI ON y "AT 0 ... 0.020 0.016 m a z 0.012 8 m 0.008 > < -I UJ a 0.004 1,000 2p00 3,000 4,000 5,000 6,000 7,000 FREQUENCY IN CYCLES PER SECOND 8p00 9,000 Fig. 23 — Schematic, insertion delay and attenuation characteristics of twenty-five 5000 cycle low-pass line filters connected in tandem. have been obtained using the usual type sections like those used in the 5000 cycle filter and then correcting for phase by an all-pass structure. Such a method would have resulted in a somewhat more expensive filter and one giving more overall delay. 3. Repeaters. — The chief sources of delay in telephone repeaters are the transformers. However, some additional low frequency delay is caused by shunt retard coils and series condensers. This is kept within negligible limits by using large values of both. Conversely inductance in series and capacitance in shunt cause high frequency delay but this effect can easily be made negligible in any frequency range in which one is interested. Fig. 25 is a schematic of the telephone repeater 514 BELL SYSTEM TECHNICAL JOURNAL used in the cable circuit previously referred to for program transmis- sion. A repeating coil and input transformer at its input and the out- put transformer at its output are shown. These will be discussed in the next paragraph. 4. Transformers. — As an example of phase distortion in transformers we shall consider those shown in the telephone repeater of Fig. 25. Here, the phase shift of a small number of transformers would be of little importance but where a large number are connected together as in long toll circuits their effect cannot be overlooked. The delay caused by a transformer is similar to that of a high-pass filter. The 140 120 100 80 2 O < Z u I- 60 40 20 h / 1 — . DELAY ■^ y ^^ i f kTTE NUA TlO^ i J 0.028 0.024 0,020 If) Q Z o 0.016 o 0.012 >- < 0.006 0.004 UJ O 1000 2000 3000 4000 5000 6000 7000 FREQUENCY IN CYCLES PER SECOND 8000 9000 Fig. 24 — Insertion delay and attenuation characteristics of twenty-five 8000 cycle low-pass line filters connected in tandem. insertion phase characteristics of these three transformers between impedances for which they were designed are shown in Fig. 26 and their insertion delay characteristics, dB/dco in Fig. 27. It will be noticed that practically all of the delay occurs below 100 cycles. The three together give at 40 cycles a value of dBjdi^ of .0008 seconds. 25 sets would give .020 seconds delay. Experience has shown that this amount of delay for speech at low frequencies is negligible whereas at high frequencies such would not be the case.-'^ 5. Attenuation Equalizers. — Attenuation equalizers introduce some phase distortion but the amount is so small that it can generally be -- At high frequencies {dB/dw)aiax. — {dB/dui)^^. must generally be kept under .005 to .010 seconds if its effect can be neglected for speech. PHASE DISTORTION IN TELEPHONE APPARATUS 515 UJ a. Hii' ^im^ P — r"i_iw.A^^ c o in C a ^=" rt o o o u 'o T3 in a o o CD 516 BELL SYSTEM TECHNICAL JOURNAL neglected. The presence of such equalizers in the circuit for program transmission did not influence the design of the phase correctors. However in considering the design of the equalizers the particular structure used was chosen on the basis of its giving a minimum amount of phase distortion. Phase Distortion in Other Communication Systems Phase distortion has for some time been considered a real problem in submarine cable telegraphy. If the highest reversal frequency of a telegraph signal is /„ it has been found expedient to correct for both 14 12 10 in UJ LU a. o LU Q Z I IT) in -2 < I a -4 -10 1 .__ — =--= / ^^"^ "^ . — — — — ' "" / / * / / / 1 k .-y A 6^ ^ Iff"" c^ / ^-,9-^ ^ ' \^ 200 400 600 600 ipOO ipOO 3,000 FREQUENCY IN CYCLES PER SECOND 5.000 7.000 9p00 Fig. 26 — Insertion phase characteristics of repeating coil, input transfornier and output transformer used in repeater shown in Fig. 25. amplitude and phase distortion over a frequency range of from zero to between 1.4 and 1.6 /„. More phase distortion can be tolerated when the signal is read by the operator from a siphon recorder slip than for automatic printing. In the case where fv is 60 cycles per second a 45° departure (at 60 cycles) from the low frequency character- istic if continued as a straight line it has been found in some cases to cause errors to be printed. While no two telegraph cables are alike all except very short ones require that some means of phase correction be incorporated in the terminal apparatus. Although the general principles of correction have been investigated mathematically and the results have been very useful as a guide in indicating the relative effects of factors involved, PHASE DISTORTION IN TELEPHONE APPARATUS 517 the practical solutions so far have been largely empirical. This is due to the comparative simplicity of the experimental method. The results of a number of adjustments in the terminating apparatus may be observed experimentally while one is investigated mathematically. This is due to the fact that one can tell, using an oscillograph, from the direct observation of the received signal when a proper adjustment is reached. It is not necessary, as in telephony, to make complete 0.0007 0.0006 0.0005 to o z O 0.0004 U UJ to 0.0003 < _l UJ Q 0.0002 0.0001 . _ OUTPUT ' TRANSFORMER *" ! t 1 1 1 REPEATING "- COIL A- INPUT , TRANSFORMER Vr' \ A 200 400 600 800 ipOO ipOO 3p00 SpOO FREQUENCY IN CYCLES PER SECOND 7p00 9pOO Fig. 27 — Insertion delay characteristics corresponding to the phase characteristics of Fig. 26. articulation tests or a large number of single frequency measurements. The advent of the continuously loaded cable made possible by the use of permalloy has simplified the problem. A discussion of detailed methods of phase correction in telegraph cable cannot be gone into here. Circuits in general use have been described in previous publica- tions.^^ Two other important places where it has been necessary to control phase distortion are (1) in circuits for picture transmission '-■* and (2) -'•' The Loaded Submarine Cable, O. E. Buckley, Bell Sys. Tech. Jour., Julv, 1925; High Speed Ocean Cable Telegraphy, O. E. Buckley, Bell Sys. Tech. Jour., April, 1928 ; The Application of Vacuum Tube Amplifiers to Submarine Telegraph Cables, A. M. Curtis, July, 1927; Automatic Printing for Long Loaded Submarine Telegraph Cables, A. A. Clokey, Bell Svs. Tech. Jour., Julv, 1927; J. R. Carson (U. S. Patent 1,315,539) and R. C. iMathes (U. S. Patent 1,311,283). -■•Transmission of Pictures over Telephone Lines, H. E. Ives, J. W. Horton, R. D. Parker and A. B. Clark, Bell Sys. Tech. Jour., April, 1925. 518 BELL SYSTEM TECHNICAL JOURNAL television circuits.^^ The necessity in both cases is similar. In picture transmission the frequencies between 900 and 1700 cycles were in- volved and a maximum departure for the system as a whole of dB/doi from a constant value of .0005 seconds in this frequency range was found permissible. In television it was considered desirable to transmit frequencies over the range from 10 cycles to 20,000 cycles such that the value of dBjdw as a function of frequency was constant about ±.00002 seconds over all but the lowest part of the range. There at the very lowest frequency ±.001 seconds was considered permissible. Bibliography Carson, J. R., and Zobel, O. J. Transient Oscillations in Electrical Wave Filters, Bell Sys. Tech. Jour., July, 1923. _ Carson, J. R. Building Up of Sinusoidal Currents in Long Periodically Loaded Lines, Bell Sys. Tech. Jour., Oct., 1928. Clark, A. B. Telephone Transmission for Long Cable Circuits, Bell Sys. Tech. Jour., Jan., 1923. Decker, Hans. Dampfungsentzerrung und Phasenverserrung, E. N. T., Vol. V, 1928. KiJPFMULLER. Die Erh6hung~der Reichweite von Pupinleitungen durch Echosper- rung und Phasenausgleich, E. N. T., Vol. Ill, 1926. KiJPFMULLER AND Mayer. Uber Einschwing Ver Gauge in Pupinleitungen und ihre Verminderung, Wiss. Ver. a. d. Siemans-Konzern, Vol. V, 1926. KiJPFMULLER. iJber Beziehungen Zwischen Frequenzcharakteristiken und Aus- gleichsvorgangen in Lineoren Systemen, E. N. T., Vol. V, 1928. LiJscHEN. Die Technik der Telegraphie und Telephonie in Weltverkehr, E. T. Z., July, 1924. Mead, S. P. Phase Distortion and Phase Distortion Correction, Bell Sys. Tech. Jour., Apr., 1928. Sandeman, E. K., Rendall, A. R. A., and Turnbull, L L. Phase Compensation, Elec. Comm., Vol. VII, Apr., 1929. Zobel, O. J. Distortion Correction in Electrical Circuits, Bell Sys. Tech. Jour., July, 1928. Appendix I Phase Shift Defined Networks with a pair of input terminals and a pair of output terminals such as lines, filters, equalizers, phase correctors, transform- ers, etc., are designed to work between a source of e.m.f. £„, of impe- dance Za, and a receiving device of impedance, Zh. The source of e.m.f. is generally spoken of as the generator and the receiving device as the load. Such a network, N, connected to a generator and a load are shown in Fig. 28. Terminals 1 and 2 are the input terminals and 3 and 4 the output terminals. For any frequency let Za = Wa be the image impedance -^ at ter- ^^ Production and LItilization of Television Signals, F. Gray, J. W. Horton and R. C. Mathes, Bell Sys. Tech. Jour., October, 1927; Wire Transmission Systems for Television, D. K. (iannett and H I. Green, Bell Sys. Tech. Jour., October, 1927. -^ The image impedance at either end of a four terminal network is given by the square root of the product of two impedances at that end, one measured with the opposite end short circuited and the other with it open circuited. PHASE DISTORTION IN TELEPHONE APPARATUS 519 minals 1 and 2 and Wh at 3 and 4. For these terminating conditions let the input current be /„' and the output current h' . The image transfer constant, 6, of the network then is la' Let d = log. d = A' +jB' (1) (2) lo- ll— Fig. 28 — Network connected between a generator of impedance Z^ and a load of impedance Zj. A' is the real part of the image transfer constant and B' is the ima- ginary part. We have h' Wa' W,' = e' = e^'\B\ (3) B' is the image transfer phase shift of the network.^^ Its value as a function of frequency gives the image transfer phase characteristic of the network. There is another type of phase shift of more frequent use. Let h be the current through the load before insertion of the network, i.e., when the generator and load are connected directly together or con- nected together by means of an ideal transformer of the best turns ratio. Let lb" be the current through the load with the network in place as shown in Fig. 28. Then h JT,= t \B' (4) B" is the insertion phase shift?"^ It will be noted that when the terminating impedances are the image impedances that B' and B" are the same.-^ In most practical cases the phase shifts as defined above are deter- mined for pure resistance terminations hence the phase shifts may equally well be said to relate the applied voltage to the received cur- rent. The angle of lag of the received current is regarded as positive. -^ 20 logio eA' gives the image transfer loss in decibels. ^* 20 logio eA" gives the insertion loss in decibels. ^' The term insertion loss and insertion phase shift is here extended to include cases where apparatus is designed to work between resistance impedances of different values. 34 520 BELL SYSTEM TECHNICAL JOURNAL Appendix II A nalytical Discussion of Phase Characteristics By means of the Fourier Integral any signal or wave whatever may be regarded as the sum of an infinite number of steady state sinusoidal frequency components which have existed and will exist for all time. Their amplitudes are infinitesimal and they are separated by dif- ferentially small amounts in their frequency spectrum. The finite wave is the sum of the infinitesimal components and is determined by their relative amplitudes and phases. The general Fourier Integral for the wave /« may be written la = S y cos {(Jit + d) dco, (5) where co = 2irf, f being the frequency, y and 9 are functions of oo. If the amplitude is altered by a constant factor at all frequencies there is no amplitude distortion. For the purpose of discussing phase dis- tortion we shall assume this factor unity. Let the angle be modified by any network by an angle B which is a function of frequency. The expression for the received wave will then be h= f y cos (co/ + 0 - B) doi. (6) Let us assume a simple case where B is proportional to frequency, i.e. B = aico. (7) Then lb = S y cos (co/ -\- Q — a\ >K 1 2 3 4 K = CHARACTERISTIC IMPEDANCE OF SYSTEM I'ig. 4 — Modified arrangement for special impedance measurements. (Open termination.) Impedance measurements are made as before with the following changes as shown in Fig. 4: (1) X/2 is bridged across the measuring trunk in place of K; (2) instead of having the output of the network short-circuited or open, the output is bridged on the input. This case corresponds somewhat to the one above in which the open termination was used on the output of the system, with the exception that the return current now traverses the system only once in its complete trip from the bridge back to the bridge and, consequently, is attenuated and delayed only one half as much as before. These impedance curves are quite similar to those obtained above and may be interpreted in a similar manner. In this case, the envelope delay of the system in seconds at the frequency {J2 +/])/2 cycles is, approximately. T. 1 2(/2-/i)' where (/2 — /i) is the length in cycles per second of one half an im- pedance cycle of an impedance curve. Fig. 5 shows the computed impedance curves resulting from measurements made in this way on the same 100-mile unit of phase corrector and also gives the correspond- ing envelope delay-frequency characteristic of this network. The case just described corresponds to the former case with the open end termination; that is, the results obtained are the same as those which would be obtained from the former case by using only one half " D. K. Gannett, U. S. Patent 1,725,756. 532 BELL SYSTEM TECHNICAL JOURNAL of the system terminated at the far end in an open circuit. Impedance curves 180° out of phase with those obtained by the last method de- scribed can be obtained by using what is equivalent to one half of the system terminated at the far end in a short circuit. The circuit ar- 3.0 2.5 « 2.0 o z o o liJ 1.5 <2 _/ _i 2 1.0 0.5 •~v, / \ S^ ENVELOPE DELAY / y \ s^ 300 X o 200 100 \, ^^ RESISTANCE y^ \ \ \ / 1 ^ \ \ \ J \J ' \ + 200 + 100 10 I 0 o -100 -200 REACTANCE N, / -^ / 1 / A / / \ / / \ / \ ^ / y J \y 400 800 1200 FREQUENCY 1600 2000 Fig. 5 — Computed curves illustrating method of Fig. 4. rangement for such measurements is given in Fig. 6. When the system being considered is balanced, the connections can be made as shown on the left. However, when the system is completely un- balanced, for example, a network built with no apparatus in one side I MEASUREMENT OF PHASE DISTORTION 533 so that all points in this side of the network are at the same potential, the measurements may be made with the arrangements shown on the right of the figure. In the latter case, the measured impedance is four times that obtained for an equivalent system by the method shown on the left of the figure; and in plotting these values for comparison with those obtained by the method of Fig. 4, one fourth of the measured values should be used. The results obtained in this way are evident from what has already been described and will not be illustrated here. BALANCED SYSTEM UNBALANCED SYSTEM SYSTEM TO BE MEASURED SYSTEM TO BE MEASURED U iij9 2ID K J K UJ o LJ Q 1 (O K= CHARACTERISTIC IMPEDANCE OF SYSTEM Fig. 6- — Modified arrangement for special impedance measurements. (Short termination.) As before, the curve passing through the points of intersection of the impedance curves corresponding to the open and short terminations should be used as the axis in determining the length of impedance cycles. For the network illustrated in Fig. 5, the resistance curves would intersect on the line +150 ohms, while the corresponding re- actance curves would intersect on the zero line. The delay obtained in this way is the delay of the system between characteristic impedances. When other impedances are bridged across the measuring trunk than those shown in the figures, the delay measured is the insertion delay between impedances having the same relation to the actual bridging impedance as K has to the value of bridging impedance shown in the figures. In most practical cases, the characteristic impedance of the network to be measured is a pure resistance and the network is designed to work between this impedance at each end. The value of resistance which 534 BELL SYSTEM TECHNICAL JOURNAL should be bridged across the measuring trunk to measure the delay of the network under these conditions is obvious from the above discus- sion. Sufficient accuracy can often be obtained by considering only the impedance curve obtained for either the open or the short termina- tion and using the axis of this curve for determining the length of the impedance cycle; e.g., in Fig. 5 (which gives curves corresponding to the open termination) using the lines +150 ohms and zero as the axes of the resistance and reactance curves, respectively. b. Determination of Envelope Delay from Phase Shift Measurements Methods will now be briefly described for determining the envelope delay of a system from special measurements of the steady-state phase shift; that is, the difference in phase shift for a definite frequency dif- ference or the difference in frequency for a certain difference in phase shift will be measured and the delay computed therefrom. Three measuring methods will be considered in which the fundamental principles involved are much the same. Practical circuits for measur- ing delay by these methods may be rather complicated from an ap- paratus standpoint in order to facilitate the measurements as much as possible. The details of these circuits are not given here, but they are disclosed in various patents. ^■-' '^' ''* The following gives the essential principles on which these methods of measurement are based: The current of the measuring frequency from an oscillator traverses two separate paths and is then combined at the receiving end of the measuring circuit. In the first path no phase distortion is introduced, while in the second the frequency is transmit- ted through the system of which the delay is to be measured. Both paths contain resistance attenuators, so that the magnitudes of the currents received from the two paths may be adjusted as desired. If now a frequency is chosen such that the phase shifts through the two paths cause the two received frequencies to be exactly out of phase, an adjustment can be made so that an observer listening with a receiver to the combined received currents will hear no tone when the two received currents are equal in magnitude. If the frequency is now changed continuously, the observer will hear the tone increase and then decrease again to zero when the two received currents are again exactly out of phase, care being taken that the two received currents are kept equal in magnitude. This means then, that for this difference in frequency, the phase shift in the system under measurement has changed a complete cycle. From what has gone before it is simple to calculate ^^ U. S. Patent 1 596 941. " h'. Nyqiiist and H. A. Etheridge, Jr., U. S. Patent 1,596,942. "S. B. Wright and K. W. Pfleger, U. S. Patent 1,596,916. MEASUREMENT OF PHASE DISTORTION 535 the approximate value of the envelope delay for the system under con- sideration. In the three patents referred to, circuits for different purposes are given. In all of these methods both ends of the system to be measured must be available to the tester. The first describes a circuit in which the method of measuring is quite similar to that just described except that by means of a reversing switch the frequency interval is found corresponding to a change in phase shift of one half cycle. This method is suitable for measuring relatively large delays. The second circuit referred to is much the same as the first except that a definite phase shift can be introduced in the path which contains the system under consideration by means of a reactance inserted between two artificial resistance lines of considerable length. The method of opera- tion is exactly the same as before except that here frequency intervals can be measured for changes in phase shift which are not integral multiples of one half cycle. This method is suitable for measuring much smaller values of delay than the first circuit referred to, but is not particularly suited to measuring very small delays. The third method is adaptable to measuring very small values of delay, such as those introduced by separate units of equipment. Here a phase shifter is introduced in the path containing the system under consideration and the change in phase shift through the system for a particular frequency interval is measured. The phase shifter for this purpose should be continuous in its operation ; and in the circuit referred to, the relative phases of the received currents from the two paths are compared by means of a vacuum tube device which indicates a zero condition on a meter when the two received currents are in quadrature. c. Direct Measurement of Envelope Delay Here the phase shift of the envelope of a modulated wave is measured under steady-state conditions and this gives a direct measurement of the envelope delay when the measuring set is properly calibrated, inas- much as the delay of the envelope of the modulated wave is closely related to the differences in phase shift for the component frequencies of the modulated wave transmitted. Before describing the details of the measuring circuits, some of the principles underlying the trans- mission of simple modulated waves will be considered; and for this purpose envelopes produced by sine wave modulations will be used. It is assumed in this discussion that the modulations in the transmitted current are repeated periodically and that the attenuation of the system used for transmission is completely equalized for all the frequency components. 35 536 BELL SYSTEM TECHNICAL JOURNAL Consider a 1000-cycle sine wave which is modulated by a 25-cycle sine wave in such a manner that the envelope just reaches zero once per cycle of the modulating wave. This wave is found on analysis to consist of three components, namely, 1000 cycles of two units amplitude, 975 cycles of one unit amplitude, and 1025 cycles of one unit amplitude. At the start, it is somewhat simpler to consider this case with the 1000-cycle component removed. In other words, the current transmitted through the system now consists of 975 and 1025 cycles in equal amounts. This value at the sending end may be conveniently written sin 975 2^ + sin 1025 2^t. The equivalent graphical expression is 2 cos 25 lirt sin 1000 lirt. Now suppose that the 975-cycle current suffers a phase change of (3975 during transmission and that the 1025-cycle current suffers a phase change of /3io25, then the analytical expression for the current at the receiving end is sin (975 livt - ^^n) + sin (1025 27r/ - ^xo2h). The corresponding graphical expression is n I oe T^ /3i025 "~ ^975 \ . / inAA^T~^ ^1025 + /3975 2 cos 25 27r/ ^ sm 1000 27r/ ^^ In comparing the graphical expressions for the current at the sending end and the current at the receiving end, it is apparent that the only changes that have taken place are phase shifts of the 1000-cycle carrier wave and of the 25-cycle modulating wave. The phase shift of the 25-cycle modulating wave represents the actual delay of the deforma- tion of the carrier wave. If the circuit is sufficiently long so that this phase shift amounts to one complete cycle, then the corresponding delay equals one period. For any other delay, the phase shift and de- lay are, of course, proportional. It will be apparent, therefore, that the delay may be represented by the following equation: ^ 2 X 25 (27r) Aoj ' where 7a is expressed in seconds, the numerator in radians and the denominator in radians per second. 7 a, the value of the delay of this envelope, is according to our previous definition substantially the envelope delay. MEASUREMENT OF PHASE DISTORTION 537 This is the simplest form of transmitted current to which the term envelope delay can be applied. This type of wave, being made up of two sinusoidal components of equal magnitude, has the important property that its envelope suffers no distortion regardless of the length and complexity of the circuit as long as it has no non-linear element and as long as the two component frequencies are transmitted with equal attenuation. Going back now to the original case of the 1000-cycle sine wave modulated by the 25-cycle sine wave where both sidebands and car- rier are transmitted, the corresponding graphical expression for this current is 2(1 + cos 25 27/) sin 1000 2^ and the corresponding analytical expression is 2 sin 1000 27r/ + sin 975 27r/ + sin 1025 27r/. Now if the three components suffer phase changes equal to ^^to, /3iooo, and /3]025, the analytical expression for the wave at the receiving end is 2 sin (1000 27r/ - iSiooo) + sin (975 27r/ - ^3975) + sin (1025 27r/ - /3io25); and there is no simple corresponding graphical expression. It will be convenient to consider this wave as being made up of two components, one being the steady component 2 sin (1000 27t - /3iooo) and the other being a variable component O ^^„ /oC T~~i /3i025 — ^975 \ • / i nr»n T^i ^1025 + /3975 2 cos 25 27r/ 7: sm 1000 27r/ r which is the same as the total current discussed above. The outstand- ing complexity in this wave is the presence of a distortion which arises from the fact that the 1000-cycle carrier wave in these two components is not transmitted with the same phase change. The phase change of the 1000-cycle current, making up the steady component, is equal to /3iooo, whereas the phase change in the variable wave is represented by fiilb + /3l025 In other words, it is the average of the phase changes at 975 and 1025 cycles. Now, if it happens that these two expressions are equal, then there is no distortion. If, however, as is the general case, these expres- 538 BELL SYSTEM TECHNICAL JOURNAL sions are not equal, then there is a distortion which may be very easily exhibited by considering the case where the difference between (/3io25 + /3975)/2 and jtJiooo equals 90°. The current which then results is modulated by 50 cycles whereas the original wave was modulated by 25 cycles. Where the difference in question is intermediate in value between 0 and 90°, the detected modulating wave is complex, but has a component equal to 25 cycles. This component gradually gets smaller and disappears completely when 90° is reached. Now, if the circuit is made still longer, the 25-cycle component in the detected modulating wave again makes its appearance but has suffered a discontinuous shift of 180° in passing through the extinguishing point. By the time the phase difference in question has reached 180°, the received wave is distortionless except for the phase shift of 180° in the envelope which is not apparent, or at least is not distinguishable from a delay of one half cycle. With this distortion in mind, it will now be apparent that the delay suffered by the modulated wave we are considering is no longer a definite quantity. However, it can be made definite for practical purposes by confining attention to the 25-cycle component of the envelope only. The distortion in question consists merely in adding other components to this one but does not shift its phase (except for the discontinuous change spoken of above). Consequently, for practical measuring purposes, if a device is used which eliminates the various harmonics of the 25-cycle current, this wave is perfectly definite for delay measuring purposes excluding the exceptional case where the fundamental component passes through zero. These two forms of envelopes have been discussed in more or less detail because of the fact that they are the simplest ones for transmis- sion without phase distortion. For this reason they have been used as the basis of the measuring devices which will now be described. The phase shift suffered by the envelope during transmission can be meas- ured by comparison with a standard of the same frequency as the modulating frequency. This will be, of course, a direct measure of the envelope delay. From the preceding discussion of some of the principles involved in the transmission of modulated waves, it is evident that the phase shift during transmission of the simple envelope considered is equal to one half the difference of the phase shifts of each of the sideband fre- quencies. This phase shift of the envelope is then the difference in phase shift for a definite frequency interval and is quite convenient for the measurement of the envelope delay. The envelope delay so measured is that for some frequency intermediate between the two MEASUREMENT OF PHASE DISTORTION 539 sideband frequencies and, although it is not accurately the envelope delay for the carrier frequency, it may be taken as such for all practical cases when the modulating frequency is taken small enough so that the slope of the phase shift-frequency curve for the carrier and the two sidebands may be considered constant. AUXILIARY PHASE SHIFTER PHASE SHITTER Fig. 7 — Arrangement for direct measurement of envelope delay. shifter shown R = SOttL = I/SOttC. In the phase Fig. 7 shows schematically a circuit for measuring the envelope delay by measuring the phase shift of the envelope. ^^ The carrier frequency is modulated with another frequency, 25 cycles for example, and then transmitted through the system to be measured. At the receiving end an ordinary amplifier-detector is used to demodulate the received wave and obtain the modulating frequency. This source can then be compared in phase with a reference frequency which is obtained from the original source. In order to avoid including the effects of the measuring apparatus itself, the change in phase shift so measured through the system under consideration should be compared with a similar measurement made with an artificial resistance line substituted for the system under test. The difference of these two will, of course, be the phase shift suffered in the system by the envelope of the modu- lated current; and the envelope delay of the system in seconds at the carrier frequency, /, is then given approximately by ^ 360 p ' '^U. S. Patent 1,645,618. 540 BELL SYSTEM TECliSlCAL JOURNAL where p = the modulating frequency in cycles per second and M = the phase shift of the envelope of the modulated wave in degrees. In order to measure the value of M, some method of comparing the phases of various currents must be used. Also it is convenient to have in the measuring circuit a phase shifter or some means of controlling the phase of the modulating frequency. The value used for the modulating frequency will vary somewhat with the frequency used for measurement and with the conditions under which the measurement is made. Of course, other things being equal, the greater the value of this modulating frequency the greater will be the frequency difference for which the phase shift is measured and the accuracy of the measurement will be correspondingly increased. This is true, however, only when the envelope delay is changing very slowly within this frequency interval. In most cases where the envelope de- lay is changing quite rapidly with frequency, it is necessary, therefore, to use as small a value for the modulating frequency as will give the required accuracy. In practice, both conditions of measurement will be encountered so that some sort of compromise value should be chosen for a particular measuring set which will do fairly well for its require- ments. \'arious modifications of this circuit for loop and straightaway measurements are given in the patent referred to. \ arious methods of modulation and detection may be used. (1) The set-up '" shown in Fig. 7 has been used extensively for loop measurements on systems, including various telephone circuits and phase correcting networks. The details of the circuit of this set are not given here, but certain phases of its makeup and operation will be discussed. A frequency of 25 cycles from a tuning fork is used for modulation. In measuring the phase shift of the transmitted envelope, a dynamometer detector and phase shifter are used as described in the patent referred to.^-' When the modulated wave as transmitted over the system is de- tected, the modulating frequency is obtained. This will, in general, differ in phase from the original modulating frequency. If the de- tected frequency and the original frequency are now put into the dyna- mometer detector, the phase of one of these frequencies can be shifted by means of the phase shifter until these two frequencies are 90° out of phase, which is indicated by zero reading of the dynamometer. The amount of phase shift which has been introduced in order to bring about this condition is a measure of the delay of the system being 1* Compare " Phase Compensation III — Nyquist Method of Measuring Time Delay da/dw," E. K. Sandeman and I. L. Turnbull, Electric Communication, Vol. \'II, p. 327, 1929. MEASUREMENT OF PHASE DISTORTION 541 measured. If the detector were balanced with a zero delay system and, then, rebalanced with the system under question inserted, the differ- ence in these readings as given by the phase shifter would indicate the delay of the system. An integral multiple of tt might not be taken care of in this measurement, but this is of little consequence. For the modulating frequency of 25 cycles, a phase shift of nine degrees in the envelope of the 25-cycle modulation corresponds to a delay of .001 second. For convenience, therefore, the phase shifter used in this set is arranged in steps so that each step corresponds to a phase shift of nine degrees, or a delay of .001 second. In order to read intermediate values of delay, an auxiliary phase shifter, which Fig. 8 — Arrangement for direct measurement of envelope delay at low frequencies. consists of a variable condenser bridged across the output circuit of the detector tube, is used and calibrated directly in steps of .0001 second. This particular delay measuring set has been found quite useful in the frequency range of 300 to about 10,000 cycles per second. The absolute value of delay, of course, is not that which is measured, but this can usually be determined from the measured value by adding this measured value to the integral multiple of .04 second, which is suitable for the case in hand. (2) For measurements below 300 cycles, the circuit arrangement shown in Fig. 8 can be advantageously used. This is based on princi- ples exactly the same as those just described but differs considerably in the application of these principles. Here a relatively low frequency must be used for modulation. One and a quarter cycles per second has been chosen because it is satisfac- tory for measuring at frequencies as low as 10 cycles and is easily ob- 542 BELL SYSTEM TECHNICAL JOURNAL tainable from a distributor driven by the 25-cycle tuning fork used in the measuring set described above. The frequency used here for modulation is exactly 1/20 that used in the other set. Modulation is accomplished mechanically by means of a commutator and resistance potentiometer arranged as shown at the left end of the figure. The commutator brushes are rotated at a speed of \}^ revolutions per second. The carrier frequency is connected to the potentiometer as shown so that the brushes as they pass over the commutator segments will pick ofif various voltages from the poten- tiometer, and on the completion of one revolution the resultant current at the output is equivalent to a cycle of complete modulation of the carrier such that both sidebands are transmitted with the suppression of the carrier frequency. The potentiometer has been designed with steps in such a way that, for all practical purposes, a modulation of pure \]4: cycles is obtained, the higher harmonics in the modulated wave being so far removed from the fundamental and relatively so small that they are negligible. As only the two sidebands of the modulated wave are transmitted here, the current which results from detection of this transmitted wave at the receiving end will have a frequency of 2}^ cycles or twice the modulating frequency. Another set of commutator brushes is re- volved over a set of segments, somewhat similar to that already mentioned, with a speed exactly twice that of the first, namely 2}i revolutions per second. The output of the detector is connected to these brushes and transmitted through the potentiometer shown con- nected to the segments of the commutator to a very sensitive gal- vanometer. The result of this arrangement is effectively a 2^-cycle modulation of the 2^-cycle current received from the detector and as a result of this modulation the current received by the meter will consist of a d.-c. component and a 5-cycle component, the relative amounts of each depending on the relation of the commutation to the phase of the received current. The brushes of the first commutator may be rotated at any instant relative to those of the second by a manual adjustment and their position relative to some arbitrary point noted. The galvanometer is arranged so that it does not respond readily to any except direct current. There is a particular position of the commutator brushes (and another 180 degrees removed from it) which will give no deflection in the galvanometer. The particular details of measurement are considerably different from those of the preceding circuit, but in principle the arrangement is much the same. With a resistance line between the sending and receiving terminals the adjustable brushes are shifted until no deflec- MEASUREMENT OF PHASE DISTORTION 543 tion is obtained in the galvanometer. Then with the system to be measured inserted between the terminals of the measuring set, the brushes are again adjusted until the galvanometer shows no deflection. The .setting in both cases can be noted by means of a suitable scale and the difference between the two settings for calibration and measure- ment is, of course, the phase shift of the envelope of the modulated wave in the system used for measurement. This scale can be cali- brated in terms of seconds so that it measures the envelope delay directly. It is evident from the above description that a one-degree shift of the commutator brushes corresponds to an envelope delay of .00222 second. This set has been found useful in measuring the phase distortion in circuits below 300 cycles per second, especially recently when con- siderable importance has been attached to the low-frequency distortion on circuits which have been developed for program transmission. (3) When small amounts of distortion are to be measured and the frequency range will permit, a higher frequency may be used advan- tageously for modulation. Such a circuit, adapted for straightaway measurements, was used for checking up the phase correction of certain circuits used for television demonstrations.'' The circuit arrangement has been described in the reference given. In this particular case, it was not expected that the distortion of the overall system including the phase correction would be very great, so that the chief point of interest in these measurements was the detection of small changes in delay over the relatively large frequency range concerned. The modulating frequency used was 250 cycles per second, this larger value being used to obtain the desired accuracy. The frequency required for reference at the receiving end was provided by sending the modulating frequency over another circuit in the same manner as that used on the circuit being measured, except that a constant frequency for the carrier was used in the reference circuit. The purpose of this was to introduce approximately the same delay in the reference circuit as in the measured circuit because of the fact that the phase shifter used at the receiving end was capable of measuring only small differences in phase. d. Direct Measurement of Delay of Envelope Instead of measuring the envelope delay, which is d^fdo} by defini- tion, it may sometimes be desirable to measure the delay of the en- velope, say, the interval between the instant of application of a sin- usoidal wave and the instant of the received wave reaching a prede- termined value. One suitable arrangement which may be utilized to 544 BELL SYSTEM TECHNICAL JOURNAL this end has been described by Herman.'^ His description, particu- larly his Fig. 2, should make it unnecessary to describe it here. Some Results Obtained by Various Methods of Measurement In the accompanying figures results will be given in graphical form of various results which have been obtained from using the measuring 600 «n a z o o ■ u 5 z < u Q Ul a. O _] > z UI 1 i'- \ // // /l ' "^ v-^ \ A \\ /<_SOLID LINE— OBTAINED FROM \\ Y RESISTANCE CURVES ABOVE \\ ■-DOTTED LINE -CHARACTERISTIC OF DESIGN N, », _^ y \ 400 800 1200 CYCLES PER SECOND 1600 2000 Fig. 9 — Results of impedance measurements on phase corrector for 200 miles of 19-ga. H-174 side circuit. devices described above on actual telephone circuits or networks designed to be associated with them. The values added to the meas- " " Bridge for Measuring Small Time Intervals," J. Herman, B. S. T. J., \'ol. \'1I, p. i^i, 1928; particularly application No. 2, p. 349. MEASUREMENT OF PHASE DISTORTION 545 38 / 36 / / o z o / / d 32 z / / \ / 5 _J Q 30 o / \ / -i u > 5 26 \ y y \ ^ y 26 V \ ^ ^ ^ 24 1000 1200 1400 CYCLES PER SECOND Fig. 10 — Envelope delay characteristic for 231 miles of 19-ga. H-174 side circuit. 120 ■ 1 1 1 MO a z o o bj z ui 60 V ■^ X ■^ . 70 1600 2000 2400 CYCLES PER SECOND Fig. 11 — Envelope delay characteristics for 19-ga. H-44-25 circuits. 546 BELL SYSTEM TECHNICAL JOURNAL ured values to give the absolute values of envelope delay shown on the curves are obtained from an approximate estimate of the delay of the system under consideration. so 1 1 48 1 1 2 46 Z o o UJ in 1 1 -I i z 44 _l Q 1 \ / UJ Q. O _l UJ UJ \ / \ / / 40 \ y / \ V ^ y 38 500 1000 1500 CYCLES PER SECOND 2000 2500 Fig. 12 — Envelope delay characteristic for 708 miles of 16-ga. B-22-N two-wire circuit. No data are given here to show the results of measurements by the method of Fig. 1. This circuit is particularly useful in the measure- ment of networks and many examples of data obtained in this way are included in another paper.'' Fig. 9 shows the results of impedance measurements made as shown in Fig. 2 on a phase correcting network which was designed to equalize the delay for 200 miles of 19-gauge H-174 side circuit for picture trans- MEASUREMENT OF PHASE DISTORTION 547 mission. In the upper part of the figure the resistance-frequency curves are shown and the envelope delay-frequency curve derived therefrom is shown in the lower part of the figure. The delay char- acteristic for which this particular network was designed is also shown for comparison and gives a rough idea of the accuracy of this method of measurement. No figures are shown here which give the results of measurements made by the methods referred to for determining envelope delay from special phase shift measurements. Although the actual methods of in a Hbo in 2 Z70 bj 111 0. S 60 / z L y' UJ h— — 50 2000 3000 4000 5000 CYCLES PER SECOND 6000 7000 8000 9000 Fig. 13 — Envelope delay characteristic of 737 miles of 16-ga. B-22-N program circuit. measurement in these cases are quite different from the impedance measurement method, the delay results are obtained in a similar man- ner, and these have just been illustrated by the figure above. Figs. 10, 11 and 12 show the results of direct measurements of the envelope delay, using the method described above which has a modulat- ing frequency of 25 cycles per second. Fig. 10 gives the measured envelope delay-frequency characteristic for 231 miles of 19-gauge H-174 side circuit. Fig. 11 gives the envelope delay-frequency char- acteristics as measured for approximately 1560 miles of 19-gauge H-44-25 side circuit and for approximately 1515 miles of 19-gauge H-44-25 phantom circuit. Fig. 12 gives the corresponding characteris- tic for 708 miles of 16-gauge B-22-N two-wire circuit. Fig. 13 gives the envelope delay-frequency characteristic for 737 miles of 16-gauge B-22-N cable circuit equipped with phase correctors for program transmission.^"* The measurements for frequencies above 300 cycles per second were made with the measuring device using a modulating frequency of 25 cycles while the measurements below 300 '^ "Long Distance Cable Circuit for Program Transmission," A. B. Clark and C. W. Green. To be presented at Summer Convention of A. I. E. E. at Toronto, June 1930. 548 BELL SYSTEM TECHNICAL JOURNAL were made with the set which used 13^4 cycles as the modulating frequency. Fig. 14 shows the measured envelope delay-frequency characteristic for a special open-wire circuit ■* used for a television demonstration between Washington, D. C. and New York, N. Y. Curves are shown for measurements on the circuit alone and for the circuit equipped with its dry weather equalizer. The curves do not appear to be as smooth as the curves shown in the above figures, but this is largely due to the difference of the scales used for plotting the measurements. However, 0.3 a in a. a < z Q oz 0.; li V O 0.1 111 o O H uj a Q. lu O (t -I a. lU lU > u. Z uj UJ (£ -0.1 / J J CIRCUIT ALONE ^ / CIRCUIT WITH DRY WEATHER EQUALIZER \ / r / \ -^- ^^y ^^y ^^.' >-. -—■ A / / / y / v_ ^ s r ^r^ ^ ^ r /\ / x/ 4.000 8.000 12,000 16,000 20,000 CYCLES PER SECOND 24,000 26.000 32.000 Fig. 14 — Envelope delay characteristic for special open-wire circuit; Washington, D. C— New York, N. Y. some of the irregularity is due to the method of measurement and the fact that noise on the open-wire circuits obscured somewhat the exact point of balance. Conclusion It has not been the intent of this paper to include all the known methods of measuring phase distortion. Various methods for measur- ing phase shift are, of course, known and these can often be used to indicate phase distortion. In a practical way on telephone circuits, the term defined as envelope delay has certain advantages, and the paper is chiefly concerned with methods of measuring this quantity. In order to avoid including information which is contained elsewhere, the methods have not been given in detail; but references have been given, when possible, to sources where more detailed information can MEASUREMENT OF PHASE DISTORTION 549 be found. In setting up any of these circuits for actual use, certain precautions must be taken which will soon be evident. One particular point that might be mentioned here is the fact that the phase distortion introduced by the apparatus necessary for amplifiers and particularly detectors varies somewhat with the amount of power being transmitted through it, and this consideration must be given weight. Keys, switches, and such apparatus may be introduced for the convenience of the tester. The amount of amplification and the sensitivity of the detectors used depend somewhat on the accuracy required of the measurement. Effects of Phase Distortionon Telephone Quahty* By JOHN C. STEINBERG This paper discusses the effects of the type of phase distortion found in low pass filters and the loaded line on telephone quality. The effects are ascribed to three factors; the first involves the slopes of the phase characteristic at various frequencies in the range of interest, the second involves the intercept values on the phase shift axis of the tangents to the phase curve, the third involves the interference caused by portions of one sound overlapping por- tions of a succeeding sound. The first factor appears to be the one of most importance. IN the engineering of telephone systems it is convenient to define their transmission properties in terms of the changes that occur in trans- mission in the amplitude and the phase of steady state sinusoidal waves of different frequency. The terms attenuation characteristic and phase characteristic refer, respectively, to the amplitude change, usually expressed in decibels, and to the phase shift, expressed in radians or degrees, as functions of frequency. That distortion which is attributable to the attenuation characteristic is spoken of as attenua- tion distortion, and that attributable to the phase characteristic, as phase distortion. To be of greatest use in evaluating a system the steady state proper- ties must be experimentally correlated with the satisfactoriness, or quality in its broad sense, of the system from the viewpoint of the individual receiving the signals. If the signals are speech, quality involves the recognizability of the speech sounds and their naturalness. If the signals are music, the second factor is the one of chief concern. A reasonably quantitative measure of the recognizability of the re- ceived speech sounds may be obtained by means of the articulation test which is described in a later paragraph. Naturalness is considerably less definite, and the procedure in this case has been to compare the distorted signals, speech or music, with the original or undistorted signals and obtain the amounts of distortion that cause just noticeable differences. The purpose of this paper is to discuss the effects of phase distortion on the quality of speech.^ Brief reference will be made also to a small amount of data that have been obtained for music. * Presented at New \'ork Section, A. I. E. E., May 19.^0. 1 A companion paper by C. E. Lane on " Phase Distortion in Telephone Apparatus" shows the types of phase characteristics found in various networks and discusses their relation to the transmission properties. For a discussion of methods of measuring phase characteristics the reader is referred to a companion paper on "Measurement of Phase Distortion" by H. Nyquist and S. Brand. 550 EFFECTS OF PHASE DISTORTION 551 Nature of Speech and Hearing Since the manifestation of phase distortion depends upon the type of signal and the method of observation, it is of interest to first consider the nature of the waves of speech sounds and certain facts of audition. Speech waves may be regarded as non-periodic in that they start at some time, take on some finite values, and then approximate zero again as may be seen from the wave form - of the word "seems" in Fig. 1. In this particular word the wave form of each sound and the transition periods are readily distinguishable. Although in other cases of con- nected speech this may not be done so easily it is usually possible to s EE M -^^-\r\/\fV^[\/^\j^^\f^^\\ 500 CYCLE Fig. 1 — Wave form of the word "seems." approximately distinguish between sounds and to ascribe to each an initial period of growth, an intermediate period which in some cases approximates a steady state and then a final period of decay. The duration intervals of different sounds vary from about .03 to as much as .3 or .35 seconds. Hearing appears to be concerned more with the spectra of sound waves, i.e., something corresponding to the amplitudes of the Fourier components, than with the actual wave form of the disturbance. For the type of steady state complex waves that the speech sound waves approximate for a considerable portion of their duration intervals, it has been observed that phase changes in the component waves cause ^Speech and Hearing, H. Fletcher, D. Van Nostrand Co., 1929. 36 552 BELL SYSTEM TECHNICAL JOURNAL little if any change in the character of the sound. A possible exception may arise for complex waves of large amplitude because of non- linearity in the hearing mechanism. It would be expected then that the observable effects of phase shift would arise from the intervals preceding and following the steady state intervals of the sound waves. For this reason the wave of a speech sound is regarded as non-periodic and when an amplitude frequency distribution is spoken of a Fourier Integral is implied. Types of Phase Characteristics The determination of the effects of phase distortion on quality involved the characteristics of the experimental system as a whole although, for convenience, the distortion usually originated in a spe- cific network in the system. The procedure that was followed was to make the system, except for the network, as distortionless as possi- ble. In most cases the characteristic of the system from the view- point of distortion, was the insertion characteristic of the network. Before taking up experimental results on phase distortion it will be helpful to briefly review certain conclusions bearing on the relation between phase shift and wave distortion that have been obtained by analytical methods.^ A phase characteristic of interest is one of form B = Bq -\- Bioi, where B — phase shift in radians and co = 1-kj. If the original wave be represented by a Fourier Integral, the expression for the received wave may be obtained by shifting the phases of all of the sinusoidal components in the original in accordance with the above expression, assuming negligible attenuation distortion. For convenience the two terms in the above expression may be considered separately. If this is done, it can be shown by inspection that a constant shift of B^ in the phases of all of the sinusoidal components gives a wave which is the sum of two waves, one the original wave multiplied by an amplitude factor cos 5o, the other a wave obtained from the original by shifting all of its components by 7r/2 radians and multiplying by an amplitude factor sin B^. If the phases of the sinusoidal components in the original are shifted by the amounts Bxw, where co = 27r times the frequency of the com- ponent, it can be shown that the resulting wave differs from the original only in that the origin of time is displaced by an amount Bx or the slope dBldw of the above expression. ^ Transient Oscillations in Electrical Wave Filters, Carson and Zobel, Bell Sys. Tech. Jour., July 1923. Building Up of Currents in Long Periodically Loaded Lines, Carson, Bell Sys. Tech. Jour., Oct. 1924. Phase Distortion and Phase Distortion Correction, Mead, Bell Sys. Tech. Jour., Apr. 1928. Phase Compensation, .Sandeman, Electrical Communication, 1929. Transient Solutions of Electrical Networks, Mason, Bell Sys. Tech. Jour., Jan. 1929. Phase Distortion in Telephone Apparatus, Loc. cit. EFFECTS OF PHASE DISTORTION 553 As will be seen later, it is convenient to regard these two operations as occurring in sequence. The second term of the expression introduces a definite time delay of Bi, sometimes called the envelope delay, and no distortion in the form of the original wave. Following this operation, the phases of all of the sinusoidal components of the delayed original are shifted by the constant amount Bq the resulting wave being the received wave. If Bo equals zero or even multiples of it, the amplitude factor sin Bq equals zero, so that, the received and original waves are identical in form. If Bo is an odd multiple of t the received wave is reversed in sign only. In both cases the received wave is delayed by an amount (IBjdw, and the wave cannot appear until this time has elapsed. For all other values of Bq the form of the received wave differs from that of the original to a greater or less degree depending upon the original wave form and the value of Bq. In this case the delay in the received wave as a whole cannot be spoken of precisely for no point on the received wave can be said to correspond to a point on the original wave. Theoretically the received wave may begin to appear at some earlier time than dB/doj, as has been shown by Mr. T. C. Fry in some unpublished work for the case of a wave having the form of a tele- graph dot. When the phase characteristics are curved over appreciable portions of the frequency range, as is usually the case in actual systems, exact statements of the above nature are difficult to make. It seems best, therefore, to confine the discussion to particular characteristics and to the case of speech waves. A qualitative picture of what happens for the type of characteristic shown in Fig. 2 may be seen by regarding it as the limiting case of a characteristic made up of a number of straight lines of different slopes, each line approximating the curved characteristic for a frequency range A/. As discussed above, the wave of a speech sound may be regarded as made up of steady state component waves of different frequency. The resultants of the component waves in various fre- quency ranges A/ are subject to the phase distortion discussed in the preceding paragraphs, that is, the original forms of the resultants are delayed by times dB/doo, and then undergo a distortion that depends upon the terms Bo or the intercepts of the straight lines with the vertical axis. As mentioned above it is convenient to regard these operations as taking place in sequence, the first introducing definite delays and no distortion in portions of the original signal corresponding to frequency ranges A/, the second introducing a constant amount of phase shift in 554 BELL SYSTEM TECHNICAL JOURNAL the component waves of each delayed portion. Thus the first opera- tion spreads the wave out on a time scale, the part of the original wave corresponding to the frequency range having the minimum slope ar- riving first and followed successively by portions corresponding to the remaining frequency ranges. The relative delay for a range A/ is given by the difference between the slope of the phase characteristic -2Tr Fig. 2 — A curved phase characteristic. in the range A/ and the minimum slope. This difference or the expres- sion [{dB/doo)f — (dB/do))min\ is spoken of as the delay distortion. The delay distortion characteristic is simply a graph of the above expression plotted against frequency. The second operation distorts the wave forms of the delayed portions corresponding to the frequency ranges A/, thus making it impossible to speak of the delay of the final or re- ceived portions or of the received wave as a whole. This may be described in other terms by saying that a network hav- ing a characteristic of the type shown in Fig. 2 may be thought of as if EFFECTS OF PHASE DISTORTION 555 it were made up of two sets of networks, the two sets being connected in series. The first set consists of a number of networks in parallel, each network passing a frequency range A/ and having a phase characteristic of the form Bxw, where J5i is the slope of the straight line approximating the curved phase characteristic in the range A/. The second set con- sists of a number of corresponding networks having phase characteris- tics of the form Bq, where Bq for a network passing the range A/, is the constant term in the equation of the straight line approximating the curved phase characteristic in this range. The phase distortion thus consists of two operations in sequence, the first introducing definite delays in various portions of the original wave, the second introducing constant phase shifts in the component waves of each delayed portion. A definite contribution to the recognizability of a speech sound may be associated with each frequency range A/ in the undistorted state. At the output terminals of the first set of networks the various portions corresponding to the ranges A/ do not combine to form an exact copy of the original wave, because of the different delays that have been introduced. It is supposed that their normal contributions to recognizability are decreased by a factor depending upon the delay distortion and the duration time of the speech sound. This factor is referred to here as the " time factor " and it would be expected to operate even though the second set of networks were non-existent. Since the constant phase shifts of the second set of networks are not all multiples of It, additional distortion will be introduced by this set of networks. To take account of this another factor called the "inter- cept factor" is introduced. As will be seen later this factor seems to be negligible for the case of speech waves due in part, no doubt, to the sustained character of the waves and to the mechanism of hearing as previously described. In addition to the above, when we deal with a succession of speech sounds, as in connected speech, a third factor might be expected to operate because the delayed frequency ranges of one sound may overlap the least delayed ranges of a succeeding sound and interfere with its perception in the manner of an extraneous noise. As will also be seen later on this factor appears to be negligible for the type of characteristic shown in Fig. 2 because the so-called noise and the signal with which it interferes do not have components in a common frequency range. When this is true, noise in general interferes much less than when the signal and the noise have components in a common range. Phase Distortion and Quality Quite aside from the recognizability viewpoint, when speech from a system having phase distortion is compared with that from a system 556 BELL SYSTEM TECHNICAL JOURNAL having negligible distortion, it is noticed that the distorted speech is accompanied by certain audible effects which appear to be extraneous to the speech and transient in character. As discussed above phase changes in the component frequencies of steady state waves cause little if any change in the character of the sound. This would indicate that the so-called audible effects of phase distortion arise in the transi- tion periods, i.e., in the period between the approximate steady state of one speech sound and that of the succeeding sound, and are due to the spreading out effect of phase distortion. Data on the amount of distortion that will cause just noticeable effects will be discussed in a later paragraph. Before discussing the various factors affecting the recognizability, it Is of interest to consider the importance as obtained from articulation tests, ^ of different portions of the duration intervals of speech sounds, and also the importance of different portions of the speech frequency range. Fig. 3 shows the effects upon sound articulation of limiting the transmitted frequency range by means of high or low pass filters in a system having neglible distortion in other respects. Although the curves do not intersect at 50 per cent nor do the articulation values of complementary filters add up to 100 per cent, they may be used with qualifications, to measure the contribution or importance to articulation of a portion of the speech frequency range. Thus the slope vs. cut-off frequency for the low pass filter curve gives a measure of the contribu- tion to articulation of a frequency range A/ when contiguously added to the range 275 to/. Articulation tests that have been made with voice operated relays give an indication of the importance to articulation of portions of the duration intervals of speech sounds. In the tests, syllables of the consonant-vowel-consonant type were spoken at intervals of about 3 seconds. A circuit having a relay adjusted so as to break contact almost simultaneously with the beginning of a syllable, was used. The contacts of a second relay formed a short circuit across the receiver. The operation of the first relay caused the second relay to break contact ■* Articulation Testing Alethods, H. Fletcher and J. C. Steinberg, Bell Sys. Tech. Jour., Oct. 1929. In an articulation test lists of syllables are spoken into the transmitter of a system having phase distortion and observers at the receiving end write down the sounds which they hear. The observed lists are compared with the spoken lists and the errors determined. The percentage of the total number of spoken syllables that are correctly observed is called the syllable articulation. A syllable is considered to be incorrectly observed if one or more of the fundamental speech sounds which it contains are mistaken. The percentage of the total number of spoken speech sounds which are correctly observed is called the sound articulation. When attention is directed toward a specific sound such as "e," the term individual sound articulation is used. It is the percentage of the number of times that "c" was spoken that it was observed correctly. EFFECTS OF PHASE DISTORTION 557 after an interval of time depending on the time constants of the relay circuit alone. The time taken for the second relay to operate represents the time clipped from the initial consonants of the syllables. Fig. 4 shows the initial consonant articulation plotted against the operating time of the second relay. If equal elements of time in the 100, 90 80 70 60 50 40 30 20 10 0 HI 1 W GH pa; 5S F UT :R TSr - -< ^^^ .^ -o- T / K /o \ J / ^ /■ y ■J ^ [' ^ 7 ^ l^ 1 1 1 1 1 1 1 1 1 t • 1 1 / / -A 100 34579 2 34 5 7 9 1,000 10,000 CUT-OFF FREQUENCY IN CYCLES PER SECOND Fig. 3 — •Importance of frequency range to articulation. duration intervals of the consonants are of equal importance, then the clipping by an amount ^T should decrease the articulation by a factor, K= {\ - ATT), where T is the average duration time of a consonant (1) In the above tests, the syllables were spoken separately. Oscillograms taken for syllables spoken in a similar manner show an average duration time of .16 seconds for the consonant sounds.^ The straight line in Fig. 4 was calculated by multipling the articulation obtained for zero operat- ing time by a factor K as determined from the above equation with ^Speech and Hearing, H. Fletcher, D. \'an Xostrand Co., Inc., 1929. 558 BELL SYSTEM TECHNICAL JOURNAL T — AG seconds. The data indicate that equal elements of time in the duration intervals of the sounds are of approximately equal importance to the average articulation of a group of sounds. It should be pointed out that this might not be the case for individual speech sounds and also for certain types of speech distortion. In the above tests a carbon transmitter was used of a type which introduced attenuation distortion. Tests have also been made with speech having negligible attenuation distortion. Although they indicated a relation of the 100 90 Z 80 O _l 70 O cr 60 < ^ 50 Z o ^ 40 o o ^30 20 10 o * — . o 0 0.02 0.04 0.06 0.08 0.10 0.12 0.14 IMTERVAL OF TIME CLIPPED FROM INITIAL CONSONANT IN SECONDS Fig. 4 — Importance of an element of time to articulation. above type the data are somewhat questionable because of uncer- tainties in the operating time of the relay. In the next series of articulation tests a nominal undistorted speech frequency range of 0 — 4500 cycles was divided into two parts by means of filters and each part transmitted through a different channel. After transmission the two parts were recombined. The phase characteristic of each channel approximated a straight line over the greater part of the frequency range. The slope of the characteristic of one channel could be increased by various amounts over that of the other channel. One channel thus introduced a definite time delay, in the sense used here, with respect to the other channel, i.e., a delay given by the dif- ference in the slopes of their phase characteristics. The observed sound articulations plotted against time of delay are shown in Fig. 5. The articulation values decrease with increasing delay and approach EFFECTS OF PHASE DISTORTION 559 the articulation of the more intelh'gible band which was also the least de- layed band. For convenience it is spoken of as the non-delayed band. 100 80 NGN- DELAYED BAND 0-3000~ DELAYED BAND 3000-4500~ io — Q- ■' "■"^ .04 .06 .12 .16 .20 100 y 80 O NON-DELAYED BAND 0-2000'\j DELAYED BAND 2000-4500~ ? ' .04 .08 .12 .16 NGN- DELAYED BAND 0-1500'v DELAYED BAND 1500-4500~ .20 100 80 NON-DELAYED BAND 750-4500^ DELAYED BAND 0-750~ ^o-^ — — - — , ARTICULATION OF T NGN -DELAYED BAND-^ 1 1 1 .04 .08 .12 .16 .20 TIME OF DELAY IN SECONDS Fig. 5 — Articulation vs. time of delay. If the decrease is due primarily to the inability of the delayed range to contribute to articulation for a time equal to the delay inter- val, then if, A 1 = sound articulation for zero delay A2 = sound articulation of the non-delaved band 560 BELL SYSTEM TECHNICAL JOURNAL the sound articulation for a delay AT should be A = A. + K{A,- A,), (2) where i^ is a factor obtained from Eq. (1) for T = .17 seconds. This value of T is used because in these tests the articulation sylla- bles were spoken as parts of introductory sentences, such as, "The first syllable is ««/," etc. Oscillograph records for this manner of speaking indicate an average duration time for vowel and consonant sounds of an order of .16 to .18 seconds. This is somewhat less than the duration time for sounds spoken in detached syllables, and somewhat greater than the duration time for sounds spoken in connected speech containing words of one or more syllables. For the latter case, the average duration time is probably more of an order of .08 to .12 seconds. The solid lines shown in Fig. 5 were calculated from Eq. 2 which involved only the time factor. The agreement between observed and calculated results indicates that the two other factors were compara- tively small in these cases. As regards the noise factor it should be noted that the frequency range of the so-called noise and that of the sound wave with which it interferes have no part in common. When this is true, as previously pointed out, the interference from noise is small. The following tests were made with networks having curved phase characteristics of the type shown in Fig. 2. One of these networks was an all pass structure made up of two types of sections, a "B" section having a critical frequency of 2000 cycles and an "A" section having a critical frequency of 2500 cycles. By using difTerent numbers of sections different amounts of delay distortion could be obtained. Fig. 6 shows the delay distortion for the conditions that were tested. The attenuation characteristic was equalized up to 2500 cycles, and a 2400 cycle low pass filter having negligible phase distortion was associated with the network. Fig. 7 shows the sound articulation values versus the number of sections. A time factor Kj for an element A/ in the frequency range 0 to 2400 cycles, may be obtained from Eq. 1 by setting Y — 0.17 seconds and AT equal to the delay distortion given in Fig. 6, for the element A/. Delays in some frequency ranges will impair the articulation much more than similar delays in other ranges because some frequency ranges are of greater importance to articulation. The importance of various frequency ranges is closely related to the slopes of the curves in Fig. 3. To obtain an effective factor K, the values of K; must be weighted in accordance with the importance of the frequency ranges considered. This may be done approxi- mately for the 2400 cycle range by averaging the values of Kf EFFECTS OF PHASE DISTORTION 561 corresponding to successive elements A/, the elements being chosen so as to represent equal increments of increase, say 5 per cent, on the sound articulation versus cut-off frequency curve for low pass filters (Fig. 3). 0.10 009 0.06 If) O z O 0.07 u UJ 0.06 Z 5 0.05 - < III o 0.02 0.01 "s \, / \ -- ATTENUATION 20 FIL. Q = 200- ATTENUATION 1 FIL. Q = 200, DELAY 20 FIL., -^ \ / ix A / \ \, 1.- 7 \ N '1 / / \ / / . — — • __^ . .,'- ..-^ 1 1 1 / f 1 y 50 40 -0 30 - Z o 20 lO 2500 500 1000 1500 2000 -EFFECTIVE BAND 20 FILTERS Q = 200 -EFFECTIVE BAND 20 FILTERS Q=200 (EiQuALfzED)" -EFFECTIVE BAND I FILTER Q = 200 (NEGLIGIBLEV 3000 3500 4000 FRECUENCY IN CYCLES PER SECOND -ALLOTTED BAND- Fig. 9 — Transmission for a filter ha\ing a slow rate of attenuation increase. desirable that the attenuation at the edge of the transmitted band should increase at a very rapid rate, and that the delay distortion at the edge should be small. The solid curve in Fig. 9 shows the attenuation distortion, i.e., the difference between the minimum attenuation and the attenuation for any frequency/, for a filter having a slow rate of attenuation increase. The delay for one filter is not shown, but it is about 1/20 of the delay shown by the dotted curve. The allotted frequency band is determined by the frequency value at which the attenuation curve of the filter crosses that of the filter transmitting the adjacent band. The attenu- ation at the crossover for the purposes of this discussion may be taken as •> For a discussion of the relation between these characteristics and the type of filter section, the reader is referred to the previously cited paper by C. E. Lane. EFFECTS OF PHASE DISTORTION 565 25 db. The effective band is the frequency range that will give the same articulation as that given by the filter. The delay distortion is so small for one filter of this type that its effects can be neglected and the ef- fective band may be determined from the attenuation curve alone. The effective band is given approximately when the area bounded by the attenuation curve and a line parallel to the frequency axis through the 25 db point equals the area under the 25 db line between the fre- quency limits zero and/, where/ is the upper limit of the effective band. 0.05 0.04 o z o o in Ji|3 ■0|TJ 5 0.03 0.02 0.01 ATTENUATION ATTENUATION ATTENUATION ATTENUATION DELAY 50 40 J) ■0 30 20 10 Z o I- < D Z 500 1000 ■EFFECTIVE BAND -EFFECTIVE BAND 1500 20 FILTERS 20 FILTERS 2000 Q = 80 — Q-200- 2500 3000 3500 4000 FREQUENCY IN CYCLES PER SECOND ■EFFECTIVE BAND 20 FILTERS Q = 20o(E;QuALrZED)~* )- ■EFFECTIVE BAND I FILTER Q"oRebl ALLOTTED BAND 200 /'NEGLIGIBLE DELAY Fig. 10 — Transmission for a filter having a rapid rate of attenuation increase. This is true because the effect of attenuation near the cutoff frequency is to reduce the contributions of the various frequency ranges by a factor that is proportional to their attenuation, for the attenuation limits considered. In the above case this factor varies from zero at 3000 cycles to unity at 2400 cycles. The effective band width is 2825 cycles. The dotted curves on Fig. 9 show the delay and attenuation values for 20 filters in tandem. The effective band width that is shown for "delay equalized" is that due to the attenuation distortion alone. It was obtained as described in the preceding paragraph except that the attenuation for zero articulation was taken as 40 rather than 25 db since there is no interference from the signals in the adjacent band. In this case the effective band due to attenuation distortion alone is 2550 cycles. The effect of delay distortion when taken into account in the 566 BELL SYSTEM TECHNICAL JOURNAL manner previously described further reduces the effective range to 2500 cycles. Thus the additional reduction due to delay distortion is small. Fig. 10 shows similar data for two filters having rapid rates of at- tenuation increase, one having coils with a (2 of 80, the other ^ having coils with a (2 of 200. For 20 filters {Q = 200) in tandem the effective band due to attenuation distortion alone (delay equalized) is 2800 cycles. The effect of delay distortion further reduces the effective range to 2650 cycles. Comparing these ranges with the corresponding ranges for the filters of Fig. 9 shows that the filters of Fig. 10 use the allotted band, which is the same for both figures, more efficiently from an articulation standpoint, i.e., the effective band is a larger fraction of the allotted band. The delay distortion, however, in the filters {Q = 200) of Fig. 10 is more noticeable. This will be seen by noting that the amount of delay near the cutoff for the filters of Fig. 9 is very much smaller than that of Fig. 10. The noticeableness of the delay distortion may be decreased by equalizing for the delay distortion, or by using filters with coils of smaller Q. Any gain made by the former method is made at the ex- pense of the minimum delay, i.e., the constant delay in the major part of the transmitted range. Any gain made by the latter method is made at the expense of the effective frequency range as shown by the effective band for 20 filters {Q = 80) of Fig. 10. Twenty filters of the type shown in Fig. 9 have about the same overall performance as 20 filters of the type shown in Fig. 10 having coils with a (2 of 80. It is evident that in the design of filters, compromises must be made between the rate of attenuation increase at the edge of the transmitted band, the minimum delay and the dela)^ distortion. The compromises that are made in an actual system depend upon many factors and their discussion is beyond the scopeof this paper. Although the time factor appeared to be the one of most importance for the phase characteristics that have been discussed here, it should be pointed out that this may not be true for all types of phase characteris- tics. Much work remains to be done for other types of phase char- acteristics, for example, characteristics which show irregular changes with frequency rather than the smooth changes of the type discussed here. It seems, however, that the main thing in securing transmission free from phase distortion is to provide a phase characteristic that is linear with frequency over the frequency range of interest. '' Q refers to the ratio of the reactance of a coil to its effective resistance. Long Distance Cable Circuit for Program Transmission* By A. B. CLARK AND C. W. GREEN The rapid growth of the telephone cable network in this country has made it desirable to develop a system whereby this network may be utilized to transmit programs for broadcasting stations over distances upwards of 2,000 miles. Such a system has recently been developed and given a trial on a looped-back circuit 2,200 miles long with very satisfactory results. It transmits ranges of frequency and volume somewhat in excess of those now handled by the open-wire circuits which are used for program work, and also in excess of those handled by present-day radio broadcasting systems when no long distance lines are involved. The paper deals first with the transmission requirements of broadcasting systems and then gives a description of this new cable system. AS discussed in two recent papers, ■• one of which was presented before this Institute, telephone circuits are now extensively used for chain broadcasting. Radio broadcasting stations covering various local areas in the United States are connected together by wire circuits so that programs are delivered simultaneously to all of them. Thus, it is possible to deliver a program to the whole nation at once. About 35,000 miles of telephone circuits are now being regularly utilized for this service and about 150 radio broadcasting stations receive pro- grams from one or more of the chains of wire circuits. Today practically all of this service is being furnished by means of open wires using voice-frequency channels. Long distance cable routes are growing rapidly and are supplementing the open-wire routes, particularly those carrying very heavy traffic. Fig. 1 shows the long distance cable routes now in use in the United States, together with the additional routes proposed for installation within the next few years. The advantages in placing some circuits in these cables which will adequately handle program transmission service were evi- dent and led to the development described in this paper. Because of the special characteristics which program transmission circuits must possess it was necessary to develop an entirely new type of cable circuit, in which the method of placing the wires in the cables, the type of loading and all of the apparatus, including amplifiers and distortion correcting apparatus for both amplitude and delay, differ radically from other cable circuits. The development was recently completed and a trial installation made in which wires were looped ^ F. A. Cowan, "Telephone Circuits for Program Transmission," presented at Regional Meeting of S.W. District of A. I. E. E., Dallas, Texas, May 7-9 1929. Proceedings of A. I. E. E., July, 1929, A. B. Clark, "Wire Line Sys terns for National Broadcasting," presented before the World Engineering Con gress at Tokio, Japan, October, 1929, Proceedings of I. R. E., November, 1929 Bell System Technical Journal, January, 1930. * Presented at Convention of A. I. E. E., Toronto, June, 1930. 567 37 568 BELL SYSTEM TECHNICAL JOURNAL •o ^ . (0 <1) (t X T) , (U o c C (/> u. 0J3 n) <0 in c o c ■n U 3 c o c !T! (r IJ X m DO LlJ (U Q. F to o -(-' (tl o r/l Ul 1 o (_> 3 O J3 'c3 biO LONG DISTANCE CABLE CIRCUIT 569 back and forth in the cables between New York and Pittsburgh so as to produce a circuit 2,200 miles in length. Tests were made on this circuit over a period of several months and very satisfactory results were obtained. It is, therefore, planned to make extensive application of this system and eventually program circuits may be provided in cable over practically all of the long toll cable routes. So as to appreciate what is involved in the design of this system there will first be presented a discussion of the transmission require- ments. Following this, the new system will be described and its more important transmission characteristics set forth. Transmission Requirements For program transmission the ideal, of course, is to provide a trans- mission line such that no distortion whatsoever will be caused to program material transmitted over the line whatever be its length. Ideally also, program pickup apparatus, radio transmitters, radio receivers and loudspeakers should be such that the program delivered from the loudspeaker should sound exactly like the original program delivered to a direct listener in the best location. To meet this ideal, however, would require that the whole audible range of frequencies, extending from about 20 to 20,000 cycles, and a tremendously wide range of volumes representing power differences of more than a million-fold be handled without any distortion whatsoever. Actually the radio art is far from attaining this ideal. It does not seem reasonable, therefore, to provide lines very much superior in transmission performance to the rest of the system since this would unnecessarily increase the cost for providing the service. However, telephone lines represent a fixed investment which must remain in service for many years in order to keep costs within reason and, furthermore, it is, in general, not practical to change the transmission characteristics of the lines once they have been installed. It is, therefore, necessary to take into account the fact that the broadcasting art has considerably improved in the past and is likely to improve in the future and provide telephone lines of sufficiently good charac- teristics to anticipate the improvements which are likely to come within a reasonable period of time. These general considerations have led to the adoption of the follow- ing as practical standards of performance for the new cable system: Frequency range to be transmitted without material distortion — about 50 to 8,000 cycles. \^olume range to be transmitted without material interference from extraneous line noise — ^about 40 db, which corresponds to an energy range of 10,000 to 1. 570 BELL SYSTEM TECHNICAL JOURNAL Some of the more detailed considerations which have led to the setting of these standards will now be given. Frequency Band Figure 2 gives some data in regard to the frequency range required for different musical instruments as well as speech. These data were obtained in the Bell Telephone Laboratories using an arrangement capable of picking up and reproducing practically the whole audible frequency range. Certain very low-frequency instruments, such as organ pipes and bass drums, were not included in the tests owing to laboratory limitations. A number of observers listened to the repro- duced material, first, when practically the whole frequency range was transmitted and, second, with either the high frequencies or low ACTUAL TONE RANGE ACCOMPANYING NOISE RANGE ''CUTOFF FREQUENCY AT WHICH 80%OF THE OBSERVERS COULD DETECT THE FILTER TROMBONE BASS CLARINET BASSOON BASS TUBA SNARE DRUM 14 " CYMBALS KETTLE DRUM-TYMPAr|ll BASS DRUM VIOLIN CELLO BASE VIOL FEMALE SPEECH MALE SPEECH II - - «^ — - .. .. 30 50 100 200 5000 10000 20000 500 1000 2000 CUTOFF FREQUENCY Fig. 2 — Summary of important ranges required for dififerent instruments. frequencies cut off by means of filters. The observers endeavored to note whether there was any perceptible effect when the filters were introduced but did not attempt to determine whether introducing the filters made the reproduced material sound more or less pleasing. Referring to the figure, it will be noticed that at the lower frequen- cies little appears to be lost by cutting off frequencies below about 50 cycles. At the upper frequencies, however, with certain of the musical instruments something is lost by cutting off frequencies above 8,000 cycles. Hissing sounds, sounds of a percussion nature and sounds of jingling keys, rustling paper, etc., appear to be most affected by cutting off the high frequencies. Tests have shown, however, that when the frequency range 50 to LONG DISTANCE CABLE CIRCUIT 571 8,000 cycles is transmitted with very little distortion within the band the results obtained are very pleasing. The ordinary observer without making direct comparison tests is unlikely to detect the absence of the higher frequencies. From the standpoint of radio transmission there will probably be some difficulties in handling the 8,000-cycle range which has been tentatively set as a standard for the cable line. Each radio station, theoretically at least, is now being allowed only a 10,000-cycle band of frequencies and, since both sidebands are transmitted, each band is fully occupied when transmitting 5,000 cycles. Since adjacent frequency ranges are not assigned to stations in the same locality, a certain amount of spreading out is, no doubt, tolerable, so that those listeners who are close to broadcasting stations should, in gen- eral, be able to pick up the 8,000-cycle range without undue inter- ference from other stations. The more distant listeners will have trouble if their sets take in the complete 16,000-cycle band required to handle, on a double-sideband basis, the 8,000-cycle program range. Letting in this wide frequency range will bring in increased interference from other stations and will also increase the atmospheric interference. In spite of this increased trouble which the distant listeners will have, it can no doubt be argued that it will do little harm for the radio stations to put out the full 8,000-cycle band. The nearby listeners, if they have very good sets, will in general be able to appre- ciate this, while the distant listeners, if their sets are arranged to receive only a 5,000-cycle band, should receive only slightly more interference from wide-band stations occupying adjacent frequency bands. Evidently, if the frequency range were doubled so as to furnish the listener with practically the whole audible range of frequencies, these radio difficulties would be exaggerated. It seems certain that, if radio stations were to handle the whole audible band of frequencies, a reassignment of frequency bands to these broad-band broadcasting stations would be called for and also quite probably these radio stations would be forced to resort to single sideband transmission. It is not sufficient merely to fix the limits of the frequency band. Limits to the allowable distortion within it must be established. Tests have indicated that it is desirable that different frequencies within the transmitted band should not suffer attenuations differing by more than about 5 db corresponding to power differences of about three-fold. The transmission delay* suffered by different portions of the fre- * "Delay" as used in this paper has the same significance as "envelope delay" used in literature on phase distortion. It is defined as d^jdu where /3 is the phase shift and CO is 2 7r times the frequency. 572 BELL SYSTEM TECHNICAL JOURNAL quency band must also be considered. This is necessary because, when transmission over long distance lines is involved, this delay tends to be different for different parts of the frequency band and the distortion produced is a function of the frequency-delay character- istics. Tests have indicated that the high frequencies, say those in the range 5,000 to 8,000 cycles, should not suffer delay in transmission over the line more than 5 to 10 milli-seconds greater than the delay suffered by frequencies in the neighborhood of 1,000 cycles. However, at the low end of the scale more delay may be tolerated: for example, 50 cycle waves may be delayed as much as 75 milli-seconds more than those in the neighborhood of 1,000 cycles without noticeable deteri- oration in quality. Requirements must also be imposed as to "linearity" of the trans- mission, that is, constancy of efficiency with different current strengths. If the transmission departs too much from "linearity" several dis- agreeable effects may be produced: (1) Spurious frequencies which are by-products of the true frequencies will become large enough to be annoying, (2) strong sounds will not be reproduced as well as weak sounds, and (3) when weak sounds are transmitted along with strong sounds the strong sounds will tend to obliterate the weak sounds. In the design of this program transmission circuit the criterion was adopted that transmission put over the circuit at the maximum prescribed volume level must not sound appreciably different than transmission put over the circuit at a considerably lower level, at which lower level the non-linear distortion is negligible. Volume Range A favorably-seated listener to a high-grade orchestra is treated to a wide range of volumes. Opinions differ as to just how wide a volume range can be appreciated by such a listener, but it seems certain that it is at least 60 db, corresponding to a power range of one million to one. The human ear can hear volume ranges in excess of 100 db, corresponding to a power range of ten billion to one. For loudspeaker reproduction it has been found that a room must be particularly quiet in order to be able to appreciate a volume range of 60 db. Rooms in three-quarters of the usual residences are probably too noisy for a volume range as great as this to be appreciated. A 40 db volume range, corresponding to a power range of 10,000 to 1, can be appreciated in most rooms where radio listening is done and is quite satisfactory for most musical selections. From the standpoint of design, the maximum volume of a wire program transmission system is limited by the requirement that the LONG DISTANCE CABLE CIRCUIT 573 program must not be allowed to spill over unduly as crosstalk into neighboring circuits which may be carrying telephone messages or other programs. The volume may also be limited by the requirement that serious non-linear distortion be not introduced by effects pro- duced in the vacuum tubes of the amplifiers or in any magnetic-core coils either in the apparatus or in the line. On the other hand, the minimum volume which a wire program circuit can handle is limited by the tendency of the noise present on the circuit to annoy the listener when the program volume is very weak. Crosstalk from other circuits into the program circuit also enters as an important consideration, since radio listeners must not be able to pick up intelli- gible conversations during those times when the program volume is very weak or when actual pauses occur in the programs. From this, it is seen that the matter of widening the volume range of a wire program transmission system involves not only added cost to keep non-linear distortion and noise within limits but also, and perhaps even more important, added cost to isolate the circuit from other circuits on the same route. From the standpoint of the radio part of broadcasting systems handling very wide volume ranges also presents difficulties. Radio transmitter and other radio equipment noises become more serious as the volume range is widened. More important, however, widening the volume range without corresponding increase in the radio trans- mitter capacity reduces the effective range of a radio broadcasting station, since this increases the tendency for the faint parts of the programs to sink below the level of atmospheric and receiver-set noises. At present it is understood that most radio broadcast programs where no long distance wire circuits are involved are being delivered with a volume range of about 30 db.^ In order to anticipate improve- ments which may come in the broadcasting art, however, it has seemed desirable to provide wire circuits in cable which will handle a wider volume range than this and, accordingly, 40 db has been taken as a working standard. This volume range appears to satisfy almost everybody with the possible e.xception of some who listen to broad- casts of symphony orchestras and the like. With the present limi- tations of volume ranges to about 30 db, there has been some complaint that much of the artistic quality and effectiveness of broadcasts of such high-grade music has been lost because of the fact that the operator manipulating the volume range control seemed to reduce the range an undue amount. 2 O. B. Hanson, "Volume Control in Broadcasting," Radio Broadcast, March, 1930. 574 BELL SYSTEM TECHNICAL JOURNAL Studies are now under way looking toward systems which will compress the volume range transmitted over the line and expand it at the far terminal, but possible applications to radio systems may be difficult since receiver characteristics need to be considered. If some volume range compression and expansion system is not employed, ability to handle a materially wider volume range can only be obtained with considerable difficulty. In the radio part of systems it will require reductions in radio transmitter noises and involve loss in the effective range of radio stations, unless higher powered transmitters are employed. In the wire part of systems it may involve the use of amplifiers and loading coils capable of handling more power, means for materially reducing the crosstalk coupling between circuits and also means for making the program transmission circuits more quiet. Description of New Cable System In this program transmission system the nominal telephone re- peater spacing of 50 miles, common with message telephone circuits, is retained. The pilot-wire regulator system which compensates for changes in transmission caused by temperature changes in message circuits is also used for the program circuits.* The diagram in the top part of Fig. 3 shows several hundred miles of program transmission circuit, illustrating how it is divided up into repeater sections and pilot-wire regulator sections and also indicating the principal pieces of equipment located at the repeater stations. As indicated on the diagram of Fig. 3, there are two classes of repeater stations, known as regulator stations and non-regulator sta- tions. At the non-regulator stations the repeater gains are maintained at fixed values while at the regulator stations they are varied under control of the master pilot-wire regulating mechanism in such a way as to compensate for the transmission variations of the cable con- ductors caused by temperature changes. At each non-regulating repeater station are placed: 1. An attenuation equalizer which corrects for the attenuation differ- ences at different frequencies (at average temperature) intro- duced by the preceding repeater section. 2. A delay equalizer which corrects for the difference in delay at different frequencies introduced by the preceding cable section. 3. A one-way amplifier introducing sufficient gain to overcome the line loss, together with the added losses introduced by the attenuation and delay equalizers. ^ A. B. Clark, "Telephone Transmission Over Long Cable Circuits," A. I. E. E. Transactions, Vol. 42, Feruary, 1923. LONG DISTANCE CABLE CIRCUIT 575 RELATIVE POWER no ujt- I- < O ir ^, trO K •* O Si t ATTENUATION _L I REGULATING NETWORK 4- I ATTENUATION _) LU 1 X I f^ 1 OJ 1 '^ 1 < o ID i t a UJ a. \ ATTENUATION I * I REGULATING NETWORK < Q. J UJ -DO. ATTENUATION tr I ATTENUATION a. I zui en O I O 01 > C O s o 'S- H RELATIVE VOLUME IN db 576 BELL SYSTEM TECHNICAL JOURNAL At the regulating repeater stations the arrangement is the same as at the non-regulating stations, except that another stage is added to the amplifiers. This stage includes a potentiometer associated with relays controlled by the master pilot-wire mechanism, the whole being arranged so as to compensate for the changes in transmission loss of the cable pairs caused by temperature changes. In the lower part of Fig. 3 is shown a transmission level diagram, from which can be noted the losses and gains introduced by the different parts of the system, for a frequency of 1,000 cycles. Cable The transmission paths are provided by means of 16 B. & S. gauge non-phantomed pairs having a capacitance of 0.062 microfarads per a. u a z o 1- < z LJ y- \ .36 1 .32 1 i 1 .28 / / / / .24 ^ - - — •" .20 / / 16 d / / / / 1? / _J to 30 50 100 200 500 1,000 2.000 5,000 8,000 CYCLES PER SECOND Fig. 4 — Attenuation-frequency characteristic for 16-ga. B-22 cable pairs at 55° F, terminated in characteristic impedance. mile. These pairs are loaded with 22-milhenry inductance coils spaced 3,000 feet apart. Present long distance message telephone circuits in cable have loading coils spaced 6,000 feet apart. The nominal cutoff frequency of the new circuit is about 1 1,000 cycles, permitting effective transmission of a frequency band extending up to about 8,000 cycles. The nominal impedance is about 800 ohms and the attenuation per mile, at 1,000 cycles and average temperature, about .24 db. Fig. 4 LONG DISTANCE CABLE CIRCUIT :)/ shows the attenuation at average temperature plotted as a function of frequency, while Fig. 5 shows the line impedance. Figure 6 shows how the cable circuit attenuation varies with tem- perature at different frequencies. As will be seen from the curves, temperature change produces effects not only in the series losses but also in the shunt losses. The series losses are changed largely because 10 30 50 1400 \ 1200 \ \ \ \. 1000 \ \ > s s \ s \^ \ ^ ■ — (/) 800 I \ RE >I5 s [t/ *h CE O 600 S s > N \ s \ N 400 \ \ 200 \ s "s V. 0 ^ "^" RE \C T/ kh CE 100 200 500 1000 2000 5000 8000 CYCLES PER SECOND Fig. 5 — Mid-coil characteristic impedance for 16-ga. B-22 cable pairs at 55° F. the resistance of the copper cable conductors changes with tempera- ture, and to a smaller degree because of changes in effective resistance of the loading coils. The shunt losses change with temperature due largely to changes in the conductance losses and, to a lesser extent, changes in the cable capacity with temperature. The conductance loss is approximately directly proportional to frequency so that it has maximum effect at the highest frequency. The effect of temper- ature on the conductance loss is opposite to the effect of tempera- ture on the series loss so that increase of temperature reduces the shunt loss. The matter of securing the necessary electrical separation between the 16-gauge program transmission circuits and the other circuits contained within the same lead sheath involved particular study. The use of shielded pairs was considered. Such use of shields, how- 578 BELL SYSTEM TECHNICAL JOURNAL ever, would very greatly increase the space occupied by each program circuit and, therefore, considerably increase the cost. By careful design of the cable and control of methods of splicing, it was found possible to avoid the use of shields. It was not found practicable, however, to make use of the phantom possibilities on the program pairs. < UJ o z 3 SEplESUOSS VABIATID'J .„^ ^ S = ^ ^ ■ ^ ■-. ^ ^ 2 ,^ s^ N V ^ **• ^ ^ ^ \ \ \ \ TCTAl! LPS5> VARIAinCiN \ 0 ^ " ■ ^^ SHUNT L0£.£ N /APIA nor I K s V -2 \ \ \ 10 30 50 100 200 500 CYCLES PER SECOND 1000 2000 5000 8000 Pig. 6 — Attenuation variation of 100 miles of 16-ga. B-22 loaded cable circuit for a temperature change from 55° F. to 109° F. The method adopted was as follows: Restrict transmission over a particular 16-gauge program transmission pair, as a general propo- sition, to one direction only. Place the program pairs assigned to transmission in one direction among the 19-gauge quads used for four-wire transmission paths going in the same direction, and the program pairs transmitting in the other direction in the oppositely- bound four-wire group. Fig. 7 shows a cross-section of a typical cable containing six program transmission pairs, three for transmission in each direction. Loading Coils The 22-milhenry loading coils used on the program transmission circuit have cores of compressed powered permalloy, which is the magnetic material now generally used in the Bell System loading coils.'* Their overall dimensions are the same as those of the loading coils for the ordinary telephone circuits in toll cables. ■* W. J. Shackelton and I. C. Barber, "Compressed Powdered Permalloy — Manufacture and Magnetic Projjerties," Transactions, A. I. E. E., Vol. 47, No. 2, Al)ril, 1928. LONG DISTANCE CABLE CIRCUIT 579 Typical effective resistance-frequency curves for the loading coils are given in Fig. 8; these curves include current magnitudes greater than those involved in program transmission service. The core eddy current losses, varying with the square of the frequency, are prin- O — MAKE-UP 19 QUADS— 16 GAUGE 6 PAIRS — 16 GAUGE 114 QUADS — 19 GAUGE SHEATH THICKNESS 0.125 INCH OUTSIDE DIAMETER — 2.6 INCHES (aPPROX) — LEGEND — PROGRAM PAIR FOUR-WIRE QUAD FOR TRANSMISSION IN ONE DIRECTION (40 QUADS) FOUR-WIRE QUAD FOR TRANSMISSION IN OTHER DIRECTION C^O QUADS) TWO-WIRE QUAD — 16 & 19 GAUGE HEATH Fig. 7 — Cross section of typical full sized cable. cipally responsible for the resistance increase at the higher frequencies. The increase of attenuation with frequency caused by these core losses is readily corrected, however, by the attenuation equalizers which, as described later, also correct for the attenuation-frequency distortion caused by other factors in the cable circuit. X o UJ o z CO tn UJ a: u > H O u / y i ll L 5 M ILS -2 0 11 ^ ^ F1 ^ 10 30 50 100 200 500 1000 CYCLES PER SECOND 2000 5000 8000 Fig. 8 — Effective resistance of 22 milli-henry loading coils used on program transmission circuits in toll cables. Owing to the low hysteresis loss of the compressed powdered per- malloy material, the non-linear distortion introduced by the loading is inappreciable within the range of volumes handled by this program system. For example, in a 1,000-mile circuit for the condition where 580 BELL SYSTEM TECHNICAL JOURNAL the power output from each repeater is 1 milliwatt (corresponding roughly to the average power when the program volume is maximum), the non-linear distortion that occurs in the loading causes an increase in the overall transmission loss of the circuit of only 1 db at 8,000 cycles, as compared to the loss for negligibly small power. At 1,000 cycles, the loss increment for the same comparison is .13 db. The Fig. 9 — 6-Coil loading case for cable program circuit, ^th actual size. harmonic production in the coils is another measure of their excellence with respect to non-linear distortion. For a 400-cycle line current of 1 milliampere, the ratio of the third harmonic e.m.f. generated in an individual loading coil to the fundamental e.m.f. is equivalent to a loss of 80 db. The current magnitude above assumed corresponds approximately to the maximum repeater output (single-frequency basis) of 1 milliwatt; the average current that flows in the loading coils is very much smaller due to the smaller average repeater output and to line attenuation. In this connection, it is to be noted that the third harmonic voltage varies with the square of the magnitude of the fundamental current, and directly with frequency. The higher harmonics are, of course, much lower in magnitude than the third harmonic. LONG DISTANCE CABLE CIRCUIT 581 For the purpose of minimizing crosstalk, the loading coils are shielded individually by placing each in a metal container. In addition, the leads to the coils in the stub cable and within the coil case are cabled in individually shielded quads, the "IN" and "OUT" leads of a loading coil being in the same shielded quad. As a result of these precautions, the crosstalk between the loading coils is practically negligible. Even at the highest frequencies involved in program transmission, the crosstalk is only of the order of 2 crosstalk units, corresponding to an attenuation of about 114 db. The shielded program circuit coils required on a given cable are potted separately from the loading coils used on the telephone message circuits. These cases are of welded steel construction. A photograph of a 6-coiI case for underground use is shown in Fig. 9. The under- ground type of case has a special protective coating supplemented by a wrapping of heavy paper. Amplifiers Figure 10 is a schematic of the amplifier circuit as used at non- regulating repeater stations. (At regulating stations an automatic transmission adjustment stage is added, which will be described later.) Front and rear views of the amplifier, which is designed for relay-rack mounting in accordance with present-day telephone practice, are shown in Figs. 11 and 12. The lower panel is the amplifier, the upper the transmission adjustment stage, which will be treated later. In the regular amplifier a standard Western Electric 102-F tube is used in the first stage and a 101-F tube in the second. The amplifier uses resistance coupling and the various coils which affect the trans- mission performance have very high inductance so as to give the device very uniform transmission performance at different frequencies. The use of permalloy for the cores of these coils makes it possible to obtain the necessary high inductance without going to unreasonable coil dimensions. The gain is controlled by 5 db and 10 db artificial lines in the input circuit with a slide-wire potentiometer for the fine adjustments. Resistances in the grid circuit of the second tube allow an adjustment of the gain at high frequencies. Increasing the re- sistance causes a decrease in gain at these frequencies. The grid potential of the tubes is obtained from voltage drop in the filament circuit. The condenser in the grid circuit with its associated resistance serves to keep noise which may be present in the filament circuit from entering the grid circuit. The ideal amplifier should give a constant gain for all frequencies over the band to be transmitted regardless of variations in magnitude 582 BELL SYSTEM TECHNICAL JOURNAL \- D Q. I- D O -^MJ^ T -JO -"MV^ ■— o^^AV^ <: T 1 nJl^ - UJ UJ Z z < o 2 -4 o I- -6 1 1 ".. ^ ! i 1 i 0 3 0 5 0 100 2C )0 5C )0 1 0 00 20 00 50 OC ) 3( )( )0 CYCLES PER SECOND Fig. 13 — Gain-frequency characteristic of non-regulating repeater without line equalizer. With respect to departure of the amplifier from linearity, the effects produced are largely caused by the vacuum tubes. Very little of such distortion is introduced by the amplifier coils. Measurements on one of these amplifiers have shown that with a single frequency output of 1 milliwatt, which is about the average power corresponding to the maximum program volume, the second harmonics are about 50 db weaker than the fundamental, i.e., differ in power from the fundamental in the ratio 1 to 100,000. Other harmonics are lower in magnitude. Non-linearity in the amplifier also manifests itself by change in gain with current strength. In this amplifier a variation in load from 1 milliwatt to a much weaker load causes a change in gain of only 586 BELL SYSTEM TECHNICAL JOURNAL about .01 db, while a variation in load from 60 milliwatts to 6 milli- watts causes a change in gain of about .4 db. The input and output coils in the amplifier and, in the case of the regulating repeater, the retardation coil also, tend to delay the trans- mission of low-frequency currents more than those of high frequency; an action which is due to the inductance of these coils shunting the circuit. As this reactance becomes less at the lower frequencies, the delay becomes greater. It can be reduced by increasing the values of shunting inductances. It is largely to reduce this effect that permalloy core coils of extremely high inductance are used, as noted .007 ! 1 .006 .005 w § 004 w .003 ' I .002 \ \ 001 \ \ s, s 0 Iv, •*. 10 30 50 100 300 500 1000 CYCLES PER SECOND Fig. 14 — Delay-frequencj- characteristic of non-regulating repeater. above. The condensers appearing in series also cause delay at low frequencies and must be given capacity suf^ciently great to keep the delay within proper limits. Inductance in series or capacity in shunt will also result in delay at the high-frequency end. However, in the frequency range covered by these amplifiers there is no difficulty in keeping this delay small enough to be negligible. The delay characteristic of one of these amplifiers is shown in Fig. 14. With 40 amplifiers in tandem, the overall delay at 35 cycles is 75 milli-seconds greater than at 1,000 cycles, while there is no appreciable difference between the delay at 1,000 cycles and the delay at higher frequencies. LONG DISTANCE CABLE CIRCUIT 587 Attenuation Equalizers As will be observed from Fig. 4, the transmission loss of the cable circuit varies considerably with frequency. Since the amplifier has a flat gain characteristic, an attenuation equalizer is called for to correct the distortion introduced by the cable. A diagram of one of these equalizers is shown in Fig. 15. In Fig. 16 is shown the loss Hh- -*MAr- ■Hh- •^ — f— ^MAAr<^^ ^ — crQir?rv-o-||- ' 2 o— 4 — o Fig. 15 — Schematic circuit of attenuation equalizer. introduced by a 50-mile section of cable at average temperature, the loss introduced by one of these attenuation equalizers and the total loss of line and equalizer with the offsetting gain introduced by the amplifier. Automatic Device to Overcome Effects of Varying Temperature Trans- mission Adjusting As the temperature of the cable changes its attenuation changes, the amount of the change being different at different frequencies. Referring back to Fig. 6, it is seen that on a cable circuit 1,000 miles long a temperature change from 55° F. to 109° F. causes changes in the transmission as follows: At 100 cycles 18 db change, power change of 63 At 1,000 cycles 28 db change, power change of 625 At 8,000 cycles 3 db change, power change of 2 When it is appreciated that in an aerial cable a temperature change of 54° F. may take place in only a day or two, the importance of compensating for this effect may be appreciated. 38 588 BELL SYSTEM TECHNICAL JOURNAL In order to compensate for this effect of varying temperature, a regulating stage is added to the amplifiers at the various regulator stations. Fig. 17 shows how the regulating network stage is added to one of the amplifiers and also shows the general nature of the regulating network circuit. Because of the peculiar and complicated way the transmission loss of the cable circuit varies with temperature, Si -o I z g !< z> z UJ ^- 26 f 24 22 20 18 16 12 10 8 6 4 2 0 Fig 10 30 50 LOSS INTRODUCEC BY LINE GAIN INTRODUCED BY REPEATER 106 206 560 1060 CYCLES PER SECOND 2000 5000 8000 16 — Attenuation-frequency characteristic of line equalizer and 50 miles of 16-ga. B-22 cable circuits. a somewhat complicated regulating network is called for. Front and rear views of one of these regulating networks are shown in Figs. 11 and 12, the upper panel being the regulating network and the lower the normal amplifier. Fig. 18 shows how the gain characteristic of the amplifier is altered by different steps of the regulating network. This is very closely complementary to the change in cable loss caused by the temperature variations and thus it will be evident that the effects of the temperature changes are largely eliminated. Delay Equalizers The velocity of transmission through a loaded cable decreases as the frequency is increased toward the cutoff point of the loading. To neutralize this effect, delay-equalizing networks are inserted in the circuit which retard the lower frequencies, thus equalizing the velocity LONG DISTANCE CABLE CIRCUIT 589 en C3 u O 0) +-» be c 3 be ID o s o be 590 BELL SYSTEM TECHNICAL JOURNAL o Q. UJ I- (/1 C H Q UJ cc cr UJ U- UJ cr I UJ cr O > ^^ --, . -^ / -^ ^_ ^ / 1 0 3 0 5 0 IC )0 2( 50 500 1000 2000 500C 3000 CYCLES PER SECOND Fig. 18 — Gain-frequency characteristic of regulating repeater without line equalizer. 4.0 3.5 (0 O z o o iij UJ 3.0 2.5 2.0 1.5 1.0 DELAY INTRODUCEDI BY LINE DELAY INTRODUCED BY LINE AND EQUALIZER 10 30 50 100 200 500 1000 2000 CYCLES PER SECOND 5000 8000 Fig. 19 — Delay-frequency characteristic of 50 miles 16-ga. B-22 cable with and without delay equalizer. LONG DISTANCE CABLE CIRCUIT 591 of transmission through the combination of cable and networks for all frequencies in the band to be transmitted. Fig. 19 shows the delay characteristic of a section of cable 50 miles in length, with and without the delay-equalizing networks. The delay is seen to be main- tained within db 0.05 milli-second of a constant value. A schematic circuit of these networks is shown in Fig. 20. With the greatest MUTUAL INDUCTANCE BETWEEN THESE TWO WINDINGS THIS COIL IN SAME POT WITH OTHER TWO WINDINGS Fig. 20 — Schematic circuit of section of delay equalizer. For a 50-mile equalizer, three kinds of sections are used which vary in the resonant frequency and in the sharpness of resonance. The first three sections are of one kind, the fourth is of another and the last seven are of the third kind. length of cable circuits which will be used in this country for program transmission, this amount of deviation per section is not sufficient to cause objectionable distortion. For a 50-mile section uncorrected, the delay at 8,000 cycles would be 0.9 milli-second greater than at 1,000. A description of these delay-equalizing networks with the theory of their performance is being presented in another paper so that a more detailed description is omitted in this paper. Office Wiring Owing to the wide frequency range transmitted over the circuit, special care must be taken with the office wiring. This is to avoid excessive variations in the losses introduced by this wiring due to changing humidity conditions. A new type of insulated cable is used in which the textile material of the insulating wires has been very thoroughly washed to remove all traces of foreign substances, so that the absorption of moisture with its accompanying increase in loss is greatly reduced.^ The office cabling is also shortened as much as possible, the outside cable connecting directly to the repeaters without * H. H. Glenn and E. B. Wood, "Purified Textile Insulation for Telephone Central Oftice Wiring," .4. /. E. E. Transactions, Vol. 48, April. 1929. 592 BELL SYSTEM TECHNICAL JOURNAL passing through the usual test board. At points in the circuit sensitive to noise interference or crosstalk where the energy level of the trans- mitted signals is very low, the circuit units are connected by means of shielded pairs. This shield is connected to filament ground, as are also the cases of the various transformers which are insulated from the supporting metallic frame. The unavoidable noise potential existing between the frame and the filament circuit cannot then pro- duce any appreciable disturbance in the circuit. Overall Perjormance of System A measurement of the transmission loss of the 2,200-mile test length of B-22-N cable circuit gave results as indicated in Fig. 21. UJ _l I O UJ a. UJ I a. z < -14 -12 -10 - 8 - 6 -4 - 2 0 +2 +4 +6 1 /c \ / / / 1 / B V / \ A B 1 A /I — H ^ y ^ > \ / A 1 / B y\f C # 1 / 1 0 3 0 5 0 1C )0 2( 30 5( 30 1,( 300 2.0 00 5,0 oc ) i i.OOO CYCLES PER SECOND Fig. 21 — Transmission-frequency characteristics of 2,200 miles of 16-ga. B-22 cable program transmission circuit. Curve A — Ideal characteristic. Curve B — • Measured characteristic. Curve C — Line without equalizers. It will be observed that over the range from 35 cycles to 8,000 cycles the transmission loss was practically the same at all frequencies, departing only about ± 2 db. For comparison, another curve (Q is given on the same drawing showing the transmission characteristic which would have been obtained if distortionless amplification had simply been added to the line with no attenuation equalizers. The delay-frequency characteristics of the 2,200-mile test length of B-22 circuit are shown in Fig. 22. Two curves are given, one for the circuit without delay equalizers, the other with delay equalizers. With respect to non-linear distortion, it was found by test that when the maximum volume was held at about — 5 db, as read on a LONG DISTANCE CABLE CIRCUIT 593 volume indicator, or about 1 milliwatt of average power, the non- linear distortion became inappreciable. As a matter of fact, occa- sional bursts up to at least 0 db were not badly distorted. It may be observed that the — 5 db volume is about 10 db less than repeaters CIRCUIT WITHOUT DELAY EQUALIZERS CIRCUIT WITH DELAY EQUALIZERS CURVE A B DELAY AT 1000 CYCLES .106 SECONDS .168 SECONDS O z o o UJ w < _l UJ o A O nft \ V CiA. > V ; k, A, A /\ r .^' ( 1 10 30 50 100 200 500 1,000 CYCLES PER SECOND 2P00 5,000 8,000 Fig. 22 — Delay characteristics of 2,200 miles of 16-ga. B-22 cable program transmission circuit. of the same nominal capacity and loading coils of similar character- istics handle without appreciable distortion under regular message telephone circuit conditions. The minimum volume which could be transmitted over the cable circuit, which was set by noise and crosstalk picked up by the program circuit, was found to be about — 50 db at the repeater outputs. This means that the volume range carrying capacity of the circuit was about 45 db, just a little more than the figure 40 db which was pre- viously mentioned as a reasonable standard for present-day conditions of broadcasting. If short bursts of music are allowed to go up to the zero volume at the repeater outputs, the system can evidently handle about 50 db volume range. Using special pickup apparatus and loudspeakers capable of handling practically the whole audible frequency range, tests have been made over the 2,200-mile looped-back circuit in which comparison was made of the transmission with and without the cable included. When an 594 BELL SYSTEM TECHNICAL JOURNAL 8,000-cycle low-pass filter was included under both conditions it was found that listeners had considerable difficulty in consistently picking a difference. In fact, the ordinary observer could not be relied upon to pick differences consistently even when the 8,000-cycle filter was not included. Conclusion This development was undertaken to provide a system for obtaining satisfactory channels for the transmission of broadcast programs in the rapidly growing cable network of the Bell Telephone System. The time required to complete such a development and the need for advance planning in the cable plant made it essential that the channels be adequate to render service for a number of years. Improvements in broadcast reproduction may be expected to continue and may very well result in changes in the present frequency allocations to give space for wider bands. The cable system described in this paper was, therefore, developed to possess transmission characteristics supe- rior to present-day radio systems, the margin anticipating improve- ments which may take place in the future. Acknowledgment The authors gratefully acknowledge the assistance of many of their associates in the preparation of this paper and particularly of Mr. H. S. Hamilton. Abstracts of Technical Articles from Bell System Sources Barkhansen Effect II. Determination of the Average Size of the Discontinuities in Magnetization} R. M. Bozorth and J. F. Dil- LINGER. When the magnetic field-strength acting on a ferro-magnetic material is changed, the magnetization changes discontinuously (Bark- hausen effect). These discontinuous changes have been examined in 1 mm. wires; an expression is derived and experimental arrangements are described for determining their average size for a given material in a given state of magnetization. Experimental determinations of the average size have been made for iron (including a single crystal and a hard-drawn wire), nickel, and several iron-nickel alloys (permalloys). The average size is greatest on or near the steepest part of the hysteresis loop. The greatest aver- age size, expressed as the volume of material the magnetization of which must be changed from saturation in one sense to saturation in the opposite sense to produce the same change in magnetization, is much the same for all of the materials examined, the extremes being 1.2 x 10~^ cm.'' for annealed iron and 45 x io~^ cm.^ for 50 per cent nickel per- malloy. This shows that the sizes of the discontinuities do not depend to any considerable extent on the size or kind of crystals. Criticism is made of previous work on the size of the coherence re- gion, the region within which the change in magnetization is confined. Although the effect of a single discontinuity in magnetization may be detected as far as 10 cm. from its source because of the eddy-currents induced, the experimental evidence is consistent with the view that the permanent change in magnetization is confined to the volume in terms of which the size of the discontinuity is measured as stated above, always less than 10~^ cm.^. Particle Size as a Factor in the Corrosioji of Lead by Soils.- R. M. Burns and D. J. Salle v. In order to determine that part which particle size plays in the corrosion of lead by soils, lead specimens were buried in sands (generally inert in character) of various particle sizes and were maintained for periods of time ranging from 8 days to 5 months at 40° C. in a closed system in which the humidity and the composition of the atmosphere were controlled. ' Phys. Rev., Apr. 1, 19.30. - Ind. and Engg. Cliem., Mar., 1930. 595 596 BELL SYSTEM TECHNICAL JOURNAL It has been shown that lead is corroded by contact with moist inert sands in the presence of air, and that the rate of attack is increased by increasing within certain Hmits the particle size of the sand, the mois- ture content of the sand, and the oxygen content of the atmosphere. Corrosion is caused by oxygen concentration cells which are set up as a result of the partial or complete exclusion of oxygen at the points of contact of metal and soil. Soil particle size influences the rate of corrosion by determining the extent of the electrode areas, and therefore the degree of cathodic polarization, of these oxygen concentration cells. Reverberatioji Time in "Dead'' Rooms} Carl F. Eyring. With the advent of radio broadcasting and sound pictures very "dead " rooms have been built, and the significant problem of just how much reverberation should be used in broadcasting and recording presents itself. The direct measurement of reverberation time or its calculation by the aid of a reliable formula, then, is an important aspect of applied acoustics. A reverberation time formula enables one to calculate the reverberation time once the volume, surface area and average absorp- tion coefficient of the surface of the room are known ; or if the reverber- ation time is measured it enables one to calculate the average coefficient of absorption of the surface treatment. A correct reverberation time formula is, therefore, much to be desired. Theories of reverberation leading to Sabine's reverberation time equation have been given by W. C. Sabine (1900), Franklin (1903), Jaeger (1911), Buckingham (1925). Recently Schuster and Waetz- mann (1929) have pointed out that Sabine's formula is essentially a "live" room formula and they have shown as we also show that the reverberation time equation varies somewhat with the shape of the room. The present paper presents an analysis based on the assump- tion that image sources may replace the walls of a room in calculating the rate of decay of sound intensity after the sound source is cut off, which gives a form of reverberation time equation more general than Sabine's; it points out the difference between the basic assumptions leading to the two types of formuht; it adds experimental data which support the more general type; and it ends with the conclusion that no one formula without modification is essentially all inclusive. The Provision of Radio Facilities for Aircraft ComnmnicationJ E. L. Nelson and F. M. Ryan. This subject is discussed by the authors from the viewpoint of the radio engineer. The periods of fundamental ■■' Jour. Acou. Soc. Amer., Jan., 1930. * Soc. Auto. Engs., Mar., 1930. ABSTRACTS OF TECHNICAL ARTICLES 597 study and development of apparatus are stated to be drawing to a close, and we are said to be well advanced toward the third and last period — that of general application to commercial flying. The discussion of radio-communication outfits is based on aircraft equipment recently developed by the Bell Telephone Laboratories for receiving weather reports and beacon signals and for two-way tele- phonic communication between the airplane and ground stations. Units of different types of apparatus for use in small mail-planes and in large transports are illustrated and described, together with tabular data of sizes and weights of individual units of both general types of outfit. Information regarding the requirements of shielding, bonding and installation is given, and the airplane factory is stated to be the place where provisions for radio installation can best be made. If suitable provisions have been made therefor, the installation of two-way radio equipment is said to be simple and inexpensive. A number of the larger air-transport organizations have made note- worthy progress toward providing suitable radio systems and the De- partment of Commerce is giving much assistance in the way of radio aids to air navigation, but a great deal of work remains to be done by the industry as a whole and numerous problems will require solution. New requirements will be encountered as the number and size of air- planes increase, but continuing radio studies promise that the develop- ment of aircraft radio communication will keep abreast of the develop- ment of airplanes. Transmission Characteristics of a Short- Wave Telephone Circuit.^ R. K. Potter. A method of observing and recording the audio- frequency transmission characteristics of a short-wave radiotelephone channel is described. These characteristics undergo rapid changes. They appear to be the result of wave interference between signals arriv- ing at the receiver over paths of different group or electrical length possibly combined with the distortion produced by a progressive change in the angle of rotation of the polarization plane with frequency over the signal band. The persistence of certain pattern shapes during the observation periods and the changes in these shapes from hour-to-hour suggest that they are the result of progressive rather than erratic dis- turbances in the transmission medium. Times when the audio-fre- quency characteristics were flat were very rare. However, a consider- ^ Proc. Inst. Radio Engineers, Apr., 1930. 598 BELL SYSTEM TECHNICAL JOURNAL able departure from flatness may occur without serious effect on the intelHgibiUty of the speech transmission. Synthetic patterns used in the analysis of the characteristics are explained and illustrated. Types of audio-frequency distortion result- ing from selective fading are discussed. The effect of frequency or phase modulation in producing distortion on such a circuit is con- sidered. Records are shown of the effect of an automatic gain control, follow- ing carrier amplitude variations, upon the audio-frequency transmis- sion characteristic. "Rapid" fading records revealing unlike fading on radio frequencies separated by 170 cycles are included. The sea- sonal variation in susceptibility of the circuit to this "rapid" fading is illustrated. The records mentioned above are for ordinary modulated carrier transmission and involve the results of interaction between the two side bands in the detection process. There are also shown records made on single side-band carrier-suppressed transmission. In this case detection does not modify the frequency amplitude relations and the record delineates directly the frequency-amplitude characteristics of the received radio-frequency band. Age Hardening Lead-Calcium Alloys.^ Earle E. Schumacher and George M. Bouton. The lead end of the system lead-cal- cium has been investigated and a constitutional diagram given. A peritectic reaction has been discovered and the solid solubility of cal- cium in lead has been determined for five temperatures. The solubility changes from 0.1 per cent calcium at 328.3° C. to approximately 0.01 per cent calcium at 25° C. Data for locating the solid solubility curve were obtained from thermal analysis, electrical conductivity measure- ments, microscopic examinations and age hardening studies. It has been shown on the basis of laboratory tests that greater tensile strengths and resistances to fatigue failure can be developed in some of the lead-calcium alloys than in the lead — 1 per cent antimony alloy. Certain lead-calcium alloys have been suggested as sheathing materials for electrical cables. Preparation of Air of Knoum Humidity and Its Application to the Calibration of an Absolute-Humidity Recorder J A. C. Walker and E. J. Ernest, Jr. An apparatus is described whereby constant flowing mixtures of air and water vapor may be prepared, in which the moisture content varies not more than 0.001 per cent by volume, over 6 Metals & Alloys, Mar., 1930. ' hid. and Engg. Chem., Api, 15, 1930. ABSTRACTS OF TECHNICAL ARTICLES 599 long periods. This apparatus has been utilized to calibrate a sensitive humidity recorder capable of continuously recording atmospheric humidities up to 2.9 per cent by volume of water vapor in air (equiva- lent to about 95 per cent relative humidity at 25° C.) with a sensitivity of 0.0016 per cent by volume (0.05 per cent relative humidity at 25° C). The use of the recorder in connection with the constant humidity appa- ratus is described and certain typical data illustrating the performance are given. Contributors to this Issue Millard W. Baldwin, Jr., E.E., Cornell, 1925; M.A., Columbia, 1928; Bell Telephone Laboratories, 1925- Mr. Baldwin has been engaged in studies of vacuum tube modulation ; more recently his work has had to do with some of the problems of picture transmission and television. David George Blattner, B.S.E.E., Kansas State Agricultural College, 1911; Assistant Instructor in Physics, Kansas State Agri- cultural College, 1911-13. Engineering Department, Western Elec- tric Company, 1914-25. Bell Telephone Laboratories 1925-. Mr. Blattner's work has been in loud speaker, public address systems, and phonograph recorder and reproducer developments. L. G. BosTWiCK, B.S. in E.E., University of Vermont, 1922; Amer- ican Telephone and Telegraph Company, Development and Research Department, 1922-26; Bell Telephone Laboratories, Inc., Research Department, 1926-. While with the Development and Research Department, Mr. Bostwick's work involved general problems on systems for the high quality transmission of speech and music; since then his work has been largely on loud speakers and loud speaker measuring methods. S. Brand, B.S., Trinity College, 1915; Yale LTniversity Graduate School, 1915-17; U. S. Air Service, 1917-19; Plant Department, Southern New England Telephone Company, 1920-23; Department of Development and Research, American Telephone and Telegraph Company, 1923-. Mr. Brand has been engaged mainly in transmis- sion development work on repeatered circuits. A. B. Clark, B.E.E., University of Michigan, 1911; American Tele- phone and Telegraph Company, 191 1-. Toll Transmission Develop- ment Engineer, 1928-. Mr. Clark's work has been largely concerned with toll telephone and telegraph systems. Lloyd Espenschied. Mr. Espenschied is in charge of radio devel- opment, assisting the Transmission Development Engineer, Depart- ment of Development and Research, American Telephone and Tele- graph Company. He joined the Bell System in 1910, having graduated from Pratt Institute the previous year. He has taken an important part in practically all of the Bell System radio developments, beginning with the first long-distance radio-telephone tests of 1915, at which time 600 CONTRIBUTORS TO THIS ISSUE 601 he received the voice in Hawaii from Arlington, Va. He has partici- pated in a number of international conferences on electric communica- tions. Frank Gray, B.S., Purdue, 1911; instructor and graduate student in physics at the University of Wisconsin, Ph.D., 1916; member of the Naval Experimental Station during the war. Mr. Gray entered the Bell Telephone Laboratories — then the Engineering Department of the Western Electric Company — in 1919 and has been closely associated with Dr. Ives in his studies on light. C. W. Green, B.S. in Electrical Engineering, University of Wiscon- sin, 1907 ; Instructor and Assistant Professor, Massachusetts Institute of Technology, 1907-17; Captain 1917, Major 1918, U. S. Army; Bell Telephone Laboratories, 1919. Mr. Green's work has had to do with the development of Carrier Telephone Systems and Voice Frequency Repeaters. Herbert E. Ives, B.S., University of Pennsylvania, 1905; Ph.D., Johns Hopkins, 1908; assistant and assistant physicist, Bureau of Standards, 1908-09; physicist, Nela Research Laboratory. Cleveland 1909-12; physicist. United Gas Improvement Company, Philadelphia, 1912-18; V. S. Army Air Service, 1918-19; research engineer, \\'estern Electric Company and Bell Telephone Laboratories, 1919 to date. Dr. Ives' work has had to do principally with the production, measure- ment and utilization of light. C. E. Lane, A.B., University of Iowa, 1920; M.S., University of Iowa, 1921; Engineering Department of the Western Electric Com- pany, 1921-25; Bell Telephone Laboratories, 1925-. During the last four years Mr. Lane has been engaged in the development of such trans- mission networks as filters, attenuation equalizers and phase correctors in the Apparatus Development Department. The five years prior to this were spent by him in the Research Department, engaged in general studies in acoustics, such as speech, hearing and loud speaker dev'elop- ment. W. H. Martin, A.B., Johns Hopkins University, 1909; B.Sc, Massachusetts Institute of Technology, 1911; American Telephone and Telegraph Company, Engineering Department, 1911-19; Depart- ment of Development and Research, 191 9-. Mr. Martin's work has related particularly to transmission of telephone sets and local ex- change circuits, transmission quality and loading. H. Nyquist, B.S. in Electrical Engineering, North Dakota, 1914; M.S., North Dakota, 1915; Ph.D., Yale, 1917; Engineering Depart- 602 BELL SYSTEM TECHNICAL JOURNAL ment, American Telephone and Telegraph Company, 1917-19; De- partment of Development and Research, 1919-. Mr. Nyquist has been engaged in transmission work particularly relating to toll cables. H. S. OsBORXE, B.S., Massachusetts Institute of Technology, 1908; Austin Research Fellow in Engineering, 1908-10; Eng.D., 1910; American Telephone and Telegraph Company, Engineering Depart- ment, 1910-19; Department of Development and Research, 1919-20; Department of Operation and Engineering, 1920-. Mr. Osborne is Transmission Engineer and as such is responsible for assisting the Associated Companies in connection with telephone and telegraph transmission and protection matters. J. C. Steinberg, B.Sc, M.Sc, Coe College, 1916, 1917; U. S. Air Service, 1917-19; Ph.D., Iowa, 1922; Engineering Department, West- ern Electric Company, 1922-25; Bell Telephone Laboratories, 1925-. Dr. Steinberg's work since coming with the Bell System has related largely to speech and hearing. H. M. Stoller, E.E., Union College, 1913; M.S. in Electrical En- gineering, 1915; Engineering Department of Western Electric Com- pany, 1914 and 1916-25; Bell Telephone Laboratories, 1925-. Mr. Stoller's work has dealt with special problems connected with electrical power machinery, particularly voltage and speed regulators; multi-fre- quency generators employed in voice frequency carrier telegraph sys- tems; and synchronization and speed control equipment for sound- picture systems. W'lLLiAM Wilson, Victoria University of Manchester, 1904-10; B.Sc, 1907; M.Sc, 1908; Cavendish Laboratory, Cambridge Univer- sity, 1910-12, B.A., 1912; Lecturer in Physics, Toronto University, 1912-14; D.Sc Manchester, 1913. Engineering Department Western Electric Company, 1914-24; 1925- Bell Telephone Laboratories; Assistant Director of Research 1928-. Dr. Wilson has published numerous papers on radioactivity and thermionics and since 1917 has been in direct charge of vacuum tube development and design and since 1925 has also been in charge of radio development. The Bell System Technical Journal October, 1930 Chemistry in the Telephone Industry ^ By ROBERT R. WILLIAMS An account is given of the activities of the Chemical Department of the Bell Telephone Laboratories. In the Laboratories chemists act chiefly as advisers and critics. They concern themselves with such problems as the theory of chemical structure as related to dielectric properties and simultan- eously attack the task of making an improved substitute for gutta-percha which renders possible a transatlantic telephone. They are interested in the colloidal structure of cotton and silk and the influence of moisture and elec- trolytes on their insulating properties. The dispersion hardening and fatigue resistance of lead and its alloys, the fabrication of platinum-alloy vaciuim- tube filaments, and of new magnetic materials such as permalloy and permin- var have required their attention. The corrosion of cable systems is due largely to stray currents but is stron:'ly influenced by chemical factors. Other underground corrosion, and particularly the slower and more insid'ous corrosion and tarnishing of indoor telephone apparatus, have justified a broad program of investigation of the corrosive factors involved in ocean, earth, and atmosphere. Related to these studies are those of protective finishes, whether metallic coatings or organic paints and lacquers. The per- manence of a great variety of materials must often be predicted as best it may without the test of service life. This interest in permanence is reflected in a program of experiments in preservation of telephone poles. THE public mind associates the chemist with glass retorts and evil smells, with war gases or the glare of furnaces against the sky. To the technical leader in industry, however, the wide distribution of chemists outside predominantly chemical enterprises has become a familiar fact. A casual reference to the thirty subject classification headings in Chemical Abstracts will serve to illustrate how widely indus- trial chemists have become disseminated and how large a volume of work they are producing. When one reflects that all engineering is essentially applied physics, and that it has become subdivided into a score of specialized fields, it seems very natural that chemistry also should have found varied appli- cations as the sum total of chemical knowledge has increased. Perhaps the day is not far distant when the term ' ' physical chemist ' ' will repre- sent to the lay mind as well-defined and distinct a calling as that of civil engineer. It is therefore only a part of a general movement in industry that has placed the chemist in a position of some importance in the telephone business. His relative importance in the communication field is small, as the industry must permanently remain essentially electrical rather ^Iiid. (3° Engg. Cheni., April, 1930. 603 604 BELL SYSTEM TECHNICAL JOURNAL than chemical. His usefulness depends primarily, not upon the number or size of the operations which are entrusted to his exclusive care, but upon the distinctive mode of thought which he contributes to a critical consideration of the methods and processes in use. As in medicine, so in telephony the chemist is an aid to progress, not a prime mover. If one endeavors to define the distinctive mode of thought of chem- ists, he is at once led to point out the fact that the chemist by his train- ing intuitively tries to account for most phenomena by a consideration of the composition of the materials involved. From his first days in the laboratory he is taught what "chemically pure" means and learns that even minor impurities may often have most important consequen- ces, good or bad. While it is obviously true that many things happen without a chemical cause, it is equally, though less obviously, true that variations of composition or chemical changes in composition are fre- quently associated with the happenings. No one is so well qualified as the chemist to ferret out such obscure correlated and often important facts. The distinctive nature of the training of chemists tends to adapt them to the role of critics. The fundamentals of the old-school physics reached a point some thirty or forty years ago when it was felt that the whole field had been fairly thoroughly combed over and all essential principles were known. These principles became embodied in formulas and conventionalized modes of attack upon the problems of engineering which have often been accepted at more than their true value by the average product of the engineering school. Chemistry, on the other hand, has never reached so high a development. Even in first-year chemistry one encounters facts not explained satisfactorily by an^^ known theory, and the reading of even a score of pages of an elementary chemistry which has passed its first printing will bring one upon state- ments which require modification in the light of more recently acquired knowledge. By comparison with applied classical physics, chemistry is a youthful science and its devotees are inclined to a juvenile disre- spect for tradition. The enormous consumption in a telephone plant of such materials as lead and copper, paper and textiles, rubber and asphaltic compounds immediately implies the necessity of the chemist for the performance of his most conventional function, that of analysis. While the Bell Tele- phone Laboratories does not undertake the systematic inspection analy- ses of the large variety of products purchased, it does undertake a great volume of analysis of such products as a referee. Such work often re- veals defects of analytical methods or defective statement of specifica- tion requirements that necessitate large numbers of comparative analy- ses to form a basis for proper amendment. CHEMISTRY IN THE TELEPHONE INDUSTRY 605 Another large portion of the work of our analytical laboratory has to do with materials used in the prosecution of research problems by physi- cists, engineers, and chemists in other groups. For example, in the course of development of such magnetic materials as permalloy and perminvar, methods for the analysis of unusual alloys have had to be devised. The utilization of microchemical and electrometric methods has often offered a way out of difficulties. For instance, it became necessary in connection with a corrosion problem to measure accurately the amounts of volatile acids in certain woods, and for this purpose a differential electrometric titration method was developed. Spectro- scopic analysis finds its use, not only in the examination of minute speci- mens of material such as the deposits on vacuum tube filaments, but also in the estimation of minute impurities in grosser products, as, for example, the presence of zinc in solder. A not inconsiderable volume of research in analytical chemistry has grown out of these problems. The most important function of the chemist in the telephone labora- tory is not a conventional one. It consists in a scrutiny of the appara- tus, equipment, materials, and processes of the industry to determine where and how chemical reactions or variations in composition are affecting functional operation . Sometimes the problems so encountered lead to extended researches pursued over a period of years to answer specific questions or to accumulate a reservoir of general information to be drawn upon as needs arise. Sometimes questions are brought to us by workers in other fields, apparatus designers perhaps, questions such as can be answered offhand, or at most require a few days or weeks of work. A considerable part of the more interesting work is published. In order to make clear how chemistry applies to telephone problems, a number of examples have been chosen from various parts of the gen- eral field. Many of the tasks enumerated below are shared by the chemists with technical staffs of other departments. They will be dis- cussed from the chemical viewpoint with only sufficient reference to the general engineering considerations to make the problems intelligible. Chemical Coxstitutiox and Electrical Characteristics of Pure Substances The most fundamental piece of work in dielectrics which we are under- taking is a study of chemical constitution of pure substances in relation to their electrical characteristics. This work is going forward under the direction of H. H. Lowry, and is expected to serve as a valuable supple- ment to similar work being conducted elsewhere, principally in univer- sity laboratories. While it is well known to all chemists that aqueous 606 BELL SYSTEM TECHNICAL JOURXAL solutions are in general relatively good conductors, and that solutions of substances in fat solvents are relatively good insulators, a much closer analysis of the latter class is necessary for telephone purposes. Among the electrical characteristics of importance are dielectric con- stant, insulation resistance, a.c. conductivity, and dielectric strength. Each of these characteristics varies over quite wide ranges, depending upon certain conditions, of which temperature and frequency of alter- nation of the electric current may be mentioned as most important. In this work an endeavor is being made to determine, for example, how the symmetry of the molecule affects the dielectric constant, and to distinguish in dielectrics the contribution of energy loss made sever- ally by the electron, the atom, and the molecule. For such a purpose it is obviously necessary to deal with highly purified substances and to begin with those of simple chemical structure. Many of these, such as hexane, benzene, ethyl ether, and alcohol, are not expected to have the slightest importance as practical insulators, but serve to show how particular atomic groupings affect dielectric behavior. For this pur- pose we have included the methyl halides in our study, making meas- urements both in the pure state and in dilute solution, of dielectric con- stant and a.c. conductivity at temperatures ranging from 100° C. to the boiling point of the liquid. The frequency of the current has also been varied from 1 to 100 kilocycles. Rubber and Gutta-Percha Parallel with this study of the theory of dielectric behavior, a number of more immediately practical problems are being prosecuted, the ma- jority of them under the direction of A. R. Kemp. One of these relates to the use of rubber and gutta-percha in submarine insulation. The latter is the classical material for this purpose, while rubber has been regarded as an inferior substitute. The supremacy of gutta-percha is due in part to its mechanical characteristic of thermoplasticity, which permits it to be extruded as a continuous insulating layer about a con- ductor, requiring nothing but the simple process of cooling to convert it into a tough, firm sheath. An even more peculiar virtue of gutta- percha is the stability of its electrical characteristics during prolonged immersion in water. By patient experiments extending over several years it has been demonstrated that the inferiority of rubber in this one respect is due wholly to its non-hydrocarbon constituents. Methods for the elimination of these foreign substances, notably water-soluble salts, proteins, and quebrachitol, which could be applied without dam- age to the hydrocarbon, have required extended study. CHEMISTRY IN THE TELEPHONE INDUSTRY 607 In connection with the water-soluble impurities, an interesting prob- lem in osmosis arose. The absorption of water by rubber is found to be a direct function of its content of water-soluble substances and an in- |IUC i Fig. 1 — Apparatus for observing rate of oxygen absorption. Rubber, asphalt, or paint film to be studied is placed in bulb A (enclosed in thermostat not shown) and apparatus is filled with pure oxygen. By operation of the manometer C, water is electrolyzed in bulb B, generating gas automatically in quantity just sufficient to maintain oxygen in buret D constantly at atmospheric pressure. verse function of the salt content of the external water. Thus a sheet of raw rubber which is produced by the evaporation of latex, and there- fore contains all the natural water-soluble impurities, will, upon immer- 608 BELL SYSTEM TECHNICAL JOURNAL sion in distilled water, gradually absorb more water than was contained in the original latex, though the sheet still retains its original form, somewhat swollen of course by water absorption. On the other hand, rubber which has been carefully freed of all water-soluble impurities absorbs but a small percentage of water under the same conditions. Fresh-water cables are more liable to degradation from water absorp- tion than cables in salt water, and if the ocean were a saturated solution of sodium chloride the problem of the use of ordinary rubber would be materially simplified. A development of primary importance in this connection has been the elimination of the proteins from rubber. It appears that the pro- tein constituents form an intricate network which permeates the entire Fig. 2 — High grade gutta-percha in shapes popular with the forest gatherers. mass. While the amount of proteins present is too small to cause a large absorption of water, they do insure that such water as is absorbed causes a maximum electrical damage by extending the water-bearing filaments through and through the material. The partial hydrolysis of the proteins and the thorough coagulation of any remaining residues by CHEMISTRY IN THE TELEPHONE INDUSTRY 609 autoclaving the rubber by steam pressure has afforded a simple but effective means of stabilizing the material electrically against the action of water. Rubber also finds many uses in aerial insulation, a field which is rather backward in its development as compared with the tire industry. The latter has received an incalculable benefit from the large amount of technical research carried on in the last ten years. To a considerable extent it has been our task to adapt technical information derived from the tire industry to the needs of the telephone. The adoption of such expedients as accelerators and anti-aging compounds has contributed a great improvement in the field of insulation. In fact, the aging of rubber in the case of insulation is obviously even more important than in the case of tires. One scarcely expects automobile tires to last for more than a season or two, but rubber insulation must often be exposed to sun and rain for ten years or more. Rubber for wire insulation must offer resistance to cutting of the wire through the insulation under severe load, such as is produced by a deposit of sleet. Our chemists have therefore been called upon to con- tribute to the development of a compression-testing machine for rubber which automatically plots the reduction of thickness of wall under an increasing compressive load. By the use of this testing machine sur- prisingly great variations were discovered and corrected in the material supplied by different manufacturers. Other useful tools in the study of rubber have been developed. A method has been devised for the direct determination of rubber hydro- carbon in compounds by means of iodine titration. Direct gasomet- ric measurement of oxygen consumption by rubber and other organic materials is performed with a special apparatus for auto- matically maintaining oxygen pressure constant at one atmosphere. Included in the program are cooperative studies of various accelerated aging tests for rubber, notably the Geer test and that of Bierer and Davis. In these ways we hope we are repaying in part our debt to technologists of the rubber industry, as well as serving our own needs. Textiles Another general class of insulating materials in which study has been well rewarded is that of textiles. It has now been quite clearly shown that textile libers serve as filaments upon which the moisture of the atmosphere is deposited, and that the electrical characteristics of the textiles are determined largely by the thickness and continuity of these water films and the conductivitv of the solutions formed bv contact 610 BELL SYSTEM TECHNICAL JOURNAL with the textiles. The presence of water-soluble impurities in textile insulations is therefore very important and their thorough removal by controlled processes of washing has resulted in vast improvements in both cotton and silk. For some reason not yet fully understood, silk is much superior to cot- ton as an insulating material over the usual range of atmospheric humid- ities. This is true in spite of the fact that cotton absorbs less water than silk at a given humidity, and that, over the usual range of use, silk is more sensitive electrically to a given increment of water than cotton. This is responsible for the extensive use of silk in the electrical industry at several dollars per pound in place of cotton, which may be had for about a tenth as much. The purification of cotton, however, has made it good enough to replace silk for a large number of purposes, especially in telephone cords, and the saving thus effected amounts to several hundred thousand dollars per year for the Bell System. In addition to this practical result, such studies as these have suggested interesting scientific possibilities in the use of electrical measurements for deter- mining the structures of colloids. Paper An allied product is paper, which is used in enormous quantities in the construction of the common type of telephone cable. This cable consists of a bundle of wires individually insulated from one another by strips of paper helically served about each. Before being enclosed in a lead sheath the bundles of insulated wires are thoroughly dried and thereafter throughout their use have to be protected from entrance of atmospheric moisture by hermetically sealing the lead covering. The functioning of the cables depends absolutely upon the maintenance of an extremely dry atmosphere within the cable. Our chemists have been called upon for elaborate studies of the effects of minute increments o- moisture upon the insulating qualities of paper and the effect of temf perature upon the electrical characteristics of paper containing various small proportions of moisture. Incident to this task it has been neces- sary to develop a humidity recorder sensitive to as little as 10 parts per million of water vapor in the air. Such a commercial recorder, pro- duced by Leeds and Northrup at our instance, is in successful use as a guide in controlling the atmosphere of cable-drying ovens. Improved devices for determining the brittleness of cable paper and for judging its predisposition to lose flexibility upon baking have also received at- tention. Still another dielectric problem is that of condensers, which are unique i CHEMISTRY IN THE TELEPHONE INDUSTRY 611 in that an insulator of maximum dielectric constant is required, where- as in most electrical apparatus a minimum dielectric constant is sought. For the sake of economy the usual telephone condenser is made of al- ternate strips of paper and tinfoil which are wound up into a compact roll, and after drying is impregnated with some form of waxy material to bring the capacity to a maximum and prevent subsequent variations with changes of atmospheric humidity. The complexity of the effects of the choice of the impregnating material upon the electrical charac- teristics of the condenser is most surprising. It might be supposed that condensers of high insulation resistance would also have a high breakdown strength, but this is by no means always the case. It has be- come evident that the nature of the interface between the individual fibers of the paper and the surrounding waxy material is of vital im- portance. Phenol Plastics A class of materials very widely used in the electrical industry for insulating purposes is well known under the general term of "phenol plastics." Cellulose acetate is another insulating material of excellent electrical characteristics and is superior to the usual fibrous materials owing to its lower water absorption. Although the uses of such mate- rials, both in massive form and as impregnants for fibrous insulation, are very extensive in the telephone field, little chemical work has been done in the Laboratories on them, partly because they are the products of a rather highly developed industry which has conducted a great deal of investigation for us. For certain types of uses, however, the phenol plastics have required some special chemical study from the standpoint of their stability. Being in the nature of condensation products formed by the elimination of water, the phenol plastics are more or less subject to the reversion of the reaction by which they were formed with the produc- tion of free phenol and ammonia. For certain uses the presence of these uncondensed constituents or hydrolytic products, as the case may be, is objectionable and has required a special investigation of means of controlling their presence. An important improvement and economy has been effected in certain textile-insulated wires by applying a cellu- lose acetate lacquer to the exterior so as to partially impregnate the textile. The film of cellulose acetate contributes a continuous smooth surface and a measure of resistance to atmospheric humidity variations. Conductivity of Copper Passing from insulators, one naturally thinks of conducting materials, of which of course copper is first in importance. It is a well-known and thoroughly tried principle that the metallic elements in the pure state 612 BELL SYSTEM TECHNICAL JOURNAL have a higher electrical conductivity than their alloys. The problem of conductivity of copper is therefore essentially one of purity, and has been fairly satisfactorily solved by the copper-refining industry. We are, however, attacking some special problems in conductivity of copper. Carbon A unique conducting material in the field of telephony is carbon. A small mass of granules of this material in every transmitter serves the all-important purpose of converting the variations of the mechani- cal energy of the voice into equivalent variations of the transmissible electric current. This the carbon does by variation of its electrical resistance with variation of mechanical pressure. No other material approaches carbon in its usefulness for this pur- pose, but there are still many obscurities about the functioning of a carbon transmitter. Decades of physical and chemical research have, however, established certain points. Transmitter carbons in general are not highly active carbons in the sense in which we have be- come familiar with that term in connection with absorptive charcoals. Gas films on the carbon, however, do play some role in their micro- phonic functioning. Of more practical importance for the present is the evidence that the microphonic effectiveness of carbon is very much dependent on the method of its preparation, and particularly the time, temperature, and atmosphere in which it is roasted. Carbon for trans- mitters is made from anthracite coal of maximum hardness and low ash content. The roasting processes are designed to produce a material of as high uniformity as possible, especially with reference to the hard- ness, compactness, and abrasion resistance of the surfaces of the finished product. It has come to be recognized that such physical character- istics of transmitter carbon as these to a great extent determine, not only its original effectiveness, but also its resistance to atmospheric disturbances and its durability under the mechanical and electrical forces exerted upon it during use. The hydrogen content of carbon has been found a useful index of the time-temperature cycle to which it has been subjected during roasting. As anthracite coal is roasted there is a progressive loss of hydrogen, which, however, does not reach zero value until a temperature above 1500° C. is obtained, at which point the material is converted rapidly to graphite. It is only at upper intermediate temperatures that satis- factory transmitter carbons can be produced. The study of the hydrogen contents of coal and carbons and of re- lated microphonic behavior has led to a definite theory that there are, contrary to belief of some authorities, only two allotropic forms of car- CHEMISTRY IN THE TELEPHONE INDUSTRY 613 bon — namely, graphite and diamond. The so-called amorphous car- bons, according to this view, are complex hydrocarbons in which the carbon greatly predominates over the hydrogen and in which, especially if the carbon has been roasted at a high temperature, the carbon atoms are in part arranged in a graphite lattice. Vacuum-Tube Filaments The filaments of vacuum tubes represent a very special form of con- ductor, the primary function of which is, of course, the emission of electrons. Vacuum tubes are used in repeater sets in all long-distance telephone lines. The filaments in these vacuum tubes consist of plati- num alloys coated with the oxides of barium and strontium. It is im- portant that these filaments should be constant in their electrical char- acteristics and have as long a life as possible. The manufacture of platinum alloys and the methods of coating them have been subjects of study by our metallurgists and chemists for several years, and as a Fig. 3 — A corner of the electroplating laboratory. result many improvements in the manufacture of the filaments have been made. Probably the most outstanding improvement has con- sisted in the substitution of platinum-nickel and platinum-cobalt alloys for the platinum-iridium-rhodium alloy formerly used as a fila- ment core. This substitution resulted in increasing the life of repeater tubes from a few months to several years. 614 BELL SYSTEAf TECHNICAL JOURNAL Ml'.TALS The metallurgical group, as far as chemical investigation is concerned, is in charge of J. E. Harris. One of his principal interests is the pro- duction in varied forms of special magnetic materials, illustrated by permalloy and perminvar. The field of magnetic materials has been so intimately connected with the fundamental physical theory of ferro- magnetism that the primary responsibility for these investigations has been lodged in a physical research group, the members of which have been responsible for fundamental inventions in this connection. The Fig. 4 — Pouring a metallic melt in vacuum by tilting a high-frequency induction furnace. problems of fabrication and composition control in the experimental work in this connection have, however, afforded much opportunity for the ingenuity of chemists. A particular case in point is that of the production of brittle forms of such special magnetic materials, the object being to permit the grinding of the metal into a coarse dust. The dust particles are then insulated by deposition of a film on their surfaces and are pressed into rings which form the cores of modified Pupin-type loading coils. This unusual objec- J CHEMISTRY IN THE TELEPHONE INDUSTRY 615 tive of rendering metal brittle has been achieved in general by the prin- ciple of introducing an impurity into the melt, which has a tendency to segregate at the grain boundaries, thus facilitating subsequent fracture. Each alloy requires some special consideration, both as to choice of embrittling agent and heat treatment and working schedules for devel- opment of proper grain size. Scarcely any metallic material has given the telephone industry more concern than the lead alloy used for cable sheath. Pure lead is too soft for the purpose and can be too easily damaged mechanically. Years ago about 3 per cent of tin was alloyed with the lead as a harden- ing agent. Tin was later superseded by 1 per cent of antimony, pri- marily for reasons of economy, and it has been reported that twenty million dollars have been saved to the telephone system by this substi- tution alone. The use of these hardening agents affords an example of the disper- sion hardening of metals, which has become familiar to the public most conspicuously in the case of duralumin. It is the belief of metallur- gists that the introduction into a molten metal of a constituent, which is precipitated out in very finely divided form upon cooling the metal, diminishes the deformability of the finished material by interposing it- self in the slip planes among the atoms of the metal. Dispersion hardening in a metal as soft as lead represents a rather extreme case, for lead at atmospheric temperatures is approximately as deformable as steel at dull red heat. It has been found that the anti- mony used as a hardening agent in lead cable sheath tends, especially under the influence of repeated flexings such as those due to thermal expansion, to redissolve in the metal and redeposit elsewhere. In this fashion large particles of the antimony grow at the expense of small particles, and the hard-worked portion of the metal is eventually de- prived of antimony content and fracture occurs. This has been the source of considerable trouble in aerial cables, especially at the bends in the cable which occur at the poles due to expansion. The working out of this problem in fatigue of metals has been a long process, but promises to bear further fruit in the develop- ment of better hardening agents. One of these, a joint development with the Western Electric Company and one which still remains to be tested on a commercial scale, is calcium, which in the minute propor- tion of 0.04 per cent has been found in laboratory experiments to pro- duce a hardening well surpassing that of 1 per cent of antimony. Another interesting metallurgical problem is that of solders for use in wiping joints in telephone cables. Somewhat to our surprise we found that some of the supposed prejudices of workmen responsible for 616 BELL SYSTEM TECHNICAL JOURNAL cable splicing were well founded and that it is scarcely practicable to make a satisfactory wiped joint with a lead-tin solder containing less than 38 per cent or more than 42 per cent of tin. New solders may well grow out of our study of why and how the old-fashioned solder works. The interests of the telephone system extend of course to many other metallic materials, notably iron and steel, brasses and bronzes, die-cast- ing alloys, etc. For the most part, however, progress in these fields has been along the lines of that of other industries. Part of our metal- lurgic shops are largely devoted to the melting, casting, and fabrication of a great variety of alloys into wire rods or sheets of specified dimen- sions for experimental trials in electrical apparatus design. Wood and Its Preservation It is a far cry from metals to wood, and particularly to wood preser- vation, which is one of our important chemical interests. The tele- phone pole is our most urgent concern. Large numbers of poles of cedar, chestnut, and southern pine are in use and the greater part have been subjected to a preservative treatment. Pine poles possess a layer of 2 or 3 inches of sapwood which is subject to impregnation under pressure, and such a treatment with creosote has long been the standard practice in the system. The problems in this field are innumerable. We will mention only a few. Given a train load of telephone poles, how does one determine the average quantity of creosote they contain and the uniformity of distribution from pole to pole and in various parts of a given pole? No two trees grow alike. Soil, climate, sun exposure, accidents in past histories such as fires in forests — all make for peculiarities of growth in each individual tree. These peculiarities reflect themselves in the absorption of creosote, so that it is entirely possible for two poles treated simultaneously in the same cylinder to differ by a factor of 5 or even 10 in the over-all creosote content per unit volume. How can one obtain a sample which will be representative of a large group of such poles? The sampling problem was approximately solved by taking a suffi- cient number of cylindrical solid cores with an increment borer and splitting these borings diagonally along a length which represents the approximate radius of the pole. Mathematically such a tapered cylin- der approaches a wedge such as would represent a true sample of the pole's cross section. The borer holes are plugged with a creosoted peg and the poles are still fit for use, so one can sample as many poles as he likes. This method is gradually being applied to a study of the content and CHEMISTRY IN THE TELEPHONE INDUSTRY 617 distribution of creosote, not only in new poles, but also in those that have been in service for varying periods of years. In order to deter- mine how much creosote is present and also its present wood-preserva- tive value, we have adopted a biological method of testing the toxicity to pure fungus cultures of creosote extracts from such old wood. An interesting story could be written about the specific resistance of various species of wood-destroying fungi to the numerous toxic agents that have been used and proposed. In passing upon the merits of a new preservative it is difficult to pre- dict its permanence in the wood when exposed to the weather. A prom- ising method, though one which does not offer the ultimate in economy of time, is the use of small twigs or saplings which are impregnated with the preservative in question and exposed in groups to the action of the weather in a fast rotting climate. In this test we are depending for acceleration entirely upon the re- duction of the dimensions of the wood, while preserving at least the more salient features of wood structure by using natural twigs or stems rather than artificially shaped pieces. In applying this principle of reduction of dimensions for purposes of acceleration, we took a leaf from our book of experiments on submarine insulation in which we had found a strong case of parallelism between the absorption of water by rubber and the saturation of a material by heat as epitomized in Pick's law. The time required to reach a given degree of saturation is approx- imately inversely proportional to the square of the thickness of the specimen. The depletion of creosote from wood appears to be an in- verse process, with of course some complications. Reasoning by anal- ogy we hope by the use of small stems to shorten the time required for depletion of creosote and other preservatives and the beginning of rot in wood by ten to thirty years — a great economy in patience. Electrochemical Investigations Another large field of chemical investigation comes under the general head of electrochemistry, which for convenience includes corrosion, corrosion prevention, and finishes of both metallic and organic (paints, etc.) types. This work as well as the analytical laboratories, is under the supervision of R. M. Burns. Among the electroplating developments has been the successful de- position of permalloy from a bath containing iron and nickel salts. The composition of the alloy containing 79 per cent nickel and 21 per cent iron can be maintained constant to less than 0.5 per cent. It is of interest to find that the alloy is deposited from the bath as a solid solu- tion and that it has desirable magnetic properties. 618 BELL SYSTEM TECHNICAL JOURNAL The corrosion work consists of both corrosion testing and fundamen- tal studies on the mechanism of corrosion processes. The corrosion tests are carried on under normal service conditions and by laboratory- accelerated processes. The results furnish guidance in engineering decisions as to the use of materials. More fundamental investigations in the field of corrosion have to do mainly with the study of the electro- chemistry of corrosion reactions, film formation, etc. It is of impor- tance, for instance, in the development of corrosion theory to determine Fig. 5 — Corrosion of lead cable sheath. (1) By direct contact with a pernicious soil. (2) Stray current anodic corrosion. (3) Acetic acid corrosion in fir duct. the effect of the environment, and of the passage of small electrical currents, upon the anodic and cathodic behavior of pure metals, and work of this nature is being carried on. Underground corrosion is an important part of the field. The most striking examples of this phenomenon are found in steel and iron struc- tures and particularly in the lead sheaths of subterranean cables. Stray currents from trolley systems often cause such corrosion and an elaborate and expensive electrical bonding system is maintained in order to minimize these troubles. The physical and chemical nature of I I CHEMISTRY IN THE TELEPHONE INDUSTRY 619 the soils and the underground waters and atmosphere often play an important role in determining the kind and extent of corrosion by stray current. Other occasional cases of electrochemical corrosion have been en- countered in which stray current, though present, does not arise from trolley-line power houses. In one large city it was found that a battery covering a square mile or more of area had been inadvertently created, such that it affected a large part of the cable system in the center of the city. The cinder fills underlying the duct runs in this area contained enough carbon to serve as one electrode, while the iron-pipe systems supplying gas and water to the city furnished the other electrode. The Fig. 6 — Experimental metallurgical shop. moist soil afforded a conducting path for a galvanic current that wrought a widespread damage to telephone cables in the area. In another and much larger area widespread injury to cables came about through the presence of traces of acetic acid in the air in wooden duct systems. The source of this acid was the wood itself, which hap- pened to be of a rather highly acid variety. The natural acidity of the wood was further increased by the somewhat drastic process of heating 620 BELL SYSTEM TECHNICAL JOURNAL which was necessary to secure a fair penetration of the wood with creo- sote. In so far as the creosote penetrated, the acid produced in heating was neutralized to a great extent by the nitrogenous bases in the creo- sote. Often, however, the total acid produced far exceeded the neutra- lizing power of the creosote bases contained in the external shell of creosoted wood. The sheath of many miles of cables underwent a par- tial conversion into white lead via the classical Dutch process which, though highly regarded by paint manufacturers, became anathema to telephone engineers. The difficulty was met by fumigating the ducts in service with a dilute ammonia-air mixture and by choosing a less acid and more easily treatable wood for future construction. Underground corrosion of other metals, notably of iron and steel, is also often serious. In the alkali soils of the southwest anchor rods for telephone poles have sometimes corroded through in a few months. Marshes represent another severe exposure for iron and steel, as, for example, in the form of loading coil cases. A newly introduced form of telephone cable for direct burial in the soil demands careful considera- tion from this standpoint. A variety of protective finishes, chiefly of asphaltic or pitchy nature, have been studied in this connection. Some remarkable cases have been noted, in which a finish that proved to have a superior protective effect in one highly corrosive soil was worse than useless in another soil which had been regarded as less corrosive in the general sense. The chemistry and physics of soils from many areas have required attention with ^the control of corrosion as an object. Particle size, saline content, and composition of subsoil atmospheres each has an influence. In a like way the telephone chemist must concern himself with atmo- spheric causes of corrosion in equipment above ground, especially in central offices. Moisture and dust contribute to electrical leakage from point to point through the complicated assemblies of electrical equipment. Corrosion products of such leakage may build up at criti- cal points and interfere with contacts, or essential though usually minute portions of equipment may be etched away. Even faint tarnishes on metallic contacts can so increase contact resistance as to imperil signal- ing. In industrial areas soot and traces of sulfurous gases add materi- ally to these hazards. Finishes It is partly to avoid such difficulties and partly for the equally utili- tarian purpose of a good appearance that metal telephone apparatus receives some special form of finish coating, whether paint, varnish, lacquer, or electroplated surfacing. In the selection of such finishes CHEMISTRY IN THE TELEPHONE INDUSTRY 621 we have drawn heavily upon the scientific work of our confreres in allied fields, but have still found ourselves faced with peculiar difficulties. A great deal of the truly excellent scientific work on finishes has been done by manufacturers with the idea of disclosing uses which will justify the sale of a particular material. But comparative data on the dura- bility of very dissimilar finishes as, for example, galvanized coatings in contrast to cellulose lacquers, are usually lacking. The fact that finishes are often used primarily for decorative purposes on relatively short lived articles has limited the study of the durability of such coat- ings. This is reflected in the fact that nearly all the scientific work in this field refers to outdoor exposures where corrosion tends to occur rapidly. Indoor exposures are commonly regarded as so mild as to be negligible. Changing fashions, as in the case of furniture, often bring an obsolescence so early as to be prohibitive were similar consideration to be applied to telephone plant. The prevalent custom of trading in one's motor car for a new model each year is a factor in another large industry involving extensive use of finishes, which tends to put great emphasis upon initial beauty rather than permanence over periods of ten to twenty years, such as must be considered in telephone plant. With the manufacturing and operating telephone companies tied into a single system, in which the Bell Laboratories' responsibility is to insure quality of product, there can be no unloading of defective appa- ratus upon the consumer, for the manufacturer is liable in the last analy- sis for defects which may appear only after years of use. These considerations have received special emphasis in the discussion of finishes because they aff^ord an excellent illustration of the point. The same sort of considerations, however, apply to nearly all the problems with which we are concerned, to such an extent that our chemical staff is widely thought of in our own organization as a group of specialists in the "permanence" of materials. Such an emphasis by the general management upon ultimate economy rather than first cost alone would doubtless be welcomed everywhere by thoughtful technical men throughout the country. It is a source of peculiar pride to the staff of the Bell Laboratories that the nature and organization of their business is such as not only to permit such an attitude but also aggressively to promote it. The Trend in the Design of Telephone Transmitters and Receivers ' By W. H. MARTIN and W. F. DAVIDSON 2 This is a report of the Joint Subcommittee on Development and Research, National Electric Light Association and Bell Telephone System. It was prepared by the Chairmen, respectively for the Bell System and the N.E.L.A., of the Project Committee assigned to this study. The report reviews the history and present trend of the design of telephone transmitters and receivers, particularly from the standpoint of their response frequency characteristics, and discusses the possibility of obtaining a re- duction in the effect of line noise by shifting their points of maximum response. It is concluded that no advantage from this standpoint is indicated inasmuch as it has been found that the distribution with frequency of the extraneous energy on telephone toll lines is approximately uniform over the more important portion of the frequency range. It is further stated that the present trend in improvement of the response characteristics of transmitters and receivers is in the direction of reducing the difference between their maximum and average response. IN the beginnings of the telephone, the outstanding marvel was that the devices used as transmitters and receivers could perform the necessary conversions between speech sound waves and electrical waves. In the application of these devices, however, it was early appreciated that the range and cost of telephone circuits were directly 1 Editor's Note: In this issue of the Bell System Technical Journal there are two papers and one report dealing with various phases of the inductive coordination problem, which have had their origin in the work of the Joint Subcommittee on Development and Research of the National Electric Light Association and the Bell Telephone System. This organization is one of the subcommittees of the Joint General Committee of the N.E.L.A. and Bell Telephone System, which has for its general objective the working out of methods of procedure whereby problems involving the physical relations between the plants of the electric supply companies and the telephone companies may be handled cooperatively on mutually satisfactory bases. The questions involved are largely of an engineering character, and to carry on that phase of the work the Engineering Subcommittee of the Joint General Committee was appointed. The Engineering Subcommittee has recommended certain broad principles of cooperation as well as the adoption of more detailed principles and practices, which were accepted by the Joint General Committee and published in 1922. As a result of further recommendations by the Engineering Subcommittee the Joint Subcommittee on Development and Research was organized. It is charged with the conduct of technical investigations, the accumulation of data, and the development of engineering methods for use in the solution of problems of coordi- nation. Its work is organized under a number of subordinate committees known as "Project Committees," each of which is assigned a certain range of subjects for study. The first volume of Engineering Reports of the Joint Subcommittee on Develop- ment and Research, containing a considerable part of the technical information thus far developed by the subcommittee has recently been published (April, 1930). ^N. E. L. A. Bulletin, Aug., 1930. 622 TELEPHONE TRANSMITTERS AND RECEIVERS 623 dependent upon the efficiency with which these instruments made these conversions. Experimental activities were, therefore, soon directed to increasing the efficiency of these instruments and especially to getting a more efficient transmitter than the forerunner of the pres- ent telephone receiver which initially was used both as a transmitter and a receiver. The outcome was the carbon contact transmitter which provided a means for drawing upon an outside source of energy in the process of converting sound waves into electrical waves and thus combined in the transmitter the function of a converter of energy with that of an amplifier. Since that time numerous important improvements have been effected in both the carbon transmitter and the magnetic receiver but the general principles of both are still employed today in the best practical instruments for commercial telephony. Both of these instruments employ vibrating diaphragms which, like other mechani- cal vibrating systems, have regions of maximum response due to reson- ance between the mass and elasticity of the diaphragm. With these resonant effects inherent in the structure, it was natural to place them in the frequency range so as to obtain the maximum benefit. In accordance with this, the reproductions of speech sounds obtained with these resonances located at different points were listened to and the judgment reached, taking into account both the intelligibility and naturalness of the reproduced sounds, that they should be placed around 1,000 cycles. It was found that a material shift in the point of maximum response of the circuit to a lower value made the output sounds "boomy" and to a higher value rendered them "thin." While precise means for measuring the effects were not available at that time, subsequent work has substantiated this choice as a wise one. Inves- tigations of the frequency components of speech sounds have shown that the principal components of about half the vowel sounds lie below 1,000 cycles and of the other half are about equally divided above and below this point. Articulation tests have demonstrated that the frequency range which covers about an octave each side of 1,000 cycles, namely from about 500 to 2,000 cycles, includes the more important frequency components in speech from the standpoint of intelligibility. The frequencies below this range are important primarily for natural- ness and those above for intelligibility and also for naturalness. The location of the region of maximum response of the telephone circuit in the neighborhood of 1,000 cycles emphasizes then this 500 to 2,000- cycle range and meets well the requirements of both intelligibility and naturalness. In addition to the diaphragm resonances there are also inherent 624 BELL SYSTEM TECHNICAL JOURNAL resonances in the enclosed cavities which are associated with these diaphragms, such, for example, as the mouthpiece of the transmitter and the cases in which the transmitter and receiver units are placed. In the present type of deskstand transmitter the several resonances are so located as to give a fairly broad maximum response in the range between 1,000 and 2,000 cycles and the resonance of the receiver has been placed so that its maximum response is around 1,000 cycles. It is seen then that the inherent resonances of these instruments have been located in the more important part of the voice- frequency range and have been utilized to increase their efficiencies in that range. The remarkable performance of the granular carbon type of trans- mitter merits some indication of its accomplishment. Its conversion of the complex speech waves into equivalent electrical waves has been improved from time to time and now the most efficient type of trans- mitter which is in general use, when energized with the direct current which it gets on short loops, has an electrical output which is more than a thousand times the magnitude of the acoustical power which is delivered by the speaker. Furthermore, it provides this conversion and large amplification at a low cost. Since the average energy given out by a speaker in carrying on a telephone conversation is of the order of 10 microwatts, the large stepup in power from the acoustic waves entering the transmitter to the electrical waves leaving it is of vital importance in affording telephone service at a reasonable cost and also in rendering the telephone system less susceptible to the effects of interference currents. While large improvements have been made in the receiver, the effi- ciency of the present instrument is very low in comparison with many other types of energy converters which it is considered practicable to use. For the receiver, the average ratio of the acoustic power out- put to the electrical power input is below 1 per cent. It is possible to increase materially the efficiency of the receiver used in commercial telephony but this would bring up the noises on the telephone circuit. Also there are limitations upon the maximum efficiency of the combi- nation of transmitter and receiver due to crosstalk between telephone circuits and to the fact that with loud talkers over short telephone con- nections the combination of present instruments is close to the point of giving uncomfortably loud sounds in the ear of the listener. In considering the performance of the transmitter in the plant, it is customary for many reasons, important among which is the battery supply circuit, to take the combination of a transmitter, a station set, a typical loop connecting the set to the central office and the cord circuit from which is supplied the direct current for energizing the TELEPHONE TRANSMITTERS AND RECEIVERS 625 transmitter. Likewise, the receiving system of the circuit may be considered to consist of the cord circuit, the loop, the set and the re- ceiver. For the connection of such transmitting and receiving systems through a distortionless trunk, the response characteristic of the overall circuit, giving the relation between the power delivered by the receiver and the power available at the transmitter, shows a variation of about 30 db in the range from 500 to 2,000 cycles with the maximum response slightly above 1,000 cycles. This characteristic applies to the type of deskstand apparatus which is now the most generally used station equipment in the Bell System. With the development of the telephone art numerous ideas have naturally been investigated for improving the performance of the transmitter and the receiver. Taking into account the various con- siderations and possibilities, the present procedure is on the basis that further improvements should come from reduction in distortion rather than from increases in the maximum response. For practical instru- ments the desire is primarily to reduce the distortion without sacri- ficing the average efficiency over the important part of the voice range. Means have been developed for reducing the distortion in these instru- ments but in general such improvements have involved material re- ductions in efficiency. For example, a very high degree of freedom from distortion is realized in the type of carbon transmitter which has been so widely used for pickup work in radio broadcasting. This trans- mitter, however, requires a powerful amplifier to bring its output to a value comparable with the type of transmitter used in commercial telephony. It has been possible also to obtain large reductions in the distortion of the receiver but here, too, large sacrifices in efficiency have attended this accomplishment. Material progress has been made, however, toward the ideal of a combination of low distortion without sacrifice in efficiency. Some of these improvements have been incorporated in the trans- mitter which is used in the handset type of station apparatus. The frequency response characteristic of transmitting and receiving systems such as described above, but using the handset instruments instead of the deskstand, shows a variation of only about 20 db in the range from 500 to 2,000 cycles, and with the handset instruments this same variation of 20 db covers the range from 500 to 3,000 cycles. The handset transmitter presents, therefore, a material advance from the standpoint of reducing distortion. The proposal has been made at various times that the interference situation might be helped by providing still more efficient transmitters. The various possible means of still further increasing the transmitter 626 BELL SYSTEM TECHNICAL JOURNAL efificiency, however, are attended by many difficulties and compli- cations. The efficiency at the point of maximum response could, of course, be increased by piling up the several resonances which have been referred to, but this would give serious distortion effects which would more than offset any increase in loudness which was obtained. In general, it has been found that any improvements which might permit higher maximum responses can be utilized to give greater benefit in the reduction of distortion. Moreover, any large increase in the trans- mitter efficiency would require measures such as the reduction of the efficiencies of the receivers in order to avoid increased crosstalk effects between circuits and uncomfortably loud transmission over short connections. When the program of the joint development and research work of the N. E. L. A. and Bell System was formulated there was the idea that in view of the resonant characteristic of the telephone receiver, some benefit might be obtained in the performance of the telephone circuits in the presence of line noise by shifting the point of maximum response of this instrument. Some cases had arisen where pronounced harmonics in the power system in the neighborhood of 1,000 cycles caused serious troubles in nearby telephone circuits and it was felt that if this condition were found to be prevalent in power circuits, some relief in the interference situation might be obtained by shifting the point of maximum response away from this region. The investi- gations which have been carried out under Project 4 of the N. E. L. A.- Bell System Joint Development and Research Subcommittee, of the noise on telephone lines in different parts of the country have shown that the average distribution of energy with frequency is approxi- mately uniform over the range from 300 to 2,000 cycles with, however, a pronounced dip in the region around 1,000 cycles. A similar de- crease in the energy of the components around 1,000 cycles is also shown by the results of the investigations made under Project 5 on the wave shapes of electrical power machinery. With this situation, shifting the maximum response of the telephone receiver away from its present location would thus in the average case be placing it in a region in which larger amounts of interfering currents are to be found. Moreover, examination of the data showing the distribution with frequency of noise currents, indicates that on particular circuits this distribution is by no means uniform but in many cases is materially higher in the region below 1,000 cycles and in other cases materially higher in the upper regions. It would not appear, therefore, that a shift in the maximum response of the telephone receiver would on the average give an improvement from the interference standpoint. TELEPHONE TRANSMITTERS AND RECEIVERS 627 Furthermore, any compromise in the instrument characteristics to favor the line noise conditions, should not have an adverse effect on the many connections on which noise from power systems is unimport- ant. As has been noted, a material shift in this maximum response would have a marked effect on the naturalness of the reproduced sounds. On the whole, then, no advantage to the interference situa- tion has yet been indicated for shifting the resonance of the receiver. In accordance with these considerations, the present effort in the development of telephone transmitters and receivers is being directed along the lines of reducing the deviation between their maximum and average responses. Any improvements in the instruments which it may be found practicable to make will be in the direction of increasing the intelligibility and naturalness of the telephone conversations, and the justification for their adoption will include their effectiveness in the presence of typical distributions of interfering currents on the tele- phone lines. Mutual Impedances of Ground-Return Circuits Some Experimental Studies * By A. E. BOWEN and C. L. GILKESON This paper describes some of the results of the work of the Joint Develop- ment and Research Subcommittee of the National Electric Light Association and Bell Telephone System on the mutual impedances of ground-return circuits. The first part of the paper deals with some experiments which were per- formed to establish an experimental background for the testing of theoretical ideas. Different theories, one involving an "equivalent ground-plane," a second a d. c. distribution in the earth, and a third an a. c. distribution in the earth, are discussed in the light of the experimental results. While none of these is adequate to explain all the observed phenomena, each has a field in which it can be made useful. The second part of the paper is devoted to a description of practical means for predetermining the mutual impedances of power and telephone lines. This involves an experimental determination of a curve of mutual impedance as a function of separation in the region of the proposed exposure and the calculation of the overall mutual impedance between the proposed lines from this curve and the dimensions of the exposure. The results of trials of this method in two locations are given which indicate that itshould be of sufficient accuracy for engineering purposes. Introduction THE magnitude of the inductive coupling between power and tele- phone lines is a factor of fundamental importance in problems of coordination to prevent interference between these two classes of lines. Accordingly, this is one of the subjects under investigation by project committees of the Joint Committee on Development and Re- search of the National Electric Light Association and the Bell Tele- phone System. It is the purpose of this paper to present the results of some work which has been done under the auspices of the Committee on one phase of this problem, namely, the mutual impedance of ground- return circuits. The mutual impedance of two ground-return circuits is determined by measuring the ground-return current in one circuit (the "disturb- ing" circuit) and the open-circuit voltage at the terminals of the second circuit (the "disturbed" circuit). The vector ratio of the open-circuit voltage to the ground-return current is then defined as the mutual impedance of the two circuits. For any normal or abnormal operating condition of a power system, * Presented at the Summer Convention of the A. I. E. E., Toronto, Ontario, Canada, June 23-27, 1930. 628 MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 629 the currents, either at fundamental frequency or at any harmonic fre- quency, in any of the lines can be resolved into components, some of which are entirely confined to the wires while another component flows in a circuit composed of all the wires as one side with the ground as a return path. The work which is described in this paper deals with the magnitude of the induced voltages on exposed telephone lines caused by the latter component. It has been directed to two ends, first, the establishment of an experimental basis for the study of the physical factors involved in the inductive coupling of ground-return circuits, and second, the development of practical methods to enable the advance calculation of the mutual impedances of power and tele- phone lines. The work is accordingly presented in two parts; first are given the results of tests made at a field laboratory in which testing conditions could be controlled, and second, tests in which the practical side of the problem was investigated are described. Cross Keys Tests and Theoretical Background Cross Keys Tests. An extended series of measurements was made at a field laboratory operated by the subcommittee near Cross Keys, New Jersey, about 20 miles southeast of Camden. A single conductor, located about 34 ft. above the ground and 8500 ft. in length, was avail- able for the disturbing circuit. For disturbed circuits, 500-ft. lengths of insulated wire were laid on the earth parallel to the disturbing con- ductor, at several separations, as shown on Fig. 1. Grounds were pro- .iGROUNO CONNECTION -4850'- DISTURBING LINE :500>< -3150'- GENERATOR OHi" LINE DATA DISTURBING DISTURBED HEIGHT 34 FT LENGTH 8500 FT SEPARATIONS 21.5 FT. 56.1 FT. 101.4 FT 200 FT. 500 FT. 1000 FT. 0 500 FT. -"^DISTURBED / LINES SCALE IN F£ET 0 500 1000 1 ■ ■ ■ ■ I ■ ■ ■ ■ I SETUP FOR GRADIENT MEASUREMENTS ■4500' GROUND CONNECTION H I GRADIENTS WERE MEASURED ALONG DOTTED LINES -ao-|ii' Fig. 1 — Cross Keys tests — experimental setup. vided at the ends of each line. Ground-return current was transmitted over the disturbing line at 60 cycles from a commerical source, or at frequencies between 100 and 1000 cycles from a vacuum-tube oscillator with power amplifier. The measuring instrument was an a.-c. poten- tiometer, equipped with suitable filters so that the observations were unaffected by the presence of harmonics in the disturbing current. At several frequencies within the range from 60 to 1000 cycles the 030 BELL SYSTEM TECHNICAL JOURNAL 0.1 0.07 0.05 (O 0.04 5 I O 0.03 LlI Z 0.02 z < I- 10 _i 0.01 < D D 0.007 2 0.005 1000 CYCLES o -o ^^ " ^ J^.^ ^ ^ k 500 CYCLES P ^ k. _ Ij ^ V \ (J u 200 CYCLES N \ \ \ ^ o 0- EXf 'Ef ^l^ IE VI- AL POINTS ^ <^ N > S s 60 CYCLES -^> ' ' ' ' ' — < L^ 1 *^ ^ > 20 30 50 100 200 300 600 HORIZONTAL SEPARATION IN FEET Fig. 2 — Cross Keys tests — mutual resistance. 1000 0.7 0.5 0.4 0.3 H 0? UJ tu u. o o If) (T LU 0.1 Q. W X 0.07 o z 0.05 ILI o z < 0.04 1- o 0.0.3 < UJ DC _) < 0.02 D 1- 3 2 0.01 0.007 0.005 ° ~ ^ ^Xi ^ V ^'Or.. °^ ^ ■~o. ^ ^ ■> \ s. \ N, ■^ ^ -^On \ \ \ ^^>-o_ \, \ ^^^ \, \ \ X, \ \ s \, \ \ — o. ^ *^ ^^ r \ \ \ \ ^^^ x,^ N k v \ \ o- EXF ER IM EN T/ VL POINTS ^ \ \ \ \ V, \ \ \ \ s T s s s, 20 30 50 100 200 300 HORIZONTAL SEPARATION IN FEET Fig. 3 — Cross Keys tests — mutual reactance 500 1000 MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 631 current in the disturbing line, the open circuit induced voltage in each of the short ground-return circuits, and the phase angle between these two quantities were measured. The mutual impedances were derived from the ratio of the induced voltage to the inducing current, in ac- cordance with the definition. The results of these tests are given on Figs. 2, 3, and 4. Fig. 2 shows the resistance components, Fig. 3 the reactance components, and Fig. 4 the magnitudes of the mutual im- pedances. 0.7 0.5 0.4 0.3 ^- 0.2 UJ UJ u. o o S 0,1 Q. W 2 0.07 I O O 0.04 z < 3 0.03 a. 2 < H =) 2 0.02 0.01 0.007 0.005 ^o, -■ ■^c. "-0 ^ s. r^e ^nQ \ \ \ V b ■' r^^Oci \, \ ^^c,,_ \ k N L \ \ 1 \ V k N V X \, \ \ \ V ~o. ^^ = ^Cv^ \ N 1 \ \ \ ^^^ ^ \ \ V \ 1 0- -E XP e;rimental p OINTS \ N \ \ \ \ s ^ 20 30 50 100 200 300 500 HORIZONTAL SEPARATION IN FEET Fig. 4 — Cross Keys tests — mutual impedance. 1000 The measurements described above were made with the object of investigating the mutual impedances of ground-return circuits in which the ground connections on the disturbing line are sufficiently removed from those on the disturbed circuit so that effects due to proximity of the grounds may be ignored. The results presented above were supple- mented by observations demonstrating that the induced voltage in a 632 BELL SYSTEM TECHNICAL JOURNAL O o u. a. u Q. in X o a < o 3.0 2.0 1.0 0.5 0.3 0.2 0.1 0.05 0.03 0.02 0.01 ^ 0.005 < o JJ 0.003 2 "" 0.002 < D 3 0.001 5 0.0005 0.0003 0.0002 0.0001 0.00005 0.00003 0.00002 0.00001 4 1 1 ~i \ \\ \ •It 1 I 1 \ \\ w \ \ \ VA \M \m \\> \\ y \ \ v\ A \ v,\ ^\ k v*i \ \ \ V \ \ > ^ ^ \ — ^ J ^L/N DFRr ■JEAT \ V " LI VE — \\ >) .EXTENSION OF LINE PERPENDICULAR TO LINE V W 200 400 600 800 1000 1200 DISTANCE FROM ELECTRODE IN FEET 1400 1600 1800 Fig. 5 — Cross Keys tests — mutual impedance gradient in vicinity of grounding electrode. Experimental curves. MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 633 parallel circuit was closely proportional to the length of the circuit and that the voltage induced in a ground-return circuit extending perpen- dicular to the disturbing line was exceedingly small. As the points of grounding on the disturbed circuits approached those of the disturbing circuit this proportionality no longer existed nor was the voltage in a perpendicular circuit of negligible magnitude. A second series of tests was therefore conducted to determine the nature of this effect and the area in which it was of importance. In these tests voltages were measured in very short disturbed circuits extended along radii converging on the ground electrodes on the dis- turbing line. At each location the circuit was made progressively shorter until the quantity measured per unit length was practically independent of the length. Thus the gradient of the mutual im- pedance, in the direction of the radius at the point of measurement was determined. These measurements were made only at a frequency of 60 cycles. 80 70 HJeo o 50 40 ui 30 O 20 5 10 I 0. -10 ^ . ^ ^ ^ / / J /underneath ^^ LINE / / ' / / ^ PERPENDICULAR "'^ TO LINE -^ '^^^ r r: EXTENSION ■ OF LINE 200 400 600 800 1000 1200 DISTANCE FROM ELECTRODE IN FEET 1400 1600 1800 Fig. 6 — Cross Keys tests. Phase angle of mutual impedance gradient in vicinity of grounding electrode. Experimental curves. The resulting observations are shown in Figs. 5 and 6. Fig. 5 shows the magnitude of the mutual impedance gradient along three radii, one radius being directly under the disturbing line, the second perpen- dicular to it, and the third along the extension of the line. Fig. 6 shows the corresponding phase angles. Under the disturbing line, as the distance from the grounding point is increased, the gradient ap- proaches a constant value and the phase angle changes rapidly from a very small value to an angle approaching 80 degrees. Along the latter two radii, however, the magnitude of the gradient appears to decrease indefinitely and the phase angles are smaller. 634 BELL SYSTEM TECHNICAL JOURNAL A more complete analysis of the results of both groups of tests is given in connection with the discussion of theory which follows: Equivalent Ground-Plane Theory. The equivalent ground-plane method of computing the mutual impedances of ground-return circuits utilizes a very simple formula and has been in common use for a number of years. A derivation and discussion of the formula together with some experimental results are given in the report published by the California Railroad Commission in 1919.^ This method assumes that the returning earth current may be con- sidered as flowing in a hypothetical plane surface of perfect conduc- tivity located some distance below the actual surface of the earth. This surface is usually termed the "equivalent ground-plane." The depth of the equivalent ground-plane below the actual surface of the earth varies in different locations from about 50 ft. to 5000 ft. or more, depending upon the character and resistivity of the earth and the fre- quency. This method is subject to the objection that it fails to represent com- pletely the observed phenomena. For instance, the method represents the mutual impedance only with a reactive term, while the experimental results indicated a substantial resistance component, particularly at the wider separations and higher frequencies. Furthermore, no attempt is made to explain the phenomena observed in the neighborhood of the ground electrodes. However, in one respect the theory leads to results comparable to those observed; the magnitudes of the mutual imped- ances as observed under conditions in which end effects are negligible can be checked reasonably well with a suitable choice of ground-plane. Comparisons demonstrating this point are made on Fig. 7, where it will be seen that the curve of experimental mutual impedance for a frequency of 60 cycles can be fitted very well by a calculated curve with a ground-plane depth of 8vS5 ft. That the depth of the equivalent ground-plane depends on the frequency is seen from the fact that to fit the experimental curve at 500 cycles requires the use of a ground- plane depth of 385 ft. Method Assuming D.-C. Distribution in the Earth. For an earth of uniform conductivity, the distribution of the current in the earth for a ground-return circuit energized from a d.-c. source has been employed by G. A. Campbell ^ to derive formulas for the mutual resistance and inductance of ground-return circuits. The mutual resistance is ex- pressed by a very simple formula which involves only the earth resis- tivity and the distances between the points of ground connection on the ^ See Bibliography. 2 See bibliography. MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 635 disturbing and disturbed circuits. For the calculation of the mutual inductance formulas and graphs requiring only a knowledge of the mutual arrangement of the wire parts of the disturbing and disturbed circuits with respect to each other and the earth are given. The mu- tual inductance is independent of earth resistivity. While these for- mulas are, of course, strictly applicable only for direct currents, it is to 04 0.3 0.2 o o >n a UJ Q. in 2 I o 0.1 0 07 0.05 0 04 o 2 0.03 < Q Ul Q. 2 0.02 < D D 2 0.01 0.007 0 005 ~--. — -. .^ ==. ^ =^ ( -O/ 0 N f^ IS<^^ N °z^ \ ^^ ^ \ ■^< N \ ^s k ^=- < '^ 1 t. Sci. =^ ^ *- S.J *J ^e*, \ •^^^ ^ ^^^^^2^ ^5 ^ V '^ ^ N •< ^ ^ T H = GROUND PLANE DEPTH IN FEET • 20 30 50 100 200 300 HORIZONTAL SEPARATION IN FEET 500 1000 Fig. 7 — Cross Keys tests. Ground plane theory. Comparison of measured and calculated values of mutual impedance. be expected that at sufficiently low frequencies the ground-current distribution would not differ appreciably from that for direct current, and hence for these frequencies, these calculated d.-c. mutual resistance and inductance should approximate the actual values. In the paper referred to, some experimental results at frequencies of 25 and 60 cycles supporting this point of view are presented. The experimental curves of Fig. 2, which were obtained from meas- urements at Cross Keys on the 500-ft. disturbed lines near the middle of the 8500-ft. disturbing line, indicate a pronounced increase in mutual resistance with increase in frequency in the range from 60 to 1000 636 BELL SYSTEM TECHNICAL JOURNAL cycles. These results have been replotted on Fig. 8, and it is apparent that for separations within the range of 20 to 500 ft. the mutual resist- ance increases rapidly in almost linear relation to the frequency. For the frequency range and circuit lengths involved in this series of tests, it would appear that a formula for the mutual resistance, based on a d.-c. distribution in the earth is inadequate. 0.07 0.06 I- 111 HI "- 0.05 O o >n a. UJ Q. 10 0.04 X o ai z 0.03 < if) cr <0.02 I- D 2 0.01 1 / / / / / '^/ / <9 / / ■■'^^ f/ f ^^. f .# r / / / / '/ / V ^EPM ^AT^o^ X -100 OFEEf J r ^ ^ y 200 400 600 FREQUENCY 8 00 1000 Fig. 8 — Cross Keys tests. Variation in mutual resistance with frequency and separation. The mutual-inductance curve of Fig. 9 has been computed according to the formulas given by Campbell, and for comparison purposes the mutual inductances derived from the mutual reactances shown on Fig. 3 are also plotted. It will be seen that the observed mutual induc- tances decrease as the frequency is increased, and that while the trend of the observed values is towards agreement with the calculated values MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 637 as the frequency is decreased, the agreement is far from good at 60 cycles, the lowest frequency used in these tests. In the immediate vicinity of the grounding electrode on the disturb- ing circuit, however, the experimental observations of mutual imped- ance gradient can be explained fairly well in terms of a d.-c. distribu- tion. The curves of Figs. 5 and 6 show that in the immediate neigh- borhood of the electrode, the gradient along any radius diverging from the electrode decreases ver}' rapidly with increase in distance from the 180 I- UJ Hi 160 o o If) £i.*o CL I 120 UJ X o oioo lu 80 U z < O 60 3 Q Z _l 40 < D (- 2 20 ^ ^ f^ ^ ^^^0 ^'i ^^3( ^ ^ ^ ^ rt K S \ V,, i» ^^^"---^ \ ■v. "^ ic ^Oo ?5 ^^$^ ^^ ^ ^ \ '■"^ ^ ^ ^ ^ 20 30 50 100 200 300 HORIZONTAL SEPARATION IN FEET 500 1000 Fig. 9 — Cross Keys tests. Campbell theory comparison of measured and calculated mutual inductances. electrode, and is approximately in phase with the current. The gra- dient along the radius under the disturbing line approaches asymptoti- cally a constant value, and beyond 300 ft. from the electrode the phase angle changes rapidly from a very small value to a value approximating 80 degrees. The gradient along the other radii, however, appears to decrease indefinitely and the phase angles are smaller. Such effects are in qualitative accord with predictions based on a d.-c. distribution, as will be seen by reference to Figs. 10 and 11. On Fig. 10 are plotted the resistance and reactance components of the observed gradient under the disturbing line, with values computed using Campbell's formulas. Two calculated curves for the resistance 638 BELL SYSTEM TECHNICAL JOURNAL component are plotted, for conductivities of 2.5 X 10~^"' and 2.5 X 10~^^ (ahmhos per cm. cube). It will be seen that the experimental values lie between these two curves, tending towards the former for a. hi a. z o I- z UJ Q < a. o u o z < o < a z < u o z < I- UJ cc -I < I- D z 3.0 -j- 2 0-1- 1 1 ' \ O "^ \ \ n 1 \ . \\ \ . w \' \ ^'5^ \ ''k A fi' 1 1 V % . 1 % \^ ^'^' \^1 k<^ ll-* ^^ ',?, \ . c\r\f\^ Vo -:c Vv" S-.t/^i c\r\c\'\ \^ Y- \y' \ v-v r\r\r\r\e. \.^ \i "h % 0.0003 ^% ^\-^ ^-.. \ ^'< r^. ^ \ \ REA CA CTA MPE MCE ELL GR>! EOF kDIENT- MULA L- -K- L^ #^ -'^CTAnc 1 •" " V\ \Hi ^ -J ^/-N \ - GRadk ^^. ■^LRIMENTAI J W*' t>o S, ~1 i?^/ V s. CiC^TiCiCW V -Co ^^^c V x^ r\f\r\rv\0 <- I.. K ~^ 0.00001 _ ^~>^^;^ir^ ^**- ^ ^ 200 400 600 600 1000 1200 1400 DISTANCE FROM ELECTRODE IN FEET 1600 1800 Fig. 10 — Cross Keys tests, electrode. Mutual impedance gradient in vicinity of grounding Comparison with calculated values. short distances from the electrode and towards the latter for long dis- tances. As in the measurements previously described, the calculated mutual reactance component is greater than the measured value, MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 639 although in this case the discrepancy is substantially smaller. On Fig. 11 are shown the phase angles of the gradient as computed from the calculated values of resistance and reactance components given on Fig. 10. Here also the measured curve falls between the two calculated curves. 90 80 70 60 UJ UJ o UJ a 50 ,40 30 20 10 -10 ^ — — CALCULATED X=2.5X I0"'3aBMHOS cm. CUBE"" -^ ^^■'' __^ , -o- • ^ ^ \ ^ ^ • / / /^ / / / / J / / / t / -MEASURED 4 / / / / f / / / / / CALCULATED X=2.5 X lO'l^ABMHOS CM. CUBE • • V L ^«3=3 200 400 600 800 1000 1200 DISTANCE FROM ELECTRODE IN FEET 1400 1600 Fig. 1 1 — Cross Keys tests. Comparison of measured and calculated (from Campbell's formula) angles of mutual impedance gradient. Since the gradient near the electrode is obviously affected mainly by the conductivity of the earth in the immediate neighborhood, and that at remote points is influenced more by the conductivity of the earth at substantial depths, the possibility that the earth in this region is not homogeneous, but stratified, is suggested. These curves seem to sup- port the conclusion that the earth in this neighborhood has at least two strata, the upper one having a very low conductivity and the lower one a conductivity approximately a hundred times greater. Further e.xperimental evidence tending to the same conclusion has been ob- tained, and will be described presently. For the present it may be pointed out that this conclusion is supported by the geological data pertaining to this region, for which an upper layer of sand and gravel from 130 to 170 ft. in depth is indicated, superimposed on a mixed structure of sand, clay, and shale, with a substantial amount of ground water. Methods Considering A.-C. Distribution of Earth Current. The problem of computing the mutual impedance of ground-return circuits, considering an a.-c. distribution in the earth has been attacked by sev- 640 DELL SYSTEM TECHNICAL JOURNAL eral writers.^ In the interpretation of the experimental results, the papers of J. R. Carson ^ and F. Pollaczek '' have been used, since a min- imum of assumptions was made in the solutions advanced by these writers. The assumptions made are that the disturbing circuit is straight and of great length,-' that the propagation constant, in ab- solute units, of the circuit is very small compared to unity, and also that the earth is a homogeneous body of fairly good conductivity. With these assumptions it is found possible to solve the fundamental field equations for the magnetic and electric fields in the vicinity of the disturbing conductor at points remote from the ends of the circuit and thence to get the mutual impedance. Physically, this method recog- nizes and takes into account the fact that in a conductor of large ex- tent, such as the earth, the distribution of alternating current will be influenced by the changing magnetic field. Qualitatively, the effects are similar to those involved in the well-known skin effect, and may be thought of in terms of a distribution of eddy currents in the earth. It is obvious that the distribution of the eddy currents will depend on the earth conductivity and also on the frequency. The resultant fields, and hence the mutual impedances, will then be functions of earth conductivity and of frequency. Presentation of the formulas and graphs giving the results of the analysis is outside the scope of the present paper, and reference should be made to the original papers for these. As an illustration of the re- sults, however, the curves of Fig. 12 have been prepared, showing the calculated mutual resistance and reactance of ground-return circuits at a frequency of 60 cycles for several values of earth conductivity, within the range of experimental values. Both the resistance and reactance components are seen to be pronouncedly affected by earth conductivity, particularly for the larger separations. In applying this theory to the tests made at Cross Keys, the pro- cedure adopted, in the absence of direct data on the earth conductivity at this location, was to choose an earth conductivity which would result in the best fit between the calculated and observed values, and to see whether a single value for earth conductivity would suffice to explain all the results. On Fig. 13 comparisons have been made between the experimentally determined mutual impedances for the 60- and 500-cycle frequencies; the curves were computed by use of the formulas given by Carson. It will be seen that in so far as the magnitude of the mutual impedance is concerned an excellent agreement can be made between ' See bibliography references 3 to 9, inc. ' See bibliography. * See bibliography. ^ See bibliography. MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 641 the calculated and observed values. However, for the best agreement it is found necessary to assume a different earth conductivity at 500 cycles than at 60 cycles. Thus, while at 60 cycles the indicated earth conductivity is 4.2 X 10~^^ abmhos per cm. cube, at 500 cycles it is 0.2 O O O a. ill 0. (O 2 I O 0.1 0.0 7 0,05 004 003 0.02 0.01 0.007 0.00b 0.004 0.003 0.002 0.001 0O00 7 00005 _ 00004 D 00003 I- D 2 0.0002 UJ o z < o < a. o z < o z < I- u} If) UJ cr 0.0001 — . , ~~^ - ^ ■•*iB^ 2jj^^ " ■*.^^ •-N. *• ■^ ■■-^ "^ ^ '■s. ^ 'V N ^\^ ^ N S \ \ N s \ \ s \ \, ""- • — — '^ — — = :: - ===•" ~ -"- -' 1^ ^ N J- ■ ^-^ s. — -, \ \ ;: ■s \ ' \ V > ^ I \ \ ^L \ S \ \ \ y N MUTUAL REACTANCE \ \ \ N S. ' \ ^ \ \ \, U^v^.^ MU1U«L KtilblANLL \ \ \^- • V" \ \ \ V \ COMPUTED USING FORMULA GIVEN IN BELL SYSTEM TECHNICAL JOURNAL VOL- UME 5,NO. 4, OCT. 1926, PAGE 539,(CAR50N) HEIGHT OF POWER LINE — h| = 50 FEET HEIGHT OF TELEPHONE LINE - h2=25 FEET EARTH CONDUCTIVITY — X- AS INDICATED ON CURVES— IN C.G.SE.MU \ \ -J \ \ ( V >, \ \ "-^ \ \ \ ^\ V . ^ V \ \ \ \ \ \ I \ 1 10 20 30 50 100 200 300 500 1000 2000 3000 5000 HORIZONTAL SEPARATION IN FEET 10,000 20,000 Fig. 12 — Resistance and reactance of ground-return circuits. Frequency 60 cycles. 2.76 X 10~^^ However, a computed curve for 500 cycles, using a conductivity of 4.2 X 10~^^ falls below the experimental curve by only 30 per cent. Table I gives the values of earth conductivity required to TABLE I. Cross Keys Tests. Earth Conductivity Giving Best Agreement Between Calculated and Measured Values of Mutual Impedance. Indicated earth conductivity from Carson's formulas ency cycles abmhos per cm. c\ 60 4.2 X 10-'' 200 3.75 X 10-13 500 2.76 X 10-13 1000 2.0 X 10-13 642 BELL SYSTEM TECHNICAL JOURNAL give the best fit to the curve of mutual impedance at each frequency. The range is not large, extending only from 4.2 X 10~^^ at 60 cycles to 2.0 X 10-13 at 1000 cycles. Turning to the components of the mutual impedance, however, the agreement is found to be not as good. Fig. 14 shows the measured values of mutual resistance and reactance at 60 cycles and at 500 cycles, also the computed values, the calculations at each frequency being made with the earth conductivity indicated in Table I. At 60 cycles 0.3 0.2 o 0.1 o m 0.07 CL 10 5 0.06 '^ 0.03 Z < Q ID CL 0.02 5 < D I- D 2 0.01 0.007 0.005 O"— ^ -":. tj^ *• ^ C ^L i- \ n. »^ c. ^^<> hC,\ > s-*-- '9^. N ^ >. ■^;y V\'+ -^^^ °v. V ^1n np— — - tc -n. 4 \ 47 ■f/ t^° Cyn " \ \ \ \ \ ■^-^^ \ > > \ \ J o- EXF EF IK/ Er -JT AL POINTS ^3 > ^. \ \ 1 xb 0.04 O --^T^- 20 30 50 100 200 300 HORIZONTAL SEPARATION IN FEET 500 1000 Fig. 13 — Cross Keys tests — Carson theory. Comparison between calculated and measured values of mutual impedance. the agreement is quite good, but at 500 cycles the departure between calculated and measured values is large. The measured mutual resist- ances are consistently lower than those calculated, while the measured mutual reactances are higher. As indicated by the above comparisons, a theory of the kind under discussion leads to results which are in quite good quantitative agree- ment with the experimental results; it is of some interest to discover whether an extension of the theoretical ideas would lead to still closer agreement. It was stated previously that the measurements around I MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 643 the grounding electrode could be accounted for on the hypothesis that the earth in this neighborhood is stratified, with a conductivity of around 2.5 X 10~^^ near the surface and 2.5 X 10~^^ in the lower depths. Qualitatively, it is to be expected that with such an earth structure the mutual resistance would be less, and the mutual reactance greater, than the corresponding values for an earth of uniform con- ductivity, since the eddy currents near the surface of the earth will be less, due to the lower earth conductivity. 0.3 0,005 20 30 50 100 200 300 500 HORIZONTAL SEPARATION IN FEET 1000 Fig. 14 — Cross Keys tests — Carson theory. Comparison between calculated and measured values of mutual resistance and reactance. Quantitatively, it would seem that a first approximation to the effect of a stratified earth in which the upper stratum has a much lower con- ductivity than that of the lower region could be obtained by assuming that the currents in the upper layer are negligible and hence that this layer can be abolished. The mutual impedances can then be worked out by the formulas applicable to a homogeneous earth, using the earth conductivity of the lower region and fictitious conductor heights, formed by adding the thickness of the upper stratum to the heights of the conductors above the actual earth's surfaces. Preliminary calcu- lations have been made using this scheme, and it was found that using a 644 BELL SYSTEM TECHNICAL JOURNAL conductivity of 5.0 X 10~^^ for the lower stratum and a thickness of 130 feet for the upper stratum an excellent agreement could be found between calculated and observed values for all frequencies. The agree- ment extended not only to the magnitudes of the mutual impedances, but to the components as well. Because of the simplifying assumption that the disturbing circuit is so long that the effects due to the ground connections at its ends can be neglected, the theory which we have been discussing is obviously in- adequate to explain the phenomena in the neighborhood of the ground- ing electrode. Practical Methods for Predeterminixg Coupling between Power and Telephone Lines The ultimate purpose of the work of the committee is to develop simple methods to enable the calculation of the mutual impedance be- tween power and telephone circuits before they are built. It is evident from the foregoing discussion that the use of any formula for the mutual impedance of ground-return circuits requires a knowledge of the con- ductivity of the earth or of the depth of the equivalent ground-plane. In the relatively few places in which tests have been made, a range of earth conductivity from 10^^- to 10~^^ abmhos per cm. cube has been observed, and reference to Fig. 12 indicates that within this range of earth conductivity a variation in mutual impedances of 20 to 1 or more may exist. Therefore, other experimental work has been done with the object of developing relatively simple testing schemes, the results TO SPIER J> FALLS ^ tJO GLENS FALLS SARATOGA- ^(^/^ ^nV^ 0 GLENS FALLS -i^^^ I ^°-^ ^ SARATOGA— MONTREAL" ' TO WHITEHALL SEE DETAIL A DETAIL "A" TWO TOLL LEADS AT ROADWAY SEPARATION SCALE IN KILOFEET 0 I L . EXPLORING WIRES SCALE IN KILOFEET 0 12 34 Tower 99 Circuit Height Frequency I Fig. 15 — Glens Falls tests. Circuit arrangements 3-Phase conductors Power 50 ft. 60 cycles Grounded Telephone 20 ft. I MUTUAL IMPEDANCES Of GROUND-RETURN CIRCUITS 645 of which could be used to predict the coupling coefficients in advance of the construction of the power or telephone line. An obvious method is to determine an experimental coupling curve by performing tests simi- lar to those made at Cross Keys, using short-length disturbed circuits and either an existing power or telephone line, or a specially laid out conductor, as the disturbing line. 0.07 0.05 0.04 O O a. in 5 0.03 0.02 0.01 0.007 - 0.005 O 0.004 z K 0.003 V < ui (£ 0.002 o z < UI O 0.001 z < S 0.0007 in tt 0.0005 -" 0.0004 < K 0.0003 D 2 0.0002 OOOOI i fe,- •: MUTUAL RESISTANCE^ N MUTUAL REACTANCE — - •• ■..^\ — ..^ > ^ \ - -*— ^ \ ^ J ^ s, V ^ *N V ^ > V S ^° rAI CAH ATFD FOR \ s Vk ^=1. 75X10-12 \ N^ \ \ \ \ \\ > >,\ \ A \ \ V \ \ \ \ \ .v \ \ V H- \ \ \ \ ■ \ \ \ \ 1 b 30 50 100 200 300 500 1000 2000 3000 5000 HORIZONTAL SEPARATION IN FEET Fig. 16 — Glens Falls tests. Experimental values of mutual resistance and reactance. This experimental curve would then be used to compute the coupling between power and telephone lines. One advantage of using an experi- mentally determined coupling curve is that it obviates the necessity of knowing or assuming a structure and conductivity of the earth; the coupling curve can be used directly without reference to any theoretical formulas. To determine the practicability of such a scheme, 60-cycle tests have been made in two locations where existing exposures were present, for the purpose of determining the accuracy with which experi- mental observations could be predicted. 646 BELL SYSTEM TECHNICAL JOURNAL Tests at Glens Falls, N. Y. Fig. 15 shows the arrangement of circuits involved in tests made at Cilens Falls, N. \ . A section of the Saratoga- Glens Falls telephone line about six miles in length was energized with ground-return current. Measurements were made of the voltages induced in short ground-return circuits laid on the ground parallel to the straight section of the telephone line. The resistance and reactance components of the mutual impedance derived from these measurements are given on ¥{g. 16. As a matter of interest the mutual resistance and reactance computed by the use of the Carson formulas for an earth conductivity of 1.75 X 10~^^ are also given. This earth conductivity gives the best agreement between the calculated and observed magni- tudes of the mutual impedances. The general agreement between the computed and observed quantities is much like that found from the Cross Keys tests. Earth return current was then sent over the power line from Spier Falls to Tower 99, and induced voltages measured in the entire exposed section of the Saratoga-Montreal telephone line, and in several parts of the exposure as indicated on the sketch. In Table II, the observed TABLE II. Glens Falls Tests Measured Mtitual Impedances of Power and Telephone Circuits and Comparison with Values Calculated from Coupling Curves of Fig. 16. Section of Measured mutual Calculated mutual telephone line impedance — ohms impedance — ohms 0-1 .0586 /68.5° .0614 r^3.3° 1-2 .0294 /52.4° .0564 /49.8° 2-3 .0476 /73.4° .0382 /69.6° 3-4 .107 /56.4° .113 /49.2° 4-5 .100 /44.4° .0117/35.8° 0-5 .347 /5 .8° .267 /55.3° mutual impedances determined from this latter test are compared with values calculated by using the experimental coupling curve given on Fig. 16. The agreement between computed and observed values is, in general, only fair, although for two of the parts, the agreement is excellent. It is thought that the rather poor check for Sections 1-2 and 2-3 is due to the inductive effect of currents set up in the ground wire on another power line which extended through these sections. With regard to the extreme departure of the measured mutual impedance for Section 4-5 from that calculated it is impossible to decide the cause from the experimental data available. A possible explanation is that it is due to a large difference in earth conductivity in this region from that in the region in which the coupling curve was determined. The MUTUAL IMPEDANCES OF GROUND-RETURN CIRCUITS 647 large difference in this section is reflected in the rather poor check in the overall coupling (Section 0-5). Tests at Massillon, Ohio. Tests made at Massillon, Ohio, were similar to those at Glens Falls, except that the arrangements were somewhat more elaborate. The layout of the circuits involved is shown in Fig. 17. The exposure is about 16 mi. long with separation between the power and telephone line ranging from a crossing to about 4200 ft., a large part of the exposure being at a separation of about 3000 ft. A set of "exploring wires," each 200 ft. in length, was laid on the ground parallel to the telephone line as shown in the detail of Fig. 17. These were arranged in four groups and were distributed over an area approx- TO 5HLAN0 N s 1 ' n 1 •^ ¥ '^1 ?• S: 2i ' S CHICAGO- ._! ^yiL.''' _^£Ll. •" PHILADELPHIA TO MANSFIELD \MASSILLON-ASHLAND I '• "T"""^ ; f ",''^^~ 1°'"^ "^'^^ • 1-^ ! ' 1 '^^>^V "'t 3PHASF LINE I DEf^iU^A" I I ;«0 CYCLES -^ EXPLORING WIRE I DFTAIL "A- S I r NETWORK -j . DETAIL A ^^^^^ « * I I . IN MILES ■" J > EXPLORING 1^ 0 as 10 < ; SCALE WIRES :2 " 1 P'. r IN FEET ->) "St " ; 0 1000 1 / It ' -■J: ^_ i lOy-^ I " "_/["CHICAGO"-j;;PHiLADELPHIA LINEfj I CIRCUIT POWER I SI ttl J 5 HEIGHT 50-80 FEET : ""r hI^ ^ & ■ FREQUENCY 60 CYCLES i 5i '-^A !— "'ImSsLIw- ASHLAND LINE J Fig. 17 — Massillon tests — circuit arrangements. MASSILLON SUBSTATION y TO ^CANTON_ TELEPHONE 22 FEET imately 1>^ by 2 mi. Coupling curves were determined from measure- ments of the voltage induced in the exploring wires for the condition of the telephone line energized with 6 amperes ground-return current, and also for the condition of the power line energized with 40 amperes ground-return current. The mutual impedances derived from the two sets of measurements are practically identical. Fig. 18 shows the resistive and reactive components of the coupling curve using the aver- age of all measurements made on the exploring wires. A comparison of the measured curves with curv^es calculated by Carson's formulas for a value of earth conductivity of 3.6 X 10~^^ abmhos per cm. cube show the same type of agreement as that observed at Cross Keys and Glens Falls. The principal reason for using such a large number of exploring wires on this particular test was to investigate the effect of local irregularities of the earth upon an experimental coupling curve and to determine the minimum number and length of exploring wires which it is necessary to use in order to be reasonably confident of the accuracy of the results. 648 BELL SYSTEM TECHNICAL JOURNAL The data indicate that if only one of the seven groups of 200-ft. explor- ing wires had been used, the maximum deviation of any one point from the corresponding point on the average curve would have been less than 25 per cent and that the probable deviation would have been less than 10 per cent. This deviation could probably be reduced by using a somewhat longer exploring wire. O O a. UJ Q. 0.1 0.07 0.05 0.04 0.03 0.02 0.01 0.007 U 0.005 < 0.004 H O < 0.003 o 0.002 z < UJ o < 0.001 11 oa u. .Q ~ O try ^^ l^ I z ^ 36 30 25 20 15 1 0 5 0 ^ 0 1000 2000 3000 4000 FREQUENCY IN CYCLES PER SECOND Fig. 1 — Response characteristic of room noise meter. The room noise meter employed is shown together with its auxiliary equipment in Fig. 3. It consists of a condenser transmitter for convert- ing acoustical energy into electrical energy, six stages of amplification for raising the level of the noise currents sufficiently to operate a ther- mocouple meter indicating device, and a weighting network, as de- CONDENSER TRANSMITTER TWO AMPLIFIER STAGES _ ATTENUATOR (key 8, DIAL) ONE AMPLIFIER STAGE FREQUENCY WEIGHTING THREE AMPLIFIER STAGES THERMO- COUPLE TYPE DETECTOR MICRO- AMMETER ^ Fig. 2 — Schematic diagram of room noise meter. scribed above, as well as certain apparatus not employed in obtaining the results reported here. The general layout of the circuit is indicated in the schematic diagram of Fig. 2. A portable battery supply and means for calibrating form the necessary auxiliary equipment. An ^Physical Review, Vol. 19, April 1927, pp. 588-600. ROOM NOISES IN TELEPHONE LOCA TIONS 657 adjustable attenuator controlled by a key and a dial is provided between stages of the amplifier so that the noise energy being measured may be brought within the range of the meter over a range of levels of 80 db (corresponding to a power range of 100,000,000 to 1). Fig. 3 — Room noise meter and auxiliary equipment. Operation of the Room Noise Meter. — -The noise meter is first cali- brated, as described below, so that its sensitivity is set at a predeter- mined value. The condenser transmitter is then placed at the spot where it is desired to measure noise, and the gain of the amplifier is adjusted by means of a key and dial until the needle of the microam- meter in the output circuit fluctuates about a given point. The set- tings of the key and dial then give a measure of the noise. In addition to the average readings obtained in this manner, readings of the fluc- tuations in the noise can be similarly obtained. As an aid in the read- ing, the microammeter scale is calibrated in decibels. The calibration of the meter in the field consists of a check on the over-all sensitivity of the instrument. The filament currents and plate voltages are adjusted to the correct values. Then a fixed percent- age of the electrical output of a standard buzzer, the current from which is measured by a thermocouple, is fed into a special receiver which is placed in a prescribed way on the condenser transmitter. The gain of the amplifier is then adjusted until the output microammeter needle reaches a predetermined point. The sensitivity of the meter will then be as shown on Fig. 1. 658 BELL S YS TE M TE CHNI CA L JO UR NA L An over-all calibration of the meter, as a function of frequency, is given on this figure. To obtain this, separate determinations were made of the volts generated by the condenser transmitter per unit of pressure, and the meter reading per volt generated by the transmitter, as a function of frequency; and the results were combined to give the values shown. Harmonics in the testing waves were reduced to such a point that they did not affect the results. After a substantial part of the survey had been completed, a check was made of the electrical por- tion of the calibration, and the changes found were quite negligible. Accuracy of the Meter. — The precision of the apparatus is substan- tially greater than the precision with which ordinary varying noises can be measured. The readings obtained for steady inputs are propor- tional to the input, with an error of less than 1/2 db, over the entire range of noise amplitudes found in the survey. The apparatus is shielded electrically. In only one case did electrical fields produce any observed errors in the readings ; this was when an attempt was made to measure the room noise near a rotary converter in a power station. The vacuum tubes are mounted in such a way that the effects of ordin- ary mechanical vibration on the readings are negligible. Comparison of the Two Methods. — In general, the meter method gives results in physical terms while the masking method gives them in terms of effects on the ear; consequently, the choice of the method to be em- ployed in any particular case depends somewhat on the use to which data will be put. It is true that the meter includes a network to simu- late the sensitivity of the ear for various frequencies; it does not, how- ever, simulate other properties of the ear, such as the departures from linearity in response by which subjective tones are produced by the ear mechanism, and the complicated way in which one sound masks an- other.*^ The meter method, unlike the masking method, avoids any errors due to variations in human ears. This advantage is offset to some ex- tent by the fluctuations of the meter needle, which make it difficult to obtain the mean reading if the noise is unsteady as is the case with most room noises. In the case of noises of a distinctly intermittent, staccato character, the warbler tone can be heard and recognized in the brief intervals when the noise is a minimum. A preliminary investigation showed that, for a noise of this sort, the relation between readings obtained by the mask- ing method and by the meter method was different from the relation obtained for a steady noise, the warbler readings being relatively lower in the case of intermittent noise. * R. L. Wegel and C. E. Lane, "Auditory Masking and Dynamics of the Inner Ear," Physical Rev., Feb. 1924. ROOM NOISES IN TELEPHONE LOCA TIONS 659 Both methods were used in the survey, because it was felt that each gave information which could not be as accurately obtained from the other, and also because the use of two methods enabled each one to be used as a check upon apparatus defects which might occur in the other. In using the masking method, data were taken by two experienced observers and corresponding measurements averaged. All meter measurements were made by one observer. Results of Survey Noise in Business Locations. — One hundred and nine business loca- tions were visited. The magnitudes of the noises measured varied from that found in a doctor's quiet office to the din of a large manufac- turing plant. Distribution curves for the noises measured are shown in Fig. 4 for the meter method and Fig. 5 for the masking method. For any point on one of these curves the corresponding per cent of all of the measurements made had values equal to or greater than the indicated abscissa value. It may be seen that with the exception of the "high" curve of Fig. 5 the curves for meter and masking methods are fairly similar in shape. The "middle" curve has been selected to represent the masking method. If there are excluded as extremes those noises which were so low that 95 per cent of all the noises measured equaled or exceeded them, and those which were so high that only 5 per cent of the measurements equaled or exceeded them, the spread of noise magnitudes is seen to be about 40 db. The standard deviation of the measurements is about 12 db. As shown on Figs. 4 and 5 the median business room noise would produce a reading of 23 db on the meter scale and a masking of 27 db on the high-frequency warbler tone, 39 db on the middle-frequency tone, and 31 db on the low-frequency tone. The average business room noise was about 2 db higher than the median. Some conception of the amounts of noise represented by these figures may perhaps be gained from the following. The extremely loud noise measured in a local station of the New York subway while an express train was passing produced a meter reading of 70 db, while the lowest noises measured in the survey, in quiet residences, gave readings near Odb. Data on noise at the business locations tested have been grouped so as to show the average differences in the room noise values obtained for different types of business and for different sizes of towns. It will be appreciated that only a very small number of measurements were in- 660 BELL SYSTEM TECHNICAL JOURNAL CUMULATIVE PERCENTAGE OF MEASUREMENTS ooooooooooo s \ \ \ \ 1 \ \ s. \ \ \ V \ \ 10 20 30 40 50 NOISE METER READINGS IN db 60 70 Fig. 4 — Results of noise meter measurements of noise in business locations. H .■=»o < -I D 5 3 20 O 10 ^ ^^ N \ \ ■<^ v\ ^ \ v V o \ V \ \ ^ \ \ \ \ \ \ \ \ \ N \ ^ <\ i^^ \ V, ^ V ^^ ^ 10 20 30 40 WARBLER READINGS 50 60 70 •THRESHOLD SHIFTS IN db 80 Fig. 5 — Results of measurements of noise in business locations by masking method. ROOM NOISES IN TELEPHONE LOCA TIONS 661 eluded in each sub-classification, and that consequently it is not safe to generalize from these sub-groupings as to room noise conditions in gen- eral. Averages of the room noise measurements for the different types of business locations are shown in the following tableJ Type of business location Masking of middle-frequency tone Meter reading Number of measure- ments Offices 42 db 34 57 35 40 24 db 18 40 21 25 34 Stores Factories Public buildings 34 18 23 Average of all businesses (weighted according to num- ber of measurements made) .... 109 The above figures show a significant difference between the noise measured in factories and that measured in other types of location. The other differences shown were found not to be significant when ex- amined in the light of the spread in values for individual locations in each class. Averages of the business room noise measurements obtained in var- ious sizes of towns are shown in the table below. Size of town Masking of middle-frequency tone Meter reading Number of measure- ments Class A (over 400,000 pop.) Class B (100,000 to 400,000 pop.) 45 db 37 " 42 " 27 " 26 db 22 " 27 " 11 " 39 18 Class C (10,000 to 100,000 pop.) . Class D (less than 10,000 pop.) . . 41 11 These figures indicate that (with the exception of Class C towns) the business noise measured in large cities was greater than that in smaller towns. This is believed to hold true despite a fairly large spread in in- dividual measurements within a given class. The exception in the case of Class C towns is explained by the fact that a fairly large percent- age of the measurements in this class were made in large factories. Room noise in business locations was observed to be quite complex in ^ It will be noted that in the results given, the difference between the masking of the middle-frequency tone and the meter reading is relatively constant. It was found that for any considerable sub-group of the measurements, this difference was not far from 15 db. This figure, of course, would not in general hold for a single noise selected at random. There was a general tendency for the difference to be somewhat larger for larger values of noise. 662 BELL SYSTEM TECHNICAL JOURNAL frequency composition. The masking effect of the noise on the middle band was greater than that on the high and low bands. In order to give an approximate interpretation of this in terms of pressures in var- ious frequency regions, account must be taken of the relative magni- tudes of threshold pressures in the three warbler frequency bands, since the masking effects were obtained by subtracting threshold settings of the attenuator from the settings made in the presence of the noise. For CUMULATIVE PERCENTAGE OF MEASUREMENTS ooooooooooo '"v ^ \ \ \ ^ ■10 0 +10 +20 NOISE METER READINGS IN db 30 Fig. 6 — Results of noise meter measurements of noise in residence locations. the middle and upper bands, threshold pressures are about the same; hence, the lower values of masking for the high range indicate that com- ponents in this range are in general relatively weak. As previously determined,^ threshold pressures at frequencies in the low band are several decibels higher than those in the other bands. Combining the values of masking for the low and middle bands with the corresponding threshold pressures, it is seen that the physical magnitudes of compo- * H. Fletcher, "Useful Numerical Constants of Speech and Hearing," Bell System Tech. Jl., July 1925. ROOM NOISES IN TELEPHONE LOCA TIONS 663 nents in the low- and middle-frequency ranges are in general not far different. The above analysis is, of course, very rough, as the whole range from 250 to 5600 cycles is divided into only three bands. Room Noise in Residence Locations. — Measurements were made in 96 residence locations. Figs. 6 and 7 show distribution curves for these measurements. Compared with the corresponding measurements made in business 100 90 5 80 a. D if) < 70 O u O < I- z LU o a. 60 50 40 < D D u 30 20 10 1 \ \\ N \ "A \r \o 03 1 ^ I. o \ \ \ \ \ V \ \, \ \ ^ 0 10 20 30 40 WARBLER READINGS-THRESHOLD SHIFTS IN db Fig. 7 — Results of measurements of noise in residence locations by masking method. locations it is apparent that the room noises encountered in residences were not only much smaller in magnitude but also varied less in magni- tude than business room noises. The average of the residence room noises is about 18 db less than the average of the business room noises, while the spread in residence room noise (using the 95 per cent and 5 per cent points on the curves as limits) is 20 db, compared to 40 db for busi- ness noise; the standard deviation of the residence measurements is 6 db, compared to 12 db for the business neasurements. Unlike the 664 BELL SYSTEM TECHNICAL JOURNAL curves for business noise, the curves for residence noise are very sym- metrical, showing similar distributions above and below the average values. As shown on Figs. 6 and 7, the median residence room noise would produce a reading of 7 db on the meter, and a masking of 12 db on the high-frequency warbler tone, 20 db on the middle-frequency tone, and 13 db on the low-frequency tone. The average was about the same as the median. The average of the room noises measured in single-family houses was practically the same as the average of the noises measured in apart- ments. Averages of the residence room noise measurements obtained in towns of various sizes are shown in the following table: Size of town Masking of middle-frequency tone Meter reading Number of measure- ments Class A (over 400,000 pop.) Class B (100,000 to 400,000 pop.) Class C (10,000 to 100,000 pop.) Class D (less than 10,000 pop.) 20 db 20 " 23 " 17 " 7db 8 " 6 " 7 " 33 14 37 12 It will be observed from this table that the residence noises measured in large cities were no greater than those measured in smaller towns. A study of the data showed that 27 of the i?) measurements made in Class A towns were made in residences which would be classed as apart- ment houses. It is possible that the noise usually associated with big cities is confined chiefly to non-residential locations, and that apart- ments on side streets are no noisier than residences in smaller towns. It should be recalled, however, that the number of measurements in each class of town was very small. In any case the data tend to show that the difference between residence noise in the large city and that in the smaller town probably is not extremely large. The measurements for Class A cities were made chiefly in Manhattan and Brooklyn with a small number in Newark. It was found, in a manner similar to that discussed above for business noise, that the average residence room noise was quite complex in fre- quency makeup, and apparently did not differ materially from the aver- age business noise in the relative amplitudes of low and high frequen- cies. Comparison of Room Noise in Winter and Summer. — Forty locations were visited both in summer and in winter and the data compared. It was found that both business and residence noises were somewhat ROOM NOISES IN TELEPHONE LOCA TIONS 665 greater in summer than in winter, the average difference being 4 or 5 db. The spread in values obtained under summer conditions was less than that found for the winter conditions. This was because the noises which showed the least magnitude, when measured in winter, were found to be higher under summer conditions, while the highest noises measured failed to show an appreciable change with season. These highest noises were largely caused by indoor machinery, and would not be appreciably modified by outside sources. The average frequency composition of the noises measured under both summer and winter conditions seemed to remain about the same as far as could be determined. Selection of Typical Room Noise and Its Reproduction The data obtained have been used in determining the characteristics of a typical room noise to be recorded on a phonograph record and re- produced for use in laboratory tests. Since the data revealed no difference between the average frequency composition of great and small noises, it has been possible to choose a single recorded noise and to vary merely the amplitude of the repro- duced noise, keeping its frequency makeup constant. The recording and reproduction of such a noise have presented pro- blems, particularly from the point of view of naturalness. It has been found difficult to reproduce a noise by simple means in such a way as to give the illusion that the noise is real, not artificial. The requirements for reproducing a noise which will be typical in its effect on the intelligi- bility of speech transmitted over telephone circuit are, however, con- siderably less severe than those for obtaining naturalness. Three main factors seem to be involved in the problem. In the first place, room noises often contain high-frequency components, undoubtedly includ- ing some extremely high frequencies. These components, while they are generally of low energy content, seem to contribute substantially to the naturalness of the sounds. The effect of these components on the intelligibility of speech transmitted over a telephone circuit would, however, be much less than their contribution to the naturalness of the noise, since the transmitted speech is generally limited to a band of not more than 3000 cycles. The frequency band transmitted by the re- cording and reproducing system was nearly twice this amount, being limited both by the mechanical characteristics of the apparatus and by the unavoidable noise generated in this apparatus, the amount of this noise increasing as the band width increases. Second, room noises emanate from a considerable number of sources located in different positions, so that in order to reproduce them with complete fidelity 666 BELL SYSTEM TECHNICAL JOURNAL List of Towns where Room Noise Survey Measurements were Made Number of measurements Business 7 32 Residential Class A (over 400,000 pop.). • • Brooklyn, N. Y. Manhattan, N. Y. 18 13 Newark, N. J. 0 2 Total 39 11 7 3i Class B (100,000 to 400,000 DOD.) Jamaica, N. Y. Yonkers, N. Y. 11 3 Total 18 0 7 0 14 Class C (10,000 to 100,000 pop.) Bloomfield, N. J. East Orange, N. J. Flushing, N. Y. 2 9 5 Harrison, N.J. 3 0 Kearny, N. J. Maplewood, N. J. Milburn, N. J. 6 6 4 0 10 1 Mt. Vernon, N. Y. 6 4 New Rochelle, N. Y. 0 2 Orange, N. J. Summit, N. J. 4 2 0 2 West Orange N. J. Total 3 41 6 0 2 37 Class D (less than 100,000 pop.). Hollis, N. Y. Madison, N. J. 7 3 Pelham, N. Y. 3 1 Richmond Hill, N.Y. 2 1 Total 11 109 12 Grand total 96 each source must be reproduced separately in its own position. On account of binaural effects in hearing, the proper locating of sources seems to have a considerable effect on naturalness. The most practi- cal method of securing an approximation to this effect in the reproduced noise is to dispose a number of loudspeakers in different places in the room, chosen by test so that false directional effects are avoided. Third, the effects of reverberation must be considered. A noise picked up in a highly reverberant room, and reproduced in another highly re- verberant room, w^ould have in it tw^o sets of reverberations. The best method of taking care of this seems to be to make artificial adjustments in the reverberation in the two rooms. Finally, there is a residual effect due to the fact that a person experiencing an actual noise is aided in his recognition of the noise by visual and other factors enabling him to refer it easily to its source; these are, of course, not present when the sound is reproduced. ROOM NOISES IN TELEPHONE LOCA TIONS 667 Conclusions and Acknowledgment While a certain amount of work on room noise conditions in tele- phone locations had been previously carried out, this survey represents a considerable advance in knowledge of room noise magnitudes. It provides data for work on the effects of noise on telephone transmission as well as furnishing certain information of wider interest. The meth- ods of measurement employed, when further developed in the light of the experience gained in this work, should prove valuable in other room noise investigations. The authors wish to acknowledge the work of Messrs. J. W. Whit- tington and R. E. Philipson of the National Electric Light Association and Messrs. J. M. Barstow and R. S. Tucker of the American Tele- phone and Telegraph Company, in designing and building the room noise meter and in carrying out the survey. Contemporary Advances in Physics, XXI Interception and Scattering of Electrons and Ions By KARL K. DARROW This article deals with a couple of aspects of one of the amplest questions of modern experimental physics: the question of what happens when an elec- tron (more generally, an electron or a proton or a charged atom of any kind) collides with an atom or a molecule. It is well known, of course, that if the electron has energy enough, it may excite or ionize the atom. There are many different modes of excitation, and often several of ionization; the variety of possibilities is wide. If any of them occurs at an encounter, the electron loses energy and speed, and may suffer a change in the direction of its motion — a "scattering," as this is called. Even if it loses no measur- able amount of energy at a collision, it may be "scattered," that is to say, deflected. The scattering and the energy-losses of the electrons are studied both on their own account, and because of the light they shed on what is happening to the atoms. ^ IMAGINE a stream of electrons projected, all with known and uni- form velocity and along the same direction, into a rarefied gas. Perhaps it ionizes the gas; if so, positive ions appear, and one may de- tect them and identify them and count them in any of various ways, without concerning oneself about the destiny of the ionizing electrons. Or perhaps it excites the gas without producing ions; if so, the atoms (or molecules) send forth light, and one may detect the excitation and identify the manner and measure the likelihood thereof, without paying any attention to the corpuscles responsible. Nevertheless, these corpuscles also must have been affected; they must have given up some at least of their kinetic energy, and if they still retain some motion, it is probably no longer in the same direction as at first. If there is ionization or excitation of the gas, there should be electrons wandering off obliquely from the stream, and moving more slowly than when they entered the gas; in technical language, there should occur "scattering with loss of energy." Even if the incoming corpuscles are moving too slowly to ionize or excite, there might be — and there are — electrons wandering off obliquely with practically undiminished speed; they have suffered "elastic impacts" with atoms or molecules, they have been deflected merel}-, or "scattered without loss of energy." And even if the incoming corpuscles are moving fast enough to ionize or excite, some may be scattered with undiminished speed while others are spending some of their energy in these operations. Also, some of 1 This article, in somewhat altered form, is to appear as a chapter in a forthcoming book on ionization and conduction in gases. 668 CONTEMPORARY ADVANCES IN PHYSICS 669 the electrons may adhere to atoms or molecules.^ One sees that there are several items of knowledge about the effects of the gas upon the elec- tron-stream, which cannot be discovered by studying the inverse effects of the electron-stream upon the gas, the formation of ions or of excited atoms; it is necessary to observe the stream itself. Even our knowledge of the latter effects may be improved by examining the former. These are the purposes of the experiments which I shall now describe. The first (and the most) of these experiments may seem rather para- doxical, in view of what I have just been saying; for they are experi- ments not on electrons of the stream, but on the absence of electrons from the stream. One sends a beam of these corpuscles (or, it may be, a beam of protons or of once-ionized potassium atoms) across a stratum of gas, measures the number which go in and the number in the emerging beam, and puts down the value of the difference as the num- ber which "vanish from the beam." "Vanish" is a good word in this connection; it is not meant, of course, that the missing corpuscles and their charges literally cease to be; it is meant simply that they do not belong among those which go straight through with undiminished speed and undeflected path, as though the gas were not there. I will say that they have been "intercepted," for this is a word which does not imply any choice among the varied possibilities of stoppage, adhesion, and deflection with or without loss of energy. Experiments on intercep- tion of fast electrons — up to 30,000 equivalent volts — ^were first per- formed at the beginning of this century; but the earliest accurate work on slow electrons — say 50 equivalent volts and downward — ^is only ten years old. The results of these experiments are very striking; but of course they yield only a small part of what is wanted. We want to know what becomes of the "vanished" electrons, which way they have gone and with what residual speed — the total number and the distribution-in- direction of those which have been scattered without loss of energy; the total number and the distribution-in-energy and the distribution-in- direction of those which have ionized or excited the atoms which they struck; and the number of those which have stuck to atoms, if such there be. Such information, as anyone would foresee, is harder to acquire. Of the distribution-in-direction of the scattered electrons, nothing was known four years ago ; and what in this last quadrennium has been ex- - Some gases being monatomic and others not, it is necessary to say "atoms or molecules" when making general statements, if one wishes to be exact; but in the following pages I shall often use either word by itself, even when the statement in which the term occurs is meant to apply to gases of both kinds. 670 BELL SYSTEM TECHNICAL JOURNAL plored is very little, compared with what remains. As for the dis- tribution-in-energy, the first step in determining it was taken some sixteen years ago. It was indeed a great step ; for it led to the discovery of the process of excitation, the transfer of energy from moving electrons to atoms which shift these latter from their normal condition into one or another of their "excited" states.^ But it was only a beginning; the method had to be much modified and refined, to make it capable of finding the answers to such questions as I have phrased above; and the modifications were scarcely even imagined as lately as four years ago. Hence the reader must not expect to be introduced to a very great body of systematized knowledge. As for experiments on protons and on other kinds of charged atoms, they too are all extremely recent. 1 begin with the experiments on interception of electrons. Suppose then that a beam of electrons is sent across a tube, having first been limited by a sequence of slits or holes so that it has a definite contour, like the beam of a searchlight, which it retains all the way across the tube if there is vacuum. Further, suppose that on the far side of the tube there is a collector just large enough to swallow up the entire beam so long as it does not spread, but no larger; or alternatively, a collector covered by a screen pierced with a hole just large enough to let the beam, or a fraction of it, pass through. Even so, the result may depend on the diameter of the beam in a way which the reader will see for himself later on; it is best to think of a very narrow pencil of corpuscles.^ When a gas is introduced into the tube, the current into the collector will decrease. The decrease will be proportional to the density of the gas, so long as this is not too great; and it will be possible to define a "cross-section of the molecule for interception of electrons" in the same general way as is the custom in many other fields. Which is to say: denote by dx the distance which the beam traverses through the gas; by Q the number of electrons which enter the gas per unit time, hence by Qe the amount of charge which in unit time would arrive at the collector were the gas away; by Re the amount which in unit time does actually arrive at the collector; by N the number of molecules per unit volume: then the cross-section in question — call it c — will be defined 2 That is to say, the discovery by experiment; it had been predicted by Bohr (for the history of these matters, see for instance my "Introduction to Contemporary Physics," Chapter yill). ■• Of course this is an ideal which can never be perfectly realized. No matter how many diaphragms may be set up in a row to narrow and sharpen the beam, there will always be transverse motions of the electrons, relatively more important the smaller the forward velocity is made. Moreover the mutual repulsion of the corpuscles will tend to widen out the beam by driving its members apart. This is one of the reasons why experiments in this field were first performed on fast electrons, then extended to smaller and ever smaller velocities as time went on and technique was improved. CONTEMPORARY ADVANCES IN PHYSICS 671 by the equation: Q - R = aNQdx. (1) The greater it is, the greater the fraction of the number of incident elec- trons which are intercepted, the greater the probabihty of being inter- cepted for any one electron; it is thus a measure of a probability of likelihood, the "likelihood of interception." The quantity {Qe — Re) is "missing current"; it is the amount by which the current to the collector drops off, when N molecules per unit volume are introduced into the tube. Nothing has yet been said, nothing has even been implied, about the fate of this lost current and about the missing electrons which presumably bore it into the gas. I have, in fact, been using the very neutral word "interception" so as to evade all implications in excess of what the data say, which is, that some of the electrons fail to persist in the beam. Not to suppose that they have been annihilated, there are at least two conceivable things which may have happened to them. They may have made elastic im- pacts against molecules, bouncing off in new directions, and being thus deflected out of the beam without suffering much change in speed. Or, they may have struck and stuck to molecules moving in other directions than that of the beam. Other possibilities are thinkable; but these are enough to hold in mind for the present.^ (The word "absorption" is used by some, especially by Germans, in the sense for which I here use "interception." It seems to me to con- vey unwanted implications, but there may be differences of opinion on this point. Much the same problem of language occurs in optics. Usually the term "absorption of light" means in practice "departure of photons from a beam of light" irrespective of whether they are actually swallowed up by atoms, or deflected without any loss of energy; but it is rather common nowadays, especially in treating of X-rays, to use "absorption" for the former mode of disappearance only, and "scatter- ing" for the latter. Since it is necessary now to distinguish two kinds of scattering of photons, the complications are beginning to rival those of electronics.) Adopting either the elastic-impact idea or the adhesion idea, we may visualize this quantity o- in a familiar way. We may conceive of the molecules, for this purpose and for this purpose only, as spheres so constituted that when an electron touches one of them it sticks — or else rebounds, whichever theory we are using. The value of a is then ^ Lenard reviewed a number of possibilities, and considered ways of distinguishing them in his brochure Quantitatives ilber Kathodenstrahlen. He made a peculiar dis- tinction between reflection of electrons from molecules, and small deviations of elec- trons by molecules; it seems to have been suggested by his work on very fast cor- puscles. 672 BELL SYSTEM TECHNICAL JOURNAL the value which must be assigned to the cross-section of these spheres, in order to make the calculated values of "missing current" agree with the observed ones. An elastic-sphere model is also used in the kinetic theory of gases: one visualizes a gas as a flock of spheres, and for their cross-section one chooses the particular numerical value which, when inserted into the kinetic-theory formula for the viscosity of that gas, gives a figure agree- ing with the measured viscosity. This is the so-called "gas-kinetic cross-section," which I will denote by o-q. One should not expect it to be identical with the quantity a which has just been defined, nor be surprised at finding differences — even differences in order-of-magnitude - — between the two. The elastic-sphere model is good for many pur- poses; but it has its limitations. The ratio of \ \ V \ N l \ S^ 1 s k 120 V 'OLTS V 180 V OLTS \J ^*»<. 1 A - \ V c B "~~~- 0 5 10 15 20 25 30 35 40 45* e Fig. 11 — Distribution-in-angle of electrons scattered from molecules and atoms of hydrogen. (G. P. Harnwell, Physical Review.) electrons which were practically not deflected at all in their collisions, obtained curves with maxima remarkably sharp, indicating several distinct and very precisely determined energy-transfers, the two most prominent amounting to 12.78 and to 9.25 equivalent volts. Yet another device is that of R. Kollath, which is much simpler than Harnwell's, but functions only for one small range of angles of scattering; it is shown in Fig. 12. The primary electron-beam passes through the slits Bi and B^, then onward into the collecting-chamber A\ those corpuscles which are scattered at angles between (roughly) 87° and 93° are able to pass between the flat metal rings L of which one 690 BELL SYSTEM TECHNICAL JOURNAL sees the traces on the plane of the paper, and go on to the annular collector X; those which are deflected through other angles are caught by the rings. A potential-drop from the rings to the annular collector precludes from reaching this the electrons which in being scattered have lost more than 15 per cent of their initial energy, so that the ratio of the currents to K and ^ is a measure of the proportion of the I Fig. 12 — Kollath's apparatus for determining amount of scattering at angles near 90° {Annalen d. Physik.) primary electrons, which in traversing a (known) distance through gas of a (known) density are deflected without more than the stated loss of energy through angles within 3° of a right angle. Measuring the ratio at various pressures of gas, Kollath determined the corresponding cross-section for a number of gases. The continuous curves of Fig. 13 are those which he obtained for three of them; the broken curves represent the intercepting cross-section as found by Ramsauer, with ordinates reduced in the ratio 1 : 20. One sees that for each gas the two cross-sections vary in much but not altogether the same way.^^ I must quote the results of the work of Langmuir and Jones, though 1' It happens that the ratio 1 : 20 is about the ratio which the electrons received by the collector K would bear to all the scattered electrons, if the distribution-in- angle of these were isotropic — which of course is not necessarily so. CONTEMPORARY ADVANCES IN PHYSICS 691 the method which they invented is so extremely different from any of those by which the foregoing data were acquired, that in this place I cannot give anything like a full account of it. Briefly: the gas is in a metal cylinder having a filament running along its axis and metal plates lOOr \J VOLT lOOr He VvOLT 100 ^^/ A CO2 / / / /volt Fig. 13 — Broken curves: cross-sections by Ramsauer's method. Continuous curves: cross-sections for scattering at angles near 90°, by Kollath's method (see text). {Annalen d. Physik.) which almost close its ends. One of these end-plates is raised to a potential some fifty or a hundred volts above the filament; and so dense is the electron-current pouring out of this latter, that almost the whole of the gas in the cylinder becomes violently ionized and shining, 692 BELL SYSTEM TECHNICAL JOURNAL and assumes the potential of the plate. I say "almost all" of the gas; a narrow cylindrical sheath about the wire remains comparatively dark, and between the wire and the outer frontier of this sheath the entire potential-rise is spread. Having traversed the frontier, the electrons shoot into the luminous zone of the gas with the energy corresponding to the full potential-rise from the filament to the end-plate. It is as though the boundary of the sheath were a grid connected to the plate; the discharge itself creates its own impalpable grid. The method, then, consists in measuring the current into the cylinder over a range of values of the potential thereof, beginning when the cylinder is at the same potential as the filament and the only electrons which can attain it are those which come clear through the luminous zone without deflection or loss of energy, and ending when it is at the same potential as the end-plate and the luminous gas (or at any rate when it is well above the filament) and electrons can reach it despite their collisions en route. By analyzing the shape of the curve, Langmuir and Jones are able to deduce the values of the cross-section for interception for the various gases they tested, and values of several other things as well; but the analysis is intricate. I shall therefore say only that for the gases neon, hydrogen, argon, helium, nitrogen and mercury, and applying to the electrons voltages ranging from 30 to 100, they found for the cross-section values departing by less than ten per cent from the gas-kinetic cross-section - _ __ ^ _ 5-00 5:20 5.40 6-00 TIME Fig. 8 — Illustrates effect of pin length and baffle on d.-c. surface leakage. mounted on wood pins in the standard manner with similar ones mounted on wood pins which have been covered with a thin metal foil, .04 03 z o a. lij a. (/) O 02 I O a. o .01 Q ' n A H 1 S; i j 1 , > 1 WOOD PINS ' COATED AND BONDED / jl >^ ^* ^ y \x / ij ^^ — ">. / '^, / \ i/ > / y / Ij / '\. -'> \^ _ ( / N / \ I f t y > N ry ^ — *. >^ '^ \ V -^ / / ^WOOD PINS \ / / / r ^ :^ r- 10 12 6 AM 10 12 2 -.— PM— ^ Fig. 9 — Illustrates effect of coating and bonding of wood pins on d.-c. surface leakage. the coatings of a pair of pins being electrically joined. Thus, both the pins and the crossarm between them are short circuited. Fig. 9 shows the results of such a test on the type of insulator shown A STUDY OF TELEPHONE LINE INSULATORS 707 in Fig. 1. At the start of the rain and during the first few hours the wood appears to reduce the leakage considerably but as the rain con- tinues and the pin takes up moisture its effect seems to decrease to nil. While similar tests on this and other types of insulators give corres- ponding results, still other tests indicate that at times, the wood con- tinues to help out for many hours. One such test covering a period of about 23 hours showed, for this same type of insulator (Fig. 1), that the wood pin reduced the leakage by 30 per cent on the average. On the other hand, similar tests for the design of insulator shown in Fig. 10, very frequently show the pin to have negligible effect. Fig. 10 — Experimental design (C. D.) While the foregoing tests have shown an advantage in favor of the wood pin, that conclusion holds only for the conditions of the test; namely, both wood and metal pins having the same diameter. When an insulator is specifically designed for a metal pin both the pin and insulator diameters can usually be made small because of the greater mechanical strength of the metal pin. The advantages of the design of small diameter, as previously discussed, may readily offset the slight disadvantage of the metal pin and the net result may be an actual gain. 3. Influence of Insulator Material. While the insulators are new, the molecular attraction between the insulator material and water plays a very important part; so important, in fact, that tests on such new insulators frequently give an unreliable basis on which to make decisions. On new glass insulators and particularly on those of 708 BELL SYSTEM TECHNICAL JOURNAL borosilicate compositions, the rain water does not seem to wet the surfaces but rather stands in individual separate drops. The phenom- enon is so marked that the surfaces have the appearance of having been oiled or waxed. Under such conditions the conducting path is broken up and is discontinuous. Naturally its conductivity is very small. It would be fine if this property of the material could be preserved. Unfortunately, along with exposure to weathering there comes an increasing tendency for the drops of water to spread out and unite to form a continuous path. Apparently, this action results from the collection of impurities on the surfaces, for such molecular phenomena are well known to be very sensitive to any contamination. The direct surface leakage has been observed generally to increase 10 or more times after only a few weeks' exposure. In one particular case where it rained the same day that the new insulators were installed, their surface leakage was observed to be less than one per cent of that measured on insulators of the same shape and material which had been aged for about two years. This study of insulators has shown that while changes in design may affect changes in direct surface leakage by a factor of, say, 10, changes in kind of material affect this leakage by a factor of two or less after long exposure. In general, then, the material may be said to be less important than the design. Purely general considerations indicate this conclusion in view of the surface nature of the phenomenon. Con- sider an insulator of a given design exposed to the elements for several years. More and more foreign matter will collect on the surfaces as time goes on. It is obvious that as this process continues, the insulator material is becoming less and less important and it is conceivable, at least, that given time enough, this surface of foreign matter would determine the insulator performance irrespective of the material be- neath this surface coating. At this point, design is paramount. However, the material might be expected to influence the aging process in some way. From this viewpoint smoothness and hardness of surface, together with chemical stability, appear to be desirable qualities. This study has been confined chiefly to various kinds of glass. In general, these have all had smooth surfaces, but they have varied in hardness and chemical stability. For example, borosilicate glasses are said to be generally harder and chemically more stable than the common alkali group. The relative performance of one sample from each of these respective groups is given in Fig. 11. Both were molded to the design of Fig. 12. A STUDY OF TELEPHONE LINE INSULATORS 709 This particular test which was made after more than a year of exposure shows only a small difference between the two glasses. A previous test (exposure of nine months) showed quite the same results while 10 z .04 t .03 LiJ Q. m O .02 O a: o I ,01 u o < < ~~~ 4 - / A / f M 1 7 \\ 1 f \ \ y \ \ L f \^ 1 CS STD. )ROSILICATE / N / r \ \ V ^BC 1 > ^. _, ^ \ i k \ \ ^ ■ ■ \ ^ "* — " "" / CS ALKALI J GLASS /> \ \ — ^ — '' ~"~ — — — ' s ^^ "^ ■^" ■ "■ ~ I: 1 23456789 TIME-HOURS (am) Fig. 11 — Illustrates effect of insulator material on d.-c. surface leakage. still earlier tests had indicated the borosilicate as quite superior to the alkali glass. Fig. 12 — Standard C. S. design for use with standard steel pin. While the foregoing tests illustrate the point made above as to the relative importance of design vs. material they cannot be considered as conclusive evidence on the relative merits of these two glasses for direct surface leakage. 710 BELL SYSTEM TECHNICAL JOURNAL Probably more important than either hardness or chemical stability, which directly affect surface leakage, is the transparency of the material which affects leakage only indirectly, through the medium of insects, apparently small spiders. These spiders build nests on the inner surfaces and seem to prefer dark or dimly lighted spaces for their homes. Therefore, the more transparent the insulator, the less attractive home it makes; and of opaque materials, probably white ones are accordingly less attractive than dark colored ones. Other factors which enter into the spiders' choice appear to be those of space and of protection from the elements. These latter factors are functions of insulator design rather than material. There has been little opportunity to study these factors on the in- sulator test lines, due to the lack of spiders. Only one specific case has been found where their effect was marked. This was a case where small borosilicate insulators were given an opaque metal coating. After this coating had been on several months the direct surface leakage increased to several times the value for the similar uncoated ones. An investigation showed spider nests under many of the coated insulators. On the other hand, several types of larger insulators have shown no such effect when coated, although some of these have been exposed for several years. These results are not conclusive. They merely indi- cate that design, as well as transparency of material, is a factor. An experience of the telephone plant in the use of opaque insulator material (porcelain) showed a serious reduction of efficiency after a few years of exposure, apparently explained by the action of insects, 4. Specific Conductivity of Film. The specific conductivity of the rain itself before it reaches the insulators is determined by the nature and amount of impurities collected in its fall. Both the kind and amount of impurities must vary greatly in different localities, for ex- ample, industrial centers as compared with open country. Then again, the amount in any given locality must vary throughout any storm on account of the cleansing action of the rain on the atmosphere. On reaching the insulators the rain will suffer a further increase in conductivity depending on the impurities it finds there. Smooth vertical surfaces should be advantageous in reducing the collection of dust. After a prolonged dry period in which the surfaces have become dust coated, the conductivity may be quite high at the start of rain. As the rain continues a certain amount of cleansing action occurs on the unexposed surfaces, depending on the splashing. A decrease in leak- age corresponding to this cleansing action has occasionally been ob- served. A STUDY OF TELEPHONE LINE INSULATORS 711 Thus, in a locality where rain is infrequent but where dew, for ex- ample, might wet the insulators, the leakage might conceivably be quite high. This phenomenon can be produced artificially and emphasized by placing roofs over the insulators, thus preventing any direct cleansing action of the rain. Roofs of sheet metal six inches in height, five inches in width and twelve inches in length were placed over insulators of the design in Fig. 1. The ends were left open to permit passage of line wire. While still new and clean these protected insulators showed less leak- age than the unprotected ones of the same design. However, after a in ti. o 3 z o a. a. ui Q. to o X 2 o a: o I UJ o < < .04 — ^^ . •** ' *■* ~" ^ .<^?ri U3 'M i '^1 ■^ % ~ 1 O^ t >' ,^ Od 1 7^ ( 1 0. ~ f. ->. .01 // V — ^ -V i / \ 1 / ■^ — " N \^ 7 V y ' ^-H -. .^^ =;; /' < - __ _ ^^ «.• -'' / ■/ \ ^ y n 6 8 PM. 10 12 4 AM.- Fig. 13 — Illustrates effect of sheds on d.-c. surface leakage. few months the protected insulators became covered with a thick layer of dust blown under the roofs by wind. Then when it rained, enough moisture reached the surfaces to wet the dust, but not enough to have much cleansing action. The surface leakage then becomes very high, as will be apparent from Fig. 13. Item C — Dielectric Absorption in Insulator Material 1. General Characteristics. When a dielectric is subjected to a varying potential field a certain amount of the electrical energy is dissipated in the material in the form of heat, depending on the nature of both the material and the field. This phenomenon is commonly called dielectric absorption. 712 BELL SYSTEM TECHNICAL JOURNAL Such a field exists between the line and tie wires on one side and the pin on the other. The insulator material lies more or less in this field and, therefore, dielectric absorption is naturally to be expected. While it is convenient to refer to a single insulator it should be remembered that two insulators are in series between wires at any one point. The chief characteristic of the leakage resulting from this phenom- enon is its variation with frequency. Approximately, it increases directly with the frequency. In the voice range its magnitude is gen- erally negligible, but may become appreciable in the carrier range, particularly at the upper end. Item C, like item B, increases in wet weather, but to a far less degree. This increase, which is brought about by the enlargement of the field m ^ .08 .06 10 g .04 O a. o ' 02 o < < UJ r • ^ X y ^' .' .» _• / X ^ X >^ .' y / ^f 3^y ,«^ y ^ *^ "^1 1 '■^1 -^ ^ -*- »* ' 0^^<^ o ^ ^ v^e^ :^o^' • ' t ^ J^ 9^ ^^ \- flint/ r,\ A«;c;> > y « j^ '^ ^ f^ ^ - -= nnoriQii i- ME :as ^UBbLi-^ CA TE GL AS >S _ - _ -- JS. z= =.= "= — — ' — ^ "" 1ASE ANGLE .15° 10 20 30 F REQUENCY- KILOCYCLES 40 50 Fig. 14 — Variation of (C) with frequency for standard D. P. design. caused by the wetting of the insulator surfaces, cannot be directly separated from the increases in the several other sources. Indirectly, however, a general idea of the magnitude of item C in wet weather can be obtained in the following two ways. Both methods are based on the observation that the increase in capacitance between the wire and pin produced by metal-coating the outside of the insulator is invariably greater than the corresponding increase caused by wetting the uncoated insulator. For example, the increase due to coating the insulator of Fig. 3 exceeded the maximum increase observed for any rain by at least 40 per cent. More commonly this figure would be 100 per cent both for this and other shapes. Therefore, if such metal-coated insulators (mounted on bonded metal pins for reasons appearing in the discussion of items D and E) be meas- A STUDY OF TELEPHONE LINE INSULATORS 713 ured in dry weather the leakage conductance so found is Hkely to ex- ceed by a substantial amount the wet-weather value of item C. Another procedure consists in calculating the leakage of the metal- coated insulator. This method requires a knowledge of the wire-to-pin capacitance and the phase angle of the insulator material. Fig. 14 shows the measured value for the insulator of Fig. 1 when molded from a clear flint glass. The phase angle of this particular glass is not known. However, the wire-to-pin capacitance is known. Using this known value, the leakage has been calculated for two measured values of phase angle for flint glasses, between which the value of this particular glass probably lies. Similarly, the measured and calculated leakage are shown for this same design of insulator when molded from a borosilicate glass of known phase angle. Neither method is capable of high accuracy, but they need not be for our present purpose of determining the order of magnitude of item C. Consider item C at 50 kc. where its relative importance is greatest. At this frequency the metal-coated flint-glass insulator gave a measured leakage which is only about 10 per cent of the total wet- weather leakage of such an insulator as commonly used uncoated. So it appears that even at this high frequency, item C is less than 10 per cent of the total leakage of this particular design and material. This particular sample of flint-glass is not the best of the common alkali glasses nor is it the worst; so this round number of 10 per cent is a fair average value to use for alkali glasses. The corresponding value for the borosilicate glass of which the sam- ple was representative is about 4 per cent. 2. Influence of Design. The absolute magnitude of item C is in- fluenced somewhat by design because of the effect of shape on capaci- tance. An idea of the magnitude of this effect can be obtained by comparing Fig. 14 with Fig. 15. Both designs were cast from the same batch of glass so that material plays a small part in comparing the two designs. In general, for a given size of pin, the capacitance is decreased by enlarging the insulator diameter, particularly at the wire groove. Similarly, for a given outside diameter, the capacitance is decreased by making the pin diameter less. In general, too, the shorter the insula- tor, the less the capacitance. Through the ordinary range of shapes the absolute magnitude of item C will not vary more than about three to one, and expressed as a percentage of the total wet-weather leakage, it is doubtful if item C, in 714 BELL SYSTEM TECHNICAL JOURNAL the worst combination of poor design and poor material studied, could reach as high as 20 per cent. The foregoing remarks apply to insulators on metal pins. When wood pins are used, the capacitance is less and C is correspondingly less. If the pin is dry, C is extremely small ; if the pin is wet, C is still consid- erably less than it is for the same insulator on a metal pin. a. _i D (/) Z u. O a o I O o I liJ o < < in .\c. ^ / / y < .10 V y' t r y / / y / / f /- /' y 08 / / ,o - / X <^ t^ r /^ ,<, ( / 06 ^ <,'^' / ^ ^' / .rt J^/ ^ ^' * y / . V ^x ^' * / / "f^/ ^' * 04 / * / t > .V >' FLINT GLASS / 02 7— ^ ^* ^ • n 10 20 30 FREQUENCY-KILOCYCLES 40 50 Fig. 15 — Variation of (C) with frequency for standard toll design. 3. Influence of Material. The range of the absolute value of C for various kinds of glass is much greater than it is for various designs. Phase angle of the material is frequently used as a criterion but Hoch^ has shown that for insulator purposes, the product of phase angle and dielectric constant is a better criterion. Using the latter, a range of more than 20 to 1 is found for the glasses studied. From the standpoint of item C alone, on account of its small relative magnitude there is little justification in going to high grade glasses, especially as these are more costly, unless other sources are first reduced sufficiently to make C really important. 'See "Power Losses in Insulating Materials," E. T. Hoch, Bell System Technical Journal, November 1922, pp. 110-116. A STUDY OF TELEPHONE LINE INSULATORS 715 Item D — Dielectric Absorption in Pins I. General Characteristics. Item D applies only to pins of dielectric material (usually wood) or to metal pins with cobs of dielectric material. Like C, D is roughly proportional to the frequency and its impor- tance is greatest at the upper end of the carrier range of frequencies. Again, like C, D increases in wet weather due to the accompanying increase in wire-to-pin capacitance. It also increases because of ab- sorption of moisture by the pin. Fortunately, due to the small value of C, particularly when dielectric pins are used, a rough measure of D can be obtained by again making use of metal-coated insulators. (/I DC o I- < _I z U- o ^ o I o a. u I u o < < .i+ .3 - ^^ — • ■^ .2 -^ "-' ^ -^ -^ ^ ^ .1 y y' - - -- ' — -" ' ,^ — ■ — o 10 40 50 20 30 FREQUENCY-KILOCYCLES Fig. 16 — \'ariation of (D) with frequency for standard D. P. design. Fig. 16 shows two measurements of the leakage in dry weather of metal-coated insulators on wood pins. These insulators are similar in shape to that of Fig. 1, and were molded of a borosilicate glass. Due to the low capacitance when wood pins are used and to the high quality of glass, item C can be neglected here. Item E, which is to be discussed in the next section, enters here to some extent, but the cross-section of the crossarm is so much greater than that of the pins that this item, while it cannot be neglected, is probably small. The measured leakage then closely represents that due to dielectric absorption in the pins under the conditions of the test, namely, metal- coated borosilicate insulators. 716 BELL SYSTEM TECHNICAL JOURNAL To the extent that metal coating gives greater capacitance than does wetting the insulator the value of D is magnified by these measure- ments. On the other hand, to the extent that the capacitance is reduced by the high quality of glass, the measurements give too low a value of D for alkali glasses. As these effects approximately balance each other, the measured values are fairly representative of D for the alkali group. The difference in the two values measured at different times is mainly due to the moisture content of the pins, the higher value corresponding to the higher moisture content. Consider the insulator of Fig. 1 made of alkali glass and mounted on wood pins. Item C will be small, very small when the pin is dry; and while appreciably larger when the pin is wet, it is still considerably smaller than its value for a metal pin of the same dimensions. D, on the other hand, is large, even when the pin is dry; so in this example, D may be said to be considerably more important a factor than C. In the case of this same design molded from a borosilicate glass, D becomes relatively even more important a factor. 2. Influence of Design. The general remarks which applied to C as to insulator diameter and length and as to pin diameter apply to D. 3. Influence of Insulator Material. For a given design, the lower the dielectric constant of the insulator material, the lower D will be. 4. Influence of Pi?i Material. Some study of this subject has been made, but the results are not sufficiently conclusive to report. Item E — Displacement Current Losses in Crossarms 1. General Characteristics. This item will first be considered as it applies to insulators on metal pins. The path of the displacement current is as follows : the current passes from one line wire through the capacitance of the one insulator to its pin, thence through the crossarm to the other pin and finally through the capacitance of the other insu- lator to the other wire. The capacitances of the two insulators are thus in series. The adjective "displacement" has been applied to the current because its magnitude is mainly determined by the insulator capacitance. The losses produced in the crossarm by this current obviously depend on the electrical equivalent of the crossarm and on the insulator ca- pacitance. If the former were a pure resistance with a magnitude small compared with the reactance of the series capacitances, then E would increase approximately as the square of the frequency. Experi- mentally, E is generally found to increase at a rate lying between the first and second power of the frequency over the range studied. A STUDY OF TELEPHONE LINE INSULATORS 717 Ill o z o < o I- z .4 O X o °^ 3 UJ o z 1000^ .2 a z o o I- z < =5 500 -J .1 a ^ ^ -- "^ ^ ^ ■— ' '^^c-!^ co^ ^ X' ^ ^ X / / ^ - i:::^ / S^AC/TA, JC r V r _ - / 10 20 30 FREQUENCY-KILOCYCLES 40 50 Fig. 17— Variation of equivalent parallel capacitance and conductance with frequency for 25 crossarms in parallel dry weather measurement. a. o »- < _i Z a. UJ Q. o X o a: O u o < < o 1 — y /■ / 4 / / / / / 3 / / « t 1 ^^^ ? ^ i*->i V/ > / / / / / - 4 ^ r' / ^ -' t ^ oC -n ^ -^ f ^A^'' .SV; pr ^ ^ ^ ^^ r / 4u ^^ _ ^ — — -" . — n -- — — • -- — — " 1 rn 10 20 30 40 50 FREQUENCY-KILOCYCLES Fig. 18— Variation of (E) with frequency in dry weather. 718 BELL SYSTEM TECHNICAL JOURNAL A knowledge of the electrical equivalent of the crossarm, together with the insulator capcitaace, permits E to be calculated. The constants of the crossarm between steel pins six inches apart, as measured on a dry day, are given in Fig. 17. From this measurement, the magnitude of E has been calculated for individual insulator capaci- tances of 4 and 20 m. m. f., and the results are shown in Fig. 18. The capacitances of most of the insulators studied lie within these two limits. Fig. 18 also shows the value of E as measured for the insulators of Fig. 19. This measured value of E was obtained as follows: The leak- Fig. 19 — British post office design. age of the insulators was measured first with the crossarms short- circuited by tying a wire from each pin to the other. Then the wires were removed and the measurement repeated. The difference between these measurements is E, if other sources are assumed to have remained unchanged. This assumption appears justified from the following test of its validity. The measurement of E, together with the measured constants of the crossarm, enables the insulator capacitance to be cal- culated. Such a calculation has been carried out and the value of capacitance so obtained checks very closely the value obtained by direct measurement. A STUDY OF TELEPHONE LINE INSULATORS 719 The curves of Fig. 18 show that E may be large in dry weather. In fact, it is generally several times the magnitude of item C in dry weather and at the upper frequency range. Therefore, E is the controlling factor in dry weather. As to the effect of weather conditions on E, the increase in insulator capacitance brought by wet weather tends to greatly increase the losses. On the other hand, the large decrease in the crossarm impedance re- sulting from rain tends to decrease the losses. As a net result of these opposing effects, E is generally less in wet than in dry weather. In fact, after several hours of rain, E has been found to be almost negli- gible. The upper curve in Fig. 20 shows E for the insulator of Fig. 1 foil- coated and mounted on steel pins with composition cobs. This curve tn- o in o r o cr o u o < < .3 .2 y 7 4 y y ^ ^ y ^ .1 ,y y ^ ,^ y* ^ -^ ^ ^ ^ ^ << ^ -" ^ ^ i-" <-i 10 20 30 FREOUEINCY- KILOCYCLES 40 50 Fig. 20 — Measured leakage due to (E). was obtained in dry weather. The lower curve of Fig. 20 shows the magnitude of E for the same insulators after \y^ hours of rain. The equivalent conductance of the crossarms varies over a wide range depending on the weather. For instance, values of d.-c. conductance 40 times as great as that recorded on Fig. 17 have been observed and the data indicated that still higher values are probable. The smallest observed value was one tenth of that recorded in Fig. 17. The experimental determination of the magnitude of E for insulators on wood pins is a difficult problem, because E cannot be readily separated from D. The importance of insulator capacitance on E has already been es- tablished, so a knowledge of this factor for insulators on wood pins en- ables an estimate of E to be made. 720 BELL SYSTEM TECHNICAL JOURNAL For example, the insulator of Fig. 1 on wood pins gives a measured capacitance in dry weather of about 4 m. m. f. E for this value is given by the bottom curve of Fig. 18. In this particular example, E is quite small in dry weather. The general decrease in E accompanying wet weather indicates that this item does not contribute much of the leakage of insulators on wood pins, at least after the crossarm is very wet. 2. Influence of Insulator Design. The wire-to-pin capacitance varies perhaps three to one for the designs covered in this study. The corresponding range in E is obviously great. Therefore, insulator design has an important influence on the magnitude of E. 3. Influence of Insulator Material. The materials studied have not shown a range in dielectric constant of more than about two to one. The corresponding range of wire-to-pin capacitance is even less; so insulator material may be said to have a relatively small influence on E. 4. Influence of Pin Spacing. Item E is naturally expected to be influenced by the pin spacing. However, the data bearing on this effect are too meager to report. Item F — Losses Due to Unbalanced Displacement Currents Flowing in External Impedances such as Crossarms, Poles, Etc. 1. General Characteristics. In general characteristics, F is very similar to E. As E is caused by a displacement current which flows directly from one line wire to the other via the crossarm in the manner already discussed, so F is similarly caused by unbalance displacement currents flowing through crossarms, poles, etc. There is not sufficient space in this paper either to present details of the theory of these losses or to describe the many interesting tests made to illustrate the efi^ect. In brief, F is due, first, to any difference that may exist between the capacitance of the insulator on one wire and that on the other wire, and second, to other unbalances in capacitance such, for example, as those caused by the presence of other wires. The first of these causes will be recognized to be very similar in nature to a second order effect of E and is accordingly small in magni- tude, at least if the same kind of insulator is employed on each wire; so F, due to this particular cause, is normally small. The second source is the more important one. F, resulting from it, is greatest in dry weather, like E. Here, F's importance is so great that transpositions in the insulator test line were found necessary, despite the line being only 200 feet in length. The dry-weather leakage of many of the insulators under test is so small that, without transposi- A STUDY OF TELEPHONE LINE INSULATORS 721 tions, errors caused by the presence of other wires have amounted to several hundred per cent. The variation of F with frequency is much the same as E. Also, like E, F is less in wet than in dry weather. The general remarks made in the discussion of Item E regarding the influence of insulator design and material apply also to F. For the well transposed lines used for carrier circuits and for the reasonably well balanced insulator capacitances that the standard construction gives, /^contributes little to the total wet-weather leakage. Item G — Displacement Currents Flowing Over Insulator Sur- faces THROUGH High Resistance 1. General Characteristics. Over the carrier range of frequencies this item is the most important source of leakage in wet weather. It probably contributes more than all other sources combined. On this account, it may not be amiss to repeat here an already known theory which fits the results of the present study fairly well. WIRE o "^AAAAA- PIN o VWV\A n/WW^ Fig. 21 — Electrical equivalent of an insulator for the production of (G). Consider an insulator, the outside of which is wet. Divide this surface into elements of area and take, for example, one of them near the bottom of the insulator. Assume a small displacement current to flow from the pin to this element through the small capacitance which exists between them. Now for this current to reach the wire, it must flow through the thin film of moisture lying between the element chosen and the wire. This film offers a high resistance to the current, not high enough, however, to seriously limit the current, but, neverthe- less, sufficient to produce heat losses. These losses when integrated over the entire insulator surface become important and qualitatively, at least, account for the characteristics of item G. Fig. 21 shows in much simplified form an electrical equivalent of this action. An inspection of this figure will throw some light on the sub- ject. For one thing, it is clear that the apparent wire-to-pin capacitance will decrease with increasing frequency. Wet-weather tests invariably 722 BELL SYSTEM TECHNICAL JOURNAL show this effect. (Incidentally, this effect tends to reduce the magni- tude of C and D, at the higher frequencies.) It is also apparent that the conductance of this simplified circuit is zero at zero frequency and a maximum at very high frequencies. In the intervening range the conductance at first rises nearly as the square of the frequency, then the relation becomes more nearly linear and finally tapers off to a final constant maximum value. R F ( R WlKtO-AAA( — 'j'— ^W^^ iji — WV 9 ^^ (/) WIREo-^VW— ^J^AA/V — '^^—^N\-^^ R R R a. o > /^' ^ , y / y / •7 .^. '/ /^ r^A f 0 voyf. ? .3 «n y^ ' ' ^n^OHMS - ■ / r-- •"" m y/ / ^* •- 0 .1 < ^ r— /. ^' -«•' < d /; r / UJ _j /, / y ^ ^ / ^ ^ y V 1 0 F RE 2 QU 0 EN CY - ^ -b '\^ ^ ^ .-J f^ / *" .^ 4 ,^ 0 ^^ \ \ \ 10 20 30 40 50 FREQUENCY-KILOCYCLES Fig. 23 — Relative leakage of D. P., C. W. and C. S. insulators as measured at Phoenixville, Pennsylvania, in moderate rain. One of the new types is designed for use on the wood pins of existing lines. It is known as the C. \V. insulator and its design is shown in Fig. 3. The pair of pin thimbles illustrated at the bottom of that figure is first placed over the two wood pins, then the insulators are screwed A STUDY OF TELEPHONE LINE INSULATORS 725 into place over the thimbles, forcing the latter well into the threads of the wood. The thimbles are constructed of thin copper and are bonded together by a tinned copper strip. The other new type is designed to screw directly over a steel pin. This is known as the C. S. insulator and its design is shown in Fig. 12. At each crossarm the two steel pins are bonded by means of a wire underneath the arm. .D / .5 / / to / P / 5 / 3 |.4 / A /' o / a. A /' ^ / / g.3 a. / / O X o /' G /< / ^ / ^ /^' i-^ / ,^ ^ / X y UJ 'i < / ^ ^ ,— - --■ -" / y y" E — • y' ^ ^ "^ ^ ^ ^ ^ " ^ X" "-' D ^ ^ 10^ — __ *-* __ — -" — -\ ■"^ C 0 ~ ^"~ ^^ ^^" ^^" ^"" ^^ b" ^^ 10 20 30 FREQUENCY-KILOCYCLES 40 50 Fig. 24 — Estimated allocation of leakage for D. P. insulator. Both new designs are molded from borosilicate glass. This glass Is more expensive than the alkali glass used in the old D. P. design. On this account, and on account of the pin thimbles in one case and the steel pins in the other, the new insulators cost more to install than did the old ones. Both of the new designs were brought out several years ago before this study had conclusively demonstrated the importance of surface 726 BELL SYSTEM TECHNICAL JOURNAL losses (item G). It will, therefore, be of some interest to discuss their performance in the light of the more complete knowledge. Performance of D. P., C. W., and C. S. Insulators. The leakage of these three types, as measured on the insulator test line at Phoenixville, Pa., in a moderate rain, is given in Fig. 23. This measurement does not give a true picture of the relative efficiency of the three types because no two of them are aged alike. Besides, the relative efficiency varies considerably with different weather conditions. However, the meas- urement will serve our present purpose, which is to analyze the total leakage of each design and thus give a perspective which could not well be brought out in the detailed discussion of the several sources of (O O u Q. 01 O X o cr o_ s' I LJ < < ^ — ■ — ^ "^ X ^' ^ ^ G / ^ C -^ — — — = — B 10 40 50 20 30 FREQUENCY -KILOCYCLES Fig. 25 — Estimated allocation of leakage for C. W. insulator. leakage. It should be pointed out and emphasized that the allocation of the total leakage to its component parts can be only very approx- imate. Fig. 24 shows an estimate of the leakage contributed by the several sources for the D. P. design. The leakage directly through the insula- tor material is negligible and item A, therefore, does not appear. Similarly, the leakage due to unbalanced displacement currents flowing in crossarms, poles, etc., is considered negligible and, therefore, item F does not appear. At a frequency of 30 kc, for example, B, the direct surface leakage or d.-c. leakage, is about 5 per cent of the total. The dielectric absorp- tion in the glass C is about 10 per cent. The dielectric absorption in the wood pins B is about 20 per cent. The crossarm losses E contribute about 10 per cent and, finally, the losses on the insulator surfaces G contribute about 55 per cent. A STUDY OF TELEPHONE LINE INSULATORS 727 Fig. 25 shows a similar estimate for the C, W. design. Here the bonded pin thimbles shield the wood pins from any electric field and thus eliminate dielectric absorption D from the pins. Similarly, by short-circuiting the crossarms, the losses occurring there are eliminated. Accordingly, both items D and E are made negligible and do not appear. Of the remaining factors, the direct surface leakage B contributes about 12 per cent of the total (at 30 kc, for example). The losses in the glass C are liberally estimated at about 5 per cent. Finally, the surface losses G contribute over 80 per cent of the total. (/)■ z u. o a u Q. (/) O I o cr o (J < it: < H 0 , — — "^ ^ — ■ — ^ "^ -"■^ G - ■ ^ — 1 — — — — 3 •B C- ^ — — = 10 20 FREQUENCY- 30 KILOCYCLES 40 50 Fig. 26 — Estimated allocation of leakage for C. S. insulator. Comparing the C. W. performance in this test with that of the D. P., we find that most of the improvement shown by the C. W. has resulted from the elimination of items D and E. Due to the single skirt design of the C. W. and the pin thimbles, item B has been increased in magni- tude. The loss in the glass C has been decreased by the use of borosili- cate glass. However, the most important item G has been only slightly reduced, and if C. W. insulators in this test were aged as far as the D. P. the C. W. might show no improvement with respect to G. The pin thimble construction tends to increase the insulator capacitance and thus tends to make G larger for the C. W. than for the D. P. design. The use of a borosilicate glass with its lower dielectric constant counter- acts this action somewhat. The estimated division of losses for the C. S. design is given in Fig, 26. The use of metal has eliminated any dielectric absorption D from oc- curring in the pins. The bonding of the pins by wire has eliminated crossarm losses E. 728 BELL SYSTEM TECHNICAL JOURNAL Of the remaining factors, the direct surface leakage B contributes about 4 per cent or less of the total at 30 kc. The losses in the glass C are liberally estimated at 10 per cent or less, while the surface losses G contribute about 85 per cent or more. In this design the absolute magnitude of B has been decreased some- what, chiefly because the small diameter of the steel pin permits a small diameter of insulator, the advantages of which were pointed out in the detailed discussion of this item. The low capacitance made possible by the small steel pin has helped to make both C and G relatively small, although most of the improvement in C is due to the borosilicate glass. The improvement in the surface losses G over the D. P. design is quite marked. For the new designs the two factors B and G are the controlling ones. In this particular test B happens to be quite small in magnitude, and would naturally lead one to conclude that B had been made unneces- sarily small at the expense of G, especially since these two items in many respects place conflicting requirements on insulator design. However, B has been observed at times to reach a value as high as one third of the total leakage at 50 kc. The necessity of engineering for such cases makes the design more reasonable, especially when it is recalled that the insulators must serve for direct current and low frequencies, as well as for the carrier range. Of the new designs, the electrical superiority of the C. S. over the C. W. design is apparent. This fact, together with economic consider- ations, has led to the almost universal choice of the C. S. rather than the C. \V. type for the field of application of the new insulators in the telephone plant. The utility of the C. S. insulators in the telephone plant will be more clearly apparent from a consideration of the reduction in attenuation which their use brings about. The losses in transmission over a pair of wires at carrier frequencies come chiefly from two sources: one, substantially fixed in magnitude, depending mainly on the resistance of the wires ; the other, quite vari- able in magnitude, depending on the leakage conductance between the wires and, therefore, on the weather.^ In the case of 165-mil copper wires on 12-inch spacing these two components of loss are approximately equal in wet weather at 30 kilo- cycles when the older type of insulators are used. The C. S. type cuts ^For a more detailed discussion of attenuation see a companion paper, "The Transmission Characteristics of Open-Wire Telephone Lines," by E. I. Green, presented at the Summer Convention of the A. I. E. E., Toronto, June, 1930 and printed in this issue of the Bell System Technical Journal. A STUDY OF TELEPHONE LINE INSULATORS 729 this variable component about in half at that frequency thus reducing the total wet-weather attenuation of these wires to about 75 per cent of its former value. For smaller sizes of wires the percentage reduction is correspondingly less. The benefits of the lesser attenuation can be utilized in the plant in various ways, depending on local conditions ; for example, in increasing repeater spacing, in employing smaller gages of wires, or in increasing the number of insulators per mile to provide for better transposition designs. In addition, the new insulators, in having reduced the variable com- ponent of loss, improve the stability of carrier circuits to a marked degree. Acknowledgment Only the electrical features of the new designs have been discussed. Closely related to these are the many mechanical problems which naturally arise in new construction. These latter problems, during the development of the new designs, came under the supervision of Mr. C. S. Gordon, assisted by Mr. J. T. Lowe. The Corning Glass Works has cooperated in molding special experi- mental insulators of various compositions. Data on the electrical properties of numerous glass compositions have been supplied by the Bell Telephone Laboratories. The writer desires to express his thanks to Messrs. F. A. Leibe, L. R. Montfort and L. Staehler for assistance in the measurements and to Mr. H. R. Nein for assistance in the preparation of this paper. The Transmission Characteristics of Open- Wire Telephone Lines ^ By E. I. GREEN Values of the primary transmission constants R, L, G, and C for open-wire telephone lines are presented, and the factors which affect these constants in practise are discussed. Consideration is then given to the constants which are of principal interest in telephone work, namely, the attenuation, the characteristic impedance, the phase constant, and the velocity of propaga- tion. Data regarding these characteristics are given for the frequency range from 0 to 50,000 cycles. NEARLY 3,000,000 miles of open wire are now furnishing toll service in the Bell System, and this total is increasing at a rate of more than 100,000 miles a year. Hence, the subject of the transmission characteristics of open-wire circuits, in addition to being of considerable natural interest, is of no little importance in many branches of tele- phone work. In the design of apparatus to be associated with the open-wire circuits as w^ell as in the engineering and maintenance of the facilities derived from them, a knowledge of these transmission charac- teristics is indispensable. The problem of determining the characteristics of the open-wire circuits has, of course, been coexistent with the circuits them- selves, and hence dates back to the beginnings of telephony. Of late years, however, there has been a very decided change in the nature and scope of the problem. This has resulted from many factors, particularly {a) the extensive application of carrier telephone and telegraph systems- and {b) the constantly increasing length of the long distance circuits. The first of these factors has extended the trans- mission range upward from about 3000 cycles to about 30,000 cycles, and may well extend it higher in the future. The second, in combina- tion with the higher standards which are now applied in long distance transmission, has required greater accuracy in the data, emphasizing especially the importance of time and space variations in the charac- teristics. Also, recent changes in the construction of open-wire lines (to be described later) have necessitated substantial additions to the data. 1 Presented at the Summer Convention of the A. I. E. E., Toronto, Ont., Canada, June 23-27, 1930. 2 See "Carrier Systems on Long Distance Telephone Lines," by H. A. Affel, C. S. Demarest, and C. W. Green, A. I.E. E. Trans., Vol. 47, 1928, pp. 1360-1386. Bell System Tech. Jl., July 1928, pp. 564-629. 730 OPEN-WIRE TELEPHONE LINES 731 There will be studied in this paper those inherent characteristics of open-wire lines which are used most frequently in telephone transmis- sion work. These characteristics are: first, the attenuation, second, the impedance, and third, the phase characteristic, with which must be coupled its near relative, the velocity of propagation. The range of frequencies to be covered is fixed on the one hand by the d.-c. telegraph systems, as well as by the program transmission circuits, whose lower frequencies extend to 100 cycles or less, and on the other hand by the carrier telephone and telegraph systems, which make the range up to about 50,000 cycles of interest. Line Construction Arrangements In order to study the characteristics of open-wire lines it is necessary to know something of the constructional arrangements which are em- ployed. The conductors most commonly used for the open-wire tele- •10-0 12 — ---12 — 5-5- ^, ^2 ^3 ^^ ^^( ^ ^6 -r^ B^ .^ ,0^ Fig. 1 — Configuration of an open-wire line with 12-in. non-pole pairs. phone lines of the Bell System are of 165-mil (No. 8 B. W. G.), 128-mil (No. 10 N. B. S. G.), and 104-mil (No. 12 N. B. S. G.) hard-drawn cop- per. These are the conductors usually employed for carrier systems. Other gages of copper, as well as a small amount of iron or steel wire, are used to some extent for voice-frequency and d.-c. telegraph trans- mission only. The wires of the lead (as an open-wire line is frequently designated in telephone parlance) are strung on poles, the normal spacing and num- bering of wires being generally as shown in Fig. 1. Starting at the left end of the crossarm, the adjacent horizontal wires are grouped in pairs, 732 BELL SYSTEM TECHNICAL JOURNAL wires 1 and 2 comprising one pair, 3 and 4 another, etc. The character- istics of the pairs are of primary interest. Phantom circuits, which are derived from two pairs or side circuits, will be discussed later. The two wires of each pole pair (that is, a pair which bestrides the pole) are about 18 in. apart, and those of each non-pole pair 12 in. apart. There has recently come into vogue a different arrangement of wires which is designed to reduce the coupling between circuits and thus per- mit a maximum use of carrier facilities. This arrangement is por- trayed in Fig. 2. In this newer configuration the separation between the wires of each non-pole pair is reduced to 8 in., and the horizontal separation between pairs is widened to 16 in. Fig. 2 — Configuration of an open-wire line with 8-in. non-pole pairs. The ordinary spacing between poles on open-wire toll lines is 132 ft., corresponding to a total of 40 poles per mile. \\'here additional strength is required, the number of poles per mile may be as high as 50, while outside of the heavy sleet area it may be as low as 30. The types of insulators employed on open-wire lines will be dis- cussed under the heading of leakage conductance. Two methods of transposing the wires are in current use. In the older of these, which is illustrated in Fig. 3, the wires are brought at the transposition pole to a "drop bracket." The transposition is accom- plished over a total distance of two spans by gradually rotating the plane of the wires through 180 deg. It will be seen that the wire con- figuration is abnormal throughout the two spans. In the newer method, the wires are crossed practically at a point. This may be done by means of two brackets known as "break irons," as illustrated in Fig. 4, or by means of a single bracket. The "point" transpositions OPEN-WIRE TELEPHONE LINES 733 preserve the nominal spacing between wires and thus avoid the ir- regularities in spacing which occur when drop bracket or "rolHng" Fig. 3 — Transposition of wires with drop bracket. transpositions are used. With the point transpositions, however, two pairs of insulators are required at each transposition point as corn- Fig. 4 — "Point" transposition on break irons. pared with a single pair of insulators at a non-transposition point. This results in an increase in the total number of insulators and in 734 BELL SYSTEM TECHNICAL JOURNAL variations in the number of insulators on different pairs because of the different number of transpositions employed. Primary Constants It should be noted here and now that the phenomena of line trans- mission are the same throughout all of the frequency range under con- sideration. Transmission over wires at high frequencies is accom- plished in precisely the same manner as transmission at low frequencies, the wires acting as the guiding medium for the energy in both cases, and the same theory may be applied to both. A review of the well-known theory for the propagation of alternating currents over wires will show that the line characteristics in which we are interested are dependent upon the four quantities known as the primary constants of the circuit. These are as follows: R = Series resistance in ohms per mile. L = Series inductance in henries per mile. C = Shunt capacitance in farads per mile. G = Shunt leakage conductance in mhos per mile. These quantities may be stated per mile of wire or per mile of circuit. In this paper all values will be per mile of circuit, or, as it is commonly expressed, per loop mile. Unfortunately the constants R, L, G, and C are by no means con- stant in practise. Indeed there could scarcely be a more fickle set of quantities. They are subject to change by a great variety of factors, of which the most important is, of course, the frequency. Hence it is evident that in order to determine the practical values of the attenua- tion, impedance, and velocity for open-wire circuits, we shall have to examine the behavior of the primary constants, R, L, G, and C. Resistance First in the list of primary constants is generally named the con- ductor resistance. The method of computing the d.-c. resistance is well known and requires no explanation here. In such computations it is assumed that the current density is uniform throughout the cross- section of the conductor. \Mth alternating current, however, the familiar phenomenon of skin effect tends to produce a non-uniform current distribution, and hence to increase the resistance. If the two wires of a circuit are close together, the effective a.-c. resistance of each wire is likewise increased by the presence of the parallel conductor, due to what is known as proximity effect. In cable conductors, OPEN-WIRE TELEPHONE LINES 735 especially when used for carrier frequencies, proximity effect is very important, but it is negligible in open-wire circuits because of the large separation between wires. The method of determining the skin effect resistance of round wires is presented in various publications, and the theoretical results have been experimentally confirmed on numerous occasions.^ Values of the 25 UJ 20 Q. O o _) DC u 15 Q. (/) I O 10 o z < t- (/) UJ ct ^^ y y y . y y y y .• y ^ ^\^1 y ^ ^^ y y .n >1 y ^ A /^ ^ y / y y ,/ / ,1? ^f' > X' ^ y r \ ^ r ,^ / r- /^ y ^ ^ ^ / ^ f __ • ^ ^ y y /' \< => ^ ^ ^ r y ^ 7 ^ ■^' / y y y \/- / y -^ / y^ / ,x y 10 40 50 20 30 FREQUENCY- KILOCYCLES Fig. 5 — A.-C. resistance of open- wire pairs at 20 deg. cent. (68 deg. fahr.). a.-c. resistance of 165, 128, and 104-mil copper pairs at 20 deg. cent., (68 deg. fahr.) determined in accordance with skin effect theory, are plotted in Fig. 5. It will be noted that the increase in resistance due to skin effect is small in the voice range, but rather astoundingly large in the carrier range, amounting at the higher frequencies to from 200 to nearly 400 per cent. Experimental evaluations of open-wire resistance are in extremely 'See "Wave Propagation Over Parallel Wires — The Proximity Effect," J. R. Carson, Phil. Mag., Vol. 41, April 1921, pp. 607-633; "Experimental Researches on Skin Effect in Conductors," A. E. Kennelly, F. A. Liws, and P. H. Pierce, ^. I. E. E. Trans., Vol. 34, Part 2, 1915, pp. 1953-2021, and "Skin Effect Resistance Measure- ments of Conductors at Radio Frequencies," A. E. Kennelly and H. A. Affel, /. R. E. Proc, Vol. 4, No. 6, Dec. 1916, pp. 523-574. 736 BELL SYSTEM TECHNICAL JOURNAL close agreement with the values given in Fig. 5.* For old wires, however, the resistance may be somewhat higher than these values. This is because of the presence of contact resistance in the twisted sleeve joints in the wires and also perhaps because of an actual decrease in the conductor diameter occasioned by corrosion. The increase in the d.-c. resistance of old wires due to these causes may be as much as 5 per cent. The corresponding percentage of increase in the a.-c. resis- tance will, of course, be much smaller. The d.-c. resistance of a copper wire varies with temperature according to the familiar formula: Ro = Roi [1 + ai (^ - k)], (1) where Ro and Roi represent the d.-c. resistance at temperatures / deg. cent, and h deg. cent, respectively, and ai is the d.-c. temperature coefficient of resistance of copper at /i deg. cent. At a temperature of 20 deg. cent, the value of ai is generally taken as 0.00393. Similarly, the a.-c. resistance R of a copper wire at a temperature / may be represented as follows: R = RAi +^i(/-/i)], (2) where Ri = a.-c. resistance at temperature /i deg. cent., . 1 dR ,-,, . p . p Ai = ^ -jT = a.-c. temperature coerhcient oi resistance ol copper at Ki at ti deg. cent. Now the skin effect resistance ratio depends upon the magnitude of the d.-c. resistance, being smaller the larger the resistance. Hence, a given change in temperature which changes the d.-c. resistance pro- duces a change in the opposite direction in the skin effect resistance ratio, so that the percentage change in the a.-c. resistance is less than the percentage change in the d.-c. resistance. In other words, Ai is less than a\. As illustrated in Fig. 6, the a.-c. temperature coefficient of resistance for open-wire pairs, starting at the d.-c. value ai, straight- way decreases as the frequency is increased, and at high frequencies approaches a value of— . An explanation for this asymptotic value is presented in Appendix I. The temperature assumed by the conductors of open-wire lines de- pends of course, upon the weather conditions which prevail in different sections of the country. In order to obtain information on this subject, * The value of R, and also that of the other primary constants, may be determined directly from open and short-circuit impedance measurements on a line short enough to avoid propagation effects. A longer line may be used instead, in which case it is necessary to correct for such effects. OPEN-WIRE TELEPHONE LINES 737 a study has been made of the Weather Bureau records for a number of representative cities in various parts of the country. The chief in- terest naturally centers in the extreme temperatures reached by the wires. It appears that on the average the air temperature will not drop below about — 20 deg. cent. (—4 deg. fahr.) on more than about 10 days per year in the colder sections of the country, while a limiting temperature of about 35 deg. cent. (95 deg. fahr.) will not be exceeded on more than about 10 days per year in the warmer sections of the country. Because of imperfect radiation, the temperature of a wire in 0.005 UJ a < a. o Z 0.004 iiJ o LJ LJ a. o u '^ 0.003 z LJ ^ 0.002 UJ O o 0.001 r ^ - ^ ^^ ::-. \ \ V \ k 7 k '< ^ p *;, •V. ^ 100 500 1000 5000 10,000 FREQUENCY — CYCLES PER SECOND 50,000 100,000 Fig. 6 — A.-C. temperature coefficient or resistance for open-wire pairs at 20 deg. cent. the sun will ordinarily exceed the temperature of the surrounding air by a small amount. A few tests have indicated that for open wires the increase over the air temperature on a warm day is not more than 5 deg. cent. Temperatures of — 20 deg. cent. (— 4 deg. fahr.) and 40 deg. cent. (104 deg. fahr.), therefore, appear to be representative values for the limiting temperatures assumed by open-wire lines. Reference to Equation (1) shows that this range of temperature gives possible variations in the d.-c. resistance of 16 per cent below and 8 per cent above the value for 20 deg. cent. The total annual change in resistance at any one place will, of course, be less than the sum of the above changes. In the Middle West, how- ever, where the weather variations are much greater than in other parts of the country, the total annual change in d.-c. resistance may be as 738 BELL SYSTEM TECHNICAL JOURNAL much as 20 per cent. This section of the country has also the great- est diurnal range of temperature, giving a d.-c. resistance variation of as much as 8 per cent. Inductance The inductance of a circuit formed of two parallel wires whose dis- tance between centers is negligible compared with their length is r 2D L = 0.64374 2.3026 logic ^ + a^S X 10~^ henrys per loop mile, (3) where the diameter of each wire d, and the distance between their centers D, are expressed in the same units, where ix is the permeability, and 5 is a factor depending on the frequency. The tendency of alternating currents to concentrate on the surface of a wire evidently reduces the magnetic flux within the wire and de- creases the internal inductance of the wire. In Equation (3) the in- ternal inductance is represented by the factor nb. At low frequencies, for which the current is uniformly distributed across the cross-section of the wire, the value of b is 0.25. For very high frequencies there is practically no magnetic flux within the wire, and the value of b is zero. Between these frequency limits the value of b is determined with the aid of skin effect formulas or tables.^ For the wire diameters and spac- ings employed on open-wire lines the change in the total inductance due to skin effect is relatively small. It is assumed in Equation (3) that the two wires are suspended in space or at a considerable distance from the ground and from other wires. In practise, the presence of other wires probably has some effect on the inductance, but for well transposed lines this effect is negligibly small. The inductance at different frequencies of 165, 128, and 104-mil copper pairs having various spacings between wires is shown in the following table. As will be seen from Equation (3), the inductance varies with the logarithm of the separation between wires. In the table the values of inductance are shown for pole pairs, which have a separation between wire centers of about 18.25 inches, and for non-pole pairs having wire separations of 12 and 8 inches. The values of inductance given in the table have been closely checked by measurements on open-wire pairs. ^ See, for example, Circular No. 74 of the Bureau of Standards, "Radio Instru- ments and Measurements." OPEN-WIRE TELEPHONE LINES 739 Inductance of open-wire pairs — henrys per loop mile Frequency- cycles per second 165-mil 128-mil 104-mil 8-in. 12-in. 18.25-in. 8-in. 12-in. 18.25-in. 8-in. 12-in. 18.25-in. 0 1,000 10,000 25,000 50,000 Infinite 0.00311 0.00311 0.00305 0.00301 0.00299 0.00295 0.00337 0.00337 0.00331 0.00327 0.00325 0.00321 0.00364 0.00364 0.00358 0.00354 0.00352 0.00348 0.00327 0.00327 0.00323 0.00319 0.00317 0.00311 0.00353 0.00353 0.00349 0.00345 0.00343 0.00337 0.00380 0.00380 0.00376 0.00372 0.00370 0.00364 0.00340 0.00340 0.00338 0.00334 0.00331 0.00324 0.00366 0.00366 0.00364 0.00360 0.00357 0.00350 0.00393 0.00393 0.00391 0.00387 0.00384 0.00377 Capacitance The capacitance of two parallel wires in space with a distance be- tween centers which is negHgible compared with their length is C = 0.019415 logio^ X 10~^ farads per loop mile. (4) It will be noted that the capacitance varies in inverse relation to the separation between wires. As in Equation (3), it is assumed in this formula that the two wires are suspended in space or at a considerable distance from the ground and from other wires. On an actual line the capacitance of a pair is changed to an appreciable extent by the presence of other wires, and to a slight extent by the capacitance to ground. The true capacitance between the two wires under actual conditions may be derived from the direct capacitances between all wires and the direct capacitances of all wires to ground.^ The capacitance is not changed to any great ex- tent by skin effect. The means of insulation and support provided at each pole have an appreciable effect on the capacitance of a pair of wires, especially in wet weather. This is due to the fact that the insulators and, under certain conditions, the pins and parts of the crossarms, act as the dielec- tric of small condensers which are, in effect, shunted between the line wires. These effects are being discussed in a companion paper.'^ The percentage increase in capacitance due to the insulators varies with different weather conditions and different types of insulators, ranging * See Technical Report No. 54 of the Railroad Commission of the State of Cali- fornia, Joint Committee on Inductive Interference, entitled "Inductive Interference Between Electric Power and Communication Circuits," 1919. ''See "A Study of Telephone Line Insulators," by L. T. Wilson, printed in this issue of the Bell System Technical Journal. 740 BELL SYSTEM TECHNICAL JOURNAL from about 0.5 to 4 per cent of the capacitance between line wires. Values of the capacitance of 165-, 128-, and 104-mil pairs in space and on a 40-wire line are given in the following table: Capacitance — microfarads per mile Wire spacing 16S-mil 128-mil 104-mil In space On 40- wire line In space On 40- wire line In space On 40- wire line 8 in. 12 in. 18.25 in. 0.00977 0.00898 0.00828 0.00996 0.00915 0.00863 0.00926 0.00855 0.00791 0.00944 0.00871 0.00825 0.00888 0.00822 0.00763 0.00905 0.00837 0.00797 As before, values are given for pairs having wire separations of 8, 12, and 18.25 in. These values include an allowance for the dry weather capacitance of the insulators. The difference between the values in space and on a 40-wire line indicates the importance of the effect of the other wures, the insulators, etc., upon the capacitance. The capaci- tance values given in the table are fairly representative of the values that will obtain on well transposed lines. The values of inductance and capacitance which have been given are based on the assumption that the nominal separation between wires is preserved throughout the entire line. This is not the case when drop bracket transpositions are employed. As has been pointed out, the wires are brought closer together at the drop brackets, thereby in- creasing the capacitance and decreasing the inductance. The amount of the change in inductance and capacitance due to this cause ranges from 1 to 5 per cent for the transposition arrangements designed for carrier system operation. Leakage Conductance The leakage conductance per unit length of circuit, which is repre- sented in the transmission formulas by the symbol G, is by far the most erratic of the primary constants. Since it is a momentous factor in the attenuation its investigation has been very actively prosecuted over a considerable period of time. The determination of the value of G for direct current is quite simple, involving merely a measurement of the actual conductance between wires for a length of circuit short enough to avoid propagation effects. For alternating currents, however, it is customary to employ an equivalent value of G which includes all of the losses suffered by the OPEN-WIRE TELEPHONE LINES 741 power transmitted over the pair except the normal PR loss in the wires themselves. This inclusion of numerous little-understood losses in the general term leakage has at times served to insulate the individual losses from analysis. Methods of determining the value of the "equivalent leakage conductance" and of analyzing its component losses are avail- able, however.^ The nature and magnitude of the different losses which occur at the insulators are being discussed in detail in a parallel paper. ^ Accord- ingly, only a brief mention will be made in this paper of the types of insulators which are now in use on the open-wire lines of the Bell Sys- tem, and of the values of leakage conductance experienced with these different types. The DP or double-petticoat glass insulator illustrated in Fig. 7 is now standard for use on all important toll circuits, except those equipped with the special carrier insulators discussed below. On a number of older circuits single-petticoat glass insulators, known as toll insulators (see Fig. 7) and double-petticoat porcelain insulators are still in place. In view of the numerous and complex sources of leakage loss, it is not surprizing that the leakage conductance for a given pair at a particular frequency varies with changing weather conditions and with the age of the insulators over a very wide range of values. Because of this wide range of variation it is possible to give here only selected leakage values which serve for engineering purposes. Values of the total leakage conductance for open-wire pairs equipped with DP insulators are plotted in Fig. 8. These values are intended to represent the highest values ordinarily obtained on an old circuit which is in a good condition of maintenance. The wet weather values have been so chosen that they should be exceeded on only a few days of the year, while the dry weather values represent the performance that should be expected from any circuit on a clear, dry day. Particular difficulty is experienced in selecting standard values of d.-c. leakage owing to the fact that the range of values encountered in prac- tise is exceedingly great. The measured values depend to a great ex- tent on the degree to which the line is kept free from tree branches, foliage, moss, broken insulators, and other possible sources of leakage. These special sources of loss, of course, represent a much smaller part of the total leakage losses at carrier frequencies. The standard values of leakage conductance are derived for a line having 40 pairs of insulators per mile. Where the number of insulators * See quencies "Methods of Measuring the Insulation of Telephone Lines at High Fre- ," by E. I. Green, A. I. E. E. Trans., Vol. 46, 1927, pp. 514-519. 742 BELL SYSTEM TECHNICAL JOURNAL differs greatly from this figure, it is necessary to correct the leakage values accordingly. Differences in the number of insulators per mile result from the use of different types and numbers of transpositions, different pole spans, double crossarm construction, etc. Fig. 7 — Types of insulators employed in the Bell System. a. Toll insulator. b. DP insulator. c. CW insulator and pin thimble. d. CS insulator. OPEN-WIRE TELEPHONE LINES 743 Considerable study has been given to methods of reducing the leak- age conductance, particularly at carrier frequencies, and two new types of insulators have been developed for this purpose. In these there is used an improved dielectric (borosilicate glass) which has a low power factor and a reasonably low dielectric constant, as well as good chemical stability. Two expedients for eliminating losses in the pins and cross- arm are employed. In the CW insulators, illustrated in Fig. 7, two 50 Q- O o a. u a. i X - i /\ -VAt^- _^ ■^ / NJJ^ -r f^ / /^' 1 .c — — - - "^ ^ ^ ^ ^ m •-' cw ^'^^^^ - T WFAT H£?- . — — / ^> ^^ , *■• ^ ... — LCS^Nt-pT. i-"— •"" mC S*" -^-9 ■s.— " u^ -- ' -" ■* 10 20 30 FREQUENCY - KILOCYCLES 40 50 Fig. 8 — Leakage conductance of open-wire pairs equipped with different types of insulators. metal shells or thimbles are bonded together and placed over the wooden pins. In the CS insulator, also shown in Fig. 7, steel pins are employed and the two steel pins of the pair are bonded together. In the past few years CS insulators have been applied to the open-wire lines of the Bell System in increasingly large numbers. With these two types of insulators a substantial reduction in the total leakage conductance is brought about. The best available figures for the limiting values of leakage conductance for these types are shown in Fig. 8. A further advantage obtained through the applica- tion of the new insulators is that of stabilizing the attenuation at carrier 744 BELL SYSTEM TECHNICAL JOURNAL frequencies. Experience has indicated that the CS and CIF insulators reduce the daily leakage (and attenuation) variations due to change in weather conditions to about one-third and one-half, respectively, of their value for DP insulators. This degree of stabilization is not indi- cated by the differences between the dry and wet weather leakage values shown in the figure, but it must be recalled that these values represent extreme conditions, while the stabilization referred to above is for average conditions. Attenuation The attenuation constant is the real part a of the propagation con- stant 7 as given in the familiar formula 7 = a +i^ = V(i? +7Xco)(G + jCco). (5) The attenuation constant is also given by the following expression a^ = 1/2 [V(i?' + L2co2)(G2 + CV) - (LCco^ - RG)]. (6) Where L-co^ is large compared to I^ and CW is large compared to 6^, it can be shown ^ that Equation (6) reduces to "^iVz + lVc- ^^^ This formula is very useful for computing the attenuation of open-wire circuits at carrier frequencies, in which case its accuracy is adequate for all practical purposes. It is frequently of value also for quick computations of the approximate attenuation of open-wire circuits in the voice range. The first term of Equation (7) represents the series losses, and is commonly referred to as the "resistance component of attenuation," while the second term represents the shunt losses, and is called the "leakage component of attenuation." It will be observed that the resistance component of attenuation varies inversely with the quantity '-^ , while the leakage component varies directly with the same quan- tity. This quantity -i /-^ , as will be seen later, represents the nominal characteristic impedance of the circuit. It is shown in Appendix II that a circuit of fixed resistance and leak- age conductance will have minimum attenuation when the ratio of L to ^ See "Transmission Circuits for Telephonic Communication," by K. S. Johnson, N. Y., Van Nostrand, 1927. OP EN- WIRE TELEPHONE LINES 745 C is such as to make the resistance component equal to the leakage component. Because of the variation of the resistance and leakage conductance with frequency, the ratio of L to C which gives minimum attenuation evidently depends upon the frequency. At voice fre- quencies the resistance component for open-wire circuits is, as a rule, considerably larger than the leakage component, so that it is generally possible to reduce the voice-frequency attenuation by inserting load- ing coils, which increase the value of L and thus reduce the resistance component at the expense of an increase in the leakage component. The amount of reduction in attenuation obtainable by loading is .100 _j ££.075 UJ a m Q o I- < Z UJ I- .050 .025 _^^ ' „Cf>T\4t_^_.. ■— "'^ ^,,,^vJtVi^TT \0^J^2iii — -^ ^■^ ^^ DpyWEATHER. [Z^ ' / ^'■■' 1 TLJiNN^^i^^^ '*'.--■ \ i 'ppy VVElATHtK__ ^i^«"^ // L . . . .'^-^'--r" T^^^^^M^"^ " / / ,J — fl ^ \MFATHER >"" \65MlL.-:;DaVVV T^'- 7 0 12 3 4 5 FREQUENCY-MLOCYCLES Fig. 9 — Voice-frequency attenuation of open-wire pairs equipped with DP insulators. evidently limited by the value of the leakage component and by the additional resistance which is contributed by the loading coils. For open-wire circuits at carrier frequencies the value of the leakage component of attenuation is quite large in comparison with the resist- ance component, and coil loading would, in general, be detrimental. At the present time the use of loading on the open-wire circuits of the Bell System has been practically abandoned. Owing to the importance of other factors, especially the line crosstalk, it is ordinarily impracti- cable to design the open-wire circuits to secure precisely the minimum attenuation at the highest working frequency. In the carrier fre- quency range, however, the wet weather attenuation of the pairs most commonly used is not materially higher than the theoretical minimum. 746 BELL SYSTEM TECHNICAL JOURNAL Values of the attenuation constant of open-wire pairs of different gages when equipped with DP insulators are presented in Figs. 9 and 10. The values are plotted in db per mile.^" The attenuation values of Figs. 9 and 10 have been determined for a temperature of 20 deg. cent. (68 deg. fahr.) and for the dry and wet .300 .250 _i i a. UJ Q. (B Q z o I- < Z .200 .150 .100 .050 A ,/ /^ / / y .<< ,«^ ^^ ^ A .^ / v^^ ^ y X < ^'T ^ ^z ^ ^A rl --^ <^ ,y ^ pi 'A 3^r ^ X .'" ^' .^ -^^>\ /^ ^- <• r'^ ^ / / .,4 ^:ii^^;..V^ ' ,, / / J >\o >^' s,- l^" ,.' ^^ / / y y ..^.^^^ ^' r^ / / ^ ?«> .9-^ ^^-- ^ / / .<* ^ ,y V f" r' / / f/ y ^' \(S ^l / /. ;/ r y ' ^ /' V * ^^ ^^ v* > /. V ^ ^ ' A V X r *• /" 10 20 30 FREQUENCY-KILOCYCLES 40 50 Fig. 10 — Carrier frequency attenuation of open-wire pairs equipped with DP insulators. weather values of leakage conductance previously presented. It will be recalled that these values of leakage are derived on the basis of 40 pairs of insulators per mile, and are intended to represent, not average values, but the highest values ordinarily obtained under conditions of dry and wet weather. Systems are ordinarily engineered on the basis of the extreme wet weather attenuation values. When a line runs through the more arid parts of the country, however, advantage is i^See "Decibel— The N'ame for the Transmission Unit," by \V. H. Martin, Bell System Tech. Jl., January 1929, pp. 1-2. i OPEN-WIRE TELEPHONE LINES 747 often taken of this fact by making the repeater spans longer than normal. A comparison of the attenuation values for a 165-miI open-wire pair when equipped with different types of insulators is presented in Fig. 11. For purposes of comparison with the normal values a curve of the attenuation due to resistance only, representing the ideal condi- tion of zero leakage conductance, also is given in Fig. 11. 10 20 30 FREQUENCY- KILOCYCLES 40 50 Fig. 11 — Attenuation of 165-mil open-wire pair for various conditions of insulation. The values of line capacitance employed in determining the attenua- tion of values of Figs. 9, 10, and 11 include an allowance for the average capacitance increase due to the insulators. The attenuation curves shown are strictly applicable to pairs having a wire separation of 12 in., but they are approximately correct for spacings of 8 and 18.25 in. When the number of pairs of insulators per mile differs greatly from the standard value of 40, a correction is applied to the attenuation val- ues. Special curves make it possible to obtain this correction conveni- ently. Curves are also available for correcting the standard attenua- tion values to take care of changes in temperature. 748 BELL SYSTEM TECHNICAL JOURNAL It should be understood that the attenuation of an open-wire pair varies from time to time over a wide range of values, and therefore it is not to be expected that the values of attenuation measured at any par- ticular time will necessarily coincide with the theoretical values. It should also be understood that the attenuation measured on an actual pair never bears the perfectly smooth relation to frequency which is shown on the standard attenuation curves, but exhibits irregularities varying in magnitude according to the irregularities existing on the line. Thus the curve of attenuation as measured on a very well transposed open-wire pair," which is delineated in Fig. 12, represents about as .20 UJ LlJ CL m Q I z o < D z u I- t- < .15 .10 05 10 20 30 FREQUENCY -KILOCYCLES 40 50 Fig. 12 — Attenuation measured on a well-transposed 128-mil open-wire pair with 8-inch spacing. smooth an attenuation curve as it is possible to obtain on an open- wire circuit. The attenuation values shown on this curve are somewhat lower than the standard values for similar pairs. This is doubtless explained by the fact that the insulators on this particular pair were new, and the further fact that the measurements were made in winter when the temperature was low. An illustration of the significance of the attenuation data given above may be of interest. One of the longest carrier telephone systems now in service extends from Davenport, Iowa to Sacramento, California, a " Methods of measuring the attenuation, impedance, and crosstalk are discussed in "High-Frequency Measurements of Communication Lines," by H. A. Affel and J. T. O'Leary, ^. I. E. E. Trans., Vol. 46, 1927, pp. 504-513. OPEN-WIRE TELEPHONE LINES 749 total distance of about 2100 miles. The highest frequency employed in this system is approximately 28,000 cycles. Using the attenuation values for a 165-mil pair at 28,000 cycles, it appears that the dry weather attenuation for the entire length of this system might be approximately 220 db or less, and the wet weather attenuation about 330 db. This means that without amplification along the line the ratio of the transmitted power to the received power might vary from 10^^ to 10^^. Since the attenuation of a repeater section is ordinarily limited to from 25 to 40 db, ten repeaters are employed to span the total distance, and in order to compensate for the attenuation variations a gain regu- lating mechanism known as a pilot channel must be used. 20 z o < z u I- H < 15 10 i rs > ^ y \ / \ / \ / > s. ^^^ ^ ^ ■— . aX y — > a D O _l u DZ is >■ Q 3 u > Q D o _l o z < a. z < 6 12 6 12 6 12 6 12 6 RM. AJVI. P.M. A.M. P.M 12 Fig. 13 — Measured variations in the attenuation of an open-wire pair. 12 In the preceding illustration it was assumed that the range of varia- tion in attenuation increases in direct proportion to the length of the circuit. Although this may theoretically be possible, it has been found in practise that the attenuation variations during any given period of time increase less rapidly than the circuit length. The reason for this is that augmenting the length of the circuit obviously reduces the likeli- hood of experiencing extreme wet weather conditions simultaneously over the entire line. A practical example of how the open-wire atten- uation varies from time to time is afforded by Fig. 13, which shows the measured attenuation changes on a line 110 miles long during the period of two light rainstorms. 750 BELL SYSTEM TECHNICAL JOURNAL One further point is of interest in connection with the subject ot open-wire attenuation. Inductive or conductive coupling between a pair and the other circuits on the line may result in the absorption of energy in these circuits. Fortunately, the losses due to this cause are small on well transposed lines. On inadequately transposed lines, however, this interaction with other circuits, in addition to producing small losses over a wide range of frequencies, may cause incredibly large losses over a narrow band of frequencies, producing what is known as an "absorption peak" in the attenuation curve. This interesting CO Q CO V) O uw 1 — I 1 50 1 2 \ \ / r'/ \ \ ■2 < y l' 40 7 J / r / ^A J / ^i / ,* ^ ^, ^ 0^ — ^ " 30 ^1 ^ A mi / ^ 4 ^H '^ G / 'A o rT- / •», f ^ c,? 1 / S^ ^ 20 > J V t^ h ./. :^ ^- V- ^ ^ y r 10 n J 10 15 20 FREQUENCY-KILOCYCLES 25 30 Fig. 14 — Absorption peaks on an open-wire pair. phenomenon is illustrated in the attenuation curves of Fig. 14, which show how two very pronounced absorption peaks on a line about 300 miles in length were smoothed out by the application of improved transpositions. The magnitude of one of these absorption peaks will be appreciated when it is realized that the received power at the peak frequency is about one two-hundredth of that at the adjacent frequen- cies. OPEN-WIRE TELEPHONE LINES 751 Impedance : The characteristic impedance is defined by the well-known formula: Zo = R + jLco G + jG ohms. CjO (8) It is doubtless unnecessary to explain why this impedance must be matched in the apparatus. When R is small compared to Leo and G is small compared to Coj, the value of Zo evidently becomes Zo = (9) This is known as the nominal characteristic impedance. It will be noted that this impedance is a pure resistance. In the carrier range the actual impedance of an open-wire pair is substantially equal to the nominal characteristic impedance. 750 500 tn ^ 250 X o -250 \. "' 1 -— ... .^ ^_ ^_ tj . 1 ^ — - — — • - — — -- ■ — — — -- - -- -- •- — -■ — -- t RESISTANCE L -165 MIL, ■128 MIL, -I04MIL, -A ^" ■S' k= ■SL.'. tA w US. IW "* rw "' ^ "■ '" M f L -104 MIL, ^ACTANCE 1 •- -128 MIL, -165 MIL. 10 20 30 40 FREQUENCY — KILOCYCLES 50 Fig. 15 — Carrier-frequency impedance of 12-in. open-wire pairs. Values of the characteristic impedance in dry weather of open-wire pairs with 12-inch wire spacing are presented in Fig. 15. These values have been derived from the standard values of inductance, capaci- tance, and leakage conductance and from resistance values at 20 deg. cent. The basis for the impedance value of 600 ohms resistance, 752 BELL SYSTEM TECHNICAL JOURNAL which has become almost a tradition in so many phases of telephone work, will be obvious from Fig. 15. The impedance curves for pairs with 8-inch and 18.25-inch spacing are similar to those of Fig. 15, the resist- ance values for these spacings being about 45 ohms lower, and 40 ohms higher, respectively, than the values shown for 12-inch spacing. Changes of resistance and leakage conductance due to changing weather conditions have very little effect on the characteristic imped- ance at frequencies above 1000 cycles. Changes of insulator capaci- tance due to changing weather conditions or the use of different num- bers or types of insulators have an appreciable effect on the impedance. Deviations from the normal spacings between wires which result from the use of drop bracket transpositions also have an important effect upon the impedance. 750 500 to ^250 O -250 V* <»4 "^ •^ -s •*/* -^ •*' >*i V -^ '^ -^ ^ «•-< -*" >*, ■*>. ■^ V S-i ^ RESISTANCE - —M •• >-»" ■*■ V*- ■*" '*■ ^ ^ (» '^ ^ '*' '^ REACTANCE 10 20 30 FREQUENCY— KILOCYCLES 40 50 Fig. 16 — Impedance measured on a well-transposed 128-mil open-wire pair with 8-in. spacing. Like the measured attenuation, the impedance which is measured for an open-wire pair is affected by the presence of line irregularities. Hence, the measured impedance is never a smooth function of fre- quency, but displays slight irregularities throughout the entire range. This is apparent from Fig. 16, which gives a curve of the impedance measured on a well transposed pair. This curve is in remarkably close accord with the generalized values of impedance, the maximum devia- tion from the theoretical curve being about 2 per cent. OPEN-WIRE TELEPHONE LINES 753 Like the attenuation, the impedance of an open-wire circuit in a narrow band of frequencies may be radically changed by interaction with adjacent circuits. These large irregularities in the impedance commonly accompany absorption peaks in the attenuation, and are, of course, due to the inadequacy of the line transpositions. Phase Change and Velocity of Propagation The imaginary component of /3 of the propagation constant is known as the phase constant because it indicates the change in the phase of the voltage and current in circular radians per unit length of line. The value of the phase constant is given by /32 = 1/2 [V(i?' + LW)(G2 + CV) + (LCco^ - RG)]. (10) If E} and G^ are small compared to LW and CW it is clear that /3 = w VLC radians per mile. (11) For an open-wire pair the value of ^ is approximately 0.035 radian per mile, or 2 deg. per kilocycle per mile. The latter figure is a con- venient one to remember. The constant ^ also enters into the familiar expression for the velocity of propagation F = - miles per second. (12) The velocity of propagation on open-wire lines approaches the veloc- ity of light, which is 186,000 miles per second. The velocity is reduced below this value by the increase of capacity due to the presence of the other wires and the insulators, by the internal inductance, and by the presence of resistance and leakage conductance. Values of the velocity of transmission for open-wire pairs are presented in Fig. 17. At fre- quencies above a few hundred cycles the velocity of transmission is, apart from the effect of line irregularities, practically constant through- out the frequency range. In the last few years increasing attention has been focused upon the phase characteristic of the open-wire circuit. One of the reasons for this is that different velocities of transmission for different frequency components in a signaling band (which are obtained when the phase shift of the circuit is not a linear function of the frequency) may give 754 BELL SYSTEM TECHNICAL JOURNAL rise to what is known as phase distortion, and it may be necessary to correct this distortion by suitable networks. ^^' ^^ 200,000 a z o o ^ 180,000 a UJ 5 160,000 z o 140,000 < o o a. o a: a. O > 120,000 t- o s UJ > 100,000 100 ^ — ■ ^ ' — ■ ^ ^ / / // / 500 1000 FREQUENCY- 5000 10,000 -CYCLES PER SECOND 50,000 100,000 Fig. 17 — Velocity of propagation for open-wire pairs. Characteristics of Phantom Circuits Phantom circuits, which are derived from two pairs or side circuits by transmitting over the wires of one pair in parallel and using the wires of the other pair in parallel as a return, have been employed in the telephone plant for voice-frequency transmission for a number of years. Their use for carrier transmission is limited chiefly by the difficulty of reducing the cross induction with other circuits at high frequencies. Phantom circuits are generally derived either from horizontally adjacent non-pole pairs or from vertically adjacent pole pairs. Thus wires 1 and 2 are "phantomed" with wires 3 and 4, wires 5 and 6 with wires 15 and 16, etc. In some of the newer transposition arrangements the non-pole pairs are not phantomed, since it has been found that the omission of the phantom permits the operation of a larger number of carrier systems on one line without excessive mutual interference. The resistance of a phantom circuit is evidently half of the corre- sponding value for the side circuit. ^^ See "Distortion Correction in Electrical Circuits with Constant Resistance Recurrent Networks," by O. J. Zobel, Bell System Tech. Jl., July, 1928, pp. 438-534. 1' See "Wire Transmission System for Television," by D. K. Gannett and E. I. Green, A. I.E. E. Trans., Vol. 46, 1927, pp. 946-953 {Bell System Tech. JL, October, 1927, pp. 616-632). OPEN-WIRE TELEPHONE LINES 755 If the two wires forming one of the sides of the phantom circuit are designated 1 and 2, and the wires forming the other side are designated 3 and 4, the inductance of the phantom circuit is L = 0.32187 r 1.1513 log^o ^^f^;fl^^* + ^5 L LfnL>zia X 10~^ henrys per loop mile (13) where Diz, D23, etc. represent the spacing between the wires 1 and 3, 2 and 3, etc., and where d, ix, and 8 have the same significance as in Equa- tion (3). The capacitance of a phantom circuit in space is C = 0.07766 X 10-6 log: 10 4DjsDuD23Du farads per loop mile. (14) ■ JW n /^ X' / 0^ /' /^ ^ /^ ^ / / ^ x' v ^ ^ if u / .^ ^ / rT^ > > g*' ^^ (D l>^^^^ >r^ ^? " '^ ° 15 %^:4lT>^ rT.U ^V" «» •' 1 J NT^V "lA^^ <' .^ z / ^2^f .^^ .U^, r1 ^ ** \- V y 5^^^^^ ^f li^ ^^ ' ^ ■' < y / z / ■^nOj:' ' ,^^V<^ ' L ^ ^ 10 / / / / y ^ N? 6 ■y r'l. - -^^^ ^ 1- A / /; • y' ■^ \6 b^ ^^>1 ^ / y / /^ ^ ,xH ..-^ / ,? / > ^ y / J / ^^ > .05 r /. < / X y ^ • ^ > 0 20 30 FREQUENCY— MLOCYCLES 40 50 Fig. 18 — ^Attenuation of phantom open-wire circuits. 756 BELL SYSTEM TECHNICAL JOURNAL Equations (13) and (14) are based on the assumption that the phantom circuit is sufficiently well transposed to secure balanced voltages, cur- rents, and charges on the four wires. The capacitance of the phantom circuit is, of course, affected by the capacitances of its component wires to the other wires of the lead and to ground and also by the capaci- tances of the insulators. The losses which contribute to the leakage conductance of phantom circuits are similar to those which have been discussed for side circuits. The ratio of the phantom circuit leakage conductance to that of the side circuit will depend upon the relative magnitudes of these different losses. Varying relations between the different types of losses give a ratio of phantom circuit leakage conductance to side circuit leakage conductance which might conceivably range between the extremes of 1 and 6. For practical engineering purposes, however, the leakage con- ductance of the phantom circuit is generally assumed to be twice that of the side circuit. 500 250 If) I O -250 ^ — _ — ._ '■ —I ■'- i V-i _ '. 1~ -~ ^ ■^ ' ■~ '. — ■"■ ^ ~ ' ^ _ . ■ — — " ■ . z •"■ ■" RESISTANCE ^ -165M1L, -I28MIL. -I04MIL. t ^ r m.M RE ACTANCE [ I04MIL, L -128 MIL. L ■165 MIL. 0— — 20 30 40 FREQUENCY — KILOCYCLES 0 10 Fig. 19 — Impedance of phantom open-wire circuits. 50 Curves of the attenuation of non-pole pair 12-inch phantom circuits of different gages are depicted in Fig. 18. In most cases the attenuation of the phantom circuit is somewhat less than that of the corresponding side circuit. Frequently, however, the advantage of lower attenuation is under practical conditions more than offset by the large noise and crosstalk effects experienced on the phantom circuits. Values of the dry weather impedance of 12-inch phantom circuits are presented in Fig. 19. It will be observed that the impedance of a phantom circuit averages about 60 per cent of the impedance of a 12- inch side circuit. OPEN-WIRE TELEPHONE LINES 757 Characteristics of Iron-Wire Circuits There now exists on the toll lines of the Bell System a small amount of galvanized iron or steel wire. Nos. 12 and 14 B. W. G., with diam- eters of 109 and 83 mils, respectively, are the gages most commonly found. Steel wire and BB iron wire are both used, the latter being of more frequent occurrence. These iron-wire pairs display such large values of modulation and high-frequency attenuation that they are generally quite unsuitable for carrier transmission. The skin effect resistance of iron wire may be computed by standard theory, provided that the values of the resistivity p and permeability fx are known. The values of p for BB iron and steel wire are about 12.8 4.0 UJ -I 3.0 a. Hi a. CO a I z o D Z Hi ^- 2.0 1.0 ^ ^ ^ ^ ^ -^ - .\^ o^ ^ ^ ^ ^ -^ \l [^ i^ ^ - " f<^ wJf^ - -* "^ y^ !■< b^^ "^ ^^ ^- -" " y y ^w^er\ -- ** x' ^ ^ 1^ ^^ £^ ,/ /^ ^ \-\U> / y y ^ y ' ^ ^ A ^ > ^ ^ / A ^ / ^ y ^ p^ 10 20 30 FREQUENCY- KILOCYCLES 40 50 Fig. 20 — Attenuation of iron-wire circuits. and 14.8 microhm-centimeters respectively. It will be noted that these values are about seven and eight times the resistivity of copper. For currents of telephonic magnitude at frequencies from about 500 to 50,000 cycles, it has been found that the value of the permeability y. of iron wire of the above sizes ranges from about 110 to 50. Study of the available data indicates that the best average value of /x at 1000 cycles is about 85 and that the effective value decreases with increasing fre- quency. Because of the high d.-c. resistance and the large skin effect ratio which results from the high permeability, the resistance of iron wire for alternating currents is extremely great. 758 BELL SYSTEM TECHNICAL JOURNAL Except at low frequencies, where the internal inductance of the iron wire is large, the total inductance of an iron-wire circuit is not far differ- ent from that of the corresponding copper circuit. The capacitance and leakage conductance of iron-wire circuits are substantially the same as for similar pairs of copper wire. Typical attenuation curves for several gages of BB iron wire in dry weather are shown in Fig. 20. These curves are based upon experi- mental results. It will be noted that the attenuation of an iron-wire circuit averages about ten times that of the corresponding copper-wire circuit. It should be understood that the attenuation values for an iron-wire circuit are subject to rather wide variations in practise, partic- ularly because of the effects of corrosion. There is some change in the attenuation of an iron-wire circuit with change in weather conditions, but this is a relatively small percentage. The impedance of an iron- wire circuit has the same order of magnitude as the impedance of a similar copper-wire circuit. Appendix I Temperature Coefficient of Resistance at High Frequencies In the skin effect literature cited in the text, it is shown that at high frequencies the a.-c. resistance R is R = 0.00979 V/i^o, (15) where Rq is given by Equation (6) and / is the frequency. Differen- tiating with respect to t, and substituting, we find that 1 dR 1 \R^i ^^ = ^^=rW^ ^^^^ and for ordinary ranges of temperature ^, = f • (17) Appendix II Condition for Minimum Attenuation The attenuation at high frequencies is given by Equation (7). Differentiating this with respect to ylj , and setting the result equal to OPEN-WIRE TELEPHONE LINES 759 zero, it is found that the condition for minimum attenuation is which is another way of saying that R IC G \L so that the resistance and leakage components must be equal in order to have minimum total attenuation. Transients in Parallel Grounded Circuits, One of Which is of Infinite Length By LISS C. PETERSON This paper deals with a mathematical discussion of induction due to tran- sient currents of the forms / = sin ut and / = e~^K Formulas and curves are developed for the calculation of the induced voltage in exposed telephone lines due to currents of the above types. Part I THE problem of mutual impedance between grounded circuits of infinite length for steady state sinusoidal currents has been treated by a number of authors, and the solution of this problem is now well established.^' ^'^ In addition to the steady state voltages induced the transient voltages are also of importance. Riidenberg ■* and Ollendorf ^ have given approximate solutions for transient voltages due either to d.-c. switching or the sudden flow of a sinusoidal current on the assumption of circular symmetry and for circuits one of which is of infinite length. Since the assumption of circular symmetry holds only for a limited set of conditions it is desirable to develop formulas for the transient induced voltages based on the exact solution for steady state conditions referred to above. The discussion in this paper will be limited to the case of parallel wires, one of which is of infinite length, and both located on the surface of the earth but insulated from it except at their ends. Dis- turbing currents of the forms / = sin oot and / = e~^'' will be assumed, A more general case with both wires above the earth's surface leads to complicated expressions for the induced voltage not well adapted for engineering use. The restriction to wires on the earth's surface results in appreciable simplification and does not introduce a serious departure from actual conditions. With these assumptions, the following formulas holding for small and large values of time, determine the induced voltage per unit length on a secondary wire 2 due to the sudden flow of a current /(/) = sin wt in a primary wire 1 infinite in length, separated from wire 2 by a dis- tance X centimeters. 1 PoUaczek, F., E. N. T., Vol. 3, 1926. - Carson, J. R., Bell System Techtiical Journal, Vol. 5, 1926. ^Haberland G., Z. ang. Math. U. Mech., Vol. 6, No. 5, 1926. * Wiss. Veroff. a. d. Siemens-Konzern, Vol. 5, No. 3, 1927. 5 E. N. T., Vol. 5, No. 3, 1928. 760 TRANSIENTS IN GROUNDED CIRCUITS 761 VM = and sin iol CO g-irXxlj I f 2f 6t' - ioh^ ttXx'- (7rXx2)2 ' (7rX.r2)3 sin cot , 2\^ Fi2(/) = — z-Y "I 7=r[cos co/ kei'(2x\7rXco) + sin co/ l^r'(2A-\ ttXw)] 7rX.r2 ^^-VttX ^-w\xyt J L / ^ _ 67rX.r^ (ttXx'')^ co/2 a;2 W-' /' ^"^ ' and for such values of time where neither of these series expansions would give very accurate results the following formulas may be used ^12(0 = -ZVTT Z^^2 cos (w/ - ^), TT \x' t\x' B tan^ = ^ 1.0 o.a f\ A r\ \ r\ \ 1 THE FUN CTION : \ 04 0.2 A 0 -0.2 -0.4 ~ n A \\ FOR D C '0 FFERENT VALUE )F r AND S. " / \ \i ( ^ - 11 V i ^ N Ni L h A 1 y ft \ V f \ \l \ W s=io- s = t - s-in"' J vl \\\ S = 10"^ s=io-^ M^ 1 \l -0.8 -1.0 -1.2 -1 ^ m 1 1 \\ ^ V^ 1 ^ c // v^ 2 3 4 6 6 7 6 r 9 10 Fig. 1 — Plot of the integral A as a function of r for different values of s. 762 BELL SYSTEM TECHNICAL JOURNAL A and B are given by 4 A = g— ^^^'-/^ cos ^d^, Jo B = f e-""^'"!^ sin ^d^. Jo With a disturbing current / = e~^^ in wire 1, the induced voltage [ / A A 1.6 1.4 1.2 // V L e THE FUNCTION; =/e--5 Sin ^ d^ Jo \ DIFFERENT VALUES OF r AND 5. = irXx^ui r= cut) // \ \ ""^ VI \ 1 ^ A \ 0.8 ^ \ \ \ II // 1 «; — \rc^ \ -^ \ 1 // / \ /- 0.4 // / \y // ^ — t \ 1 1 ^ N 0.2 S=IOy \ / 1 / \ \ \ /// ' 1 B 0 -0.2 -0.4 -0.6 "^ \ ^ — y ^ \ J <^ 0.8 -1.0 234 56789 10 r Fig. 2 — Plot of the integral B as a function of r for different values of 5. TRANSIENTS IN GROUNDED CIRCUITS per unit length in wire 2 for small values of time is given by 763 1 i^-w=^(^ ^/3« _■ ^-'r>^V< ) 7rA.\ 24/5 4_ 36/3/6 + 12/32/7 4.^3^8 r_j2 2f__ |_ ttX-V" (tt (7rX.r2) .2\4 + 500 40.0 200 100 eo 60 40 20 10 8 f I //; // /' THE - FOR II 1 1 FUNCTION ■■ C = l-t- e )F r AND 5 / // / J 1 S=10''AND less- / / 5 = 10' S = l----..^ / / .'^ J / ^^^ ^ X — / 4 6 8 4 6 8 4 6 8 10"' ~ ■ " ~ 10-' 1 - '10 r Fig. 3 — Plot of the quantity C as a function of r for different values of s. and for values of time such that the above series can not be used by 1 Fx2(/) TT Xjc2 (Ce-^' - e-^'^'i'), where C= 1 + r ,-(jrXj: Wf)+£rf^. Finally, the induced voltage Zi2(/) due to a unit step current in wire 764 BELL SYSTEM TECHNICAL JOURNAL 1 is determined by The functions A, B, and C are plotted in Figs. 1, 2, and 3 for some values of the parameters often to be found in practice. In these formulas X is the ground conductivity in electromagnetic c.g.s. units, X the separation between wires in centimeters, / the time in seconds, and j = V— 1. The functions ker' and kei' are related to the Bessel function of the second kind for imaginary arguments de- fined by G. N. Watson, " Bessel Functions " as follows ker'(2) ±ikei'(s) = - j=^'''Ki(zj^'i^) Values of these functions are tabulated in Table I of " Bessel Functions for A-C Problems " by H. B. Dwight A. I. E. E. Trans. 1929 pp. 81 2-820. The induced voltage is in units of abvolts per cm. which is trans- formed to volts per mile by the factor 1.61 X 10~*. Part II The second part of this paper will be devoted to a discussion of the theory leading to the above results. Consider a system of two wires, 1 and 2, wire 1 being of infinite length, parallel with each other, with the heights hi and Jh above earth and separated by a distance x. The general problem is to calculate the voltage on wire 2 as a function of time due to the sudden flow of a current in wire 1, this current being zero before / = 0 and /(/) thereafter. Let the voltage on wire 2 due to a unit current step, that is, a current equal to zero before t = 0 and unity after / = 0, be denoted by Znit), then the voltage due to a current I{t) is given by Vnit) = j^J' Z:,{r)I(t - T)dr.' (1) The fundamental quantity thus necessary in the solution of the problem is Zuit). This quantity completely determines the voltage Vi2{i) for all types of disturbing currents. Zio{t) may be written as a Fourier integral : 1 r+'° pj<^t Zi2(t) =i- V- Zi2(co)(/a,, (2) where Zuioo) is the mutual impedance for periodic earth currents and ^ Carson, Electric Circuit Theory' and the Operational Calculus, page 16. TRANSIENTS IN GROUNDED CIRCUITS 765 is given by ^ Zi2(co) = 2jco log -^ + 4a; [Va^^ + J - M]e-^*i+''^^''^ COS fxx^la d/x. (3) P Jo where a = 47rXco p" = ^Vh + W^ X'^ Zii{ 7ZAX = lO ^^ ^=^ ^^^ ^^ 0.1 02 03 0.4 SECONDS Q5 Q6 07 08 Fig. 4 — Plot of the voltage Zitif) on wire 2 as a function of / for two different values of separation or conductivity as given by the product ttXa;-. 7 Bell System Technical Journal, Vol. V, page 544, October, 1926. 766 BELL SYSTEM TECHNICAL JOURNAL I am indebted to Dr. F. H. Murray of the American Telephone and Telegraph Company for the following solution of Zi2{t) as given by (2) : ZM where and 2h M\^y[\^t ^i 2\\Mi^ e ' + ' Mi" ^^1^^ nr IT Ml'' erfc V/ + Mo erfc i/sVx V/ (4) \M\ = \Mi\ = \M2\ Ml = {h - jx)\^, Mi = (h -j- jx)\ir, h = hi + ho erfc Z = 1 - erf Z = 1 _2_ r^ Vx Jo ,-t2 dx. Taking the limit of equation (4) as h approaches zero there results 1 ^12(0 =-^,(1 -e-^^^'i^), ttXx (5) which formula is of fundamental importance in the present analysis.^ This equation is also plotted on Fig. 4 for two different values of irXx^. Assuming now I{t) = sin co/ formula (1) gives sm Fl2(/, a IcV' Jo where a = ir\x^, (3 = jco. 'r + e .-0t r e-^^M+Pr^r], (6) 8 This formula can readily be checked in the following manner. The mutual impedance between wires on the surface of the ground is Zi2(co) = J 7_ K.iyx), where y = yATrXjco and Ki is the Bessel function of the second kind with imaginary argument defined by Watson, "Bessel Functions." Replace jco by p, and interpret the function of p so obtained according to opera- tional methods. The first term is independent of p and therefore of t. The second term is transformed according to the equivalent ayJpKiiayJp) = e-cV^', given as pair 922 in G. A. Campbell's paper "The Practical Application of the Fourier Integral," Bell System Technical Journal, Oct. 1928. Equation (5) is then immediately obtained. TRANSIENTS IN GROUNDED CIRCUITS 767 The integrals appearing in (6) are apparently not known in closed form. Series expansions holding for small and large values of time may be derived however. By successive integration by parts we obtain : t/O g-(a/r)+/3r^^ = g-(a/0+fl' a a a 24/5 + 36^/6 + 12,32/7 + ^3/8 a' + (7) e^' appearing on the right hand side of equation (7) is cancelled by g""^' appearing before the integral, and similarly for the first term in brackets in equation (6). In the complete expression for the voltage odd powers of /3 cancel and we have: Vnit) = sin o)/ CO IT \X^ g-ir\xyt 2/3 irXx^ 24/5 + 6/^ - co2/« {wXx^y - 12a;¥ + For large values of time equation (6) is written as (8) sm ut . CO + CO /»« „-(.alT)-\-PT /»oo p-{alT)-pT "I (9) where the integrals between zero and infinity correspond to the steady state condition while the integrals between / and infinity give the tran- sient distortion. The integral between / and infinity may be evaluated in a manner quite similar to that used above. The result with plus sign for ^ is / » g-(a/T)+^T dr = f-W0+/3< I /6 b- 12a2 1 /24 36a: iza^ a- 6a o2 (10) The integrals between 0 and infinity are evaluated by: i '00 ^— (a/T)=t/3« (11) in which the real and imaginary parts of the right hand side may be expressed by the ker' and kei' functions by the relation already given. 768 BELL SYSTEM TECHNICAL JOURNAL The complete expression for the voltage is: SI n (Jit , 2Vco Viiit) = — ^-^+- — p=[cos w/ kei'(2W7rXco) + sin co/ ker'(2.rVxXw)] 7rA.V xVttX a—v\x^it 1 1/6 67rXx2 , (ttXx^) + ^^^:^^H- W^2 aj2 Xt^ t^ ' f« • (12) For such values of time where neither of the formulas (8) or (12) give very accurate results it is necessary to perform mechanical integration. In so doing it is convenient to introduce a new variable ^ of inte- gration. Let ^ = wT, and the integrals become Jo ^c/o = - r r e-'i^ cos ^d^ ± j r e-^^ sin ^d^ , where Now let .y = r = aco = ttXcoX"^, 1 , -, = cot. J ^^-^^ A{s, r) = f e-'l^ cos ^d^, (14) Jo Bis, r) = f e~''^ sin ^d^, (15) Jo and formula (6) becomes where , . sin wt V^2 + -B2 tan^ = -^- (17) The values of the functions A and B are given in Figs. 1 and 2 for some values of 5 and r which are frequently met in practice. Assuming finally /(/) = e~^' equation (1) gives after simplifications Fi2(/) = -4-2 (^~^' - ^~^'''") + ^2 f e-^^^'l^'+^^dr. (18) For small values of / the series expansion (7) may be used. The TRANSIENTS IN GROUNDED CIRCUITS 769 result for this case then becomes VM = ^Ae-'- - e-'^") P „-Tr\xi jt e 2/3 + /3/4 , 6/4 + 6/3/5 + 132/6 + irXx^ (7rXx2)2 ' (7rXx2)3 24/5 + 36/32/6 4. 12^2^7 _|_ ^3^8 uxx^y + (19) For large values of time, introduce a new variable ^ — /3r of integration in the integral in (18). Then ^^^(^^ = ■^2iC^-^' - ''~''^'") (20) where C= 1 + r e-(^i^^+id^,\ *^" ^ (21) s = xXjc2/3. J The values of C are given on Fig. 3 for important ranges of r and 5. For 5 equal to and less than 10~2, C is for practical purposes inde- pendent of s. I am indebted to Dr. F. H. Murray, and Mr. R. M. Foster of the American Telephone and Telegraph Company for valuable suggestions during the course of this work, and to Miss R. Pedersen who carried out all the numerical calculations. Impedance Correction of Wave Filters Development of Impedance Requirements By E. B, PAYNE The present importance of wave filter impedance correction arises chiefly from its relation to crosstalk in carrier systems. Briefly, it appears that line transpositions, an effective remedy for many types of crosstalk, are less satisfactory when directed against the so-called "reflected near-end cross- talk" and "reflected far-end crosstalk" produced when waves reflected from the junctions between lines and repeater equipment of carrier systems induce currents in neighboring systems. The expense of the elaborate transposition scheme necessary for a substantial reduction in these types of crosstalk makes it desirable to diminish the amplitude of the reflected wave as far as possible by the improvement of the impedance match between lines and repeaters. A detailed study shows that this is most conveniently done by terminating the filters in the repeater by sections whose image impedances at one end match the main body of the filter, while at the other they ap- proximate constant resistances, matching the terminal impedances. The development of appropriate filter terminating sections has passed through a number of stages. The earliest filters gave reflection coefficients as great as 50% to 60% in the useful transmission band. The invention of "m-derived" and " .r-terminated " filters, plus a number of more or less empirical schemes, made it possible to obtain reflection coefficients ranging from 10% to 15% in the useful band. Recent progress has resulted chiefly from the development of a series of sections, the simplest of which is equivalent to the w-derived type, while the others, of progressively increas- ing complexity, give progressively better approximations to the ideal characteristic. The use of the more complicated sections has made it possible to reduce filter reflection coefficients to the order of 2%, or even less. At present the chief limitation appears to be the difficulty of manufacturing filters with sufficient precision to allow the theoretical characteristics to be realized. The paper is illustrated by figures showing the various stages of this progress as they are exemplified in actual designs. THE rapid increase in the demand for long distance or toll telephone service in recent years led to the introduction, about 1920, of carrier systems as a means of securing more intensive use from long telephone lines. The growth of these circuits has resulted, still more recently, in the multiplication of the number of carrier systems in use and in the close association of several similar or different carrier systems on a single pole-line. This development raised a number of totally new engineering problems and demanded careful reconsidera- tion of many other questions of comparatively small importance in earlier systems. Among the factors thus brought into prominence by carrier system development, the chief, for the purpose of this paper, is the impedance mismatch between telephone lines and repeaters or terminal apparatus. The components of a complete transmission system, such as the line 770 IMPEDANCE CORRECTION OF WAVE FILTERS 771 itself, various transmission networks, amplifiers, modulators, electro- acoustic apparatus, etc. are quite dissimilar physically and as we might naturally expect, these physical differences manifest themselves in many instances as pronounced dissimilarities in the forms of the im- pedance-frequency characteristics. For example, the characteristic impedance of a uniform line varies smoothly with frequency, but that of a wave filter changes abruptly as we go from the transmitting to the attenuating range. In spite of the possibility of changing the general impedance level by the insertion of a transformer such inherent "incompatibilities of temperament" between the characteristic im- pedances of the various components of the telephone circuit must lead normally to impedances which resemble each other only in narrow frequency bands and which may differ widely over large and important portions of the frequency spectrum. In default of some method of extending the range of similarity, most long telephone circuits will exhibit wide impedance mismatches or irregularities at numerous junc- tion points. TERMINAL. APPARATUS Z| Z2 LINE — o o REFLECTION COEFFICIENT^ Z1-Z2 Z|+ Z2 Fig. 1 — Junction of line and terminal apparatus illustrating impedances which determine the reflection coefficient. In voice frequency circuits or in carrier circuits which are not in close physical association, impedance irregularities are of importance only insofar as they affect transmission efficiency.* In addition to modifying the current which proceeds onward toward the receiving device, however, an impedance difference at any junction produces a reflected wave which retraverses the circuit toward the sending end. A convenient measure of this second effect is found in the "reflection coefficient" which may be defined as the vector difference of the two impedances looking both ways from any junction divided by their vector sum (see Fig. 1) and is equal both in magnitude and phase to the ratio between the reflected wave and the wave originally propa- gated. The effect of reflection of considerable magnitude on transmission is slight. Indeed relatively large reflection may actually improve the transmission characteristic of certain circuits. In voice frequency circuits and in carrier circuits which are not operated over lines in close * In two wire repeatered circuits reflection causes echoes which are one of the limit- ing factors of such circuits. These circuits are, however, outside the scope of this paper. 772 BELL SYSTEM TECHNICAL JOURNAL 2000 2000 100 10 12 14 16 18 20 FREQUENCY IN KILOCYCLES PER SECOND 22 24 26 Fig. 2 — Impedance and reflection coefficient of an early carrier telephone system. IMPEDANCE CORRECTION OF WAVE FILTERS 773 physical association, transmission is the only consideration. Conse- quently relatively large reflection coefficients are not objectionable in such circuits. Curve I of Fig. 2 shows the reflection coefficient of an early carrier system which was not intended to work with other systems of the same type. Large as these reflections appear relative to present day standards they did not seriously impair the transmission of the system. It is also true that appreciably better results could not have been obtained with the design technique available when the filters for the system were developed. As the increase in the demand for long distance traffic made it necessary to associate systems on the same pole line the situation became radically different through the introduction of a new factor crosstalk between systems. Crosstalk between systems at carrier frequencies is inherently large and the methods of reducing it expensive. The reflections due to the mismatching in systems increase the cross- talk between them by introducing a type of interference known as "reflected near-end crosstalk." This type of crosstalk can be made negligible only by making the impedance mismatching in the two systems very small. Since "near-end crosstalk" contributes heavily to the cost of the arrangements for reducing crosstalk between carrier systems the substantial elimination of impedance irregularities between lines and the filters and associated apparatus composing the terminals of systems becomes of great economic importance. As this need for reducing and ultimately eliminating these irregular- ities appeared a series of improvements in design technique have been developed, each better than its predecessor, which have culminated in a technique which appears to be adequate for the purpose. It differs in many essential features from the others and leads to a new type of filter section which is not of the standard recurrent ladder type. It is the purpose of this paper to give some idea of the relation of cross- talk to impedance mismatching, show how the successive stages of the filter development have grown out of the system requirements and finally to present an outline of the final technique. The accompanying paper. "A Method of Impedance Correction." by H. W. Bode gives this technique in detail. Impedance Irregularities and Crosstalk The ultimate relation between reflection and crosstalk between two lines which are associated with a number of others on telephone poles is extremely complicated. An idea of the principles which underlie the relationship may be obtained by considering only two of the circuits and assuming that the others have been temporarily removed.* If these * Since these two circuits consist of two pairs of wires, there is a potential phantom 774 BELL SYSTEM TECHNICAL JOURNAL two telephone lines parallel each other, as in Fig. 3, they will be electri- cally coupled through mutual inductance and capacity. Currents flowing in one will consequently produce crosstalk currents in the other. When the subscriber at the west end of the line {A) is talking, waves initiated by his voice will cross from line {A) to line {B) at adjacent points along the entire length of the lines. Crosstalk entering line {B) at a typical point may traverse four chief paths. It may (1) flow directly back to the west end, (2) flow onward to B" and be reflected back to the west end, (3) flow directly onward to the east terminal, and (4) flow backward to B' and thence, by reflection, to B". There are of course an infinite number of other paths involving multiple reflections but the reduction in amplitude caused by successive a' 1 A" 1 9 ie 1 b' 1 b" ^ _x ! - ' If __ _^_|h __ T 'I 1 west east terminals terminals key: — typical path of "far end" crosstalk TYPICAL PATH OF REFLECTED "NEAR EN D" CROSSTALK Fig. 3 — Diagram illustrating relation between impedance mis-matches and crosstalk in carrier systems. reflection and line attenuation makes these negligible in comparison with the others. The first two of these four possibilities of crosstalk production can be eliminated immediately. Modern carrier systems are so designed that conversations going in one direction are carried by one band of frequencies and those travelling in the opposite direction by a different band. Currents entering the west terminal, whether they follow a direct path such as AefB', or are first reflected at B", making the typical path A'efB"B', are therefore eliminated by the filters in the terminal office. Crosstalk currents of the third type ("far end cross- talk") following the typical path A'efB" cannot be eliminated since they fall within the frequency band used by the subscriber at B" for listening. We may observe, however, that these currents traverse the same length of line in travelling from A' to B" no matter what the point (such as ef\n the diagram) at which they cross from one line to the other. Since both lines are alike crosstalk currents will be attenuated and shifted in phase by the attenuation and phase shift of a single circuit which, as far as crosstalk is concerned, constitutes a third circuit. The effect of this circuit on the crosstalk of the other two is by no means negligible but con- sideration of it is omitted herein in order to simplify the presentation of other im- portant relations fundamental to crosstalk. IMPEDANCE CORRECTION OF WAVE FILTERS 775 full length line. Moreover the components of this type of crosstalk due to magnetic and capacitative coupling are nearly out of phase and so one tends to neutralize the other. As a consequence of this equal effect of the line characteristic on all the components which reach the receiver at B" the resultant crosstalk can theoretically be eliminated at all frequencies when only two circuits are present by a single transposition (crossing the wires) in the center of either line. Crosstalk currents of the fourth type, "reflected near-end crosstalk" following such paths as A'efB'B" and AghB'B", cannot, however, be disposed of so easily. The length of line traversed by the component currents which make up the resultant crosstalk depends upon the point at which they cross from one line to the other, and they will therefore be affected in various fashions by the line attenuation and phase shift. The transposition scheme required to eliminate crosstalk resulting from these currents will consequently depend, at any frequency, upon the length of the line and upon its phase and attenuation characteristics at that frequency. Complete elimination of crosstalk of the fourth type cannot be secured, even for two circuits over a finite frequency band, from a finite number of transpositions. When other lines are adjacent to the two we have considered the problem of reducing "far-end" and "near-end" crosstalk by trans- positions is still more complicated. With a number of lines it is no longer even theoretically possible to eliminate far-end crosstalk by a single transposition. It is, in general true, however, that the cost of a transposition scheme adequate for far-end crosstalk is still much less than that of the elaborate system of transposition required to reduce near-end crosstalk to tolerable values. From an economic standpoint therefore, the cost of transpositions required for near-end crosstalk is usually the main feature to be considered. Impedance Correction an Economic Means of Controlling Crosstalk Another method of reducing this near-end crosstalk, and one which experience has shown to be much cheaper than an elaborate transposi- tion scheme is found in the reduction of the reflection coefficient be- tween the line and the repeaters. Obviously the magnitude of the reflected near-end crosstalk depends upon the amount of the impedance mismatch at the junction between the line and the terminal equipment (e.g. at B' in Fig. 3). It can be made as small as we please, even with a very simple transposition scheme, if the reflection coefficient at line- repeater junctions can be sufficiently reduced. No serious mismatches would occur if the impedances of repeaters and terminal equipment were that of the modulators or amplifiers, since at carrier frequencies 776 BELL SYSTEM TECHNICAL JOURNAL the impedances of modulators, amplifiers and telephone lines approx- imate constant resistances. The interposition of filters between lines and modulating or amplifying apparatus, however, normally produces large reflection coefiicients. Since the filters in addition to being the apparatus immediately responsible for mismatching, are also inexpen- sive and easily controlled in comparison with the line, they furnish the most promising field for the reduction of reflection coefficients. Relation between Actual and Image Impedances The reflection coefficient which determines the amount of crosstalk exhibited by the system involves directly only the line impedance and the actual impedance characteristic of the filter system. In order to understand the peculiarities of the actual impedance of a filter, how- ever, it is necessary to give prior consideration to its characteristic, or image impedances. The image impedances of any transmitting device are defined as the impedances with which the device must be terminated at both ends if the impedances looking both ways at each pair of terminals are to be matched. In other words, they are the impedances with which the structure must be terminated if no reflections are to occur. Filter sections of different physical configuration and with different attenuation characteristics often have the same image im- pedance characteristics. Practical filter designs are therefore usually composite structures containing several different types of sections. Internal reflections are avoided by so choosing the arrangement of the section that the image impedance characteristics at all section junctions are matched. Under these circumstances the image impedance charac- teristics of the complete structure are the same as those of its terminat- ing sections. The image impedance characteristic of typical low pass filter sections is shown in Fig. 4i-A . A corresponding curve for band pass filters is given in Fig. 4-B. The image impedance characteristics are given only for the transmitting bands of the filters since, as previously indicated, the filters themselves suppress crosstalk in the attenuating regions making it unnecessary to control impedances outside the transmitting range. The associated equipment, such as lines and modulators, with which the filters are terminated, are approximately constant pure resistances, and may be represented in the trans- mitting range by the block type characteristics drawn over the curves. The relation between the image impedance properties of filters and the actual impedance presented by a repeater or terminal to the line can be understood from the simplified circuit diagram of a typical carrier terminal (Fig. 5). Upon examining the figure we note that there are a number of junctions at which rounded filter image im- ^ IMPEDANCE CORRECTION OF WAVE FILTERS in pedance characteristics such as those shown in Fig. 4:-A or Fig. 4-5 face the block type image impedances of the same figure. In the system of Fig. 5, for example such junctions occur at B, C, D, E, and to some ^0 ^-^— ^^^ UJ y^ O ^*'**'«v*,,^^ O / z ^'^^^,^^ z / < ^^^^^ < / o ^**^^ Q / \ LU ^Sw U 1 \ a. >v Q. / ^° \ 5 N. 5 \ UJ O < A \ UJ O < f B ^ 5 ^ 2 c 1 f ■~i -fm f? FREQUENCY FREQUENCY -^ ^0 / , \ \ UJ o \ O z \ z < \ < Q I Q U 1 UJ Q. 1 Q- 5 c \ 5 ^.° u UJ O < 5 < 5 D C \ fi fm ^- FREQUENCY FREQUENCY Fig. 4 — Image impedances of "constant-^" and "m-derived" low-pass and band-pass sections. Figs. 4-A and 4-B — "constant-^" sections. Figs. 4-C and 4-D — "m-derived" sections. extent at yl. It is evident, of course, that reflected waves will be produced at these points, and since impedance differences occur at several junctions a wide variety of multiple reflections may exist. D HIGH PASS FILTER ^ T LINE \ ^ ^ E 1 LOW PASS FILTER 1 1 1 UPPER BAND PASS FILTER LOWER BAND PASS FILTER TO VOICE FREQUENCY CIRCUITS MODULATOR DEMODULATOR Fig. 5 — Simplified circuit diagram of a typical carrier terminal. Upon reaching the line terminals, all of these reflections combine with the wave originally propagated to determine the actual current enter- ing the structure. The reflected waves are of course diminished in 778 BELL SYSTEM TECHNICAL JOURNAL magnitude in traversing the filters intervening between the line term- inals and the junctions at which mismatches occur. The attenuation of most filters within transmitting bands is so small however, that the waves may be of appreciable magnitude even after several reflections. The filter phase shift within these frequency bands, on the other hand, is large and varies rapidly with frequency. The reflected waves may therefore combine at the input terminals in almost any fashion, and the efi^ect they produce upon the input current will vary rapidly and violently as we proceed along the frequency scale. With given line impedance and voltage, however, the actual impedance of the terminal is related in simple fashion to the actual current entering it. The im- pedance, therefore, shows correspondingly wide fluctuations. The extremely irregular impedance and reflection coefficient characteristics of Fig. 2 exemplify the effect of reflections from the further junction points of the system. Another example is furnished by the curve of Fig. 7, which shows the reflection coefficient between the actual imped- ance of the filter of Fig. 6 when terminated in the line resistance, and that resistance. The humps of the curve come at frequencies whose phase shift is such that the wave reflected from the far end accentuates the departure of the near end image impedance from the desired value. The valleys correspond either to points at which the image impedances are ideal or to values of filter phase shift which cause the reflections at the two ends of the filter to correct one another. Terminal Impedances Best Corrected by Special Type of Filter Section Close impedance correction of these complicated characteristics seems hopeless. In order to keep the problem within manageable limits it is necessary to destroy the reflected waves at their source by preventing mismatches at all junctions between filters and other ap- paratus within the transmitting bands. The technical problem can consequently be reduced to the construction of a new type of filter section for use at terminations, the image impedance of the new section showing at one end a close approximation to the block type terminating impedance characteristic of Fig. 4:-A or 4-3 (i.e. a constant resistance) while at the other end it has the conventional rounded filter image impedances also shown on these figures, thus matching the standard sections forming the main bulk of the structure. Early Improvements Methods of approximating these characteristics to some extent were already available when the need for reducing crosstalk by impedance correction appeared. The first and longest step in this direction was IMPEDANCE CORRECTION OF WAVE FILTERS 779 made by O. J. Zobel with his invention of "m-type" sections.^ The schematic of a typical low-pass ^ filter terminated with these sections is shown in Fig. 6. The terminating networks are enclosed by the (wm _r: — Fig. 6- -A typical low- pass filter terminated in "w-derived" sections; m The reflection coefficient of this filter is given on Fig. 7. .512. broken lines. At one end these sections match the normal filter image impedance, as in Fig. 4-^. The approximation at the other end to the ideal block type characteristic of Fig. A- A is shown by Fig. 4-C. The actual reflection coefficient of the filter of Fig. 6 is given on Fig. 7. 35 I- Z Ol U a. LU 0. UL bL lU o o z O U UJ _l ti. UJ a. 30 25 20 15 10 m= 0.512 \ ^_^^ \ A / y \j I ' V pj ^c 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500 6000 FREaUENCY IN CYCLES PER SECOND Fig. 7 — Reflection coefficient of low-pass filter shown on Fig. 6. ^ See Bell System Technical Journal, Jan. 1923. ^ The m-type sections are applicable to all types of filters, low-pass, high-pass, or band-pass, and giv-e very similar results in all cases. For example, the curve of Fig. 4-D can be considered as being a combination of two curves like that of Fig. 4-C, with a slight distortion of the frequency scale. If we allow for this distortion in scales the approximation to the ideal characteristic over a given percentage of the transmitted band is the same for low-pass and band-pass filters. 780 BELL SYSTEM TECHNICAL JOURNAL A modification of m-type sections, leading to the so called x"-termin- ations,^ is used when filters must be connected in parallel. The modifi- cation consists essentially in the elimination of the final shunt branches of the m-derived sections at the paralleling junction. Their places are taken in the transmitting band of either filter by the impedance of the (mm Fig. 8 — Schematic of x-terminated filters — showing the way in which the parameter "x" determines the impedance which is added to each filter. attenuating filter. A simple combination of low-pass and high-pass filters, having x- terminations at their common junction and m-type sections facing their load impedances is shown in Fig. 8. The termin- ating network for the low-pass filter consists of the impedance AB and that of the high-pass filter while the network for the high-pass filter 2 Ai U a. ZLU go- UJ -I u. LU TRA NSMIT' 1 rED BANDS 20 1 II 11 1 1 II II 1 / ^x^ /' ^ 10 / N \ \ ,/' \ ^ / ^•x 1 V V ^ 0 ^ ^ 1 v^ ■ vr=> ^^ 12 16 20 FREaUENCY IN KILOCYCLES PER SECOND 24 28 Fig. 9 — Reflection coefficient characteristic of parallel low-pass and high-pass filters from the type " C " carrier telephone system. I — Using x-terminations. II — After the addition of a simple correcting network to the x-terminated filters. is the impedance CD and that of the low-pass filter. The reflection coefficient of a typical pair of low-pass and high-pass filters from the Type C carrier telephone system, terminated similarly to the filters of Fig. 8, is shown by Curve I of Fig. 9. These methods were supplemented by a number of more or less empirical schemes. For example, x-terminations, as Zobel described 3 See U. S. Patent No. 1557230, issued to O. J. Zobel. IMPEDANCE CORRECTION OF WAVE FILTERS 781 them, could be used only with complementary filters (i.e. low-pass and high-pass, or band-pass and band-elimination). In U, S. Patent No. 1616193 R. H. Mills specifically applies the method to band pass filters. Mills, proceeding from the fact that the adjacent sides of two L 1 Q it n^^ W — ^w- 7L IT 1L Fig. 10 — Schematic of band-pass filters with auxiliary network, coefficient of these filters is given by curve I of Fig. 11. The reflection band pass filters behave somewhat like complementary filters, while complementary filters are absent on the further sides of the bands, shunted simple networks, approximating the impedances of the missing complementary filters, across the parallel filter system. The filters of the Type D carrier telephone system ^ incorporated this device. 30 z u o a. Ui Q. 25 20 Z o o u z o I- o u _J \j. us a. 10 TRANSMITTED BANDS \ 1 \ '^ /- / V \ I \ \ \ A \ k / \ \, \j a/ \ / \ \y^ ^ / v . jn m \ s/l 6 7 8 9 FREQUENCY IN KILOCYCLES PER SECOND 10 Fig. 1 1 — Reflection coefficient of a set of band-pass filters. I — For partially corrected .v-terminated filters. II — For uncorrected x-terminated filters. Ill — For filters using the termination of Fig. 12-B modified for parallel operation. * The general engineering features of this system are discussed in the Transactions of the A. I.E. £., Vol. 48, No. 1, pp. 117-139. 782 BELL SYSTEM TECHNICAL JOURNAL The filter schematics are shown on Fig. 10, the auxiliary network being enclosed by broken lines. The performance of the filters was further improved by choosing terminating resistances differing somewhat from the nominal or mid-band, value of the filter image impedance. As a result of these two modifications the reflection coefficient charac- teristic shown by Curve I of Fig, 11 was obtained. Without them, the reflection coefficient would have been that given by Curve II. The great improvement of filter impedance characteristics resulting from these devices is evident from a comparison of Figs. 7, 9 and 11 with Fig. 2. Instead of the reflection coefficients of 50 per cent or 60 per cent found in the earliest filters, the technique allows us to obtain reflection coefficients of the order of 10 per cent, within the frequency range of interest, for filters operating alone, of about 15 per cent for pairs of complementary filters in parallel, and of about 20 per cent for systems of parallel band-pass filters. These results were satisfactory for several years. The continued evolution of carrier sys- tems toward higher and higher energy levels, however, and the constant increase in the number of systems in intimate physical association with one another, gradually made such standards inadequate. The reflec- tion coefficient standards demanded by the severe crosstalk require- ments of these systems have ranged from 2 per cent to 10 per cent in recent filter designs. It became evident some years ago that if these stringent reflection coefficient requirements were to be met a new ana- lytical technique, more general and more powerful than its predecessors, would be necessary. A New Technique and the Results of its Application to Impedance Correction Such a technique has been developed. The method is essentially a generalization of the processes by which Zobel's "x-terminated" filters were derived. It leads to a series of filter sections, the number of which can be extended as far as is necessary to secure a satisfactory approximation to the desired image impedance characteristic.^ The generalized configurations of several sections are given on Fig. 12. The a's and ^'s of this figure are design parameters, Zn, and Zik refer to the filter with which the section is to be used. By choosing Zik and Zih appropriately the terminations can be adapted to any type of filter structure, whether low-pass, high-pass or band-pass. The simplest of these sections can be shown to be equivalent to an "m-type" structure, and will naturally give the same results. A * For a detailed discussion of the theory underlying this technique see, "A Method of Impedance Correction," appearing simultaneously in this journal. IMPEDANCE CORRECTION OF WAVE FILTERS 783 l^LiZ, K|Z| Zi -ll K2Z2 o 0 l^aZ, K,Z, K3Z2 2Z2 ai 7. -» ^I, K2Z2 0 — — < — 0 I2 B ^^aZi iaiz, /-i 1^1 Z| K3Z, 2Z2 21l i ^2Z2 K4Z2 0— — 0 Fig, 12 — Generalized schematics of terminating sections. A — An "jw-derived" or "single-branch" network. B — A "2-branch" network. C— A "3-branch" network. 784 BELL SYSTEM TECLINICAL JOURNAL typical image impedance ^ characteristic of the next more compHcated structure (Fig. \2-B) is shown on Fig. 13. The vertical scale of this figure has been made considerably larger than that of Fig. 4 in order N UJ I.UA 1.03 .02 1. 01 O 1-00 z < I- g 0.99 Q Z o U 0.98 0.97 0.96 0.95 13L 1 ' nr 1 y' / y ;^ / / V ^^-t: --' "^ > / / / 1/ / w 1 > y _ y \ \ \ 1 1 1 1 li 0.04 0.03 0.02 0.01 -0.01 N X UJ O z < 1- Q. LU u -0.02 ^ -0.03 -0.04 -0.05 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 X Fig. 13 — Typical image impedance characteristic of a 2-branch termination (Fig. 12-5). I — Real component. II, III and IV — Various possible imaginary components. 24 UJZ OUJ ou o^ uz Ul ~ _l u. a. 0 16 10 jo— — 1 c 1 G 1 jc T " HH 00 <" ' T' 0 CUT-OFF FREQUENCY 1 J LL _ 12 14 16 18 20 22 24 26 FREQUENCY IN KILOCYCLES PER SECOND 28 30 32 Fig. 14 — Schematic and reflection coefficient of a low-pass filter with 2-branch terminations. ^ The fact that the so-called "image impedances" of Figs. 12 and 16 contain slight imaginary components, in defiance of the fact that the image impedance of a reactive network is never a complex quantity, is traceable to the method used in analyzing the structures, which will be more fully understood from the discussion in the accompanying paper. The curves actually represent the impedance of the section when they are terminated in the filter image impedance. The reason for showing several curves for the imaginary component will be brought out later. IMPEDANCE CORRECTION OF WAVE FILTERS 785 to bring out the departure of the image impedance from its ideal value more clearly. A low pass filter to which the termination has been applied is shown on Fig. 14. The terminating sections are enclosed by the broken lines. The resulting reflection coefficient is given on the same figure. It will be observed that the reflection coefficient over 93 per cent of the transmitting band is less than 2.5 per cent, or only about one-fourth that of the analogous m-type terminated filter shown in Fig. 6. The application of the structure to band-pass filters is illus- trated in Fig. 15, which represents a portion of the redesigned type -aKKKJM rw^ E £ z < 2 |[— ^^5"(KKr a. < Fig. 15 — Band-pass filters with 2- branch terminations modified for parallel operation. The reflection coefficient of these filters is given by Curve III, Fig. 11. "Z>" system. The configuration shown in Fig. \2-B has been modified in these networks to adapt the filters for parallel operation. There are no separate terminations at the receiving ends of the filters since with these very simple structures reflection at the receiving ends could be taken into account by a slight adjustment of the terminating network facing the line. The improvement produced by the new networks in the reflection coefficient characteristics of the system is evident from a comparison of Curves III and I of Fig. 11. The terminating section shown in Fig. 12-B which is one step beyond the "m-type" section in complexity, is adequate in most situations. When a severe reflection coefficient requirement must be met almost up to the filter cutoff, however, it is necessary to resort to the more com- plicated configuration of Fig. 12-C The image impedance character- istic of this section, when its parameters are adjusted for an operating range extending over 97.5 per cent of the theoretical transmitting band, is shown on Fig. 16. The reflection coefficient actually obtained when sections of this type, but with somewhat different values of the design parameters, were applied to a high-pass filter is shown by Fig. 17. The filter configuration is given on the same figure. In this figure the 786 BELL SYSTEM TECHNICAL JOURNAL susceptance controlling network of the high-pass filter at the paralleled end is composed of the coil and condenser in series across the input terminals and the susceptance of the low-pass filter; at the other end 1.03 o N O 1.02 1.0 I Z 1.00 o Q 0.99 z o o 098 0.97 ^ ^> y y \ ,/vi '^ ci "^^ N. \l / / / t \ \ "\ \ \ \ / / / / / j \ \ N / V I 1 V ^ 0.03 002 0 01 o < Q. ■0.01 O 10 10 -0.02 -003 0 0.1 0.2 0.3 04 0.5 0.6 0.7 0.8 0.9 1.0 X Fig. 16 — Typical image impedance characteristic of a 3-branch termination (Fig. 12-C). I — Real Component. II — Imaginary component. « -ft 10 12 14 16 18 20 22 24 FREQUENCY IN CYCLES PER SECOND 26 26 Fig. 17 — Schematic and reflection coefficient of a high-pass filter with 3-branch terminations. of the high-pass filter the three-element two-terminal network controls the susceptance. The conductance controlling sections at either end are composed of the first series condenser and the first shunt coil and a IMPEDANCE CORRECTION OF WAVE FILTERS 787 portion of the second series condenser. The maximum reflection coefficient over about 95 per cent of the nominal transmitting band is slightly greater than 1 per cent. We can summarize these quantitative results in the rough statement that each of the three stages in the prog- ress from the most primitive filter section to the relatively complicated network of Fig. 12-C appears to reduce the reflection coefficient ob- tainable over a given frequency range by a factor of about three or four. Impedance Correction for Filters Operating in Parallel The modifications which must be made in these sections in order to adapt them for use with filters which must operate in parallel are simi- lar to those which were made in adapting "m-type" sections to this service. The final branch of each termination is omitted, its place being taken within the transmitting band of the filter to which it be- longs, by the impedance of the parallel, attenuating, filters of the system. The parallel filters cannot however be relied upon to simulate the missing branch, even in this frequency range, with great accuracy. If we wish to preserve the high standards achieved by the terminations in other circumstances, therefore, it is, in general, necessary to in- troduce an auxiliary network in shunt with the circuit as a whole to improve the approximations to the missing branches. When this is done the reflection coefficient of the complete system is substantially identical in any transmitting range with that which would be obtained from the corresponding filter operating alone. The thorough exploitation of the possibilities of these auxiliary net- works leads to a marked improvement in the performance even of the well-known .r-terminations. The reflection coefficient characteristic of a typical pair of high- and low-pass filters has already been shown by Curve I of Fig. 9. The high value of the reflection coefficient of these filters is largely due to the fact that neither filter in its attenuating range supplies quite enough admittance to take the place of the missing shunt branch of the other filter. The addition of a simple tuned cir- cuit resonating between the transmitting bands to compensate for this deficiency in admittance reduces the reflection coefficient to the level shown by Curve II. The results for x-terminated band-pass filters are even more striking. Fig. 18 gives the susceptance at the line terminals of a set of three filters for several different conditions. The suscep- tance should ideally be zero. Curve I gives its value when no aux- iliary network is added, Curve II, the level to which it is reduced by the auxiliary network suggested by Mills, and Curve III the charac- teristic which can be obtained with the help of a more elaborate aux- iliary network. These curves can be given quantitative significance if 788 BELL SYSTEM TECHNICAL JOURNAL we notice that the deviation represented by Curve I at the point marked by the arrow would lead to a reflection coefficient of about 50 per cent even if the system were otherwise ideally terminated. Curves II and III, under the same conditions, represent reflection coefficients of about 29 per cent and 3 per cent respectively. o N U o z a UJ o ■ D ZA 2.0 1,6 1.2 0.8 0.4 -0.4 -0.8 -1,2 -1.6 -2.0 -2.4 NOTE:- THEORETICAL CUTOFFS INDICATED BY VERTICAL BARS ALONG FRE- aUENCY AXIS. I If \ II, 1\ \\ V. \ I ni, l^ ^ ") ^ Ij ™v t^ '^^ ^^ m ^ N ^ \ <5 \ \\ s \ \^1 1 \ \\ \ 0.9 1.0 1.2 1.3 1.4 FREQUENCY IN 1.5 1.6 ARBITRARY 1,7 UNITS 1,8 1.9 2.0 2.1 Fig. 18 — Susceptance correction of x-terminated band-pass filters. I — Uncorrected Susceptance. II — Susceptance after the addition of a simple auxiliary network. Ill — Susceptance after the addition of a more elaborate auxiliary network. Improvements in Filters for Use with Modulator and Demodulator We have hitherto restricted our attention to the impedance charac- teristics of filters within their transmitting bands since it is only in this range that impedance irregularities in the circuit can produce crosstalk. When a filter operates in conjunction with a modulating device, however, a high modulator efficiency with low distortion de- mands that the impedance of the filter to the untransmitted side band be low (or high) and nearly constant. All of the correcting networks IMPEDANCE CORRECTION OF WAVE FILTERS 789 we have thus far described produce sharp changes in reactance of the attenuating region and are therefore unsuitable for such circuits. In spite of their poor characteristics within the transmitting band, there- fore, it has hitherto been necessary to use mid-shunt image impedance terminations of the primitive "constant-^" type at junctions between filters and modulators. Curve I of Fig. 19 for example, shows the TRANSMITTING BAND "*~ 'to infinity FREQUENCY ATTENUATING BAND Fig. 19 — " Constant-yfe " and "m-derived" type Image impedances in the attenuating range. I — "m-derived" type image impedance. II — "Constant-^" type image impedance. impedance characteristic of an "m-type" section beyond the cutoff in comparison with Curve II, representing the impedance of the "con- stant-,^" type section in this range. The network configurations shown in Fig. 12 however represent only one of two possible classes of sections which can be developed as a result of the general analysis given in the accompanying paper. The other class is radically different in configura- tion. Networks of this second class may be advantageously substituted for the "constant-^" sections formerly used with modulators. The network characteristics in their attenuating regions approximate those of the "constant-^ " sections and while they are not quite as good in the transmitting region as the characteristics furnished by the networks of Fig. 12 they are much better than the characteristics of the "constant k" sections of Figs. A-A and 4-5. 790 BELL SYSTEM TECHNICAL JOURNAL Attenuation of Impedance Correcting Sections Reduces Net Cost of Impedance Correctioji The economic aspects of filter design demand some sort of an evalua- tion of the cost of improving filter impedances. While the terminat- 24 20 Z 16 if) O -I 12 Z o a. Hi z INSERTION LOSS OF ■^"m-DERIVEo" SECTION (m=0.3l9) PLUS -5- CONSTANT -k" SECTION ^ \ ^ ^ _( ? - 1 V ACTUAL INSERTION LOSS OF NETWORK 1 0.9 1.0 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 2.1 X Fig. 20 — Insertion loss of a 2-branch termination. 24 20 SI ? 16 I \^ 1 V v^ INSERTION LOSS OF ^"m-DERIVED" SECTION • (m = 0.450) PLUS -^"constant -k" SECTION 1 \ ^^^^ \ "" ^ CO en O -J 12 / _ Z o ^ 8 UJ CO z 4 0 -4 V I ACTUAL INSERTION LOSS OF NETWORK / 0.9 1.0 1.2 1.3 1.4 1.5 X 1.6 1.7 1.8 2.0 2.1 Fig. 21 — Insertion loss of a 2-branch termination. ing sections shown in Fig. 12 are rather complicated, their cost is dis- counted considerably by the fact that they contribute appreciably to the attenuation of the structure as a whole to undesired frequencies. IMPEDANCE CORRECTION OF WAVE FILTERS 791 Moreover, their attenuation characteristics can be varied within fairly wide limits without appreciably affecting the impedance characteristics we obtain. If we make allowance beforehand for the attenuation of the terminations, therefore, the number of sections making up the main body of the filter can be correspondingly reduced. These rela- tions are illustrated by Figs. 20, 21 and 22 which are drawn for termina- tions having the configuration of Fig. \2-B. The network attenuation is compared in each case with the attenuation of the most nearly equiv- 36 32 28 24 Si T3 ?20 o 2 o i- in z 16 12 \ \ V V .INSERTION LOSS OF -^ "m-DERIVEo" SECTION . (m = 0.63l) PLUS ^"CONSTANT -k" SECTION 1 \ ll ^\ 1 — ACTUAL INSERTION LOSS OF NETWORK-^ / 1 1 1/ 1 10 II 1.2 1.3 1,4 15 1.6 17 18 1.9 2.0 2 I X Fig. 22 — Insertion loss of a 2-branch termination. alent filter structure. The corresponding modifications in the im- pedance characteristic of the network are shown by Curves II, III and IV of Fig. 13. The real component (Curve I) of the impedance is the same in all cases, since the adjustment of the attenuation character- istic was produced entirely by manipulating the final series branch of the network, which has no effect on this component. When low- and high-pass filters are involved the terminating networks contain one more element than the suggested filter equivalent. This much must be conceded to the cost of impedance correction. It will be observed, however, that the remaining elements contribute almost as much at- tenuation as they would in standard filter sections. Indeed at fre- 792 BELL SYSTEM TECHNICAL JOURNAL quencies remote from the cutoff the attenuation of the network consid- erably exceeds that of the filter equivalent. The attenuation produced by the auxiliary (susceptance) networks used in conjunction with paral- lel filters is not so easily evaluated in terms of a standard filter equiva- lent. Since these networks produce peaks of attenuation just beyond the filter cutoff, thus enhancing the selectivity of the systems, they are however, in some respects particularly valuable. We can summarize the economic aspects of impedance correction in the statements that a severe impedance requirement will increase the number of elements (coils and condensers) required for an average filter used in carrier circuits by about 15 per cent or 20 per cent, and that the corresponding increase in the cost of the filter as a whole will be about 10 per cent or 15 per cent. Practical Limitation to Impedance Correction The fundamental limitation on the correction of wave filter imped- ances is practical rather than analytical. In other words it depends upon the accuracy with which it is possible to manufacture filters. All the curves so far exhibited have been based on the assumption that the filter elements, coils, condensers, and resistances, have the 25 Z 20 LU (J a. UJ Q. 15 I- z UJ G ii. u. UJ O u z o I- o LU 10 V \ ^m ^ — - ■^-.^,11 ^ ^^ ^ ^ ^ -. 8 9 FREQUENCY IN KILOCYCLES PER SECOND 10 Fig. 23 — Effect of element variations on a filter reflection coefficient. I — Reflection coefficient when all elements have their exact design values, II — Envelope of reflection coefficients of the best 99 per cent of the filters. Ill — Envelope of worst possible reflection coefficients. IMPEDANCE CORRECTION OF WAVE FILTERS 793 exact values ascribed to them by the design formula. The Hmitations of manufacture, however, demand that elements be permitted to deviate from their mean values by as much as 1 per cent or 2 per cent. These element deviations in general degrade the performance of filters by increasing their reflection coefiicients. The increase in reflection coefficient is largely independent of the initial reflection coefficient, that is, it is about the same for a filter whose normal reflection coefficient is very small as for one of inferior theoretical design having a large normal reflection coefficient. Curve I of Fig. 23 shows the reflection coefficient of a particular filter when all of the elements have their design values; Curve III, the envelop of maximum reflection coefficients for this filter 100 80 to z o Seo —I UJ a. S 40 o a. a. 20 \ \ 0.00 02 % ■ 2 4 6 8 10 12 14 REFLECTION COEFFICIENT REQUIREMENT IN PER CENT 16 Fig. 24 — Distribution of the reflection coefficients of a given filter at one frequency in terms of the percentage rejected in meeting any requirement. when the coils are permitted to vary ±1.5 per cent and the condensers ± 0.8 per cent from their design values; Curve II the maximum reflec- tion coefficient for the best 99 per cent of the filters manufactured to these limits. The last curve is based on a priori probability computa- tion which assumes that the distribution curves of the element devia- tions follow the normal law. A distribution curve of reflection coeffi- cients for another filter at a particular frequency plotted in terms of the per cent of filters that would be rejected in meeting any requirements is given in Fig. 24. Studies of this sort are too extensive to be included in this paper but they have shown that the advances in the technique of impedance correction here recorded are well ahead of the practical limitations of the problem. A Method of Impedance Correction By H. W. BODE This paper gives a theoretical treatment of some recently developed wave filter terminating sections whose application is discussed in the accompany- ing paper on "Impedance Correction of Wave Filters." The sections consist primarily of non-recurrent ladder networks which operate, over the transmission bands of the associated filters, as transformers whose ratio varies with frequency. The transformation ratio of the network is specified, as a function of frequency, by a power series containing a limited number of terms and the design procedure therefore depends upon the construction of power series approximations to the ratio between the resistance of the filter proper and the desired resistance. A separate network is added to secure control of the reactance component. An increased number of terms in the power series, and therefore an improved approximation to the desired transformation ratio, can be obtained by increasing the number of branches in the network. The method thus leads to a series of sections of pro- gressively increasing complexity and with progressively improving imped- ance characteristics. By an inversion of the analysis a second series of sections can also be obtained. The paper is chiefly devoted to a discussion of these two series of filter sections, but other possible applications of the method are also described briefly. THE analysis of transmission circuits with which telephone engin- eers are famiUar is an outgrowth of the general physical theory of the propagation of wave disturbances in continuous media. Problems analogous to the analysis of a smooth transmission line are found, for example, in optical and acoustical theory and in the theory of the vibrations of a taut string. The situations of most importance from the standpoint of general physics are those in which the continuous medium extends indefinitely in at least one direction. Since, moreover, this is also the simplest case, it has been customary to base our trans- mission analysis upon the analogous concept of an infinite line with distributed constants. The analysis of such a structure, since it depends upon only two quantities, the characteristic impedance and the propagation constant, is of course very simple. An actual telephone transmission circuit, however, is by no means an infinite structure containing distributed constants. Many lines, for example, are loaded. Whether loaded or unloaded, they do not ex- tend indefinitely, but are interrupted by terminal apparatus and inter- mediate repeaters. Each of these, moreover, contains a miscellany of apparatus, such as modulators, transformers, amplifiers, filters, equal- izers, by-pass circuits, and the like, having little physical resemblance to a continuous medium. 794 A METHOD OF IMPEDANCE CORRECTION 795 This physical contrast between an ideal continuous medium and an actual physical telephone circuit does not necessarily mean that the application of the wave theory to circuit analysis is a difficult matter. To a first approximation we can determine the response of a circuit merely by adding together the propagation constants of its various constituents. Unfortunately, however, the diverse components of a typical circuit usually have characteristic impedances which are widely different functions of frequency. Thus, for example the im- pedances of the amplifiers and modulators in most telephone systems are nearly constant pure resistances. Non-loaded lines approach such a characteristic at high frequencies but at low frequencies their im- pedance is usually large and may have a considerable reactive com- ponent. Loaded lines depart from a constant resistance at high fre- quencies as well. An even more complicated characteristic, consisting of a varying resistance in the transmitted band, changing abruptly to a pure reactance as we pass the cutoff, is exhibited by a wave filter. In addition to the normal propagation constants of the circuit, therefore, we must take account of reflection effects at all of the junctions between these various types of characteristic impedance. In a long circuit con- taining impedance irregularities at many junctions, moreover, we must give consideration to an enormous variety of waves which suffer multi- ple reflections from a number of junctions. This complicated system of factors may make life burdensome to the man who must evaluate them, but since they are seldom large enough to grossly affect the transmission characteristic of a circuit, they usually play otherwise a secondary role in practical transmission analyses. They do, however, blur the original clarity of the wave picture and from the standpoint of theoretical simplicity at least, therefore, they should be eliminated. For this purpose we should have at our disposal a network whose im- pedances at its two ends could be assigned arbitrarily to match the impedances actually present at any junction. The networks which form the subject of this paper were developed to eliminate reflection effects which, in addition to being a nuisance from the theoretical standpoint, were attended by serious practical conse- quences as well. The engineering problem involved is described in the paper on "Impedance Correction of Wave Filters" by E. B. Payne appearing simultaneously in this Journal. Briefly, it appears from the discussion in that paper that impedance mismatches at the junctions between terminal or repeater equipment of carrier systems and the line give rise to reflected waves which may produce cross-talk in neighboring systems. This cross-talk can be reduced as much as we like by means of line transpositions but the required transposition scheme is so ex- 796 BELL SYSTEM TECHNICAL JOURNAL tremely complicated and expensive that the reduction in the ampUtude of the reflected waves by improvement of the reflection coefficient at these junctions is of considerable economic importance. The imped- ances of the terminals and repeaters at the junctions at which reflec- tions occur are chiefly determined by their filters, which are the ap- paratus immediately facing the line. A detailed study of the relation- ship between the actual input impedance of a filter and mismatches of characteristic impedance which may occur at further junction points in the circuit shows that by far the simplest method of obtaining a low reflection coefficient at the line terminals is to produce a match of characteristic impedances at all junction points of the filter system. Fortunately speech currents beyond the transmitted band of the filters carry so little energy that the reflection coefficient of the structure in these ranges is of no importance. The technical problem therefore reduces to the construction of a new type of filter section for use at terminations, the new filter section having an image impedance within the transmitted band which at one end matches that of the standard sections forming the main body of the structure and at the other ap- proximates a constant resistance, matching the terminating im- pedances. Of course the new filter sections must also be so chosen that they will not impair the transmission properties of the system. This immediate problem has been solved. It still leaves unsettled, however, the question as to whether we can devise a type of network capable of correcting for reflection effects not only at these particular junctions but also at any other impedance irregularity in the circuit. Such a structure would transform one arbitrary impedance character- istic into another preassigned characteristic without decreasing the transmission efficiency of the circuit, much as the familiar attenuation equalizer changes the attenuation characteristic of a circuit by a pre- assigned amount without changing its impedance and without greatly affecting its phase characteristic. The mathematical analysis under- lying the sections which have been developed for filter impedance cor- rection is easily extended to a much broader class of terminating im- pedances. Judged from a purely formal standpoint, therefore, the networks appear to be a long step forward in the development of such a general impedance equalizing device. Unfortunately, it seems certain from other considerations that much of the promise thus inherent in the formal mathematical analysis may not be realized in practical applications, but since the network has been thoroughly studied only in its application to filters, its precise limitations are still uncertain. In the discussion which follows the general method of impedance cor- rection is first sketched briefly, and is succeeded by a detailed treat- A METHOD OF IMPEDANCE CORRECTION 797 ment of its application to filters. Some of the probable limitations of the method in other applications are suggested near the end of this paper. The analysis used in impedance correction can also be applied to the construction of networks having transmission properties somewhat like those of the familiar wave filter. In contrast to the usual filter theory, developed, after the analogy of wave propagation in contin- uous media, from the conception of an infinite recurrent structure, how- ever, it leads to networks which are not recurrent and are not divisible into separate sections with matched image impedances. In its present state of development the analysis is unquestionably much less powerful than the established theory. Since it may be of interest as an example showing at least the possibility of an alternative approach to filter de- sign, however, it is discussed briefly at the conclusion of the paper. General Impedance Correcting Process If no transmission requirements were imposed upon electrical structures, a wide variety of networks might be used for impedance correction. For example, we might make up deficiencies of impedance or admittance by a simple two-terminal network in series or in shunt with the circuit. In almost all circuits, however, we are interested in securing minimum transmission loss, that is to say, maximum energy in the receiving impedance, throughout the frequency bands containing the transmitted signals. The energy which goes into a system term- inated by a correcting network depends only upon generator and the corrected impedance, both of which are specified by the conditions of the problem. We can increase the energy delivered to the receiving device, therefore, only by reducing the amount absorbed in the correct- ing network. Obviously the best possible condition is found when the correcting network is composed of pure reactances. Unless either the resistance or the conductance of the circuit happens to be ideal, how- ever, impedance correction cannot be obtained by a simple two-term- inal reactive network. For this reason, the impedance correcting structures which have been developed are four-terminal networks of pure reactances. Control of the resistance or conductance component is gained, not by the direct addition of resistance, but rather through the use of the network as a sort of variable transformer, whose impedance ratio changes as we go over the frequency range. In such a circuit the insertion loss of the network is determined entirely by the ratio of the energy drawn from the generator by the original and the corrected im- pedance. Ideal dissipationless network elements are, of course, not available in practice. Except for the possible influence of this factor. 798 BELL SYSTEM TECHNICAL JOURNAL impedance correction, since it normally means an improvement in the match between generator and load impedances, should result in a slight increase of transmission efficiency. Reciprocal Impedance Relatioyis at Terminals of a Reactive Network Our restriction to networks of pure reactances allows us to make use of a principle by means of which the impedances measured at the two ends of the network under certain terminal conditions can be recipro- cally related to one another. The theorem will be given here since it is of frequent application in further discussion. Referring to Fig. 1, let us assurne that the impedance measured at terminals cd, with an im- pedance Zi connected to terminals ab, is equal to Z2, as is shown in the diagram. The theorem is concerned with the impedance Z looking into terminals ab when Z2, the conjugate of Z2, is connected across cd. Let us suppose that the generator e in Z2 produces a current i in Zi. Then, by the usual principle of reciprocity, the generator e when inserted in Zi will produce the current i in Z2. In the first case the power entering e^ - the network is obviously -r^ and the power flowing from it into Zi is 4iV2 e^R i^Ri. In the second case these powers are (i?i + Ry + (Xi + xy and i^R2. Since the network is non-dissipative the power entering the network equals the power leaving it in both cases. 4i?2 e^R = i^Ri = i'R, {R^ + Ry + (Xx + xy Upon dividing the two equations and simplifying we find : (R, - Ry + {X, + xy=o which can be true only if: R = RiandX = - Xi In other words, Z is the conjugate of Zi. We can state this result in the following words : A network composed of pure reactatices will have a given impedance, Z\ at one pair of terminals when an impedance Zi is connected to a second set of terminals if, when the conjugate of Z\ is connected to the first pair of terminals, the impedance measured at the second pair of terminals is the conjugate of Zi. A METHOD OF IMPEDANCE CORRECTION 799 This theorem can be appHed immediately to the problem of filter impedance correction discussed in the introduction. The networks re- quired for this problem were defined there as sections which within the transmitting band would have image impedances matching the line at one end and matching the image impedance of the main body of the filter at the other. If we represent the filter proper and the line by Zi and Z2 in Fig. 1, these image impedance requirements reduce to the a o Fig. 1 — Diagram to illustrate the reciprocal properties of impedance correcting networks. statement that the network must be so chosen that an impedance match exists both at ab and at cd. Z\ and Zi for this particular circuit are however, pure resistances, and therefore equal to their conjugates, within the required frequency range. The theorem shows that an impedance match will be obtained at cd provided an impedance match exists at ab, and vice versa. ^ If we please, therefore, we can consider that our problem is that of obtaining a network which, when termin- ated by a filter, has an actual impedance equal to a constant resistance. On the other hand we can start with the resistance and attempt to build up a network whose impedance matches that of the filter. Both the first or "direct" and the second or "reverse" methods of constructing terminating networks for filters are considered in the next section. With either procedure the resulting networks have both required image impedances and can be used, when properly connected, either at the line or the receiving end of the filter. The "correction" of one impedance to match another and the construction of a network having given image impedance characteristics are therefore inter- changeable conceptions. Separate Correction of Real and Imaginary Components of Impedance or Admittance The image impedance method of defining the properties of the term- inating network is a convenient one when we are concerned with the operation of the structure in the transmission system as a whole. The methods used in designing the network can, however, be described * See also Feldtkeller's paper, "Uber einige Endnetzwerke von Kettenleitern " in the Elektrische Nachrichten-Technik, June 1927, for a very similar use of this property of reactive networks. 800 BELL SYSTEM TECHNICAL JOURNAL more simply if we reject the image impedance statement of the problem in favor of its alternative. For the time being, therefore, we will as- sume that we are attempting to design a reactive network having a preassigned input impedance when terminated by a given load imped- ance. In accordance with a principle originally stated by O. J. Zobel,^ this problem of impedance correction can be simplified if we consider separately the resistance and reactance of the corrected circuit. To be more explicit, since a reactance in series with the circuit will change its reactance without changing its resistance, it is simplest to consider first the construction of a network which will produce the required resistance characteristic. Of course the reactance characteristic furnished by such a structure will not in general be ideal, but we may be able to correct it to the proper value by the later addition of a series reactive network. Quite obviously, it is equally easy to base the analy- sis upon admittances and construct first a network which will give the required conductance characteristic and make up any faults in its susceptance characteristic by a final shunting branch. This division of the network into two separate structures is, of course, not a necessary one and in view of the extremely limited range of reactance or susceptance characteristics which can be compensated for by a final, physically realizable, two-terminal reactive network may seem scarcely desirable. An alternative procedure in which this divi- sion is not attempted is mentioned in the concluding section. The reason for assuming separate correction of the real and imaginary com- ponents of impedance and admittance in the present discussion is simply one of convenience. The difficulties which might be antici- pated in the design of the final reactive compensator do not appear in filter impedance correcting problems, at least. On the other hand, the division has the advantage that it makes each step simple and allows us to meet fairly severe impedance requirements with a small number of variables. As we shall see later the method has the further advan- tage in its application to filters that it lends itself readily to the modifi- cations necessary when a number of filters must operate together. The Resistance or Co?iductance Controlling Network Since the characteristics of two-terminal reactance networks are well understood, the construction of the final reactive branch demands no 2 See, for example, U. S. Patents No. 1,557,229 and 1,557,230 where he applies it to "x-terminated" filters. The method of this paper is in some respects merely a generalization of that analysis. The relation of "m-derived" sections and '^x- terminations" to the filter terminations developed in this paper is indicated in the following section. In this connection, the previous work of R. S. Hoyt on loaded lines should also be mentioned. See this Journal, Vol. 3, p. 414, 1924. A' METHOD OF IMPEDANCE CORRECTION 801 extensive discussion. The problem of designing a four- terminal re- active network which will transform one arbitrary resistance or con- ductance characteristic into another arbitrary characteristic must, however, be treated with more respect. The configuration which has been adopted for this purpose is shown in Fig. 2. The quantities of REACTANCE CONTROLLING NETWORK r 1 r- RESISTANCE CONTROLLING NETWORK CORRECTED IMPEDANCE 2 -TERMINAL REACTANCE NETWORK ian-|X •— id3X I idnx IA4X IdiX IdgX IMPEDANCE TO BE CORRECTED .J L. Fig. 2 — Generalized schematic of impedance correcting network. the general form iajX are analytic representations of the impedances of the series branches and admittances of the shunt branches. The a's are constants whose choice determines the particular resistance or conductance controlling properties of the structure, and :x: is a function of frequency. Since the series impedances and shunt ad- mittances are all proportional to x all of the series branches will have a given physical configuration and all of the shunt branches will have the inverse configuration. For example if the series branches are inductances the shunt branches will be capacities, while x, of course, will be proportional to frequency. By using other series arm configura- tions we can obtain a considerable variety of networks. Each such network, it will be noticed, is similar to a "constant-^" filter in physical configuration. The appropriate network in any particular situation is that one which resembles a constant-^ filter transmitting the frequency range of interest. The property of this network configuration which makes it partic- ularly suitable as a resistance or conductance controlling device is the fact that in most instances the modification it produces in the resistance or conductance of the load can be expressed as a single polynomial. To be more explicit, when the load impedance is of a certain mathemat- ical type, which includes the impedances in which we are most inter- ested, the resistance or conductance of the corrected structure is given by a formula of the following sort. F(x) Rior G) = Ao -j- Aix -\- AiX'^ +•••-!- AnX'' 802 BELL SYSTEM TECHNICAL JOURNAL in which the A's are constants involving the arbitrary quantities ai, a^, etc. which specify the network elements. The quantity F{x) is usually either the resistance or conductance component of the load impedance, and in any case is a quantity entirely determined by that impedance. In order to secure the proper resist- ance or conductance from the corrected structure, therefore, it is merely necessary to choose such values of the constants Aq • - • An that the polynomial satisfies the equation . , A , A Fix) Rior G) with sufificient accuracy when R is given the desired value of the cor- rected resistance or conductance. The problem of approximating a given curve by a polynomial of given degree is a well known one in mathematics and such general methods as expansions in power series or Legendrian harmonics exist for its solution. We can, therefore, consider that the choice of these constants presents no particular diffi- culty. Even without the help of these general methods, however, the problem is so simple that suitable approximations can be obtained by cut-and-try methods. These polynomial coefficients Aq • • ■ An are merely intermediate parameters which specify the values of the elements in the network implicitly but do not give them directly. In order to determine the relation between these coefficients and the actual element values it is necessary to make a direct computation of the impedance of the net- work in terms of the a's and sort out the various powers of x in the resulting expression. Each of the quantities Ai--- An is thereby expressed as a function of the a's. Our next step must then be to de- termine values of the network elements by solving the set of simul- taneous equations relating them to the numerical values of the poly- nomial coefficients. In accordance with the procedure we have adopted, the design is completed by the computation of the reactance or the susceptance of the network, and its adjustment to the desired value by the addition of a suitable final branch. The discussion of the application of the method to filter impedances given in the next section will illustrate the process in detail. Prooj of Properties of Ladder Type Resistance Correctors As we observed in a previous paragraph the ratio of the load resist- ance or conductance to the corrected resistance or conductance can be expressed in this simple fashion as a polynomial in x only when the load impedance belongs to a certain mathematical class. Appropriate A METHOD OF IMPEDANCE CORRECTION 803 load impedances are those whose real components can be written as the square roots of rational functions^ of x and whose imaginary com- ponents are rational functions of x. We can make this conclusion plausible by direct inspection. It is obvious that the general nature of the mathematical expression for the impedance of the network can- not change radically as we add successive branches. When we add a series branch, however, the reactance is increased by a,.r, while the resistance is not altered. The functional form of the impedance then will be unchanged if the reactance was originally an algebraic function of X. But, since we must add shunt as well as series arms to the net- work the functional forms must be symmetrical whether taken on an impedance or admittance basis. By analogy, therefore, the suscept- ance also must be a rational algebraic function. The susceptance B is expressed in terms of R and X, the resistance and reactance, by B = XI{R? + X2), but X (and therefore X"^) has already been fixed as a rational algebraic function and R^ must have a similar form if the whole susceptance expression is to be such a function. This conclu- sion, since it applies equally at any part of the network, must, of course, be valid for the load impedance also. This argument is sufficient to indicate what sort of a load impedance might have the property for which we are looking — that of allowing the change in resistance or conductance produced by the insertion of the ladder network to be expressible as a simple polynomial. In order to show definitely that this type of load impedance will have that property it is simplest to begin by finding out whether the relation holds when the network consists of a single branch. In accordance with the previous discussion, the load impedance will be taken as \F2(x) G2{x) ' Fiix) where Fi{x), F2{x), Gi{x), and G-^ix) are polynomials in x. Upon multiplying and dividing the resistance expression by ■^F2{x)C{x), where C{x) is a new polynomial so chosen that when the product F2{x)C{x) is divided by G^{x) the quotient is a polynomial, the load impedance is transformed into ^lFl{x)F2{x)C^{x) . Gijx) _ F{x) .Gi{x) Fiix) C{x) "^ * GaCx) ~ Fiix) C{x) "^ * GgCx) * F{x) is a new symbol, written for ^|Fl{x)F2{x)C^{x), and, as we shall ^ Including as special cases real components which are simply rational functions, without the square root. 804 BELL SYSTEM TECHNICAL JOURNAL proceed to prove, it is the common numerator of all of the resistance and conductance expressions throughout the network. Let us suppose now that the first branch, iaix, of the network is added in series. The admittance after its addition is 1 F{x) Fix) .Gi(x) . F2ix)C{x) ' 'G^ix) F^i ^)^^^)[^ + G^ix) + aix — I- F2{x)C{x) ( -~-~-^ aix F2ix) C{x) \FAx) I F^ix) + ) \G2ix) G,{x) . \ 2 + GiX Upon remembering the way in which C{x) was chosen we observe that the expressions in the denominators of the conductance and susceptance fraction and in the numerator of the susceptance fraction reduce to polynomials. So far we have been able to show that the impedance of the load and the admittance of the network after one branch is added can be so expressed that (1) their imaginary components are rational functions, (2) the numerators of their real components are equal to F{x), and (3) the denominators of their real components are polynomials. It is also possible, however, to show that if these statements hold for the im- pedance and admittance at any two consecutive junctions they will hold also at the next following junction. Referring to Fig. 3, let us ian+iX Yn + i Zn -n-i-2 ^Sn-t-2^ / _F_ • NrvH\ / F , ■ Nn\ ian-ix I lanx Fig. 3 — Impedance and admittance relations at « + 1st branch of network. suppose that the impedance after n branches of the network have been added is __ F{x) . 7V„(x) ^" D„{x)'^' Dn'{x) and that the admittance after w + 1 branches have been added is Fix) iV„+i(.v) "^"^^ Dr^+.ix) '^ ' Dn+lix) A METHOD OF IMPEDANCE CORRECTION 805 We wish to show that the impedance after the addition of the w + 2nd branch is Zn +2 Fix) + i Nn+2{X) Dn+2{x) Dn+iix) The various N's and D's, of course, represent polynomials. The denominator of the imaginary component of Z„ is accented, to indicate that it is not necessarily equal to the denominator of the real com- ponent. The denominators in the Yn+i expression, however, have been given the same designation, since they are equal in the expression we have set up for the admittance at the terminals of the first network branch. This fact is not essential in the proof which follows, but its use somewhat simplifies the procedure. Direct mesh computation gives Fix) Zn+2 — Dn + 1 F' , M+i + Dn-\-l Dn+1 + al^2xWn+i + 2a„+2X — I- [7V„+1 + an+2xDn+i'] D. n+1 F' _^Nl+r Dn+1 Dn+i J + al+2X^D„+i + 2an+2X Since a„+2 is arbitrary, the resistance component will have the speci- fied form only if ^"+^1^, + N' n+l D n+l - is a polynomial in x. If this condition is satisfied the reactance expres- sion can obviously be put in the required form. In order to examine the denominator of the resistance expression more closely we state iV„+i, and Dn+i in terms of iV„, Dn, and Z)„. Direct mesh computation, again, gives F„+i = Fix) -[S + ^^ I + a2+i.x-2Z)„ + lan+ixDn A'n d: — I an+lX -\- jy Dn :.D r F'^ N'^ A N ' ^" D^ + 77^ + «»+i-^''^'' + 2an+,xDn '-^ 1 = ^n [-, +^J + al+ixW, + 2an+ixD„^ 806 BELL SYSTEM TECHNICAL JOURNAL and Nn +1 Dn+\ Dn f2 D •2 ^2 -1 yy^ Substitution of these values for Dn+i and A^„+i reduces the expression for Z„+2 to ^ _ F{x)^ Nn+l 4- a„+2XZ)n+l — I Dn + a^+2.v2-C>n+l + 2an+2XNn+l We have, however, assumed that Dn, Dn+i, and iV„+i were poly- nomials. The sums of the quantities constituting the numerator of the imaginary component of Zn+2 and the denominators of both com- ponents are therefore also polynomials, and, consequently, Zn+2 is written in the specified form. The rest of the proof follows the usual argument from mathematical induction. In brief, we have established directly the fact that the formula holds when the network has no branches, or only one branch. Knowing that it holds for these two cases, we conclude from the above reasoning that it holds when there are two branches. If it is valid for one branch and two branches it must also be valid for three branches, and so on. Therefore the formula holds generally. It will be observed that we have considered the admittance, rather than the impedance, when a series branch is added, and the impedance, rather than the admittance, when a shunt branch is added. Quite obviously the cases not considered are of little interest. If the analysis is stated in terms of impedance a final series branch contributes nothing to the resistance and can be considered as part of the reactance cor- recting network, while an analysis based upon admittances would similarly have no use for a final shunt branch except as a constituent of the susceptance correcting network. The general formula does hold, however, for these cases also. For example the addition of a series branch simply changes one rational function, representing the reactance at the terminals of the previous shunt branch into another rational function. The fact that the impedance at the terminals of the shunt branch falls into our general form is therefore sufficient to prove that the impedance after the series branch has been added can be written in this form also. This indicates, incidentally, that an alternative form of the proof we have been considering, based upon the impedance A METHOD OF IMPEDANCE CORRECTION 807 and admittance relation at a single junction, can be developed. Using the previous notation, the impedance Z„ will be in the required form if Z„+] is in that form, and not otherwise. Instead of assuming that the impedance at one junction and the admittance at an adjacent junction can be fitted into the formula, therefore, it is sufficient to assume that both the impedance and admittance at a single junction satisfy the formula in order to show that the impedance and admittance at the next succeeding junction satisfy this formula also. Application of General Analysis to Filter Impedance Correction The reciprocal property of the impedances at the terminals of a reactive network indicates two possible methods of applying a ladder network of the sort we have been describing to the correction of wave filter impedances. We can either terminate the network by the filter impedance and adjust its parameters to match the line impedance, or we can consider that the load impedance of the network is a constant pure resistance, representing the line impedance, and attempt to pro- duce a match at the filter terminals. These two methods of procedure lead to distinct results, since in one case the reactance or susceptance correcting branch adjoins the line, while in the other it adjoins the filter. Both are, however, admissible under the general mathematical specifications we have set up for the load impedance of the resistance or conductance controlling network and both lead to reasonably satisfactory impedance correction. The fact that a constant pure resistance is an admissible load im- pedance for the ladder network is easily established by inspection. The rational function Gi{x)IG2{_x), representing the imaginary com- ponent, reduces to zero, of course, while the rational function Fi{x)/F2{x), whose square root represents the real component becomes a constant. A filter image impedance within transmission bands is similarly a pure resistance. As a function of frequency it may be defined as the geometric mean of the open and short-circuit impedances. An open or short-circuit filter, whatever its configuration is, however, simply a network of pure reactances. The open and short-circuit impedances are therefore rational functions of frequency and the image impedance they define falls within the scope of the mathematical specification we have set up for the load impedance of the correcting network. Terminating Networks of the First Type While both of these methods of approaching the problem lead to satisfactory impedance correction, other considerations to be discussed 808 BELL SYSTEM TECHNICAL JOURNAL later recommend that one in which the filter is taken to be the load impedance for most designs. This approach will therefore be con- sidered first and in greatest detail. We will, moreover, limit ourselves to image impedances of the "constant-^" type. Practical filter designs of course are usually composite structures containing several types of sections. The image impedances at the junctions between the sections are however, nearly always of the "constant-^" type and our restriction to image impedances belonging to this class does not, therefore, seriously limit the field of application for the network. Notation The image impedance of a mid-series terminated "constant-^" filter is usually written as V 1 + ^'' 4Z2, that of a mid-shunt terminated filter as 1+ ^'^ 4Z: 2*; where Zu and Zo^ represent in each case the series and shunt impedances of the " constant-y^ " filter, and Zo( = VZu-Z2fc) is a constant which can be chosen arbitrarily to fix the impedance level of the circuit. We will find it convenient to represent the way in which the various branches vary with frequency by a new quantity x, defined by the relation Zifc .„ — = tZnX. In a low pass filter, for example, x = f/fc, in a high-pass filter x = fjf, and in a band-pass filter X Upon making use of the relation ZiuZ^k = Z^ the formulae for mid- series and mid-shunt "constant-^ " image impedances can be written as ZoVl - x2 and Zo/Vl - x^.^ This method of representing the image impedances suggests that * In terms of the usual filter notation this x = V — Uk- f fm ~ f ^|■ -4. A METHOD OF IMPEDANCE CORRECTION 809 the configuration of the resistance or conductance controlling network be so chosen that the impedances of its series branches and the admit- tances of its shunt branches are proportional to x. In other words the series and shunt branches of the correcting network should be similar physically to those of the "constant-^" filter. The complete network is then that shown in Fig. 4. It is similar to that of Fig. 2 REACTANCE CORRECTING NETWORK CORRECTED FILTER IMPEDANCE ^i-aaZoX lanX La,ZoX lZ|k(=lZoX) tapX FILTER IMPEDANCE TO BE CORRECTED ^-(=.1^x) Fig. 4 — Generalized schematic of first or "direct" type of filter terminations. except that the explicit introduction of the factor Zq into the expres- sions for the series and shunt branches reduces the a's to constants of proportionality which can be fixed, once for all, for all "constant-y^" filters. Following the analogy of ordinary filter structures it will be assumed that the first branch of the network is in series when the filter proper is mid-series terminated, and vice versa. It is then easily shown that the preceding general formula for the resistance ^ of the system reduces, both for mid-series and mid-shunt terminated filters, to R = ZoVT x^ 1 + Aix^ + A2X' + ^„.x2"' when w is the number of branches in the network. It will be observed that odd powers of x are missing. The possibilities of manipulating this expression to secure desirable resistance characteristics are obviously determined by the number, n, of variable terms in the denominator of the expression. Since n is, however, also equal to the number of branches of the resistance or conductance controlling network, and therefore determines both the cost of this network and the extent to which the resistance or conduct- ance can be made to approximate a given curve, it offers a convenient basis for differentiating between the various structures. The sim- plest cases, and the only ones of practical importance in contemporary filter design, are those for which w = 1, 2, or 3. They are illustrated in Fig. 5 and will be taken up in order. Our first step will be the es- tablishment of the algebraic relations between the element values Oi . . . a„ and the parameters yli . . . ^„for each of these three cases. * Assuming that the final branch, IfianX is in shunt as in Fig. 4. When the analysis is stated in terms admittances the results are precisely similar, except for an obvious change from Zo to 1/Zo. 810 BELL SYSTEM TECHNICAL JOURNAL The analyses are stated in terms of conductance and susceptance, since in this form they are most conveniently applied to the impedance correction of systems of parallel filters, which constitute a large propor- tion of practical cases. The formuke and curves can, however, be used directly in analyses stated in terms of impedance if we merely replace conductance and susceptance by resistance and reactance and write Zo in the numerator rather than in the denominator whenever it GENERAL CONFIGURATIONS OF TERMINATION SUSCEPTANCE ANNULLING NETWORK (n = i) (n=2) APPROPRIATE CONFIGURATIONS OF SUSCEPTANCE ANNULLING NETWORKS K|Z|k I t^2Z2k K|Z|k K3Z2K 1 K2Z2k 1 m. (n = 3) ^S KiZik K2Z2k — c:^ K3Z|k K4Z2K Fig. 5 — Configurations of 1, 2 and 3 branch terminations. appears. After the relations between ai . . . o„and^i . . . ^„ have been determined we shall proceed to a discussion of methods of choos- ing values oi Ai . . . An giving a suitable resistance or conductance characteristic. The final steps are the computation of the element val- ues of the network from these values of the polynomial coefficients, the calculation of the resulting reactance or susceptance characteristic and the design of a final branch giving the complete structure the desired reactance or susceptance characteristic. A METHOD OF IMPEDANCE CORRECTION 811 A nalytical Relations between Polynomial Coefficients and Element Values Case I — w = 1. The general analysis shows that the conductance of the system must be expressible in the form G = 1 Vl X -2 Zo 1 + Aix^ A direct mesh computation of the network of Fig. S-a gives 1 Vl Zo 1 - (1 - a{')x' From which, by comparison of coefficients, ^1 = - (1 - fli^) or ai = Vl + Ai. The susceptance characteristic is given by Zo 1 + Arx-" It can be annulled exactly by the reactance iX = ^ Zo.r + -. — = I — Zr. -\ /.^k ax laix \ zfli / ai where Zik and Z^k are, as before, the series and shunt impedances of the " constant-)^" filter. If the conductance and susceptance controlling portions of the network are combined the resulting structure is identical with a half section of the conventional "m-derived" type. We have merely to replace Oi by m. Single branch conductance controlling networks therefore contribute nothing new to filter impedance correction. Multiple branch networks, which can be considered, if one pleases, as natural extensions of the "m-derived" scheme, must be looked to for the solution of impedance problems for which standard sections are inadequate. Case II — n = 2. A direct computation of the network shown in Fig. S-h gives ^_ 1 Vl Li — Zo 1 + (02^ - 2aia2).x-2 + a^\a,^ - l).v^' 812 BELL SYSTEM TECHNICAL JOURNAL whence Ai = a^ — 2aia2, A, = a^Ka^ - 1), 1 ± Vl +^1 + ^2 V(i ± Vi + ^1 + A^y - A2 ai = v(i ± vrr^i + Aif -"3Z The upper of the alternative signs usually gives the better reactance characteristic. The susceptance of this network is B = — a^x (1 _!)_(! _,,,),. 1 + Axx" + A2X^ Case III — n = 3 The general conductance expression is g^± Vl - x^ where Zo 1 + Aix"" -\- A2X' -\-AzC<^' Ai = ai^ + laias + a^"^ — la^a^ — 1, Ai= a^ai + 2^203 — 201^02^3 — laiChfi^, Az = a^a^ai — a-^a^. These equations can be reduced to fll + «3 — dxChfl-Z = ± Vl + ^1 + ^2 + Az, fli + cfs = Vl + ^1 + 20203, 010203 = V/I3 + a^cii. from which Vl + ^1 + 20203 - V^3 + aiai = ± Vl + ^1 + ^2 + Az. Upon examining the form of the radicals on the left we see that O2O3 is determined by the intersection of a parabola and a hyperbola. A METHOD OF IMPEDANCE CORRECTION 813 Once 02^3 are known the individual values of Ui, a2, and as can be found directly from the previous equations. The two radicals on the left side of the equation must be taken as positive in order to secure positive elements, which is the same as saying that the two conic sections must intersect in the first quadrant. The square root on the right hand side may be taken either as positive or negative, the susceptance characteristic obtained with the negative sign being usually preferable. It is also possible to eliminate two of the as directly, obtaining the equation [Ai" - iAoAs - iAsW + 8^3 Vl +^1 + ^2 + ^3^1^ - [2Ai^ + lA^Az - 4^i^3]a,2 - 8^3 Vl + ^1 + ^2 + ^3 a^ + [(^2 + ^3)2 + 4^3] = 0, which can be solved by standard methods. The former method is shorter, however. The susceptance is given by „ 5o + B,x^ + B^x^ B = — X 1 + Aix"" + A^x^ + Azx* ' where Bq= ai -\- az — a2, Bi = a2 -\- a^a^ — aia^ — laia^az, B2 = a-^a'^az — a^az. Methods of Choosing Power Series Coefficients Having developed the relations between the power series coefficients and the network elements we are now ready to consider methods of choosing the parameters to fit given impedance requirements. Upon rewriting our equation for the real component of the network admit- tance in the form 1 + AxX^ + A2X^ + • • • AnX"^ = vr^ X 2 Z.G we see that the problem reduces to the approximation of the ratio of :^Vl — x^ to the desired conductance G, both of which are known, by means of the polynomial \ -\- Axx^ -\- . . . + Anx"^". In most practical designs the desired filter impedance will be a constant re- 814 BELL SYSTEM TECHNICAL JOURNAL sistance. It is then convenient to rewrite the equation as AnX^'') = Vl - X\ where i?o denotes the desired constant resistance. The problem thus becomes that of simulating Vl — x^ in the range 0 < x < 1 by means of a polynomial in x~ of degree n, and if we assume that the parameter Zo can be chosen arbitrarily the polynomial is completely unrestricted, since the constant term as well as the coefficients of the various powers of X can be taken at pleasure. There are several ways of proceeding from this point. The simplest makes use of the binomial theorem. Upon expanding Vl — x"^ with the help of this theorem we reach the relation 1 1 2 1 4 1 ^ 8 16 Equating corresponding powers of x gives Zq = Rq, Ai= - 1/2, A2= - 1/8, ^3= - 1/16, Using n branches in the conductance controlling network it is possible to take the first n terms of the binomial expansion into account. The elements corresponding to these values of Ai, Ai, etc. can of course be found by the equations derived previously. The results are summarized in the following table. TABLE I Number of Branches A, A. Az fli flo az 1 2 3 - 0.5000 - 0.5000 - 0.5000 0 - 0.1250 - 0.1250 0 0 - 0.0625 0.7071 0.97679 1.00308 0 1.6507 1.96227 0 0 1.62715 The conductance characteristics corresponding to these choices of parameters are shown on Fig. 6. The curve w = 0, which corresponds A METHOD OF IMPEDANCE CORRECTION 815 to the "constant-^" type image impedance, has also been added for comparison. It will be seen from the curves that these values of the coefficients Ai . . . A„ give very good approximations for small values of x, but inferior ones for values near unity. It is preferable in most designs to sacrifice something at the lower end of the characteristic in order to secure better performance in the higher range. N UJ O z ^ o o a z o o tr o 1.2 1.0 0.8 0.6 O N UJ o z < I- (0 in 111 a 0.4 0.2 ^~~ .^^ ^ ^ ^ N V \ \l 1 0.1 0.2 0.3 0.4 as 0.6 0.7 0.8 0.9 Fig. 6 — Resistance and conductance characteristics secured from the binomial expansion. The advantage of an approximation distributed over the band is gained by an expansion in terms of Legendrian harmonics. These functions are discussed in standard reference books, such as Byerly "Fourier Series and Spherical Harmonics" or Whittaker and Wat- son "Modern Analysis." It is important to mention here, how- ever, that they are simply polynomials. Any polynomial such as |5(1 +^ix2 + AnX^") can be broken up into a linear combina- tion of even ordered harmonics, and, conversely, any linear com- bination of even ordered harmonics can be reduced to the form ^ (1 + Aix^ -j- . . . Anx"^"). It is therefore easy to convert an ex- pansion in terms of even harmonics into a power series of the sort with which we are directly concerned. The property of these functions of most interest here is the fact that, for an expansion of any given degree, they give the best "least squares" approximation to the desired function. In the range between x = 0 and x = 1, therefore, the approximation they furnish is much better for most purposes than that given by the binomial theorem. The expansion of \1 — x'^ in 816 BELL SYSTEM TECHNICAL JOURNAL terms of Legendrian harmonics is given on p. 184 of Byerly as vr^. = 5[ip.(,,_5(i)(i)Va.)-9(^)QJ=P.w Upon replacing the harmonics by their values in terms of x, — Poix) = 1, P,{x) = \ {^x^ - 1), F^ix) = I (35x4 - 30x2 _^ 3)^ o 16 p^(x) = j2 (231x4 - 315x4 + 105x2 3), and sorting out the various powers of x, values of the coefficients Ai . . . An are secured, and from these the actual element values are found by means of formulas developed previously. The following table summarizes the results TABLE II Number of Branches Ro Ko A'l K2 K, 0 1 2 3 1.273 0.9699 1.011 0.9948 0.7855 0.7855 0.7855 0.7855 0 - 0.4909 - 0.4909 - 0.4909 0 0 - 0.1105 - 0.1105 0 0 0 - 0.04986 Number of Branches Ai A2 ^3 ai a2 fls 0 1 2 3 0 - 0.7142 - 0.3236 - 0.0461 0 0 - 0.4884 + 0.4958 0 0 0 - .7162 0 0.5546 0.8986 0.9597 0 0 1.593 1.924 0 0 0 1.565 The quantities Ko . . . Kz are the numerical coefficients of the cor- responding harmonics. It will be observed that with this method of determining the network parameters Zq is not quite equal to Ro. When the analysis is based upon impedances instead of admittances the ratio Ro/Zo should be replaced by Zo/Ro. The conductance char- acteristics secured by this process are shown in Fig. 7. A METHOD OF IMPEDANCE CORRECTION 817 It is, of course, always possible to dispense with these general methods entirely and make an empirical determination of the design parameters. The particular requirements of specific design projects 1,3 1.2 I.I 1,0 I ° N O z < (- If) a. O '^ N u o 0,9 0.8 0.7 0.6 0,5 O D Q Z 0.4 O O 0.3 0.2 0.1 n=o ~— ■v. \ \ n=2 ^ -^ > ^ -WT^ \, _ \ n= 1 "\ \ REAL COMPONENT ^ \ \ ^ . \ \ \ 0.1 0.2 0.3 0.4 0.5 0.6 X 0,7 0.8 0.9 1.0 Fig- 7 — Resistance and conductance characteristics secured from expansion in terms of Legendrian harmonics. are thereby given the fullest recognition. This method was used in constructing the sections described in the accompanying paper. Even when the empirical method is adopted, however, the networks de- termined by the general expansions, particularly that in terms of Legendrian harmonics, should be valuable as starting points. 818 BELL SYSTEM TECHNICAL JOURNAL In most designs it is desirable to make the maximum departure from the ideal characteristic within the operating range as small as possible. A method of doing this for the 2-branch networks has been developed. The method assumes that the Zo of the filter has been taken equal to the terminating impedance, which assures a correct conductance at the point x = 0. The manipulation of the parameters Ai and A2 allows us to secure the desired value of conductance at two additional points. The result is a two looped characteristic, similar, if we make allowance for the difference in the assumptions regarding Zo, to that already determined for this network by means of the Legendrian expansion. The requirement that the maximum departure from the ideal within the operating range be a minimum is equivalent to saying that the amplitudes of the downward and upward loops must be equal. It can be shown that a 2-branch conductance network will satisfy this condition if 27 . 1+^1+^0 = (^l2 - 4.42)2. In view of the relations which have been developed between Ai, Ai, Fig. 8 — Design chart for 2-branch termination. A METHOD OF IMPEDANCE CORRECTION 819 and ai, a^ this condition can also be written in the form ^ (1 - ai^) = a22[4(l - a,a,y + a,K\ - a,a,)J. A second condition upon these quantities is found by specifying the range within which the impedance is to remain as flat as possible. The results of computations to determine this relationship are given in Fig. 8. x^ in this diagram signifies the highest value of x in the operating range. Fig. 8 also gives the maximum departure of the conductance characteristic from its ideal value as a function of Xq. Numerical data taken from these curves should of course be confirmed by the equations given herewith before they are used to specify element values. Susceptance Correcting Networks ^ Once the conductance controlling portion of the network has been determined by one or another of these methods our general procedure calls for the computation of the susceptance characteristic it furnishes and the design of a final shunting reactance network which will annul o N X UJ o z < (- Q. UJ o w 3 10 cc o 1.4 1.2 1.0 0.8 0.6 |n° z < < UJ a. 0.4 0.2 n = 3 y / /, / A / ^ ^ ^ ^^ -^ 0.1 0.2 0.3 0.4 0.7 0.8 0.9 1.0 0.5 0.6 X Fig. 9 — Reactance and susceptance characteristics secured from binomial expansion. ^ This section gives only a general description of the characteristics required of the susceptance correcting networks and the configurations which have been found appropriate for them. The design of these networks may be conveniently approached by means of the formulae contained in R. M. Foster's article "A Reactance Theorem," in the Oct. 1924 issue of this Journal. 820 BELL SYSTEM TECHNICAL JOURNAL this susceptance to a suitable approximation. Fortunately the characteristics required of this network are of a type which can readily be obtained with physically realizable elements. The curves of Figs. 9 and 10 represent the susceptance characteristics required for the UJ u z < u < cr o '^ N X UJ u 0. llJ o it) D 1.4 1.3 1.2 I.I -In i.o 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 n = M T / //n = 3 / / f IMAGINARY COMPONENT // / K / 1 A f / ^ y. y y -^ y y^ ^ ^ > ^ 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Fig. 10 — Reactance and susceptance characteristics secured from expansion in terms of Legendrian harmonics. Legendrian and binomial expansion networks. Empirically deter- mined networks give very similar results. The general configuration of appropriate susceptance correcting networks can be determined from an inspection of these curves. For example, if we assume that a low pass filter is in question, which means that the variable "x" A METHOD OF IMPEDANCE CORRECTION 821 is proportional to frequency, the desired susceptance curves will be recognized as being approximately those which would be obtained from tuned circuits resonating slightly beyond the cutoff. Since a tuned circuit can be considered as being a series combination of the series and shunt impedance of the "constant-^" filter, any such correcting network designed for a low-pass filter can be adapted to 1.6 1.4 1.2 o N 1.0 z < 0.8 0. UJ o D if) 0.6 0.4 0.2 \ \ n //I 1/ jy // / 0.2 0.4 0.6 X 0.8 1.0 Fig. 11 — Susceptance correction of a 3-branch termination. I — Desired susceptance. II — Susceptance actually obtained. another type of "constant-^" structure by replacing inductances and capacities by the homologous impedances of the other filter. This simple combination of series and shunt impedances is, as we have previously seen, capable of giving exact susceptance correction when the conductance controlling network contains only one branch, but it is not, in general, sufficient for 2 and 3 branch networks. Indeed, no physically realizable reactive network will cancel the susceptance 822 BELL SYSTEM TECHNICAL JOURNAL furnished by these more comph'cated structures exactly. Close approximations however can be obtained by modifying the "tuned circuit" characteristic slightly through the introduction of extra elements. Suitable configurations for 2 and 3 branch networks have already been given in Fig, 5. They should furnish susceptance characteristics at least as good as the corresponding conductance characteristics. An example of the susceptance correction of a three branch network, using the configuration of Fig. 5-c, is shown in Fig. 11. Curve I represents the ideal susceptance characteristic, Curve II that actually obtained. Impedance Correction of Paralleled Filters An interesting modification of the process of susceptance correction occurs when a number of filters are to be connected in parallel. Since the impedance of an attenuating filter is almost a pure reactance the conductance component of a system of parallel filters at a given fre- quency is furnished almost entirely by the filter in whose transmission band that frequency lies. If the system as a whole is to have the correct conductance throughout each transmission band, therefore, every filter must be given the conductance controlling network which would be appropriate if it were operating alone. While the process of conductance correction is thus exactly the same for multipled and individual filters, the process of susceptance correction of paralleled filters must be modified somewhat to take account of the susceptance component furnished by the attenuating filters. A single susceptance network will serve for the whole system. We have merely to compute the susceptance characteristics furnished by the various filters termin- ated in their conductance controlling networks and annul them through- out every transmission band by a two terminal network in parallel with the system as a whole. An example of the application of the method to a pair of parallel complementary filters having 2 branch conductance controlling networks is given by Fig. 12. Curve I in this diagram represents the susceptance of the transmitting filter, Curves II the susceptance of the attenuating filter for several different choices of its cutoff frequency. Curves III the susceptances of the corresponding auxiliary networks, and Curves IV the net result. A series combina- tion of the series and shunt impedances of either filter ^ resonating at the geometric mean of the cutoff frequencies was chosen for the ^ Since the filters are complementary the series impedance of one is similar to the shunt impedance of the other, and vice versa. By choosing the resonance frequency of the auxiliary network symmetrically with respect to the two filters, as we have done, all of the susceptance relations become symmetrical, and the network functions as well for one filter as it does for the other. A METHOD OF IMPEDANCE CORRECTION 823 auxiliary network. By using two resonant arms with closely adjacent resonance frequencies still better susceptance correction could have been secured. Filters which must operate in parallel are usually given x-termina- tions. Since an x-termination can be thought of as being a one element 1.2 1.0 0.8 0.6 04 o N 0.2 o 0 < t- 2;-o.2 o tr> z> -0.6 -0.8 -1.0 -1.2 •1.4 / A / ') y ^ h /^ f > A ni ^ A V \ N, ^} IZ N \ \ \ \ \ \ \ \ 0.2 0.4 0.6 X 0.6 1,0 Fig. 12 — Susceptance relations at the line terminals of a pair of parallel complementary filters having 2-branch conductance controlling networks. conductance controlling network the method we are discussing can be applied here also. It is interesting to note that the introduction of an auxiliary susceptance controlling network considerably improves the performance even of this well known circuit. The susceptance rela- tions at the line terminals of a pair of parallel complementary .r-term- inated filters are shown in Fig. 13, the arrangement of the curves being similar to that of Fig. 12. The improvement can be estimated from the magnitude of the auxiliary susceptance. 824 BELL SYSTEM TECHNICAL JOURNAL The auxiliary network improves the susceptance of parallel band pass filters even more than it does that of complementary filters. Curve I of Fig. 14 represents the susceptance of a typical uncorrected set of band pass filters. The first step in the improvement of this characteristic is due to Mr. R. H, Mills, who suggested that networks whose impedances resemble that of filters above and below the actual set of bands be added to the system. This reduces the susceptance + 1.0 +0.8 - +0.6 +0.4 + 0.2 o U Z a. -0.2I- UJ o W -0.4 -0.6 -0.8 -1.0 -1.2 -1.4 1 r 1 "A" iigii /} n ~ * llf-ll / / / / in ^ / i< k ly ^ y ^ 1 V \ =-'V N \ s \ \ \ \ \ I .2 10 1.2 •(■iSfi) Fig. 13 — Susceptance relations at the line terminals of a pair of parallel complementary x-terminated filters. to the level shown by Curve II. Curve III gives the completely corrected characteristic. The auxiliary susceptance correcting net- work consists of a number of tuned circuits in parallel, one resonating between each pair of successive bands, together with one resonating above the topmost band and one resonating below the lowest band. The insertion of the auxiliary network has the further advantage that it produces peaks of attenuation near the cutoffs of the filters, thus enhancing their selectivity. A METHOD OF IMPEDANCE CORRECTION 825 o N 2A NOTE:- THEORETICAL CUTOFFS INDICATED BY VERTICAL BARS ALONG FRE- aUENCY AXIS. 2.0 I 1.6 1.2 °-. 0.6 v^ \ 1 0.4 m. V^ ^ \ % V t ™x 0 t '-^V n\\ m x^ \. N V- "^X -0.4 \v A \ ^ ^. w N V 'V \ -0.8 -1.2 -1.6 \ \ I^ \ \ \ V\ 1 \ \ -2.0 -2.4 0.9 1.0 1.2 1.3 1.4 1,5 1.6 1.7 FREQUENCY IN ARBITRARY UNITS 1.8 1.9 2.0 2.1 Fig. 14 — Susceptance correction of a set of parallel x-terminated band-pass filters. I — Uncorrected susceptance. II — Susceptance after the addition of a simple auxiliary network. Ill — Susceptance after the addition of a more elaborate auxiliary network. Reverse Method of Designing Terminating Sections Hitherto we have assumed that the load impedance of the terminat- ing network was the filter image impedance, and our procedure has consisted essentially in determining an adjustment of the network parameters which would make its input impedance a constant pure resistance. As we have already seen, however, it is equally legitimate to assume that the network is terminated in the line resistance, and determine parameter values which will produce a match between its impedance and that of the filter. This assumption leads to the circuit arrangement shown in Fig. 15. Upon examining what happens to the general expression for the resistance of the network when the load impedance reduces to the 826 BELL SYSTEM TECHNICAL JOURNAL constant pure resistance, R^, we easily find that it turns out to be n ^ ^0 1 + AiX" + AiX' + . . . + AnX"'' ' where n is the number of branches in the network. Odd powers of X are missing, just as they were when the network was terminated in a filter impedance. REACTANCE CORRECTING NETWORK IdaX I LdnX ia|X idax TERMINATING' RESISTANCE Fig. 15 — Generalized schematic of second or "reverse" type of filter terminations. Our problem consists in matching this expression to the filter Upon assuming that i?o = ^o, for simplicity, . . ^„ which will impedance, ZoVl — .%•- we see that it reduces to the selection of values of ^i secure approximate satisfaction of the equation 1 1 +^i.v- + + ^n.V-" Vi^ X- Two empirical ^ choices of these parameters have been made, one ^ Our previous methods of approximation, in terms of Taylor's series and Legen- drian harmonics, are of course available here also. In addition, if we rewrite the expression as Vr^T^ I? (1 + ^i.r2 + . . . + Ar^x^-) = 1 -fvo the left hand side appears as a linear combination of the associated Legendrian functions Pi'(x), Pz'ix), . . ., defined by the general formula Pn'{x) = Vl -X-^Pn{x), ax where P„(x) is the usual Legendrian function. The problem can therefore be con- sidered as that of approximating unity by a series of the associated functions. These methods of approach differ chiefly in the relative weights which they ascribe to various portions of the frequency band. Judged by this criterion neither of the first two methods is very satisfactory for practical applications. The Taylor's series expan- sion, of course, is best in the neighborhood of x = 0. The "least squares" property of the ordinary Legendrian functions, on the other hand, tends to produce rough equality in the numerical values of the departures from the desired function in various portions of the frequency band. From the engineering standpoint, however, it is the percentage departure from the desired impedance, and not the numerical departure, which is of interest. This type of approximation therefore leads to a relative over-emphasis of the region near x = 1, where the desired function 1/V 1 — x- is large. The approach by means of the associated functions, however, avoids this objection, since the approximated function is in this case a constant, and leads to characteristics substantially as good as those obtained by means of the empirically determined parameters discussed in the text. A METHOD OF IMPEDANCE CORRECTION 827 iagZoX FILTER ' - 4^ " < < < Zo idsX Zo ld|X < Vl-X2 < < < < < Fig. 16 — A 2-branch termination of the "reverse" type. 1 FILTER ia4ZoX id2ZoX < < < -. ZoVl-x2 Zo idsX Zo idiX < < < < < < Fig. 17 — ^A 3-branch termination of the "reverse" type. 1.0 0.9 0.8 M° 0.7 u o z < t- O 0.6 < UJ Q 2 0.5 < o X 0.4 O z < 0.3 W (0 UJ q: 0.2 + 0.1 -0.1 >Kir F RESlSTAi-.w^ . N \ \ \ , \ \ , 1 / ,/ ^ ' ~ REACT/ ^NCE -^ •' 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Fig. 18 — Impedance characteristic secured from the network of Fig. 16. 828 BELL SYSTEM TECHNICAL JOURNAL when the network contained two branches, and the other when it contained three. In both instances the appropriate reactance or susceptance annuHing networks were found to be simple arms, similar to the series or shunt branches of the remainder of the termination in physical configuration. The complete networks are shown in Figs. 16 I.I 1.0 0.9 0.8 -|n° X , o 2 < y 0.6 < UJ cc a z < 0.5 N° UJ (J Z < (0 V) ui a. 0.4 0.3 0.2 + 0.1 -0. RESIS TANCE \ \ / _. BEAC" rANCE 0.1 0.2 0.3 0.4 0.5 0.6 X 0.7 0.8 0.9 1.0 Fig. 19 — Impedance characteristic secured from tiie network of Fig. 17. and 17, where the final branches, Zo/iasX in Fig. 16, and iatZoX in Fig. 17 are the susceptance or reactance annulling networks. The values of the various parameters are given in the following table. TABLE III n A, A, As ai 02 as at 2 3 + 0.0505 + 0.9114 + 1.6508 - 1.8488 0 + 3.2823 0.7973 0.6733 1.6186 1.466 0.904 1.835 0 0.925 A METHOD OF IMPEDANCE CORRECTION 829 If a perfect match were secured at the filter terminals then, by the reciprocity principle, a perfect match should be secured at the line terminals also. In order to evaluate the performance of the networks, therefore, the impedances they present to the line were computed. The results are shown in Figs. 18 and 19. Comparison of Direct and Reverse Networks At first glance the curves of Figs. 18 and 19 seem to show that while networks of the reverse type produce a good impedance match over a moderate fraction of band they will be much less successful than the structures previously described at frequencies very near the cutoff. This apparent advantage in favor of the networks first described is discounted considerably however by the economy of elements resulting from the relative simplicity of the reactance or susceptance controlling networks used with terminations of the second type. If we adopt as our standard in comparing the two types of networks the total number of elements each requires, rather than the number of branches they contain, the advantage of networks of the first type becomes much less impressive, if it does not actually disappear. More important considerations recommending the first type of terminations in prefer- ence to the second for most practical designs appear to be the greater ease with which they can be designed to meet a given reflection coefficient requirement, resulting from the relatively smaller number of branches they contain, the greater ease with which they can be adapted to filters which must operate in parallel, and the fact that the attenuation they contribute to the total filter suppression is usually more useful than that furnished by terminations of the second type. Under certain circumstances, however, the second type of terminat- ing sections have a definite advantage over the others. When a filter operates in conjunction with a modulating device a high modulator efficiency with low distortion demands that the impedance of the filter to the untransmitted side band be low (or high) and nearly constant. In spite of their poor characteristics within the transmitting band it has hitherto been necessary to use mid-shunt image impedance termina- tions of the "constant-^" type in these circuits. Impedance correcting sections of the first type are not suitable for this service because the complicated susceptance and reactance annulling networks at their line terminals produce sharp changes in reactance in the attenuating region. The outermost branch of terminations of the second type, however, is of simple configuration and if we choose it to resemble the final branch of a mid-shunt terminated "constant-^" type filter, as has been done in the sections shown in Figs. 16 and 17, we will secure 830 BELL SYSTEM TECHNICAL JOURNAL an impedance characteristic beyond the band almost as good as that of the "constant-yfe" filter. Within the transmitting band, of course, its impedance is much better than that of the normal filter section. Attenuation Characteristics of Terminating Sections In the practical application of either type of terminating section some others of their characteristics, such as their transmitting efficiency and the effect produced upon them by parasitic dissipation of energy in the network elements, are also of importance. The transmission characteristics of the networks can be determined roughly by com- paring them with standard filter sections. Let us consider, for exam- ple, the two branch termination shown in Fig. 20. If we neglect for / \ \ ^aaZik 1 1 1 1^1 Z|k 1^3 Z2k 2Z2k ^^2Z2k ^^-O o \ /\ / ^OF m- -derived'^^ection'^ 4 OF '"constant -k" SECTION Fig. 20 — Figure illustrating approximate transmission characteristics of 2-branch terminations. the moment the third element of the susceptance correcting network, the remainder of the structure can be divided, in the manner indicated by the broken lines, into two portions, one of which resembles half of a "constant-/^" section and the other half of an "m-derived" section in physical configuration. The transmission characteristic of the actual network is substantially similar to that which would be furn- ished by standard filter sections of these types. The mere fact that the network functions as an impedance corrector is, of course, sufficient to show that it will transmit efficiently frequencies within the nominal transmission band of the filter. Beyond the transmission band the attenuation characteristic would be almost exactly coincident with that of the suggested filter equivalent if it were not for the extra element in the final shunt branch. The extra element produces an anti- resonance in this arm somewhat beyond the resonance and near the anti-resonance point the attenuation is somewhat less than that which would be secured from ordinary filter sections. On the other hand the A METHOD OF IMPEDANCE CORRECTION 831 extra element considerably increases the admittance of the final shunt arm, and therefore the attenuation of the network, at frequencies remote from the cutoff. In spite of these modifications the analogy to standard sections is a fairly trustworthy guide to the attenuation of the networks. Several examples are given in the accompanying paper. Since the ideal pure reactances contemplated by the theory are not physically available these conclusions must be modified somewhat in practical designs. As we might expect, however, unavoidable dis- sipation of energy in the network elements will alter the transmission characteristic of the correcting device about as it would that of an ordinary filter. In the attenuating range the effect can be neglected. In the nominal transmission band absorption of energy in the termina- tion will reduce the transmitting efficiency of the circuit somewhat, but the loss in efficiency is no more serious than it would be in standard filter sections having the same general configuration. Parasitic resistances in the network elements may of course affect the impedance as well as the transmission properties of the circuit. Since the structure is used primarily because of the impedance char- acteristic it furnishes, possible changes in impedance, caused by varia- tions in the phase angles of the network elements, are of particular interest. Changes in impedance produced by dissipation of energy in the correcting networks, are easily estimated when the complete circuit with whose impedance we are concerned can be considered as a network of ordinary resistances, inductances and capacities and when dissipation affects the phase angles of all reactive elements equally. It can be shown that in such a network the change produced by dissipa- tion in the resistance of the structure is proportional, to a first approxi- mation, to the derivative of its reactance characteristic with respect to frequency, and that conversely the change in the reactance character- istic is proportional to the frequency derivative of the resistance characteristic. The explicit formulae are: AX=-/.f, where /is frequency and d the dissipation constant (defined as ratio of resistance to reactance) for each reactive element. A filter, with its terminating sections and load resistance, is a net- work of resistances inductances and capacities to which the theorem applies. It seldom happens of course, that all of the reactive elements 832 BELL SYSTEM TECHNICAL JOURNAL of the structure actually have the same dissipation constant. It is usually sufficient, however, to assume that "(/" in the above formulae is the average of the dissipation constants for coils and condensers. When well designed impedance correcting networks are used the react- ance and resistance characteristics of the structure will be approxi- mately constant over the operating range. The derivatives occurring in the above formulae will consequently reflect only the presence of slight ripples in these characteristics about their mean values. The slopes of these ripples will usually be quite small. We can therefore conclude that moderate amounts of dissipation will have no appreciable effect upon the impedance of a properly terminated filter. The chief exceptions to this rule occur in low pass filters, where, at low frequencies the assumption that the dissipation constant is small is no longer satisfied. In attempting to extend this principle to broader problems in impedance correction it is, of course, necessary to bear in mind that the analysis holds only for networks of resistances, inductances and capacities. We cannot expect the same results when the load im- pedance of the circuit has some arbitrary variation with frequency. For example, if we take the load impedance as the image impedance of a dissipationless "constant-^" filter and assume that parasitic re- sistances occur only in the termination, we will find that dissipation does change the impedance of the circuit. The circuit impedance will be insensitive to dissipation only when we include the complete structure, and not merely the terminations, in our analysis. More General Problems of Impedance Correction This general method of impedance correction having worked with reasonable success in its application to "constant-^" wave filter impedances, it is natural to inquire whether it can be applied to other problems with equal ease. Further possibilities for example might include the correction of other types of filter impedances, or the cor- rection over extremely wide frequency bands for the efi^ects of leakage inductance and finite mutual inductance in transformers, or the reduc- tion of actual transmission line impedances to constant resistances. All of these possible applications assume that the impedance correcting device is a 4-terminal network, transmitting useful signal energy to its load impedance. When terminated by such an element as a simple resistance, however, it might also be used as a 2-terminal network, forming one branch of a complete system. By appropriate adjustment of the impedance controlling parameters the network could, theore- tically at least, be given a wide range of impedance characteristics. METHOD OF IMPEDANCE CORRECTION 833 We might, for example, use it to approximate a pure resistance varying in an arbitrary manner with frequency, which would be a valuable impedance element in certain circumstances. None of these possibilities has been investigated in detail, and naturally the measure of success which can be achieved with any one of them will depend largely upon the precise conditions of the problem. The mathematical form we have specified for the load impedance of the network is so broad however that if we were to consider only this aspect of the situation we might conclude that the scope of the structure is well nigh universal. For example, the impedance of any finite network of resistances, inductances, and capacities can be written in the appropriate mathematical form. Even when the load impedance is not described in the required manner, either because it is empirically determined or because it has the wrong theoretical formula, the type of algebraic expression we have been considering is so general that it can always be matched approximately. Unfortunately, the range of application promised by this rather superficial mathematical discussion may be severely restricted by other considerations. In the general case, for instance, the number of terms in the denominator of the resistance expression will be greater than the number of branches in the correcting network and it will not be possible to choose them all arbitrarily. Moreover, even when the correct number of conditions have been imposed upon the power series coefficients we have no assurance that the resulting system of simultaneous equations between coefficients and element values can be solved, or that the solutions, if obtained, will always correspond to physically realizable elements. Finally, we may observe that although no difficulty was experienced in the reactance or susceptance correction of filters, it seems probable that, in view of the limited range of char- acteristics which can be simulated by physically realizable reactive structures, a straightforward application of the general method of resistance correction will often leave us with a reactive characteristic which cannot be corrected. These difficulties may occasionally be overcome by slight modifica- tions in the design process. Among other possibilities for example, we can adjust the lowest powers both in the denominator of the resistance expression and numerator of the reactance expression ^ to desirable values, obtain an approximate value for the effect of higher powers in both expressions by a trial computation and readjust the coefficients of the lower powers to take account of these previously neglected terms. 8 Since the denominator of the reactance is equal to that of the resistance, whose value is prescribed by the requirements, the reactance expression can be determined completely from its numerator alone. 834 BELL SYSTEM TECHNICAL JOURNAL Difficulties appearing in a direct application of the impedance cor- recting process may also be avoided if we adopt the reverse method of impedance correction suggested by the theorem on reciprocal im- pedance relationships. The method has already been applied to the construction of alternative filter impedance correcting sections. Similar alternative configurations can be built up in any impedance correcting problem if we consider that the structure is terminated by the conjugate of the desired impedance and adjust its parameters to produce the conjugate of the given impedance. Since the desired impedance will in general be a relatively simple function of frequency, this alternative procedure at least avoids analytical complexities. In spite of these possibilities however it seems probable that the method will fail in many situations. It seems best adapted to such problems as that of filter impedance correction, where a transformation must be made from one fairly simple characteristic to another simple character- istic. An attempt to apply it to more difficult problems should result, at best, in very complicated networks. Transmission Properties of Impedance Correcting Networks The close relationship between the impedance correcting properties of our networks and their transmission characteristics has been mani- fest from time to time in the previous discussion. The networks used at filter terminations, for example, transmitted freely within the range in which they functioned satisfactorily as impedance correctors but attenuated other frequencies. That this will be true in general is easily seen by inspection. Within the range in which a desired im- pedance characteristic is obtained, of course, our previous argument from the principle of conservation of energy is alone sufficient to show that the networks transmit with the maximum efficiency compatible with the impedance requirements imposed upon the circuit. On the other hand, it is evident from the filter-like configuration of the net- works that at frequencies remote from the operating range of the networks, where the parameter "x" becomes large, the structures will ordinarily introduce attenuation. From the impedance standpoint this means merely that for sufficiently large values of x the polynomial approximations upon which the analysis is based no longer hold, and the resulting mismatch between the generator and network im- pedances diminishes the amount of power which can enter the struc- ture. When the impedance correcting analysis is stated in a slightly modified form, whose possibilities have not as yet been completely investigated, the impedance and transmission characteristics of the A METHOD OF IMPEDANCE CORRECTION 835 circuit are still more firmly related. Thus for example the attenuation of the structure beyond its operating range results chiefly from the readily computed departure of the resistance or conductance char- acteristic of the network from that of the generator. It is also pro- duced, in part, however, by the failure of the reactance or susceptance correcting network to annul in this range the imaginary component furnished by the resistance or inductance controlling network and the effect of this factor is less easy to determine. In the modified analysis it is often possible to do away with the distinction between the two types of networks. The complete insertion loss characteristic is then embodied in a single polynomial expression. In the modified form, moreover, the analysis may often be used to determine the phase as well as the attenuation of the circuit. Granted these results, it is but a short step to the conclusion that the impedance correcting analysis offers a possible approach to the design of filters. While it is usually true that the networks will attenuate frequencies beyond the region in which impedance require- ments have been set, the amount of the mismatch which produces this attenuation, since it depends upon the impedance correcting parameters, is still more or less under our control. By suitable adjustments of the correcting network, therefore, we can design a structure to meet attenuation as well as impedance requirements. A particularly interesting situation occurs when the load impedance is a constant pure resistance.^" As we have already seen, a load impedance of this type satisfies our mathematical specification and it can therefore be used with a ladder network. Since a perfect impedance match already exists in the circuit an inserted network can be called an impedance correcting device only by courtesy. Unless the network contains so many branches that the mathematical complexity of the problem is overwhelming, however, it is possible to so manipulate the impedance correcting parameters that the network impedance matches the generator impedance approximately over a certain frequency band but is very poor outside this range. It follows from our previous discussion that the network will transmit frequencies lying within this band efficiently, but will attenuate other frequencies. Networks designed in accordance with this method therefore function as filters. They differ from conventional filters in several respects, however. For example they are non-recurrent, they cannot be divided into discrete sections with matched image impedances, and they do not possess definite cutoffs. " This circuit arrangement was first investigated by E. L. Norton and W. R. Bennett, who developed a complete analysis for a number of particular cases. Abstracts of Technical Articles from Bell System Sources A Space-Time Pattern Theory of Hearing} Harvey Fletcher. The pitch of a tone is determined both by the position of its maximum stimulation on the basilar membrane and also by the time pattern sent to the brain. The former is probably more important for the high tones and the latter for the low tones. The loudness is dependent upon the number of nerve impulses per second reaching the brain and possibly somewhat upon the extent of the stimulated patch. The experience called by psychologists "volume" or "extension" is no doubt identified with the length of the stimulated patch on the basilar membrane. This extension is carried to the brain and forms a por- tion of excited brain matter of a definite size. It is then this size that determines our sensation of the "volume" of a tone. The low pitched or complex tones have a large "volume" while the high pitched tones have a small one. The psychological experience called "brightness" may be identified with the sharpness of the peaks in the vibration form of the basilar membrane as suggested by Dr. Troland. The high tones give the sense of brightness while the low tones the sense of dullness. The time pattern in the air is converted into a space pattern on the basilar membrane. The nerve endings are excited in such a way that this space pattern is transferred to the brain and produces two similar space patterns in the brain, one on the left and the other on the right side. Enough of the time pattern in the air is sent to each of these stimulated patches to make times of maximum stimulation in each patch detectable. So when listening to a sound with both ears, there are four space patterns in the brain produced, each carrying also some sort of time pattern. It is a recognition of the changes in these pat- terns that accounts for all the phenomena of audition. The Theory of Probability: Some Comments on Laplace's Theorie Analytique? E. C. Molina. This paper is concerned with an answer to the questions "to what extent will one conversant with the Theorie Analytique be in touch with the present status of probability theory, and how sound a foundation will he have found therein for statistical applications of the theory?" ^Jour. Aeons. Soc. Amer., April, 1930. '^Bulletin, Amer. Math. Soc, June, 1930. 836 ABSTRACTS OF TECHNICAL ARTICLES 837 In answer to the first question emphasis is laid on the virtual identity between Laplace's generating function and the Cauchy-Poincaire characteristic function, on the close approach of Laplace's analysis to the form of the Fourier reciprocal equations and to the explicit presen- tation by Laplace of the Ilermite polynomials and related Gram-Char- lier expansion. In answer to the second question, the author submits Laplace's contributions to the probability of causes and points out the distinction drawn by Laplace between the meaning of the word limit when used outside the domain of probability theory and its meaning when the word is attached to the observed frequency with which an event happens. As evidence that the Theorie Analytique is in advance of much recent literature, and on account of its great practical value, the La- placian method of dealing with integrands involving factors raised to high powers is outlined. In this connection attention is called to a Laplacian differential equation which contains, as a special case, the differential equation from which Karl Pearson has derived his famous system of frequency curves. Method of Enhancing the Sensitiveness of Alkali Metal Photoelectric Cells.^ A. R. Olpin. A technique is described for sensitizing alkali metal photoelectric cells to light by introducing onto the metal surface small amounts of dielectrics, as oxygen, water vapor, sulphur vapor, sulphur dioxide, hydrogen sulphide, air, sodium bisulphite, carbon bisulphide, etc., or some organic compound as methyl alcohol, acetic acid, benzene, nitrobenzene, acetone, etc., or some organic dye as tropaeolin, rosaniline base, eosin, cyanine, kryptocyanine, dicyanine, neocyanine, etc. The marked increase in electron emission from the cathodes of cells so treated is due primarily to an increase in response to red and infrared light. Vacuum sodium cells have been produced, yielding photoelectric currents as high as 7 microamperes per lumen of white light of color temperature 2848° K and caesium cells yielding far greater currents. The response of these cells is proportional to the intensity of the exciting light even for light of longer wave-lengths than that to which the cell responded before treatment. Spectral response curves are similar for all cells using the same metal as cathode. These curves dilTer from the curves for the pure metal by the appearance of a new selective maximum at lower frequen- cies. This newly appearing maximum resembles the regular maximum for the untreated metal and is due to the presence of the sulphur and air. Changes of approximately 0.8 volt are common. ^Phys. Rev., July 15, 1930. 838 BELL SYSTEM TECHNICAL JOURNAL The validity of Einstein's equation precludes the possibility of ex- plaining the new maximum in the spectral response curve for a treated surface by a "Raman shift" of the incident light frequencies, even though the separation of these maxima is equal to certain well-known vi- bration-rotation frequencies of the dielectric molecules. It may be that the natural frequency of the alkali metal atom is diminished by the vibration frequency of the complex atom in which it is incorporated. The Lindemann formula for the frequency of the selective photo- electric maximum [Iw = (ne'^/mr^)^, primitive though it seems in the light of modern theory, has always given values for the pure metals in close agreement with experimental determinations. The n term is determined by the valence of the substance, a choice of unity being used for the monovalent alkali metals corresponding to an electron revolving around a singly charged ion. A choice of 2, 3, — for divalent, trivalent, — substances corresponds to electrons revolving around doubly, triply-charged ions. Under certain conditions the alkali metals manifest different valencies, such for instance, as those exhibited in the oxide series Na202, Na^O, NasO, Na40. These compounds can be prepared in vacuum and are light-sensitive. Spectral response curves for such cells exhibit all the selective maxima always separated from it by the frequency of a well-known line in the vibration-rotation spectrum of the dielectric molecules, usually the 1.5ju line so charac- teristic of oxygen-hydrogen, carbon-hydrogen or nitrogen-hydrogen linkages. The long wave limit shifts an amount agreeing with the separation of the maxima. With a cell so designed that the cathode could be sensitized in a side chamber and then slipped into its proper place (thus keeping the anode free from light-sensitive materials), stopping potentials were obtained for electrons, liberated by monochromatic light, from a sodium cathode before and after treating it with sulphur vapor and air. For light of wave-lengths ranging from X35C0A to X8000A falling on the treated cathode, the electron retarding potentials are found to vary linearly with the frequency of the exciting light, thus establishing the validity of Einstein's photoelectric equation for composite surfaces. From the slope of the straight line depicting this relationship, the value of Planck's constant h is found to be 6.541 X 10^", significant to three figures. An almost identical value is obtained for untreated sodium. The apparent stopping potentials, or voltages at which the photo- electric currents become zero are the same before and after the sulphur and air treatment. The voltage at which the current just saturates is always greater after treatment than before. This is a measure of the change in contact potential of the cathode called for by the Linde- ABSTRACTS OF TECHNICAL ARTICLES 839 mann formula when the value of n is chosen to agree with the valence of the metal. Data are presented showing this condition to be general for the alkali metals, a maximum response to red or infrared light being dependent upon the formation of a subvalent compound, as a suboxide. Attention is called to seemingly analogous phenomena in the fields of photoelectricity, photography, fluorescence and absorption. Some Problems in Short-Wave Telephone Transmission.^ J. C. ScHELLENG. In this paper are discussed certain phases of short-wave telephony, primarily, though not entirely, from the point of view of the transmitter. The field strengths which the transmitting station must provide at the receiver are considered. Typical data are given showing results obtained in transmission from Deal, New Jersey, to England. This is followed by a discussion of requirements and limitations of the transmitting antenna. The gains which arrays may reasonably be expected to provide are considered. The phenomenon of non-syn- chronous fading at nearby points is examined as to its bearing on the dimensions and performance of directive arrays. Other directional properties of the transmitting medium are also considered. Attention is then directed to the transmitting equipment, particular attention being given to the high-power part of it. Requirements, rather than circuit details, are emphasized. These include stability of operation, flexibility, and freedom from amplitude distortion, and phase and fre- quency modulation. The results of tests in which some of these matters were considered quantitatively are given. A Chronographic Method of Measuring Reverberation Time.'' E. C. Wente and E. H. Bedell. Reverberation time measurements are generally made with the ear and a stop watch in the manner devised by Prof. Wallace Sabine. Surprisingly consistent results can be ob- tained by this method in a reverberation chamber, where the rate of decay of sound is slow and where disturbing sounds are absent. But such measurements present difficulties if the room is noisy or if the reverberation time is short. Also it is recognized that uncertainties may be introduced because of the fact that the threshold of hearing varies between individuals and with time in the same person. It was with the object of overcoming these difficulties that the electrical method described in this paper was devised. This method does not differ essentially from that of Sabine except that an electro-acoustical ear of controllable threshold sensibility is substituted for the human ear. *Proc. 7.i?.£., June, 1930. ^Jour. Acous. Soc. Amer., April, 1930, Contributors to this Issue H. W. Bode, A.B., Ohio State University, 1924; M.A., 1926. Bell Telephone Laboratories, 1926-. Mr. Bode has been engaged in the study of transmission networks, such as wave filters, attenuation equalizers, and phase correctors. A. E. BowEX, Ph.B., Yale University, 1921; Instructor, Depart- ment of Physics, Yale University, 1921-24; American Telephone and Telegraph Company, Department of Development and Research, 1924-. Mr. Bowen has been engaged mainly in work on problems of inductive coordination of power and communication systems. I\ARL K. Darrow, B.S., University of Chicago, 1911; University of Paris, 1911-12; University of Berlin, 1912; Ph.D., University of Chicago, 1917; Western Electric Company, 1917-25; Bell Telephone Laboratories, 1925-. Dr. Darrow has been engaged largely in writing on various fields of physics and the allied sciences. Some of his earlier articles on Contemporary Physics form the nucleus of a recently published book entitled "Introduction to Contemporary Physics" (D. \'an Nostrand Company). A recent article has been translated and published in Germany under the title "Einleitung in die Wellen- mechanik." Ward F. Davidson, B.S.E., L'niversity of Michigan, 1913; M.Sc, 1920; Westinghouse Electric and Manufacturing Company, 1914-16; L^niversity of Michigan, teaching electrical engineering, 1916-22 except for two years war duty; Brooklyn Edison Company, Inc., Director of Research, 1922-. He has devoted much time to theoretical and experimental study of underground power transmission cables. C. L. GiLKEsoN, B.S., Massachusetts Institute of Technology, 1922, M.S., 1923; Transmission and Distribution Department of the Phila- delphia Electric Company, 1923-26; National Electric Light Asso- ciation 1926-. Mr. Gilkeson's work with the N.E.L.A, has been in connection with the research program of the Joint Subcommittee on Development and Research of the N.E.L.A. and Bell System. Much of his work has been associated with problems of low frequency induc- tion. E. I. Green, A.B., Westminster College (Fulton, Mo.), 1915; Uni- versity of Chicago, 1915-16; U. S. Army, 1917-19 (Captain, Infantry); B.S. in Electrical Engineering, Harvard University, 1921; Depart- ment of Development and Research, American Telephone and Tele- 840 CONTRIBUTORS TO THIS ISSUE 841 graph Company, 1921-. Mr. Green has been engaged principally in work on Hne transmission problems and multiplex transmission systems. R. G. McCuRDY, B.S., University of California, 1913; Technical staff, Joint Committee on Inductive Interference of the Railroad Commission of California, 1913-16; American Telephone and Tele- graph Company, Engineering Department, 1916-19; Department of Development and Research, 1919- ; Noise Prevention Engineer, 1930. Mr. McCurdy's work has been chiefly on problems of inductive coor- dination of telephone and power circuits. W. H. Martin, A.B., Johns Hopkins University, 1909; B.Sc, Massachusetts Institute of Technology, 1911; American Telephone and Telegraph Company, Engineering Department, 1911-19; De- partment of Development and Research, 1919-. Mr. Martin's work has related particularly to transmission of telephone sets and local exchange circuits, transmission quality and loading. E. B. Payne, B.S. in E.E., Mass. Inst, of Technology, 1917; Ar- tillery Corps, U. S. Army, 1917-18; Engineering Department, Western Electric Company, 1919-25; Bell Telephone Laboratories, 1925-. Mr. Payne has been engaged in loading coil design and more recently concerned with the development of wave filters and allied transmission networks. L.iss C. Peterson, E.E., Chalmers Technical Institute, Gothenburg, 1920; Technische Hochschule, Charlottenburg, 1920-21; Technische Hochschule, Dresden, 1921-22; Signal Corps, Swedish Army, 1922-23; American Telephone and Telegraph Company, 1925-30; Bell Tele- phone Laboratories 1930-. Mr. Peterson's work has been concerned mostly with problems relating to inductive interference. R. R. Williams, B.S., University of Chicago, 1907, M.S., 1908 Research Chemist, Bureau of Science, Philippine Islands, 1908-15 Bureau of Chemistry, LT. S. Department of Agriculture, 1915-18 Engineering Department, Western Electric Company, 1918-25; Bell Telephone Laboratories, 1925-. Mr. Williams has done extensive research work on submarine cable insulation. Since 1925, as Chemical Director, he has been in charge of the Chemical Laboratories of the Research Department of the Bell Telephone Laboratories. W. J. Williams, C.E., Rensselaer Polytechnic Institute, 1905; Assistant Engineer with Dr. W. L. Robb, 1905-06; Instructor in Physics and Electrical Engineering, R.P.I., 1906-15; Professor of Electrical Engineering, R.P.I., 1915-. Mr. Williams is a Consulting Electrical Engineer and has been Technical Adviser for the National Electric Light Association since 1923, on work of the Joint Subcom- mittee on Development and Research of the N. E. L. A. and Bell System. 842 BELL SYSTEM TECHNICAL JOURNAL Leon T. Wilson, Ph.B., Yale, 1915; Graduate Student and In- structor, Yale, 1915-17 and 1920-22; Signal Corps, U. S. Army, 1917- 19; E.E., Yale, 1919; Department of Development and Research, American Telephone and Telegraph Company, 1923-. Mr. Wilson's earlier work was chiefly in the radio field and included the develop- ment of a thermocouple type of voltmeter for radio frequencies. With the Department of Development and Research his work has dealt mainly with insulator problems of open-wire lines.