UNCLASSIFIED AD NUMBER ADA800206 CLASSIFICATION CHANGES TO FROM unclassified secret LIMITATION CHANGES TO Approved for public release distribution unlimited FROM Distribution authorized to U S Gov't agencies and their contractors Administrative Operational Use OCT 1944 Other requests shall be referred to Office of Scientific Research and Development NRDC Div 13 Washington DC AUTHORITY Secretary of Defense memo dtd 2 Aug 1960 Secretary of Defense memo dtd 2 Aug 1960 THIS PAGE IS UNCLASSIFIED S G OVERN MENT IS ABSOLVED FROM ANY LITIGATION WHICH MAY ENSUE FROM THE CONTRACTORS IN FRINGING ON THE FOREIGN PATENT RIGHTS WHICH MAY BE INVOLVED WRIGHT FIELD DAYTON OHIO ARE MISSING UV ORIGINAL PAGES 7 I SECRET NATIONAL DEFENSE RESEARCH COMMITTEE _ OFFICE OF SCIENTIFIC RESEARCH AND DEVELOPMENT T ' • DIVISION 13 SECTION___ OEMsr 435 FINAL REPORT ON PROJECT C 43 Continuation of Decoding Speech Codes PART I SPEECH PRIVACY SYSTEMS INTERCEPTION DIAGNOSIS DECODING EVALUATION THIs DOCUM ENT CON ALNS ILNFORMATION AFFECT ING THE NATIONAL DFVSE OF TIM UN Tr STATES WrTrIN THE MEANING OF THE ESPIONAGE ACT U S C 50 31 AND 32 Mrs TrLANSMIMON OR THZ REVELATION OF rr3 OONTEN•S •N ANY ' ANXER TO AN UNAUTHORJW PERSON IS PROHIBITED BY LAW S R D NO SECTION 1 0 COPY NO 4 2 DAT Oct 12 1944 BELL TELEPHONE LABORATORIES INCORPORATED NEW YORK N Y SECRET Communioations Division National Defense Research Committee of the Office of Scientific Research and Development Section 3 Division 13 flNAIT REPORT ON PROECT C43 Continuation of Decodinma Speech Codes PAKT I SPiCHE PRIVACY SYSTEMS INTIRCd•IMON DIAGN'OSIS DICODIVO EVALUATION October 12 1944 I SECRET Contract No Ol r435 Contractor Western Aleotrio Company Inc 120 Broadway New York 5 N T Project Supervisor C H G GRAY Teohnical Report Prepared by W KOVf•G BILL T•LAPEON3 LABORATORIlS INC 463 West St New York 14 N T 4 TABLE OF CONTNTS EiSTORY oF PROacT 1 ZUTRODUCTION TO 2 TEGSICAL WORT CaA PTE I IN•C 7TI0 1 1 Types of Radio Systems 3 2 Intercepted Signal Quality 3 3 4 Reoeiving Sets Types of Radio Transmission 4 4 5 Reoordin 5 6 Decoding Tools 6 ORA MR II THE SO E0TO0RAPR 1 History 7 2 Operation 7 3 Level Comprussion 11 4 Possible Inproversts 11 S Amplitude Representation 14 C11APR III SPISCH'SCML YMT1 15TIOflS 1 Systems Involving Single Ucaulation 15 2 Systems Involvin Double Modulation 16 3 Triple Modulation Reentrant Band Shift 17 4 5 Bandsplitting Systems Tine Division Lultiplex 17 18 6 Systems Using Tape Recording 19 7 8 Co binations of Tine and Frequenoy Scrambling Wave Form Lodifioation 21 22 9 Masking Systems 22 10 Voooaer Systems 24 11 Channel Liuzng Systems 24 12 Summary 26 CHAPTER IV DIAOGT0SIS OF U1TE'rTF 1 'easurezents 2 3 • i on Speotrograze Illustrations of Soranbled Speech Systems Not Illustrated II SSYSTEM 27 27 30 tilL 'TABIZ Q C01 T 3TS Contid 4 secCU R 7 set ONGRYPTNOWE• C TOOLS AND O oTHOS or Yunotional Equivalent 1 Captured 2 Compromise Deooding Methods 34 3 Automatic Decoding 37 CEAPMTR VI CRYPTOW APHIC TOCIS AND LZMT I 33 S a Program Determination Matching Spectrograms 41 41 3 Latohing Variable Area Patterns 47 4 Matohin Osoilloarams 5 indicator Methods 47 49 6 Application to Table 1 7 Determination of the Uessage CHAPTEBR VII PRACTICAL BVALUATIO0 52 54 OP PRIVACY SYSTMIX 1 Cracking Tim 37 2 ilonrepeated Code Systems 57 3 Code Ansalyss a8 4 Yield Evaluation 58 TABIL I SPZBCH SCAULING DMCXS 2AlLE II NONCRYPTOGRAPHIC DECODING IMMMS TABlLE III 103 II IN PART LIST OF PRELIUIQY REPO3 105 APPENIX 106 LIST OF ILLUSTRATIONS Figure 1 D165529 Sound Spoetrograph 8 2 Illustrating the Operation of the bound Speotrograpt 9 3 Spectrograws of Normal Speech 10 4 Amplitude Representation by Dot Spacing 12 5 Amplitude Representation by Contours 13 6 Modulators 15 7 Single Modulation 16 8 Split Phase Multiplex 16 9 10 11 12 Double Uodulation Reentrant Inversion One 7orm of Split Band Systea One Form of Tias Division Uultiplex 16 17 18 19 13 Variable Subband Delay 20 14 One Form of Time Division Scramble 20 15 Speed Wobble 21 16 Time Inversion 21 17 Wave Multiplication 22 18 Level Modulation 22 19 Subband Level Modulation 23 20 Noise Masking Using Two Channels 23 21 Masking Noise Applied at Receiving End 23 22 23 vocoder Systex Channel Mixing 25 25 24 Subband Channel Uixins 26 25 Calibration of Spectrograph Soales 28 26 Tine and Frequency Measurements 29 27 Illustrating Aided Tracking 33 28 Band Shift Filter 34 29 Variable Bartd Paso Filter 34 30 Peak Choppers and Compressor 35 31 Rectifiers 35 32 Illustrating Action of Rectifier 35 33 One Forn of Superposition Decoding 36 34 35 Directional Disorimination 37 Automatic Decoding Total Eneruy 37 36 Sidebands 37 in Two Position Displace ent System Automatic Decoding Energy Frequency Distribution 38 38 Matohing Speotrograph Patterns of Hoarepeated Code TDS 42 IV 37 LIST OF ILLUSTRATION S Cont'd Figure 39 Illustrating Inversion of Time and Frequency Scales in $ Spa ot rorama 43 40 Matohina Spectrograph Patterns of Twodimensional Scramble 44 41 Latohing Variable Area Patterns of Nonrepeated Code TDS 46 42 43 Osaillographio Traces of Vocoder Channel Signals Showing Effect of Rectification on Normal and Band Shifted Speeoh 48 44 51 46 Band Shift Detector Adaptation of Spectrograph for Decoding Switched Split Band Scramble Illustrating Repeated Code Lultiplication Systes 47 48 Illustrating Uodulation Sidebands Sinple Inversion 81 683 Alternate Inversion 65 50 Fixed Displaoement 67 51 Wobbled Displacement 69 52 Reentrant Inversion 71 53 54 Fixed Split Band Soranble 73 Rapidly Switohed Split Band Scramble Ixample 1 75 55 Rapidly Switched Split Band Scramble Example 2 77 56 Tine Division Multiplex 79 57 Tim Division Uultiplez with Noise Ohanal 81 58 Subbands Variously Delayed 83 59 60 Time Division Scranmbling TDS Combination of TDS and Rapidly Switched Split Band 85 87 61 llonsynohrnnoue Combinations of TDS and Split Band Scramble 89 62 Test for Twodimonsional Scramble 91 63 Speed Wobble 93 64 Baokwards 95 65 Ulultiplioation 97 66 Time Division Channel UIxing 67 Subband Channel Uixing 45 S49 4• % Scramble • 50 51 53 99 101 RISTORY OF PROTECT a This report covers work carried out for the National Defense Research Committee un der Contract No O Mer435 with the Western Electric Company Laboratories Iny Ina by the Bell Telephone proided with suitable equipment and could oh tain trained personnel Based on the needs of the military this project was thrice extended Under the guidanee of the Subsection on Speech Secrecy Seotion 13 3 of N D R C the Early in October 1940 there was set up in the Communications Division of IT D R C a subcommittee on Speech Secrecy This group was to consider both the sorambling and un scrambling of telephone signals It was soon recognized by thea that the decoding problem was of primary importance both as a means for evalu ating privacy systems for possible use by the Services and for decoding possible enemy signals Rualizine that the ear has very limited capabilities for anslyzing scrambled speech Mr R K Potter invented the sound speotroaraph to provide speech patterns which could be interpreted by the eye Early in 1941 a rouGh laboratory model of the sound spectrograph became available in Bell Telephone Laboratories Inc Throujh Dr 0 S Buckley chairman of the Speech Secrecy Section of N D R C arrangements were made for a demon stration of the sound spectroGraph to various N D R C representatives including Dr 7 Bush emphasis was placed at any given time on what was deemed to be most urgent This is reflected in the subject matter of the Preliminary Reports which were issued from time to time and which form the appendix to this report In addition to the specific investigations covered by these Preliminary Reports much work was carried on as the basis for more general coverage of the field of interception diagnosis decodina and evalua tion of speech privacy systems described As a result of the above demonstration and subsequent Committee action Project C32 the forerunner of Project C43 wae organized in the fall of 1941 with the immediate objective of producing a sound speotrograph in such form that it would be useful for diagnosing and decoding speech sorambling systems In Pro jest C32 Privacy Cracking a finished model of the sound spectrograph was constructed and its applioation to decoding work was successfully demonstrated to representatives of the Army Navy and V D R C Upon the termination of C32 on Febru ary 1 1942 it was decided that the work initiated under that project should be continued Accordingly Project C43 Continuation of De coding Speech Codes was authorized for one year effective February 2 1942 The project anticipated some routine decoding the production of duplicate equipment to be used by the Army and Navy intelligence servioes and further studies of decoding tools and methods At that time the Army and Navy military officers were relying almost entirely upon this project to furnish the above services until they could be mentioned above dedodint equiptent was developed moneae equiment ded to the was lavy dvlp and no dels furnished Army and This decoding equipment included 1 two models of the sound spectrograph 2 a variable area pat tern machine and 3 equipment for decoding two new enemy privacy systems intercepted by the project personnel at Point Reyes California In each case Army and Navy personnel were in struoted in the operation and maintenance of these equipment s Interceot activities of the project personnel included 1 the study of recording submitted early in the project by the Federal Cozzunications Commission 2 ezploratory work at the Bell Telephone Laboratories experimental radio reoeiving station at Holzdel New Tersey and 8 exploratory work and routine interception of radio telephone transtissiour at the American Telephone and Telegraph Co radio receiving eta tion at Point Reyes California Reports of the results of the above studies and recordings of intercepted material were submitted directly to the interested military authorities Lany speech privacy schemes were sub mitted through N D R C during the course of this project These were studied and evaluated This work led directly to the continued improve ments of the sound spectrograph and the develop ment of supplementary decoding tools and techniques Ls the Army and Navy became able to carry on decoding activities themselves with the aid of equipment and information furnished by N D R O as the result of work outlined above the activity on this project gradually decreased the Preliminary Reports covers all phases of the work on this project and constitutes a re The technical repott which follows together with the Appendix which includes all of feranms work for future studies of speech pri vacy systems INTMROUOTION TO TZONHIOAL •0RT This report sunmarises the results of about three years' experience in diagnosing do coding and evaluatinG speech privacy systeas submitted for study on this project by the Army Navy and Ir D R O Some of the results of these studies have been described in a series of Pre liminary Reports which were issued from time to tiM to cover specific studies A great deal of acounulated experience however has never been reported in this manner This final report therefore is intended to make available infer mation both positive and negative which would have to be accumulated by another croup if they analysis and cryptography which apply to tele graph types of comunication very little has been written on speech privacy systems or deood ing methods applying to then Two moderately comprehensive articles have been published One appeared in the Post Office Engineers Journal October 1933 The other appeared in the Brown Boveri Review for Decenber 1941 The latter has been reproduced and discussed in Preliminary Report No 5 It covers a number of basic types of scrambling systems and in addition discloses one that is new The immediate pressure behind these studies was caused of course by the War The material here recorded should therefore be of service if a similar energency should arise in the future To keep up with the everohanging art of oomzunication these studies should be continued under Government auspices during peace time cover rather completely speech soranbling methods in wLioh the original seecoh is transmitted with its parts nodified displaced or interchanged When more detailed technical inforaation is de sired reference may be made to the Preliminary Reports These are separately bound and fora an appendix to this report All of this material should be helpful in the development of prao In contrast to a rather extensive literature on code and cipher systems on crypt tical effective privacy systems and the eval uation of the security which they afford The present report is intended to were to embark on a similar project 2 L SINT' CEPTTON Speech privacy systems may be used in connection with radio telephone systems or with nals can not generally be picked up at great distances and whatever equipment an interceptor wire systems The unauthorized interception of wire comnunioations in wartime however is be yond the scope of the present report This chapter will therefore be confined to radio interception problems and expands the material in Preliminary Report No 25 The decoding techniques to be described subsequently of might use to oraok the privacy must also be mc bile Furthermore the decoding equipment must be operated by military personnel a large num ber of whom may be required if the enemy is making extensive use of mobile privacy 'kether it is worth while to attempt to decode a large number of small mobile communications is ques tions Chapter VII 1 Radio telephone systems range in size and complexity from high power point to point stations operatinS over great distances to the low power short range sets carried by indivi Intermediate types of radio systems are used for the higher echelons of comand For such applications the radio equipment is semi mobile It can be transported in trucks and set up very rapidly and may have a conviderable range For such applications a high degree of of inherent privacy in mobile equipment it should be noted that the very mobility of such systems adds to the security because the sig In this connection it might be noted that it is very desirable to be able to hear both course apply to wire as well as radio connumic i tionable as discussed in greater detail in Types of Radio Systems dual soldiers The high power systems are usu privacy is required and a truckload of equipment ally designed to operate between specific points right be justified because the enemy could af assigned feiote frequencies They are ford to devote considerable time personnel and qu e using specific with elboat whichrone aonsineras Sequipped with elaborate fixed antennas which equipment to decoding the kind of messages which are usually of the directive type Privacy would be transmitted over such systens equipment associated with such terminals may be as large and complex as desired to achieve vir 2 Intercepted Signal Quality tual secrecy A major consideration in such systems of course which adds to size and oom decoding the material from this point on will plexity is that the privacy must not degrade be written from the point of view of the unau the quality of the received speech to any appre tiable extent h thorized rather than the authorized listener It is first of all desirable to get a good signal On the other hand any one can iu as free as possible from interference There are tercept these hiGhpower signals at great dis several reasons for this First the process tanoes where he can have a wellequipped which unscrambles the speech also scrambles any centralized deoodina laboratory with no limits noise such as static which has been superposed tion on the size and complexity of the decoding on the scrambled signal This changes the time equipment he micht bring to bear This labora or frequency distribution of the noise breaks up tory can be adequately manned by a relatively harmonic relationships etc thereby increasing few highly trained decoding specialists not the interferin4 effect of the noise Second the necessarily members of the arred services decoding is apt to be less perfectly accomplished than at the authorized terminals which tends to no_ ortrast wit ts sitain the make the speech harder to understand Finally lot' power short range radio sets used in nil there are usually language differences which tary operations are severely restricted as to still further add to the difficulty of under size and tieight and these restrictions also ap standing the message Conversations can be oar ply to privacy equipment The smallest privacy set submittedtoed on under extremely unfavorable conditions a 10 inch cube and was designed for mobile ap by people speaking their own language but noise and poor quality rapidly degrade the intelligi plications like tanks planes and command cars bility of a language foreign to the listener ih•ile it is difficult to achieve a high degree 3 4 sides of the conversation without interruptions in order to follow the context In the case of in general the point to point systems this will directions two the because require two receivers are transmitted over separate channels at differ eant frequencies If the two outputs are mixed for listening or recording however it shouldbe kept in mind that the noise on the weaker signal will be superposed on the stronger signal and say seriouis it Putting the two signals on seriously degrade two headphones will improve this situation be cause noise in one ear does not seriously affect the intelligibility of a signal in the other ear This problem does not arise in the case of the smaller radio systems because these are gener ally operated on the basis of switching betieen transmitting and receiving conditions on the sa carrier frequency al i Lev'_ i A Usthods of obtaining a good signal are the sane for the interceptor as for the in tended receiver A few of the important oonsid erations are listed here further information on o any or all of them can be had from radio refer once works 1 Point to point systems usually on tat iterept station he intercept diectve the atenas emply Semploy directsive antenas threo near along or be located therefore should line of the radio been 2 In locating ate tioms to intercept radio transmissions in the HF raeng account should be taken of the skip die tances of the frequencies involved Better sig neals will sometimes be obtained y moving farther away from the transmitter rather than closer 3 The use of directive antennas directed to wards the transmitter being monitored will in prove the signal to noise ratio by discriminating against noise which is nondirectional These antennas of course should be designed for the frequency and polarization of the signal and properly coupled to the receiving set 4 Stronger radio signals will be received if the antennas are located in the open with no trees or other obstructions in the foreground This is particularly important in the VU range 5 Radio signals increase in intensity as the height of the antenna above the immediate foreground is increased particularly for VU transmission Thus better results are obtained with the anten nas located on high masts or on hills overlooking the foreground in the direction from which the signal is arriving If the signal is in the VHF range and other measures are inadequate it may even be desirable to consider receiving the sig nal in an airplane and recording it or retrans mittiag it for decoding 6 Noise improvement can generally be obtained by keeping the receivrig equipment away from sources of man made noise such as ignition systems and power lines 3 ' Reoeiving Sets with regard to the receiving sets a distinctioa must be made between the various as tivities of an intercept station One important activity is searching for possible enemy trans mission channels The object is to deternine all the channels in use the location of their ter minals the type of business transacted and prelimieary form Qf privacy chanel thei special uesred any Some whether all of on most e preliminary is eace on the channel Soa searcheportisos te are described in Prelivi nary Reports has 2 and 23 If no privacy is used other than the usual commercial types it is unlikely that information of military impor taene is transmitted over the channel and it may not be necessary to monitor it continuously If a new privacy system is located however it is very likely to be worth monitoring and deood in$ continuously or the searching and scanning activi ties the ordinary oomeroial sets of the com munications type equipped with a beat frequey oscillator will serve very well for all types f transuission Zveu the suppressed carrier type can be handled very well provided the sig l nal is fairly strong It may require continual umnual adjustment of the looca oscillator but sufficiently good reception can be obtained to determine the nature of the channel Oases of extreme spread band transmission can also be hndled in this % or If a particular channel employing sup pressed carrier is deternined to be worth memi tortn6 continuously then a single sideband receiver will give improved reception These receivers are equipped to amplify the partly suppressed carrier or supply a new one with great stability and they may provide as such as 15 db improvement in signal to noise ratio in some oases They also permit selecting either the upper or the lower sideband of double side band systems which may be of advantage in cases where interference occurs on one or the other sideband of such systems However these re oeivers are not suitable for searching 4 Types of Radio Transmission A knowledge of the types of radio transsmission which may be encountered is very important to the personnel of an intercept sta tion Bxperience has shown that without such knowledge the nature of intercepted signals nay be completely misinterpreted It is possible to mistake certain norma types of transmission for new systems or conversely to fail to recog nize new systems which should be monitored at once The commonest type of transmission the double sideband type in which the Sis carrier in transmitted along with the sidebande which are usually about 3 000 cycles in width and are located immediately adjacent to the crrrier These are readily demodulated by the ordinary receiver This is true even if the carrier is rapidly wobbled provided the wobble does not cover too great a frequency range Such wobbles are sometimes used in combination with simple inversion to prevent reinverting with a locally supplied carrier at the edge of one aideband In the socalled spread band system some or all of the sidebands are displaced from the carrier Demodulated signals of this tym will cover an audio frequency range greater than 3 000 cycles usually as high as 6 000 cycles It is essential therefore that the receiver be capable of handling such a band To obtain the intelligence the signals must be further demodulated as described subsequently Bl in Table I In the ordinary transmissions described above the carrier level is high compared to the speech sidebands In order to avoid loading up the transmitter with carrier and thereby permit radiating a higher sideband level many channels operate on the suppressed carrier basis In this system the carrier is either eliminated completely or transmitted with greatly reduced level To demodulate such sienals properly the weak carrier must first be greatly amplified or a new one supplied locally If this is not done the siGnals will demodulate themselves around whichever component in the sideband happens to be predominant producing thoroughly scrambled speech which can thereafter not be restored This condition can be recognized by its ohurac teristio sound to the ear toGether with wide syllabic fluctuations of the meter which ordina rily indicates the carrier level With suppressed carrier systems usu ally only one of the speech sidebands is trans mittod However a second sideband transmitting a second speech channel is sotetimes added usually displaced from the carrier by about 3 000 cycles to avoid crosstalk between thechan nels This is called twin channel operation and gives on demodulation an audio siGnal covering about 6 ko The two channels must be separated and placed in their normal positions by the methods previously cited under spread band systems The above systems are the main types of radio transmission used commerolally with amplitude modulation In addition in the VE range and above there are frequency modulation systems and also pulse modulation systems both of which require receivers specially designed to handle their particular types of signals This is too large a subject to cover here and refer ence must again be made to the radio literature Finally it should be mentioned that in addition to speech a great deal of telegraph addition to s of are telerah transmission will e a be found Thde There several types of telegraph signals including hand keyed such as Morse code or machine keyed such as Boshne and teletype Any of these types may be transmitted by keying the carrier or by keying a tone modulated on the carrier The marks and spaces may be represented by changing the azpli tude on off or by changing the frequency twotone Finally since telegraph requires a much smaller band than speech it is often operated on a multichannel basis that is a voice channel will be divided into a number of telegraph channels In addition there are facsimile transmission systems which also may be operated on an AZ or Fn basis If a new sig nal is encountered whose nature is in doubt these possibilities should be kept in mind for further investigation when the need arises 5 RecordinA The sane considerations discussed in section 2 above which make it desirable to ob tain a good intercepted signal apply also to recording and reproducing scrambled speech In addition to the requirements as to quality and noise there is an even more serious one con erning speed regulation In general systems designed for a high degree of privacy require a high degroe of synchronization and in many oases ordinary recording methods are not good enouih not only in long time average speed regulation but in the steadiness of the instan taneous speed In the case of some of the sys tens described in Chapter III for instance the requirements are so severe that even the best conmercial recorders will not meet them The best solution of this problem is to decode before recording This will be possi ble in many oases although it may sometimes entail the loss of parts of the messase while adjustments are being made or the code is being determined It happens that some of the systems described in Chapter III which impose the severest requirements on speed regulation B3 in Table can be handled in this way Ahen prepared to diagnose and decode intercepted ene my signals must be equipped with a considerable variety of special tools These should include of course such wellknown devices as oscillo graphs and oscilloscopes amplifiers oscilla tors modulators rectifiers fixed and variable filters and a supply of components for con structing special circuits that may be required Some of the less wsllknown devices whose ma ture and usefulness will be made clear in subse quent chapters include magnetio tape or wire recording and reproduoina equipment in the form of loops with multiple pickups commutators for sweep or timing circuits variable speed drive mechanisms channel shifters the variable area pattern machine and the sound spectrograph There should also be models of the more iapor tant types of existing speech privacy systems Finally and perhaps most important of all then should be stationed at the intercept location a Group of highly trained technicians who should be thoroughly familiar with radio transmission problems radio facilities oryptanalytio pro this method is feasible even poor quality re corders such as those designed to record a Vast deal of material in a izall area nay be good enough In some of the systems to be described it will not be possible to decode before record img It happens however that in the case of the Only kIown system for which this is tre the requirements as to quality 73 in Table I and speed can easily be met by good commercial type recordings The matter of convenience or ease of use of the reproducing system is very important in decoding work In this respect also the require sents are different for different privacy sys tems The recording systems using the embossing process for instance are convenient because they produce no thread and they require little attention However they all suffer from poor tracking during reproduction which can be exceed ingly burdensome especially where the material must be reproduced many tines over Recording magnetically on wire is attractive from the standpoint of convenience and also quality but backtracking is very timeoonmmuing and laborious cedures ad diagnosing and decodin methods If these technicians are not oonversant with the lanCuaGe encountered in intercepted communic itreesshudeooiuoay tions interpreters should be continuously available The best solution at the presentton writing appears to be disk recording on acetate with a machine capable of recording at v rious speeds Low speeds oa be used where quality need not be too good and a long record is de sired Higher speeds can be used where better quality is needed Such recording systems are commercially available 8 Decodina Tools Even with a l of the special tools and sth to a s e ial even it all of personnel decodin in many instances will be a difficult problem and patience and painstaking effort will bu required to obtain useful infor mation from scrambled speech Unless the needs have been anticipated the enemy may have secret In addition to the facilities discus sed above an intercept station if it is to be oornunication for a considerable period of time as a direct result of unpreparedness 6 THE SOUND SPCTROGWRPH This chapter is devoted to the sound 2 Operation spectrograph its history method of operation and capabilities The sound spectrograph anal yzes speech in terms of its three basic dimen slones frequency amplitude and time and portrays the analysis in the form of spectr e grams These are helpful in understanding the complexities of speech and what various scrambling methods do to speech to make it un intelligible 1 S• A schematic diagram of the sound epee trograph is given in figure 2 together with a description of the method of operation There is described produced by the operations i l s r t o a p t e n w i h s o s t the i h b in illustration a pattern which shows by its light and dark areas how the intensity in the signal varies as a function of both time and frequency It is the fact That both time and frequency variations are simultaneously displayed which makes spectrograts so valuable for decoding work History Scanning filters of various widths can be used for different purposes If the filter is wide it will give an analysis which is lim ited in the amount of detail it can portray in the frequency dimension but it will respond quickly to changes in amplitude with time and will therefore give sharp tine resolution The narrower the filter the more frequency detail is shown in the spectrograms at the sacrifice how ever of some of the time resolution With all the filters thus far used the shift in frequency range from line to line is only a fraction of the width of the filter Successive lines in the spectrogram therefore do not represent successive frequency bands They represent fre quency ranges which overlap by a large fraction of their total width The density of the pat terns therefore changes very gradually along the frequency dimension kind of patterns produced by this In arch 1941 an early laboratory model of the sound spectrograph was demonstrated to Dr V Bush as an instrument that with fur ther development might be useful in studies of telephone privacy It was appreciated at that time that the need might arise for intercepting communications in scrambled speech and decoding then It was also appreciated that new soramb ling systems might be encountered and that means would be needed for dia nosing such systems For such a purpose the unaided aar has very limited capabilities Such things u oscillo grams which show the wave form also provide little in the way of clues as to the mechanism by which the wave forn was ohanged Project 032 the forerunner of 043 was therefore organized in the Fall of 1941 and its immediate objective was to produce a sound spectrograph in such a form that it would be useful for diagnos ina and decoding speech scrambling systems isbThe About a month before the attack on method of analysis is illustrated in figure 3 Pearl Harbor patterns that could be used for The upper spectrograz in the figure was made with decoding work were being produced with a bread board model and the first finished model of the spectrograph was available by the end of that year Additional models of the spectrograph have since been built for the use of the armed services incorporating improvements in opera tion and in ruggedness The most recent model is shown in figure 1 The spectrograph has been used in studies of various privacy systems submitted by the Army Navy and IT D R C for the purpose of evaluating the degree of security a scanning filter about 300 cycles in width The separate words can be plainly distinguished The vowels are distinguished by dark bands with vertical striations The consonants are in gem eral less intense and show a different type of structure It will be noted that the dark bands are different in the different vowels and they change not only from one word to the next but also within each word Analyses of this type therefore graphically portray changes in the energy frequency distribution of a complex sig studies it became evident that inprovements in the spectrograph would be useful in this work Accordingly a calibrating circuit was built into the spectrograph and control circuits were udced in the form of an applique unit emphasized however that the relative intensi ties of the various components of this particular consonats differ the s of Sample of speech notably the consonants differ to a far greater extent than would be judged by whiot they afforded In the cource of t ose nal with both time and frequency 7 It should be RECORDER UNIT 2t4 RECTIFIER UNI7 UNITS INTERCONNECTED AND READY FOR USE Figure 1 D165529 Bound Speotrograph p 4 u 0i4N 4 r4 4 IX F4 b t4 h6 W4 94 cct0 ix 0 0 0 9 2 o Q C F4 r 0w 3 0 11012O l q 0 0 Z 4a in 4 01 0 40 94 0 3 01l4 9 o 4 E43 AVH 1 1000 U 01 144 UI I4 43 W4 A 4 '% g 00 to4 a 4 cc i I j48 ' 4 vw aa the relative blackness of their patterns In mess the lowest level components will be invi other words a very large amount of level con sible Conversely if the level preesion is incorporated in these patterns as will be described in the following section The lower spectrogram in the figure sholhs the same words analyzed with a filter only 45 cycles wide This filter is narrow enough to resolve the individual harmonics of which vowel sounds are uomposed The harmonices con sist of the fundamental voice pitch together with both odd and even multiples of this fre Quenoy Sone of the harmonics are stronger than the others because they are reinforced by res onance in the oral cavities as the words are fumred It will be noted that the dark areas in these patterns correspond in frequency and in trend with those in the upper spectrogram The fact that vowel sounds consist of discrete her nonics causes the vertical striations in the patterns made with the wider filter 'henever the filter is wide enough to pass several har monies at once these harnonics beat with each other and form maima and minima in the output of the filter The frequency of the beats ocr responds exactly to the frequency of the voice pitch It will be noted in the 45 cycle spec trogran that the harnonics rise and fall in fre quency from moment to moment This reflects the chaniCng pitch of the voice known as inflection Inflection is normally used in conmected speech and this fact is of assistance in decoding work because the spaoina and trend of the individual harmonics in spectrograms provide important clues in diasenoting privocy systems as will be ceostrte d in suseprevacyhtoeras willbe demonstrated in subsequent chapters is so adjusted that the low leiel components appear in the pat tern the high level components will severely overload the recording paper In order to show both the high and low level components ocourina in speech therefore it is necessary to com press the instantaneous signal into a much mar rower range In the earliest models of the spec trograph the marking amplifier shown in figure 2 was given a compressing action by means of a thyrite varistor across the grid of the output stage Whenever the output of the scanning fii ter was low the gain of the amplifier was ef fectively raised from an average condition and whenever the output was high the gain was effec tively lowered This tended to equalize changes in level with both frequency and time Vore re cently the compressor has been replaced by de vices which can exercise certain types of discrimination in controlling the instantaneous gain of the marking amplifier These devices are known as control circuits They provide patterns with better resolution in both time and frequency than can be obtained nith the compres sor The patterns shown in figure 3 were made with these control circuits in operation The circuits are desoribed in Preliminary Report IO 27 4 Possible Improvements %49 spectrograph patterns he under gone continual improvement in the course of this project but they can probably be still further improved The control circuits thus far pro duced are by no means the final word Circuits 3 Level Gomprbscion of this type can be adapted to affect the pat terns in various ways and it is conceivable that different control circuits could be devel oped for decoding different types of scrambles In normal speech there is a tremendous change in level from moment to nouent particu larly in the level of consonants as compared to voifels There is also a considerable difference in the average level at low frequencies as con pared to high frequencies This latter differ One definite line of improvement con corns the time resolution I any scrambling methods as will be seen subsequently pro present models of the spectrograph contain chap ing networks for this purpose There are how ever changes from moment to moment in the relative levels of hiCh and low frequencies in different speech sounds which cannot be corrected by shaping networks The facsimile paper on which spectrograms are made has a range of be tileen 10 and 15 db The range of levels in speech greatly exceeds this value This means that if the averag'ý level is adjusted so that the highest conponents appear at maximum black speech in the time dimension The process of analyzing the scrambled signal in such a way as to obtain high frequency resolution tends to obscure the signal at these sharp boundaries This is a basic situation which affects not only the spectrograph but all types of analyzers In order to obtain a high degree of frequency resolution a narrow filter must be used The narrower the filter however the longer its response and decay time that is the output of the filter cannot be made to change as rapidly once can be corrected by predietortion and duos sharp discontinuities of the scrambled ii 9 r 1-5 th ihwqf arnjm Iii-t 5 The-e spectrogram illustrate one nethod or mlitnden in such In that they can be interpreted quantitatively They are made up of discrete dots ell equally block The dots ere closely opened in the dark regions and widely opened in the light region There in oefinite quantitative relation between the dot spacing at any point end the level of the align-l at that point The level at any point could therefore be deter-ind by mooring the dot opening with suitable equipment These patterns show the results of exploratory mt undoubtedly further work would improve the representation Figure 4 - Amplitude Reprelentetion by Dot Spacing p A 4 4 54 4 1 45 P4 Q4D 104 43 1600 14 e4 4 044 k ASQ 0 400 1 Q Ck C lui zI in level as the instantaneous level of the Big n l This oauses strong components to spill over across the time boundaries Ixamples of such spillover can be seen in the upper speotro graze in figures 54 55 59 and 60 In general fore Le measured by measuring the dot spaoing with suitable equipment and comparing it with a scale showing dot spacing vs level Another type of representation if shown in figure 5 Here the levels are repre the recognition of various privaoy systems but it does interfere severely where spectrograms are to be used for decoding work Several po sible remedies for this situation have been de smntitg topographical variations in contour aw The contour lines each represent regions in which the signal reaches a particular fixed level The lines may be spaced so as to repre this spillover does not interfere greatly with vised which are recorded in Chapter TI 5 Aazlitude Representation In the patterns thus far discussed the instantaaneous intensity of the signal in repre sented by the ligteness or darkness of the trace in the spectrograns This representation is in herently nonlinear and practically inpossible to make quantitative Yor some types of work it would be highly desirable if the amplitules could be represented in such a way that they could be interpreted quantitatively Figure 4 shows a spectrogram which upon close inspectini will be seen to be madeup of discrete dots The dots are close together in the dark portions of the speotrogram and far ther apart in the light portions The dots themselves are all of equal blackness The spac in of the dots is in fact quantitatively related to the instantaneous level of the signal The level at any point in the speotrogras can there sented by the type of technique used in repre sent steps of any desired number of db or any number of volts In the lower spectrogram the spaces between the contour lines have been filled in with various densities of dot spacing This permits instant recognition of equality of level in different portions of the signals quantitative mplitude representation may or may not prove useful in decoding work lor certain kinds of signal it should prove use ful because it provides another disansion be sides time and frequency which can be used for determining continuity or discontinuity in the signal In other cases however it may prove useless because changes in level have arbitra rily been introduced into the scranmble The developments mentioned above es phauize the fact that the sound spectrograph is a highly flexible device and its capabilities along any line can be ereat 7 y inoreased by add ing features designed for tae specific pur'pose in Rind 1I CHAPTER III SPEECH SCRAMBLING UETHODS In this chapter we shall examine a wide variety of speech scrambling methods in or der to become familiar with the devices which might be used alone or in combination to make up speech privacy systems Some of these sys tems are in commercial or military use others exist only on paper mostly in the form of pa tents or patent applications It is not in tended to include all the variations of all the different methods but rather to cover basic scrambling methods with their most important variations in which the original speech is transmitted with its partb modified displaced or interchanged The two main dimensions of speech which are operated upon to make it unintelli gible are the frequency dimension and the tine dimension Scramblingsystems usually depend on rearranging the components of speech in either or both of these dimensions In general it may be said that those that operate on the frequency dimension alone are capable of the best quality in the raproduced speech A com plate list of the systems covered in the dis cussion is given in Table I together with other data concerning them Most of them are illus trated by means of spectrograms which will be discussed in Chapter IV INI OUT CARRIqER A Figure 6 RO CB CARRI 1 Systems Involving Single Uodulation Modulators pass filter It is then modulated with a fre quency of 3 000 cycles This produces an upper and a lower sideband of which only the latter is passed by the output filter The system is called inversion because the high frequencies in the original speech appear as low frequen cies in the output and the low frequencies in original speech appear as high frequencies At the receiving end the inverted signal in A basic device in privacy systems is the modulator One form of modulator shown in figure 6A consists of four copperoxide varistor units between two balanced coils The carrier frequency is fed into the midpoints of as some cases the coils ofn the te coils omilsed as shown shown In in somue cas tthe can be omitted as shown in figure LB Figure 7 shows the method of produc passing through an identical system in the same direction is reinverted back to normal ing simple inversion In this and in succeeding illustrations the numerical values are not neces sarily the best values for practical peration but they serve to illustrate the manner in which the device operates speech A very commonly proposed variation of this system involves using a variable fre quency instead of the steady 3 000 cycle carrier We might vary the frequency continuously or in discrete steps It should be noted however that the cutoff of the lowpass output filter In the system shown in figure 7 the speech band is limited to 3 000 cycles by a low 15 LO P P O P I 3000 1igure 7 Single Modulation ts fixed which limits the variation permissible in the carrier frequency A wide variation would either permit too much of the upper side band to get through or would cut off some of the lower sideband Two proposals of this type are discussed in Preliminary Reptrts No 8 and 20 If the modulator in figure 7 is of the type shown in figure SA speech can be scrambled by introducing instead of the 3 000 cycle car rier a square wave whose changes fron positive to negative value are irregular in tins Each one of the reversals in the carrier wave causes a reversal of phase in the speech wave The pat tern of these irreCular reversals moy be arrara so that the speech becomes unintelligible it the reoeiving end a coding wave must be intro duced which is exactly in step with the one at the receiving end with proper allowance for any delay there may be in the transmitting channel A two ochnnel system usinw one modula tor for each channel is shown in figure 8 In this system the carrier fed into both modulators is the same in frequency but 6iffers 90 degrees 0 dgres ii reqenc he amein bt tiffrs can be in phase Two separate speech ohannls transmitted by this method without substantial mutual interference but both sidebands as well as the carrier must be transmitted At the receiving end the carrier must be split into two components with the proper phases Each compo nent will demodulate its own portion of the sig nal and thereby separate the two speech channels Naturally one of the channels may consist of noise or spurious speech from a reoording or the like which tends to mask the real message if the signal is demodulated with an ordinary set Tias soheme was origia lly proposed as a multiplex system but an obvious variation is to divide a single speech band into two halves with filters and then transmit the two halves on carriers differing by 9U degrees in phase Figure 8 Split Phase Multiplex 2 Systems Invo vinag Double kodulation Figure 9 shows a much more flexible system Here the vignal is modulated twice with a bandpus filter between the two medula tore With this arrangement the carrier fre quency fed into the second modulator can be varied in several ways In the illustration two carrier frequencies are shown for the second modulator If the 8 kovalue is used the output consists of the speech band right side up but displaced from its normal position by 2 000 oy oleo If the 16 ko value is used the output conjists of the 3 000 cycle speech band inverted and displaced by 3 000 cycles We might use these two values alternately at short intervals or we might have the carrier vary continuously back and forth say betwesu 13 and 16 ko Another variation is to use a multiplicity of values say 500 cycles or 1 000 cycles apart not between 10 and 13 ko for this illustration and switch between these values in a regular or irregular sequence A disadvantage of these systems of course is that the truameis ion channel need ' to be wider than that usually afforded by radio sets or telephone lines In all of these systems the speech is sp MOD Figur 9 Double 161 MOD Modulation LP restored by passing through identical equipment in the opposite direction Figure 10 shows a system of modulators and filters for accomplishing this kind of band shift The first modulator is followed by a high Triple Modulation Reentrant Band Shift which selects the upper sideband pass filter from 3 000 cycles to 6 000 cycles This is com 3 bined with some of the original signal which ranges from 0 to 3 000 cycles The second modula tor is fed with a carrier frequency of say7 kc which inverts the whole band This is followed by a band pass filter passing the range from 3to 6ko A see end modulator with its carrier frequency placed at the lower edge of the 3 to 6 ko input band moves inverted down to the usual the whole band still range of 0 to 3 000 cycles The upper sideband of this modulation step is removed by the output A variation of this arrange lowpass filter ment isto allow the 7 kc carrier to vary in dis crete steps according to some regular or irregu lar program or vary it continuously between the limits of 6 to 9 ke This provides a variable band shifting arrangement without using more than the normal 3 000 cycle transmission channel Going back to figure 7 suppose the carrier frequency were made 4 000 cycles in stead of 3 000 but retaining the 3 000 cycle input and output filters The output would then be an inverted sideband ranging from 1 000 cycles to 3 000 cycles that portion of the sideband above 3 000 cycles would be cut off by Since however there is a the output filter 1 000 cycle gap at the lower edge of the trans mitted band the portion which would be cut off might be modulated down and sent by the filter along with the rest of the sign4 in this lower part of the spectrum In other words the par tion of the sideband which would otherwise dis appear above the upper edge of the transmitted band might be made to reappear at the bottom 3 KC PAD BP 36KC MOD MO MO LP KC 3 uD 7 KC 3OKC Figure 10 Reentrant Inversion 4 ing the modulators are all alike passing the Bandsplitting Systems band from 3 100 to 3 650 cycles A It will be seen that the uppermost modulator in figure 11 privacy system in wide commercial with its carrier of 6 100 cycles will invert the use known as the split band system involves speech band and displace it by such an amount that the frequency band which originally occupied the space from 2 450 to 3 000 cycles will pass In other words this modula through the filter tor in combination with its band filter selects the uppermost of the five subbands from the in put signal Similarly the lowest modulator in selects the combination with its band filter lowest subband from the input signal The out puts of the band filters all occupy the same frequency range but they all came originally splitting up the whole speech band into a number of subbands wnd shifting those around out of their normal positions in the frequency spectrum Figure 11 shows one manner in which this can be acconplished The numerical values are chosen so that the band from 250 to 3 000 cycles is divided into five subbands each 550 cycles wide Th uweech band is fed to five modula tors in purullel The five band filters follow 17 INPUT OUTPUT 3WITCM 10e L 410 ORTR O31 AlI10 pi Eayt 11OeFrmo Figure 46 fequncy iaiarl the fromdiferet angs outpu modlatos tht eah'o ar no dsignd ylsed accepts_0 thPadfoC 0 t 6 shifts111 atclrbn itto oaio nte oupt h fv easginCnoohwe sn ec0scn mateeoe labeled MSSW'E 5 im i iiom MODi e iiinalipe T0U i i nwihms parte iflocuyn fIqec rnge aresnWvrasnl nylnho ahsga en rnmte sse th ie h necedin nydesre tme Thn igh b ilusraed y homooth mnma wth hefiv output leads Thseutn uptwl lwy inl onceo OA the semet ofP coert opet g fro to3000mttr Arpdyrttn buhpcsu h a ra0 inlsoeatrt ohe racpal alsadteewllb ooelpig ubns 1760 Anadto lsto f frqece is as The S IT0 maybeFhagued as Often Fince oe quliyhweer hebrshmstmaeatles roa o peneon Oan astehgetf arei intrete herin r Tma Divstesrahrohn ytmw k ulitipez ode tal bang s isi the o coo1 willu codulaine oaseve to thegne fc thae tpeeh his can a ivsionl s chanel transm Ttim havcept been chaned asrofte as100 to36 cycer aro desi ered Ont fnreqprien re tds occupying thesae a allal nueerofaub nwhc lchtsho ing that itt tof th prcioeid spee a qualtsity vethese stogt e lie cyclefrequency range anrconnecti idso tas box beands asin istpossible tosift a aoert ditri e coapid erte rithout n prductinls oy eutents oAourapidlyhrotatincobruthtor kseferrie frome 250 to 3p 00ec t frigureie indcaedinth dawngfo te ecndse o e highst cane to mn roa 12 whic io s prsicoilar to 11 qeny n hetrneitedsinas hi18az 4600' BP 5 3600ý 0P FILteRs ALL ALIKE 300 Figure 12 TO'38OO' One Form of Tine Division Multiplex accomplished by feeding all the output modula loop over pulleys or attached firmly to the tors with the same carrier and connecting each modulator to a commutator segment In this illustration there are four 600 cycle subbands covering the range from 400 to 2 800 cycles It has been shown mathematically that the output of this system consists of sidebands around a perimeter of a disk The reoordimg is done by means of small electromagnetic polepieces The signal is picked up by similar polepieces which may be placed at a distance from the re cording polepiece depending on the amount of delay desired The outstanding advantage of the frequency corresponding to the rotation of the brush and also sidebands around frequencies corresponding to odd harmonics of the rotation frequency Each sideband however contains components from each of the subbands It has also been shown that the total channel width re magnetic tape system for this type of applica tion is that the signal nay be erased and the recording medium be used over end over again The quality of this type of transmission can be quired for good transmission need be no greater 2igure 13 shows a rather simple pri vacy system using magnetic tape The input sig nal is passed through a 3way pad whereby it is impressed on a band filter and also recorded on the magnetic tape It is picked up by equally spaced polepieces each associated with a dif fersnt band filter With the arrangement shown in figure 13 the band from 0 to 1 000 cycles is transmitted without delay The band from i 000 to 2 000 cycles is transmitted with say 100 nilliseconds delay and the band from 2 000 to 3 000 is delayed 200 milliseconds At the re then that of the original signal To increase the privacy of this sys ten one of the subbands nay be replaced by a band of noise This can be filtered out at the receiving end Obviously this systen requires a high degree of synchronism between the two ends 6 Systems Using Tape Recording Leaving the frequency substitution systems for the tine being we tFill introduce a rignalsigal device which permits thantha operating ofthe on the time scale The 3iost versatile uur device for this pose is the tape recording and repro qucing system This takes the form of a tape of lils magnetic alloy a few thick cither run sa a made very good with proper design ceivin end the scrambled signal is passed acysysem an identical singnageti system tae through in the Te sane iputsig way ex cept that the'two extreme band filters are im magnetic tfrchonged In this way the band which received no delay in transmission is given maximum delay in the receiving machine and the band which re qrrer polepieoes However the number of segments need not be the some as the number of pole pieces A switch is provided whereby any seg ment may be connected to any polepiece With this system the speech is EAASF PAD cut up into time elements corresponding in length to the spacing of the polepieces These time ele ments are transmitted in a scrambled order For instance 6 successive time elements which we might label 1 2 3 4 5 6 might be transmit ceived maximum delay in transmission is given zero delay in the receiver In this way all the bands are delayed the same amount and the speech is restored to normal ted in the order 2 4 1 3 6 5 The possi of TDS coding are far too complex to bilities cover here Analytical discussions are given in Preliminary Reports Nos 3 and 6 The general requirements for all TDS systems may be stated as follows 1 Each element of the once and original speech must be transitted only once 2 The sum of the delay in the trans mitting machine plus the delay in the receiving machine must be equal for all elements With This system alone does not provide any high degree of privacy but it can be con these two requirements fulfilled it is obvious that the speech comes out of the receiving machine in its normal order It is delayed Figure 13 Variable Subband Delay bined with other systems as we will see subse however by an amount equal to the sut of the quently transmitting and receiving delay An important class of scrambles in volving magnetic tape is known as tine division scrambling TDS A simplified diagram of this system is shown in figure 14 There is a re cording polepiece and a number of pickup pole pieces There is also a comutator driven in synchronism with the tape The length in time of each segment of the commutator is in general equal to the delay between successive pickup At the receiving end there are several ways of handling the scrambled signal 1 The pickup polepieces can be used as recording polepieces and the signal picked up by an addi tiona polepiece shown at the right in figure 14 With this arrangement the connections be tween the commutator and the polepieces are the same in the transmitting and receiving machines The sigal can be recorded with the sae polepiece used in the transmitting machine and the connections between the polepieces and the aegments rearranged for receiving by a pushto talk relay 3 The codes can be restricted to a particular class called selfconverse codes These have the property of being selfdecoding ERASE that is the same code which scrambles the speech in the transmitter restores it in the re ceiver An important variation of this system is called Interlace In this system the num ber of segments on the commutator is doubled The odd segments are connected to the pole pieces according to one code and the even seg ments are connected according to a completely independent code The reason for this device is to increase the difficulty encountered by the the enemy in trying one code after the other to Figure 14 One Form of Time Division Scramble find the right one particularly if the total number of codes available is pmall With the interlace system the total number of combinations possible is equal to the square of the number of codes The rotating commutator shown in figure 14 results in a repeated code that is each rotation produces the same scramble It is possible to substitute for the commutator and switch arrangement shown in figure 14 a more complex arrangement whereby the speech is scrambled in a neverrepeating manner There are several ways of accomplishing this Perhaps the simplest way to represent it is as apunched tape which permits the polepieces to be con nected to the output one at a time in any desired order permissible under the restrictions outlined above the motion of the tape and is in the same direc tion Therefore the relative motion of the tape and the polepieces is the reverse of that used in recording This is the same as running the tape backwards for reproduction IN IN •SAME DIRECTION AS TAPE BUT TWICE AS FAST i FLEXIBLE LEAD TAPE WHEEL OUT 0N Figure 16 ___ 7 Time Inversion Combinations of Time and Frequency Scrambling Obviously the two kinds of systems Figure 15 Speed Wobble Another way of utilizing magnetic tape to scramble speech is shown in figure 15 Here the pickup polepiece is oscillated back and forth along the tape mechanically With this arrangement or other variations equiva lent to speech changes the speech time scale is alternately compressed and expanded The frequency scale is correspondingly expanded and cp te r thvelyran euea a io difficulty compressed respectively 4ith the orrangenent shown in figure 16 speech is broken up into time segments each of which is transmitted backwards The uotion of the pickup polepieces ib twice ua great au described in the previous sections can be used together For instance some of the time ele ments of a TDS system might be inverted accord ing to a regular or irregular program The next more complex step is to combine the band'split ting system of figure 11 with the TDS system The codes of the band splitting system might be fixed or might be switched in synchronism with the IDS elements the time scale of the scrambled speech not being further broken up If they are switched nonsynchronously however the time dimensions will be further broken up as will be seen subsequently Combinations of nonrepeated code TDS and rapidly switched split band coding can be made to afford a very high degree of pri vacy The two kinds of coding of course must not be so interrelated that one furnishes clues for the other If for instance a certain pole piece were systematically associated with a certain spiit band code the total privacy of the combination might be impaired rather than en hlanced A coding method for avoiding this ifiut No 'I is described in Preliminary Report A very special kind of scranble is ul oduced by a system which consists functionally of figure ii rapidly switched in tandem with 91 4 figare 13 with 5 bead#% followed by an addi tiou figure n1 Thia is not the simplest form Of tbO system but it serves to illuastrate MWLTIPLIgR MULTIPt I R the DrinetiPe Two frequency Lcrambles with a time ahift in between produce a partilcular kind of twodimensional soramble in which the speech is brcsen up into both time and frequency ale merts Eaoh of these closestm may be shifted both in tim and in frequerny so as to be out of proximity with other elements with which they wrer ori inally associated either in tim or in frequency Another way of accomplishin% this kind ot scramble would be a mombination of ceiving end a reciprooal of the coding wave is system in each subband A twodimensional sys tea has been described in the Brown Boveri Lrti ale reprodured in Freliniaa7 Report Nlo 5 and anaelyzed in Preliminary Report No 9 It is capable of a very Mg n degree of privaoy storing the original speech Naturallr the coding waves at the two ends of the system must be in close agreement otherwise there will be considerable background noise in the decoded speech rapidly switched splýi band with a separate TDS For the saxe of corpleteness two other ' freiuency shiftinG will be mentioned although as far as is known they exist only on paper Suppose a saple of speech were recorded on tape and then reproduced at twice its normal speed It would occupy only half the time it took to speak the words but its frequency ranwe would be twice the normal raueg Let the upper half of the expanded fre queney rane be separated by a filter and wodu lated down to the normal range and used to fill up the unusse time The directly opposite but a alo ou system would involve reproducing re corded speech at half its normal speed the fre quency range wcold then be only half the ormnal range Alternate seotions therefore could be modulated up to fill the unused frequency space thereby keepin the total tranisittin6 tin sub stantially unchanged In both of these systems there would be a delay equal to the length of one time element sPIKC'4s S XC U C W Yigure 17 derived and Wave N ltltplioasion used as a multiplier thereby re syste ms involving ti a lgrse 18 Level Modulation AMother method for changing the wave form is aho n in figure 18 The essential fea ture of this system is an amplifier whose gali can be varied rapidly with time Drastic ohanms in the level of speech if they occur rapidly enough will make the speech unintelligible The level ohanges night be made according to sow progras or they tight be made to follow the speech wave itself For instance extreme com 8 Wave form Lodirf•otion Thichfrequsncy fans werae cshided a in which frequ ency banse were shifted around or time elements were rearranged There are a few privacy systems whioh make speech unintelligible privcy wich ystes akespeeh uinteligble of the wave form One modification direct a by of these is shown diagramwatically in figure 17 It depends upon a process whereby two waves are multiplied together that is the instantaneous amplitude of the resulting wave is the product of the amplitudes of the two input waves One of the input waves to the multiplier is speech The other is a complex oodinr wave If the coding wave is sufficiently complex the result ing scramble is unintelligible At the re pression or expansion could be used Corre sponding gain changes of course must be lade at the receiving end A variation of this system is shown in figure 19 Here the speech band is first divided into subbands and these are individu ally oubjocted to level changes according to separate program 9 One of the first schemes which is likely to occur to a person considering how to 22 nak speech private is to add noise or other disturbing signal to the speech and remove it at the other end in other words to mask the speeh He wi U find however that it is nec essary to use very high levels of maskin sig nal in order to hide the intelligibility This of course nmaes it difficult to subtract out satisfaotorilyi the difficulties are such that 0 masking systems are more likely7 to be found on wire lines than on radio A few speculative masking systems are outlined below One form of masking system is shown in figure 20 In this system two telephone lines are used At the sending end noise is added to the speech in a mixing pad and the com bination is sent over line 1 The noise lone is seat over a second line end it is used at the receiving end to cancel the noise trans mitted with the speech by simple subtraction This system has the advantage that the noise oan be completely random However sinc thb enemy might take tape from both lines and thereby be able to make the sene subtraction a variation of this system consists in distorting the noise GENERtATOR I CONTROLCKYS Figure 19 Subband Level kodulation DISTORTION DISTORTION Figure 20 Noise Masking Using Two Ohannels in sone predetermined manner before sending it over the second line At the rooeivino end this distortion is first nullified so that the noise nay be subtracted ' aturally the form of distortion uust be unknown to the enemy It can of course be varied from moment to moment SPEECH Another masking system is shown in figure 21 which uses only one line In this system noise is added to the line at the re ceivinz end instead of at the sending end Again the noise can be perfectly random Since the noise is Generated at the receivinZ end the process of cancellation can theoretically be made very exact This system however cannot be used for radio at all because the level of the MIXINGd M c Figure 21 Masking Noise Applied at Receiving 23I noise decreases with distance from the receiving stations while the level of the signal increases The interceptor therefore will get good speech signals if he is close to the transmitter With sa e level as the level of the 04eeeh in the corresponding band at the transmitting end This is accomplished by separating the signals in the various ohannels detecting them and us small variable gain amplifiers in their respective ohannels ing the resulting fluotuating do to control the telephone lines this differential can be kept to Another simple masking system its sig the on superposed tones of have a sequence receiving the At end transmittin4g the nal at end sharply tuned band elimination networks can be synchronously switched so as to remove the tones from the listener's ear Similarly short spurts of noise oovering the whole frequency band can be applied at the transmitting end and shorted out at the receiving end The spurts Ott be made to occur at irregular intervals ac Te noise is of two types dependin on whether a voiced or unvoiced sound is to be simulated For an unvoiced sound it is a hiss like thermal noise For a voioed sound it is a bun which consists of a series of harmonics covering the whole frequency range A separate carrier is used to transmit information for operating this part of the system At the traus mItting end the pitch used by the talker is mea oc ding to a neverrepeating program Both of thtse systems involve the loss of small portions of the speech either in the time scale or the frequency scale sured and this information is used to control the pitch of the buss sound The absence of a pitch signal switches the hiss sound into the system A system described in Preliminary Re port No 4 might be classified as a masking eye ter although it might be better classified a means of communicating without the enemy's as knowledge This system by itself of course in not private since the enemy can build a similar system and use the signals to regenerate speech must be achieved by operating on the Privacy channel signals One method is to permute the 10 channels at short intervals accordiag to a prearranged program Another method is to put a TIS system into the line or into each channel separately A still more effective method of this type is to apply a twodimensional scramble such as was described in an earlier section to the channels so that signal elements are dis placed in both time and frequency Vocoder Systems The vocoder system which has been de 1 scribed in the Bell System Technical lournal 2 and the Acoustical Society ournal nay be made the basis for privacy systems of various kinds The system is shown schematically in figure 22 At the transmittiw end the speech is passed through a series of band filters the outputs of which are individually rectified to form a fluc tuating do signal These signals are individu ally modulated in such a way that they Oan all be sent over a single trnsmisesion path At the receiving end synthetic speech is manufactured in accordance with the signals transmitted over the line A source of noise which covers the whole frequency range is passe through a set of band filters similar to those at the transmitting end The output of each of these filters is controlled so that it is the 11 Ohannel Ilizing Systems Thus far the methods we have examined apply to a single transmission path There is another class of privacy system which depends on using a multiplicity of paths This is of course inefficient if only a single message is to be transmitted However the method d wh a number ofh two oase s exapplbed an pointsandmatnumber of wo nages meseses would normally be transmitted over these channels simultaneously 1 Bell System Technical 3ournal Volume XIX Page 495 October 1940 2 Journal Acoustical Society of Anerica Volume 11 Paee 169 October 1939 24 L NE '4IHTANWTSFSNPKI Figure 22 Voooder System Figure 23 shows one frem of channel mix ng system Here three channels are shown connected to the three seszente of a oommuta tor Three brushes on this commutator are om neeted to the outgoing oehnnels whioh ore thereby caused to pick up one channel after the other on a time division basis Each channel contains parts of messaGes from al l three ohannels Thea commutator of course is too simple to be very effective aind would in practice be replaced by a permnuting siiitoh capable of switchiag accord ing to e nore complex progr One or more of the channels nay be fill2ed up with noise or spurious speech from a recording or other simi lar source 2 CHANNEL I CHNE CANl __________ 3 tj CHANNEL 3 r'iGre 23 channel 1LixiZ8 An analogous system which divides the messages on a frequency basis is shown in figure 24 Here each ohannel is passed through three band filters which divide the speech into sub bands Each of the outgoing channels contains@ subbands from each of the inoomi•g channels To increase the privao7 a perzuting switch is CHANNELI shown which rearranges the subbands on a time division basis If only one message is to be tranemitted the other channels can be filled in with noise or spurious speech 12 CHANNgLS S nary SWTCH The above examples cover fairly comn pletely the range of schemes that might be uied to scramble speech at audio frequencies In subsequent chapters we will examine each system from the decodina standpoint To facilitate reference to the various systems they are sum marized in the attached Table 1 This table also refers to the places where methods of do coding the various system• are discussed CHANN9l A Figure 24 26 Subband Channel Vixing DIAV'OSIS 0 U Mra SYSTIU Before discussing the diagnosis of speech privacy systems it should be pointed out that facts concerning the origin of unknown sig asle are often very necessary to their correct interpretation Such things as the frequency strength and direction of the signals the lo cation and type of receiver and the manner in which the signals were recorded can be very important pieces of data That is why it was stated in Chapter I that interceptors should be equipped with complete knowledge 4f the various kinds of radio systems and transmissions used by both sides including jamming and radar signals as well as telegraph and facsimile signals Some of these signals particularly if transmit ted with asppressed carrier can give extremely puzzling results if demodulated with an ordinary radio set These possibilities should be taken into account if signals are found which do tnot seea to fit into the classes discussed below As stated in Chapter 1I the spaeotro graph is of tremendous assistance in recognizing the nature of an unknown scrambling system The ear can usually recognize the presence of time discontinuities It can also usually recognize the peculiar quality which results fro band shifting systems The exact nature of the scramble however is usually impossible to es tablish with the ear Xven scrutiny of the wave form may yield no clue The strikinly graphio analysis provided by the spectrograph however usually takes the mystery out of the scrambling method immediately Speech privacy systems having frequency subbands will show horizontal discontinuities or boundaries in their spectrograms Similarly systems employina time division will show verti cal boundaries A considerable variety of eye teas display both horizontal and vertical bound aries liow to tell these different sorabling systems apart is the subject of the disoussian and illustrations in this chapter 1 spectrograph can be established The spectro graph is equipped with a calibrating device which consists of means for producing a complex Wave rich in harmonics from the 60 cycle power supply Spectrograms of this wave made with both the 45 cycle filter and the 330 cycle fil ter are shown in figure 25 If the power fre quanoy is known the horizontal and vertical striations in these patterns provide the time and frequency scales If the power frequency is not known the scales may be established by the fortulas given in the figure This involves ad ditional measurements with a stop watch The application of this method to 11 ke speotrograns is not explicitly stated in the figure A value of I for this condition can be found by the same formulas This establishes the time scale for the 11 ko spectrograns or the frequency scale the same pattern is used as for the %1 5 ko speotrograms However each horil zonttL striation is labeled with a frequency ob tained by multiplying the normal frequenoies by the ratio of the two K'se 7igure 26 shows how these scales can be used to measure the time and frequency bound aries in a scramble It will be noted that for aeacurizn the time elements speotrograms made with the 300 cycle filter are beet because they have sharper time boundaries 7or measuring frequency boundaries the seas filter must be used as was used in obtaining the scale It may be noted here that in present models of the spectrograph the wide filter has a different absolute location than the narrow filter and therefore should not be used to estimate the frequency of components or boundaries Illustrations of Scrambled Speech Spectrograns illustrating a lar e nun bar of privacy system scrambles are shown in fig ure 4 to 87 which are segregated at the back of this report In so far as possible these Measurements on Spectrograms spectrograms were obtained with actual working models or systese In some cases they were made with a laboratory setup simulating the systems under scrutiny In a few cases also the illus trations were made by cutting up spectrograms and rearranging the parts It should be noted in these latter oases that the boundaries are Since an important part of the diagno ale procedure consists in detemining the length of time elements and the location of frequency boundaries let us first exa %ins the procedures whereby the time and frequency scales of the 27 4 IN The speotrogazsrIlo with the narrow filter shows al2 the odd harmonics of tate 60cycle input to a special harzonic generator this isa a portion of a spectrogram made with the wide filt er The striations Below represent a beat note of 120 cycles At the left 1 a ortion out off and inverted The fact that the harmonics can be as other shifted positions# illusatrates the line@rity of the tre lined up in this as qu enoy scale At the right is a portion cut off and shifted downward by one component Since they are 2U harmoncs the base line will fall exactly between two harmonics if it represents exacilY zero frequency if the power frequency is exactly known$ both the time scale and the frequency scale are determined by the two Patter above If the power frequency is not known the time scale factor can be determined by the inches per second and the frequency by P aA equation X L a TotaLl length Of the spec troiram circumference of the recording drum Ft Number of rotations of the drum In T seconds s a Number of striations in X inches Figure 82i Calibration of Spectrograph Soales 28 29 - - Upper Spectrogram 45 cycle Filter Lower Spectrogran 300 cycle Filter The frequency boundaries are determined by comparing them with the harmonics of the calibrating lave These are all 120 cycles apart but the lowest is only 60 cycles from the base line The frequencies of the harmonics therefore are given by the above formula The element length is best determined by using the 300 cycle filter thich gives sharp time boundaries comparing them with the striations obtained by making a spectrogram of the calibrating wave with the 300 cycle filter Each one represents l lZOth of a second Ten of the above elements cover 70 striations The length of each element is therefore 1 10 1 70 120 seconds Figure 25 - Time and Frequency neasnresents nub-L77 unnaturally clear and sharp because in praotice any discontinuity causes a transient which tends to obscure the true speech along the boundaries It will be noted that some of the spectrograms in the illustrations were made with the 45 cycle filter and some with the 300 cycle filter depending on what features were to be brought out most clearly The spectrograph with which these illustrations were made was equipped with the control circuits mentioned in Chapter II and described in greater detail in Preliminary Report No 27 The illustrations are therefore clearer and sharper than those included in Pre liminary Report No 25 Yurthe•rore a larger number of privacy systems are included than in the latter report Each of the illustrations contains not only reproductions of speotrograns but also written material describing the features whereby the different scraebling systems can be reeog nized It was intended that these illustrations should be selfcontained in so far as possibli for easy reference It will be noted that in some cases the spectrograms alone are not sufficient to determine the exact nature of the scramble Cm tain systems completely destroy the typical her monis structure of speech leaviag structureless patterns which cannot be interpreted This in dicates a distortion of the wave form One of these systems which had a repeating code and a synchronizing pulse could be resolved by the oscilloscope as shown in Chapter VI figure 46 No general rules however can be given for diagnosing this type of system 3 Systems Not Illustrated Azaeination of Table I shows that there are a few scrambling systems which are not re presented in the illustrations These will be discussed in the following paragraphs In moat cases the appearance of the spectrogram pattern which would result can be visualized bý analogy with other systems The phase reversal sti tei A4 will produce a scramble indistinguishable fr the multiplication system 51 provided the phase reversals occur at irregular intervals and about as rapidly as the crossovers in the coding wave involved in El It is believed but not known for certain that they would have to occur about that often in order to make speech unintelligible The split phase system AS involving carriers 90 degrees apart was tried out in the laboratory The output appears just the same as if two speech channels or a speech channel and an interfering noise were simply superposed and then modulated with a single carrier The stepped displacement system BS is rather easy to visualize There will be time boundaries with two or more discrete conditions of displacement Obviously there are a great number of possible seQuences inoludlng the poe sibility of some of the coditions consisting of inverted displacement The irregular wobbled displacement B4 will of course be similar to B3 except that the wobble pattern will not be as simple The continuously varied reentrant dis placement 0 is praotically impossible to sim ulate artificially e@ was done with 01 If 01 is thoroughly understood however the appear ance of a wobbled instead of stepped reentrant condition is not difficult to visualize Ronrepeated code TD 73 will have oeeedoeT wilhe the same general appearance as repeated code TD It may or may not have the synchronizing pulse There will of course be no regularity in the patterns such as was pointed out in YL Ts plus inversion 01 is not diffi cult to visualize Some or all of the elements might be inverted as in AS The systems listed in 05 and 06 will both show equally spaced time boundaries cor responding to the length of the elements In 05 the harmonies would be spaced much farther apart than in normal speech and show greater slopes and curvatures Alternate elements would show rather consistent differences in frequency distribuxtion and in the degree of slope or our vature In G6 the harmonics would be spaced abnormally closely and show very little slope or curvature Words and spaces would be very long There would be a horizontal boundary in the middle of the band and the patterns in each half would appear like complete spectrograms with vowel and consonant structures apparent In both of these systens if the elements were out apart they could be rearranged to form con tinuous speech with the time and frequency scales compressed or expanded from the normal condition see Chapter VI Section 2 Level modulations E2 and 3 would hardly show up in spectrograns because of the level compression incorporated in the epeotro graph This has been verified experimentally 30 In J1 and 2 if the noise were suffi oient to mask the speech efteotively the speech oculd not be seen in speotrograms J3 is easy to visualize as is also N4 If the noise spurts are sufficiently close together however they tay produce a pattern like Hi J5 as far as is know exstsonlyon per Channel known exists only on paper In vocoder types of scrambling systems the speotrograph would show only the chanel sig male which might be either amplitude or fre scramble this type quenoy modulated osoillograma of the For wave form of of each separate channel signal provide the best mems for diag noeis and for decoding A sample of such oecil lograms which was obtained from an actual vocoder system is shown in figre 42 in Chapter 3' VI on decoding methods The various methods of scrambling such signals Kl 2 3 4 will pro duos discontinuities in these traces which are easy to visualize A sample of E 5 has not been available mizxing L3 can be done in various ways and at various speeds It will not be very easy to recognize if done rapidly No actual systems are in use as far as is known It is felt that the above illustrath and discussions cover the known scrambling meth ode fairly thoroughly It is hoped that with their help any system which might be encountered in the future can be recognised Deoodin of course is another matter which forms the sub Jeot of the next two chapters ati V XONCRYTDOGUPHIC TOOTS AQ Beginners in the study of privacy eye teon never fail to be amazed at the difficulty of scrambling speech suffiojently to destroy the intelligence The ear can tolerate or even ig nore surprising amounts of noise nonlinearity frequency distortion misplaced components gaps superpositions and other forms of inter ferenoe We can therefore very often obtain partial or even complete intelligence from a privacy system by partial or ipe4rfeot decoding and this in turn oa often be accomplished by operating on the scramble in some way which the designer did not contemplate Incidentally the fact that the ear is such a good decoding tool in ombination with 12•OTDS of this is simple inversion In this case the soranbled speech is quite unintelligible to di root listenin•g but if we kow it is inversion we can find the inversion frequency very quickly by trial Another example is the split phase system A5 The phase shifting network in the captured set could readily be adjusted to demodu late either of the two overlapping sidebands t Slightly more complicated systems are those with a simple program Again with a cap tured set or its equivalent it is usually easy to find the is program by trial The only possible difficulty in keeping step with the sending end particularly if there is no synchronizing pulse An example of this is a wobbled band these noncryptographio methods makes the produo tion of privacy systems very difficult Scrazb lin systems which look very effective 6n paper sometimes turn out on trial to degrade the intel ligibility very little although the scrambled speech usually sounds unpleasant LMost methods if they are pushed to the point where they do succeed in hiding the intelligibility are im possible to restore with good quality There are in fact very few speech privacy systems which achieve a high degree of privacy with ac ceptable quality displacement BN It for instance the wobble is sinusoidal with the frequency and the sweep limits known the problem is to keep in syn chronism In this connection a device might be mentioned which is familiar in gunfire control circles namely aided trackin With thia system changes in both frequency and phase are made simultaneously This is illustrated in figure 27 Suppose we find ourselves slightly out of step with the signal By rotating the adjusting handle forwards or backwards we ean get back into step Suppose this adjustment was These noncryptographic methods are very important because they may reduce the delay in obtaining the intelligence substantially to zero Furthermore they may render completely futile the most elaborately irregular code oha n ing hystems hich could be handled only with the greatest difficulty by straight cryptographic methods A number of noncryptographic methods are given below Some of then of course result in poor quality but the saving of time labor and equipment may be very great Each of the noncryptographic ethods has been given a desig nation whioh appears in Tables I and II and at to catch up is an indication that the motor is slow Therefore sons of the otion of the ban dle required for catching up is used by means of gearing to change the frequency drivin the actor The gear ratios are chosen to suit the particular problem With this method it is possible to got the beginning of the following paragraphs in which they are discussed These designations in the forward direction The fact that we had morAT NG A ATOR ADJUIs NG CHANG PIANoL• POSITION Di OtRENTIAL AA-M SYNCHRONOUS o'ro should not be confused with the designations ap pearing in the text which denote privacy wthods 1 Captured Set or YtLnotiona l Equivalent 1 With many privacy eystems all that we need in order to listen in is a captured set or its functional equivalent built from knowledge of the soranbling rethod An extreme example CH4ANGE FREQUENCY VAf A5 OCILATOMR igure 27 Illustrating Aided Tracking bands are variously delayed F1 Conceivably# these delays could constantly be changed with tim according to a never repeating prcgraz however would be futile because with a band filter we need only listen to one of the lesti bands disregarding the others band is Tory narrow the intelligibility may be practically complete Similarly in band split ting systoes if the switching is not rapid Dl we can follow one of the bands around the fro range The lowest or second lowest band the best Another example is theA usually is tone sequence 3 instead of trying to filter out one tone at a time as it occurs 3r5 can leave all the filters in ll of the time and still have enough speech coming through to yield the intelligence into step with and stay in stop with systezs such as alternate displacements and regular wob bles NetodsThis onornis 2 Decdin The methods outlined in this section have all been tried at least in the laboratory Their success however naturally depends to s020 extent on the switching rates and similar variables It is possible therefore that a method might prove sauccessful against aquensy scramblin g system which seems to be in the same general class as the one that was tried in the laboratory A special case in which the rejection UPP90 CUTOFF 13KMO moo 0 To I$KCd I To f I ligure 28 Band makes decoding easier concerns those systems such as A5 which depend on carrier phase to mix and then separate components There is no phase requirement imposed on the demodulating carrier unless both sidebands are transmitted There fore either sideband of such a system my be 0 Shift Filter suppressed with a filter and the remai ning sideband demodulated with a carrier of any phase The two signals in the sideband will then be AI IF mo A Mtrated 1 T T iTrelocated Figure 29 Variable For purposes such as those outlined P 1 CON R TT T0 in figure 28 Wiith this devices a band of adjustable width can be taken from any par tion 'of the signal frequency range 03000 and in any other portion of the sane fro quency range either straight or inverted One form of band shift filter is described in Pre limitary Report Noe 11 It coneists essentially of a double modulator such as was described in Chapter Ilk but with a band filter of variable width If the frequency location of the band is not to be changed the switch in figure 28 should be in the lefthand position One form of variable band filter is shown in figu re 29 ncranoh rvdueu Ti olhsas systems such as the multiplication system Ell and the T system El a oretitms it is expedient to listen to a scramble only part of the time Some of the simpler coding programs can sometimes be broken down in this nanner by trial For instance if a codine cycle has N elements we can listen to every Nth element atln ake whatever adjustments ca are needed to macke this sound natural o then listen to the next adjacent element and Band Pass Filter A Take for example a system A 2 which involves inversion about a numnber of fre4uenoiee in succession If these frequencies arc not too far apart we can choose a single frequency some where in the middle range and demodulate the whole signal with this one frequency The re a ultin g band will be right side up but displaced by varying amounts not exceeding half the total range This has been found to be quite intelli gible provided the switching rate is not too high or the range of frequencies too wide expedi ent to k With som sytm it is o oly o th frquecyrance lisen potin rather than the whole range An outstanding ex ample of this is the system in which the sub 34 adjust the syten so that these elements blenA properly This attack applies for istance to a system in which several different displace zents are used B2 A oaptu ed set of course is the easiest way of selecting every Nth ele %ant because it is usually easy to make the other time elaeznts inoperative A nother useful devioe is the limiter peak chopper In this sam class is the orn or pressor These are illustrated in figures 30 A • • They all tend to equalize the waoces sive lobes of a complex wave The peak chopper simply chops off any peak which exceeds a oer tan instantaneous volteage The compressor operates more gradually and leaves the waves well rounded If straight speech is put through any of these devices distortion products are generated because the wave form is radically to esooth the output and give the envelope wave The rectifyin action which we want here simply tbkes all the negative lobes of the signal amd turns them over As in the case of the limiter straight speech put through a rectifier of this type is about 95 per cent intelligible In the privacy system designateaA4 the phase of the speech signal is reversed at short irregular intervals If this signal is now rectified all the negative lobes will be ad•e positive and the resulting wave will be indi tinguiehable from rectified straight speech ex cept for alight discontinuities at the points where the reversals occurred in the privaoy ys tern This is illustrated in figure 32 There fore a simple phase reversal system no matter PAI modified It is found however that this kim of distortion damages the intelligibility very little These devices should be useful eagains t any privacy system in which sudden changes of level occur A good example is the subband level modulAtion system 03 A separate lim iter or compressor in each of the subbanLds will tend to sooth out the level variations and make the speech intelligibleo moo A nother nonlinear device is the resti A fier Two forms are shown in figures 31 k and • The reotifier as used here should not be oonfused with the detector The latter device also rectifies but then it has a time constant incorporated in the output circuit which tends jigun 31 SI Rectifiers t I T ORIGINAL SIGNAL V I I n PENIOD I OMRT RID I • V I I I COMIPRESSOR Figure 30 i GRID FROM P INTODE 5 I Peak Choppers and Compressor Figure 32 35 I OO RICTILECI iV I ORIGINAL REC¢TIFIED I Illustrating Action of Reotifier a how irregular should yield to rectification ox sept that distortion in the transmission process tends to obane the Wave form and thereby degrade the quality of the resulting speech It should be noted that the multiplication process Hi also results in a phase reversal every tine the ooding wave passes through zero It has been found that reotification tends to make this kind of scramble more intelligible also I connections is made from each of the bandpais filters to the output modulators whereby each of the bands in the signal is placed in the desired bands in the output Steps should be taken of course that these crossconnections do not in terfore with each other An amplifier after each band filter for instance will perform this function Figure 33 illustrates a simpler case of superposition applied to a system using 2 band shifts Es A very useful noncryptographic device is superpositico For inatane suppose we had a threeohannal Miing system such as U or 2 If we simply listen to all three channels simul taneeouly we will hear three conversations at once or possibly one conversation with two noises superposed Experience has Shown that under such conditions it is usually easy to con centrate on the desired channel and ignore the others Wo AC woodials It nay be noted here that superposin timedisplaoed elements does not appear to be successful For instance if all the segments of the commutator in a TDS machine are connected to all the polepieces the output will be straight speech with several sorazbles superposee This has been found to be completely unintelli gible In certain cases which have been met in Project 043 the privacy sets are equipped with or similar means which were intended to provide an easy method for obtaining a large number of different codes In soe eases the different codes may not be Sufficiently differ ent to be mutually private That is while there may be literally millions of different oombinatiots it sometimes happens that theresar thousands of combintios which will decode ma terial scramlled with one of the combinations Various degrews of quality of course will re ult fro thepailorncretdoig theme partial or incorrect decoding sitfo operations However as long au intelligibility can be extracted the codes cannot be considered mutually private In such cases it is possible with a captured machine to simply manipulate the dials systematically or unrystematically and listen to the result When the speech begins to sound Somewhat natural systematic trials of eas dial in turn will sometimes steadily improve the quality Something of this sort could be done with simple TDS systees also except that the use of interlaced codes makes this somewhat more difficult LP swa13 188C • Figure ° 33 One form of Superposition Decoding Another form of superposition is illuis trated by the rollowing Consider a split band system D2 in which 6 different codes are used in a neverrepeating sequence This would be rather difficult to handle by cryptographic means Suppose however we had 6 separate do coding unite each set to decode one of the 6 codes If the sorazbled signal were fed into all 6 of these decoders sizultaneously one of then would always have straight speech in its output The other 5 would be sorambled If these 6 outputs are all superposed we will hear straight speech with 5 sorazbles superposed This straight speech can be understood quite easily It will be noted that the unwanted com ponents in this kind of superposition are de rived from the wanted components and always vary in level simultanoouely with the wanted components it appears that under these condi tions they do not do much darAge ¶ hl In certain oases where there are a large number of codes possible but only a few of these codes are good codes from the standpoint of direct listening it would seem reasonable that asy code applied to the scranbled signal should turn the good code into a poor code In the 5band split band system for inttanee there are soe 3840 possible codes but oly 12 or so are considered really good Any code in the de coding machine therefore should decrease the rivaey for direct listening This has been tried in the laboratory but has not been pushed to the point of determining whether it could The split band equipment illustrated in figure 11 of Chapter III is adapted for this kind of Suiperposition A multiplicity of cross 36 compete with the superposition ethod It is mentioned because the idea Bay possibly apply to other systems which may be encountered • • A very speoialized device which ap Dlies to wire line communication only should be mentioned here because it is not very well known It distinguishes between the two directions of transmission over wires In the masking privacy systex 12 for instance the clear signal in one • 'jyD LUP 81ASID RELAY AD direction is masked by noise sent in the other direction The device illustrated in figure 84 however discriminates against the noise allowing the speech to be heard It requires a small series resistance which is built up by a stepup transformer to the line impedance The Figure 35 Automatic Deooodi occurs be used in each subband NOW _____ ___ The same method can be used for level Instead of modulation systems H2 and H disconnecting the receiver the high level por tions of the signal cause the receiver to be connected to a parallel path containing the re quired amount of loss to equalize the levels In the case of subbani level modulation HS of course a separate device of this type must secondary is connected to the other side of the line The direction of discrimination depends on the phasing of the transformr windings MC Total Inargy SPttCH ONLY Figure 34 Directional Discrimination 3 Automatic Decoding Whether speech is intelligible or un intelligible is a purely subjective matter However the uthod of ma ing speech unintelli gible involves aking physical changes in the speech wave Certain kinds of physical changes can be detected quite readily by objective means and utilized to undo the scramble automatically h The system just described operates on a total energy basis Sometimes it is possible to obtain a switching rignal on the basis of ' energy frequency distribution Consider for in stance the system using two different displace ments B2 The alternate positions of the speech band are illustrated in figure 36 In one position the band is right side up an oc cupies the range from 2 to 5 ke The alternate position is inverted occupying the range from 3 to 6 ko Since most of the energy in the speech band is concentrated in the low frequency part of the original spectrum most o• the time the system illustrated in figure 37 can be used to decode this material automatically The sig nel is applied to two band filters one covering the range 2 to 3 ko the other passing 5 to 6 ko The oututs of these band filters are reotified individually and fed to the two windings of a Obviously the moat elaborately irregular code program is completely futile if this kind of decoding can be applied A A very simple example of this is showm in figure 35 Suppose the system consists of uort spurts of noise applied in an irreguler aanner It has been pointed out that the noise must be high in level compared to the speech in order to uask the speech Therefore if the signal is applied to an amplifier detector con nested to a relay or electronio switch the relay can be so biased that it operates only on the noise spurts The receiver is nomentarily disconnected from the line whenever a noise spurt 37 TIME 0 Figure 36 I I 5C € I I REQUENCY Sidebands in Two Position Displacement System IOKC other way Consider a system in which the band is kept right side up but in w6bbled over a range sufficient so that demodulation with Ae intermediate carrier frequency will not give an intelligible signal Suppose the wobble follows an irregular nonrepeating program The is proposed decoding The signal rshed is ispressed on a network l TONU 6 6 KC 1 MOD IN •following IKC having 'a very steep rising loss characteristic If the speech band were not wobbled this net work would simply tend to make the lowest har monic of all voiced sounds the strongestoompont•a With the wobble the same thing will be true ez sept that the level of this component will undergo severe fluotuati oso Therefore the ro sulting signal is subjected to some form of automatic volume control and also a limiting action tendin to derive a single frequency Forgetting voice inflections for the derived frequency would fluctate up moment and down this POLAR RELAY 7igure 37 Automatic DeoodingEnergy requeney Distribution polar relay Obviously the relative energy in the2 fl te wiolel7 be polar differenati sorte 2 lthe entd 2 falters will be different for the 2 i diplaeaens an th rely •Ar •7•Lllbe operated alternately in the i 2 directions thereby wi lde r to r ie S• • •is SImade figu n ea in pe t the dutomaticalty conde automatically connectinG the proper carrier to the input mohlator in figure 33 to put the speech band in its normal position Obviously this will not be infallible but with displace Sednts as different as the oes used in the i2 lus tration it should operate sufficiently well to yield most of the intelligence of the messaee Naturally •' i the smaller the physical difference between th 2 positions being distinguished the ilb Ewvr hr mor f looeain m•ore false operations there will be Eswaver this method is instamtaseous even with an irreg ulerly switched system whereas cryptographie methods would be very tineconsuming 2 Another variation of this general technique night be mentioned for the sake of coa pleteness elthough it is somewhat more speoula tive Consider a privacy system which depends on speed changes M4 Changes in speed cause changes in the pitch of the voice Suppose we apply this signal to a circuit which measures the voice pitch This teohzique has been worked out in conneotion with the voooder The output of this to is a of varying frequency Th a a motor the speed change which used circuit i d h t a rstead e motor is part of the drive of a magnetic tape recording and reproducing system through which the signal is passed As the motor speed is to change the tape speed changes in such a direction as to tend to keep the derived fre • quency constant This takes out not only the speed variations but also the voice inflections However a monotone is quite intelligible rc e mup nsd f t ha o tiw e inlc n t this rgvett n voiu in frequency in step with the band wobble In fact it could be used as a subearrier in a dou ble modulation decoder to demodulate the signal to approzimately the correct position in the frequency range It will be in error however by an amount equal to the instantaneous voice pitch Possibly this amount of error will not prevent the signal from being intelligible we• know that this a ount of displacement does not destroy the intelligence of otherwise normal seech If it is desired to correct for this error two methods suggest themselves One possible method is to subtract from the derived frequenoy by a modulation process an amount equal to the average pitch of the voice being monitored This will leave a small fluctuating error Another possibility is to derive the ao tual instantaneous voice pitch by difference tone methods as in the previous section and subtract this amount fron the derived suboarrier fre quency If the displaced band is inverted in of right side up a similar procedure can be used with a network of opposite loss oharac terietics Obviously this method in either case will correctly demodulate only the voiced sufficient sufficient not someki o carryover catryove eof If not sooze kind of feet migt be incorporated in the system to vent sudden ohaages in the suboarrier frequency and thereby tend to hold over correct demodula tion for short unvoiced sounds also As men tioned above this method has not been tried but is felt to be worth recording because of the been tried out is intended to apply to irregu 1cr band displacements or wobbles B4 which great difficulty of handling irregularly wobbled would be exceedingly difficult to handle any systems by any other method 38 j Another rather speculative automatic method might be mentioned because some for of the method might prove useful against certain aultiplication systems such as Hl The code wave in the particular case encountered was re peated many times per second and there was a synchronizing pulse ahead of each cycle If the signal is applied to a synchronized cathode ray oscilloscope with a highly persistent screen a definite pattern appears because the codingwave always passes through zero at the sao time Also the speech energy tends to average out sf ter a few cycles so that the pattern reflects the aplitude of the coding wave It is quite conceivable that this pattern on the screen could be scanned optically and used to generate a de coding wave for atcratuially unscrw lUing the signal Obviously if the coding wave is changed periodically a new decodin wave is automati cally produced The only requirement is that rectifier or limiter difference tones will be generated which will lseo be multiples of 100 cycles If however the speech band is dis placed from its normal position in any way the difference tones will not coincide with the speech components If for instance the whole band has been displaced by 50 cycles then the speech components will be 150 250 850 etc The difference tones generated by a nonlinear system will be 100 200 300 etc If we now take a second difference between the output of the nonlinear system and the original components there will be generated multiples of 50 cycles The lowest component of this series will be lower than the pitch of the voice This will be true regardless of how far the original band has been shifted except for the special case where the shift happens to be an exact multiple of the voice pitch Since however the pitch is con stantly varying this coincidence is of very the coding wave persist long enough to form an average pattern on the soreen brief duration Theoretically at least a low pass filter with a cutoff lower than the normal range of voice pitch can be used as a clue to deternmn whether a speech band is in its proper location The method then would consist in having sveral decoders in parallel but listen only to the one which did not generate a onpo nent in the low pads filter L Another variation of automatic decod ing methods might be termed parallel automatic because two or nore complete decoding units are used in parallel but only the correct one is ap plied to the listening receiver To emphasize the difference between this method and the one previously discussed we will use the same ex ample namely the system with two band die placements Referrins to figure 33 suppose instead of the parallel modulators there were two complete units in parallel inoluding tbe band filter the second modulator and the out put filter One of the units is fed with the 8 ko carrier the other with the 16 ko carrier Each unit will have straight speech in its out put half the time and the other half of the time will have inverted speech displaced by 1 000 i The above illustrations will serve to show the possibilities of noncryptographie types of attack on privacy systems When a new system is encountered this' type of attack should be given serious consideration because of the say ing in time and equipment Naturally as pointed out above straightforward cryptographic attack can be made to yied'a better quality signal However experience has shown that the ear can become familiar with certain kinds of distortion and learn to extract the intelligence more and cycles A 1 000 cycle low pass filter can then be used in a device similar to figure 37 to switch the listener to whichever one of the de coding units has the straight speech For the particular system used in the illustration ther does not appear to be any particular advantage of one method over the other However the latter system coe be applied in oases where the other method might not be feasible j The parallel automatic method can be made to give a different type of switching sig nal For instanoe we might make use of the harmonic relationship between the oozponents of speech when the speech band is in its normal po eition If the voice pitch happens to be 10C cycles say then all the harmonics will be mul tiples of 100 cycles If this speech is put Sthrough a suitable nonlinear system such as a more readily with practice In general noncryptographic methods require that the signal as received be of fairly good quality In some oases the saving in time labor and equipment would be so great that if the signal as rooeived is too poor to pernit nonoryptographio attack the most reason able thing to do is to move the intercept eta tion to get a better signal In Table I there is listed for each privacy system the type of nonoryptogrBphic at which might apply It should be euphasized once more however that the method which sue cends at one switching speed nay fail at another The list therefore should be taken only as a reoomendation of systems which should be consid ered The nonoryptographic aecoding methods are suamarized in Table II S ntack 39 nYPTWRAPHIC TOOLS AM •2TEOlS uous changes with time presents formidable teoh nical difficulties at the authorized as well as the unauthorized terminals A cryptographic decoding method in volves 1 actually determining a code which will undo the scrsmble 2 restoring the speech by means of this code In the case of repeated codes this can sometimes be done rather quickly An example is the repeated code TDS system The actual codes used can be found in about 15 in utes Having found the code we can set it into our reoeiving machine and thereafter listen to the sprech directly In the case of zonrepeated codes every bit of the mestage must be hadled individually It may take a thousand or even a hundred thousand time as long to decode as it did to speak the words It may take hours or even days to obtain the intelligence from a short message meanwhile other messages will have been sent and it is obvious that we get farther and farther behind The only way we could avoid this is to have approximetely as many toeams workin in parallel as the ratio of decodiza time to message time which of cuarse is impractical if the ratio is large 2 Matching Spectrograms iiing In cases where the scrambling system involves rearrangement of the speech elements in time or in both time and frequency the basic method for determining the codes involves out tirg up spectrograms along the element bounda rise and rearranging the elements so ae to restore the straight speech An example is shown in figure 38 The criterion for rearrang ins the elements is that there should be con tinuity at the boundaries This continuity includes the position and direction of the indi vidual harmonics the position and direction of the resonance areas and in general the aMli tude as represented by the darkness or lightness of the patterns The pieces are numbered before the matching process begins and when the match has been copleted the numbers on the progran De t erminati pieces determine the code The simplest cases to handle are those involving a program which ca be determined di If the scrambling process involves in version of the time or frequency scales straight reotly from spectrograms by inspection or mea surements The reentrant inverlion system 01 nig ht be used as an example Supose a nultipli city of displacemnts were used in some irregu lar sequence Discontinuities marking the inversion frequencies appear in the speotrogram and once they have been determined by masure mants on a large number of spectrograms the progran can thereafter be determined quite read ily by using a template This template can be marked directly with the settings of the decod ig naehine which will restore the speech to its normal position speech can be restored for matching puposes by making two spectrograms as shown in figure 39 Present models of the spectrograph include means for making a aeohanically inverted pattern as well as a normal pattern The spectrogran at the top of figure 39 shows a normal pattern Directly below it is an inverted pattern of the sam material A mechanically inverted pattern is indistinguishable from a pattern produced by electrical inversion of the speech Similarly if the whole inverted spectrogram is turned through 180 so that the base line is at the bottom and towards the observer the result is indistinguishable from the case in which the speech is transmitted backwards Therefore if an element in the scramble is inverted it nay be recovered as straight speech for matching purposes by cutting the element from the meohan i oall inverted pattern If an element has been transmitted backwards it can be restored to nor mal by cutting it from the inverted pattern and rotating it 180' as described above If it is both backwards and inverted it may be restored by cutting it from the regular pattern and turn ing it around Another exa ple involving a program would be one like B2 in which two different dis plaoements were used alternately with the inter vals irregular in duration Here the time boundaries will be quite apparent and they can be measured with a suitable time scale It In all likelihood chnges of the above types will occur in discrete steps for practioal reasons The use of a program involving co•tin 41 4' 1 3 1 m m m mm - 1 my T fr m - 7 v j kW r Inversion If tho 30ml contains inverted 01mm than will appear right side up in a mechani cally imam-teal spectrogram Tho tint m1 m1 ho mmodghy rotating tho non-nt- 130 demos Rota tho positions of the bane 11m- 1n tho onnplu above Figure 39 - Illustrating Inversion of Tim and Frequency Scale in Spectrum nu-t - u- J I Ar u W4- nuns 0 men-J q mvds manna - wmw 3 - Ind-P MW It has been found from experience that matching is facilitated b% oi1laring the spec trograma by a factor of about 2 to 1 Hot only is the increased size easier to handle but the heavy photogra tio paper is much better to han dle than the facsimile paper The latter is delicate it texture and its surface is easily at aimed In this connection it should be noted that the process of enlarging the spectrogras does not appreciably affect the decoding tine in the case of nonrepeated code systems There will of course be an initial delay but in general the matohing time will be controlling and out up Sp•ctrograms o be made enl•a•re faster than they oan be matched If it is found necessary however to use spectrograas for matching purposes regularly then it might pay to adopt the technique described in Preliminary Report 'o 13 for producing large spectrogram photographically more serious of the two effects It causes energy frox a strong element to spill over into the adjacent following ele8ent in the spectrO gram This difficulty is unlikely to cause trouble in any application of the spectrograph except decoding Therefore it is felt that %eans for alleviating this difficulty shonld be recorded here A small amount of exploratory work has been done along these lines but the eibodiment of this isprovement in a speoltrograph has not been accomplished because the need was not sufficiently pressing in Project C43 The basic idea for avoiding the ob To facilitate matching appropriate means should be used for handling the elements It has been found that a slightly adhesive sur face is advantageous In the illustration of figure 38 this surface was provided by coating the boards and also the baoks of the elements with ordinary rubber cement This is also the case in figure 40 This latter example shows a twodimensional scramble Horisontal strips of rubberised board were provided for atcincelo• tis ais suitable th Bristol for matching along the tine axis souring effects of spillover is to permit the spillover to take plane in such a way as to be subsequently removable ior instance suppose a sample of TDS were recorded on the tape and suppose the speotrograph were equipped with a suitable switching arrangement such that only every alternate element wu reproduced The spillover from each elemnt would then occur in a blank area and it could subsequently be trim med off leaving a sharp clear boundary A second'spectrocra could then be made of the alternate elements again trimning off the spill over$ A logical extension of this idea wic al extesion of th ide which would save acme time would be to have two lsannitc filters and use then alternately switching seans Roth the inputs and by the outputs of the filters would have to be Once a system has been thoroughly diagnosed certain numerical properties of the codinc process will be knowt iAyvotage should be taken of this knowledge to supplement and check the matching process Lxzaples are given in Preliminary Reports Nocs 10 14 22 and 26 switched and the two switches should be sepa rated by the appropriate tine delay to take ac count of the tranmission time through the filter A third variation of this idea which requires loes equipment is to make one speo trogram in the usual manner and then make a second spectrogrma with the machine running backwards The spillover always occurs into the leading edges of the elements in spectrograms Cutting the first spectrogram in the proper h is pcrga ntepoe utn will places result in clear sharp righthand edges on each element but earh lefthand edge e pillover Cutti the see e obscured by will d end spectrogram in the proper laces will give clear lefthand edges on eaoh eleaent wlatches than edge on e lement rothe cle could then be ads between elements from the normal and backwards spectrograms in such a way as to utilize the good edges of the elements The two examples thus far cited of speotrogra matching were artificially produced by cuttina up spectrograme of straight speech and the boundaries are therefore clear and sharp frequency and he In practice the In time pactce ineand reqenc boundaries bondaies will be obscured by transients as may be seen in the illustrations accompanying Chapter IV tre quency bondaries are filter cutoffs and they are marred by overlap or underlap and by phase distortion This however is not as serious as the transients occurring at the time bound aries There is a basic difficulty here arising from the desire to obtain a hiGh degree of frequency resolution which entails the use of a narrow scanning filter The response and decoy time of such a filter is appreciable in In other respects it is to be expected that the patterns produced by the speotrograph can be improved For instance studies have soranbling systems preseeted in such a way that they pan be inter comparison with the element length in many been made which show that The decay time produces the 45 amplitudes can be re I111 i lilo-5 5 51 33 llti cltf 1 5 31 3 9 1 1 51 I ll Ill slur vii-ill tiigi Ill a i lbl cigl io 2 Iiiuli 60 lion- 1 i 1131 11 It Egg-tag 'l p I it Ian - i it It I bgi u 10 0 1 iail iil l7 Pin 33 3 1 1111 1 $3111 it it ll-unl- ll ll t ilk 3 ht-Jun KENS 0 it in h-zl I ik 1 355 3 I Rivas #5955 wrung on Enid-Evan news aw preted quantitatively This is an improvement over the rather indefinite shades of gray in the usual spectrograzs It would provide another criterion for matching In sowe oases however this might loo a handicap lor instance in TDS systems the pole Pieces ea not all of equal ot ticiency The amplitudes of adjacent speech elements are affected by this change in effi ciency and they might net appear to natch whom they really should This condition of course might be aggravated intentions lly as part of the privacy feature of the system On th whole however it looks as though amplitude represen decoding work tation should beean impr'ovement in simply related to the TDS elements No diffiI culty however was foundin matching the varia ble area patterns to find the TDS code This is described in Preliminary Report go 19 This report also describes a scheme for nullifying the effect of split band coding on the wave form This consists of modulating all the frequency bands down into one frequency band Changes in the split band aode will then have no effect on the wave form of patterns produced in this %enter@ It was also proposed at one time that the use of a whisper or monotone instead of nor mally7 inflected speech would increase the pri 8 _Ustoh1t Varifble Irea Patterns vaoy of MS systems Again this is true in that found was it but terns of spectrograms hs ben fund or omepurpsesit variable area patterns could be matched almest pattrnsoffes iethaien fouandae tawaeforsme v aptide or as easily for whispered speech us for normal r caberti d that wave trornatern offe was actually speech and with the monotone it raei moe akdiecl apectorasThey can be plyd o17 eaie Thsi1ecibdi8 lniayRpr produce the original speech Intrinsically IO 6 wae form patterns are not as good as spectre grans for diagnosing frequency shifts and the Another feature of the variable area likes However they present the time scales patterns which might be useful is that the pat move graphically and they are not subject to torus have characteristic shapesi Usually they transients at time discontinuities such as the look like a serises of damped oscillations with spillover effects discussed in the last section the highest amplitude at the beginning of each fundamental period This should enable the The particular type of wae form pat t or t us n f o ful u d wm s a v ri a ab e a e a at re c o g n itio n of cas es in w h ic h #pee ch is s tr an periodicity wrask usefuld tern founla tosth ataiabed are patn mitted backwards The characteristic proper a whether pattereqeicytbaldso be iusedt pictures Variable area patterns are more die t rpe o whtheioreunybadin tinotive to the eye than osoillographic traces They forn geometric designs that catch the eyeon Towa rd the end of prjojet 04 8 it came and facilitate matching The manter of produc ing and playing back these patterns is described to be rifoonized that there would be consider the pro able advantage in using a compressor in in Preliminary Reports Ies 1 7 and 12 An duction of variable area pattefrus This tends example of variable area patterns in process of to bring out low leve sound The distortion matching is sh wa in figure 41 taken from Prolsinry No 26 of Rpor the wave forms resultitg fro a instantaneous 0ompressidon is immaterial if th 7 are to be tsed Variable area patterns of this type only for matchning This kind of compression have been found particularly good for decoding however should be sharply distinguished from TDS systems especially repeated code systems automatic volume control action The latter is It will be noted that amplitudes are clearly represented in these patterns By matching multiplicity of cycles of a repeating code eye relatively slow acting and it is obvious that in TDS systems for instance it would alter the amplitudes of certain elementc in such a way a tea simultaneously it is possible to take ad vantage of this amplitude representation even to nake matches impossible 4 Matching Oscillograms though the wave tori itself might be obscured by other features of the privacy Ssytem For in stance the use of split band coding was once It was stalted in the previous sfection that osoillographic traces could be used instead proposed to increase the privacy of F11D systems This coonbination would be much more private than plain TDS if Judlgid on the basis of matching spectrograms particularly if the split band of variable area patterns although in general there will be a disadvantage There is one type of privacy system however for which osoillo graphic traces are required namely vocoder codes were rapidly switched at intervals not systems The signals in vocoder channels are 4' I I I i I 4 K 4 V 1K I £ S b I I er L ' I t4IR i ' z 3 fiAr4 h 6 II ' I I jf Jc F V ___ Ft Q I I V 9 ____ ______ 48 __ jut A variation of this method which hu been suggested but not tried and which should be much faster is as folles Reproduce tW•e re oorded sample through a lowpaus filter say 2500 oycles Pass it through a Ma machine end tudes to vaty sinultaneouly in the several track It has been found that if the signals in the various channels are perauted even with the sharp edges resulting from artificially pro duced scrambles the number of mismatohes tends to be about 40 percent This means that a acoo der system with its channele permuted at short intervals provides a rather difficult privacy system to decode outoff View the output of the highpass fit ter on a cathode ray oscilloscope whose sweep is synchronised with the TDS cycle Transients will occur at the boundaries of elemearts which do not belong together These will generate fre quencies higher than the outoff of the high pass filter and will appear as pulses on the scope The absence of a transient will indicate either that the elements belong together or that no energy was present Again a syetematic cycle roiy of codes should place most of the elements cor reetly noted that there is Li tendency 9 Ibe gi essentially fluctuating dc signals after they re modulated down to their normal frequency lo cation They can best be examined in the form of ozoillographio traces ligure 42 shows a set of uaistortea vocoder channel signals It will for the ampli enhances the value thealueofosi of otoillographic traces enanceso aphitra oes of of this type Without compression the lower ax li tudes are obscured by the width of the traces Instantaneous compression makes changes in the magnitude or direction of the traces apparent in the lower level sounds The patterns shown in figure 42 were produced in this nanner 5 Indicator Methods In the rollowing methods a visual in dioation is obtained denoting which of several possible choices puts the speech eleoents in their proper order These methods of course are applicable only to sues where the possible number of choices is not overwhelmingly great A natural example of a visual indication occurs in the illuetration of TDS in figure 59 When ever two originally adjacent speech elements re main adjacent in the scramble the two elements are not separated by a time boundary in the pe trogram Elements which do not belong in adja cent positions have a boundary resulting from discontinuities in the harmonics and from spill over effects The absence of a time boundary can be taken as an indication that the two adja cent elements belong together To make use of this effect the following procedure is suggested Record a saz2le of the soranble on a loop of tape Reproduce this sample through a TDS machine and make a spectrogroa noting any ad jaoenoies which occur Chamge the code in the TDS machine and make another speotrogram again noting adjacencies A systeatic set of codes shoul beore advaeeules ante which aet ofodes should be wpoked out in advance which explore Sall the possible combinations of elements At the end of such a cycle of operations it should be possible to place a large percentage of the elements correctly This oa be applied to non repeated or repeated code TDS then through a highpaut filter with the same Another example of the indicator moth od is the following Suppose in a split band D2 system 6 known codes are used in an irregular sequence and it is desired to determine the se quenoe The following procedure is suggested Record a sample uad reproduce it through a do coding machine equipped with one of the proper doeodos and make a epectropaz Certain ele ments in the spectrogram will be seen to be nor Sal speech These elem4nts Of cousel are the ones to which the particlax esde applies It is much easier to determine whether a particular element consists of straight or serambled speech than to determine which particular code was used Repeat this procedure with each of the other five codes Iach element can thereby be iden tified with a particular code varation of this procedure which should gve more positive results is as follows The output of the decoding machine used as above is rectified before makin the spectroram Rectifying normal speech does not add iarmonie componet whereas rectifying speech which con cpo talc s band shifts results in inharmonic c nents This is illustrated in figure 43 The upper speotrograa shows rectified straight speech This looks perfectly normal except that the frequency range is somewhat more completely covered with harmonics than is the case in nor zal speech The second speotrogram shows a se queies of split band scrambles The third epectrograz shows a similar salple rectified with none of the elements decoded Rectifying the undeooded elements results in a complete snear in the speotrogran compared to the reoti fled straight speech Properly decoded elements will stand out more clearly against the back ground of rectified scrambled speech 49 - 1 50 g 71 ECO nu Min l n- u au- 1 Straight' speech Rootified an I 'tf h 1 111311 Code Split Bani Scramble 94 -baJM- - L- #wu uh 3 - Similar to 110 2 Rect1fied Figure 43 - Shaving Effect of Baotifionthn on Horn and Bani Shifted Spoanh $1353 1 ti 1 43 Emit- W NON Now Ng o100 L g VICE oft RELAY I TE PW A AMROi CN APLWINiI SAIICk ITItC SCRAMOh N WO U k N IILE 60WD OLTA P4 0 2 C A SAAWLIK OF SCIRAM5LIED S K c ON T'APE fCSIL TING PATTERN CODE AWrTCMINO POINT I L TMOCLAO ii rCNII a0FKUN1 4 APPL ICATION TO SPLIT BAND DICOOING I ____ 41 ______ 4SILENT i ture 0 I 4 II it F1 I £IPM9iNT N 1 CAM 1I INTERVAL 14ICTE TRUE POM74 N 44 Band Shift Detector Yigure 45 A daptation of ipeoitrograph for Decoding Switched Split Band Scranbie knother variation of the indicator 45 Tb output of the lowpass filter is fed method was touched upon in Chapter V It con sista in sublecting the scrambled speech to a to the marking saaplifier Whenever the output 'Of the lowPUs filter is Beroc there will be no ones tone lower t'han the pitch of the voice In cessively at each of its 10 values or 5 if ure inort ance of omonnt are fich scrambed s invrswiong not is Tetie usure wisp osed Rpal tevies win ioll relte a d bec awaernc aus o 51m tec f t te ltr agien tins thwese If asetheng trces ihe ln ofqu upt a th oc Th n the ih o fact each lowe lie larepresetsatsilent intwervach setingle blan ince i decterminetfor each elseent is a foe beused to therv iru wteilisem contnoksomethisgindicathe init popr reuncylocwtich freuncy band it these settings were the correct ones If none of the marks for a particular element are came rox that verted The spectrograph night Ie used to speed up the analysis process as illustrated in figure tem has not actually been tried in this complete form but enough work has been done to show that 7X j2 51 it is possible to make use of the presence of inharaonic components in som such mazzr It cated noncryptographic methods apptar most o' mgt the elements in a 2dimensional scramble be identified as to frequency location S syetem yield very poorny to 7 noncrytograhi attack For repeated code systems however the code cat readily be de termined by zatching either speetrograas or variable area patterns# taking advantage of the numerical properties of the codes These zeth ode are covered in Preliminary Report No 14 Nonrspeated code systems however have thus far been found exceedingly diffliult to handle although the methods of Sections 9 8 and 8 above apply Efforts in this direction are described in Preliminaery Report No 28 appears therefore that a substantial fraction of One other possibility of this type might be menticced Variable area patterns of rowel sounds have characteristic oonfigurations These configurations depend on their harmonic structure and a disturbance of this structu Se should chamge these Patterns in a recognizable nanner For instance if the components are inharemnic there will be no perio4oioty at the fundamental pitch rate It might therefore be possible to use varitble area patterns which can be produced nuch nore rapidly then spec trograma as indicators along the lines of the above discussion 6 Application to Table I In this section we will examin e the application of oryptodraphic rethods to the specific sora•blizg systeos listed in Table I In this table the systems which might require cryptographic attack are indicated by a refer i ence to a page in this section The follo wU paragraph numbers refer to privacy systeas in Table 1 iA4 Among the systems listed under single modulation the only one that might require oryp tographic treatment is the phase reversal system This system is a special case of the multipliea tion system which will be treated later B4 08 Among the double and triple modu lation systems the irregular contirious displacements have not been handled by non cryptographic methods It might be necessary to take a continuous series of speotrograms to determine the displacements as a function of tins This might some day be done continuously and instantaneously in which case conpensatind frequency changes might be mude continuously by hand to decode the material systeMs i band splitticd Aoow the Dl e i can od es switched be slowly or the fixed solved by inspection as discussed in Chapter IV on diagnosis If the code is rapidly switched however single elements seldom contain suffi oient information to deterUins the codes If the switsaing sequence is a repeated sequence it nay be worthwhile for the sake of quality to determine the sequence end get in stop with it In this cooe the nethods described in Section 5 above should be of assistance If the switching sequence is never repeated the indi reatonable 7 r 74 Speed variations according to sote preliminary laboratory tests are rather ineffec tive in maeking the intelligence of speech un less the variations are exceedingly wide and rapid Technical difficulties then become so great that this appears to be an unlikely pri vacy system by itself Small variations in speed heover might be used to make speatro grams of TDS system more difficult to match In this case however it will be unnecessary to determine the speed variation program if the MS scramble can be removed 01 0 G8 Coabinatione of TDS and froe quency scrambles are interesting fro the cryptographic standpoint Since repated code TS systeus were found easy to break it was proposed to add various forms of split band scrambles It was argued that the continuously changing frequency scrambles would alter the shapes of variable area patterns so that they could not bm matohed Furthermore the changing frequency scrambles would sake speotrogrus un suitable for matching especially if the split band codes were switched nonsyschronously compared with the •8 boundaries Each tine the frequency code was twitched a new vertical bourd ary would appear in the speotrogram and in com bination with the TS boundaries the spectrograms would be very severely broken up in the tinm scale It was found hovever as d esussed in Preliminary Report No 19 that if the TDS code is a reopated code the frequency scrambles can be practically ignored in matching variable area patterns Having found and removed the TS code the remaining frequency scramble can be solved by noncryptographbi methods In the case of nonrepeated IDS how ever the addition of split band coding would increase the difficulty considerably provided that the two coding systems do not provide clues to each other The most promising method for Figure 46 Illustrating Repeated Code Uultiplioation System handling this system appears to be to determine the split band codes first by the methods of Section 5 above If the split band codes are then rtnovrd the remaiing scramble can be hand led aa straight TDS Another possible method is to make variable area patterns with all tht do oodes superposed as described in Suotion St in Chapter V The resulting patterns however gardless of the value of speech signal at the mosent If several cycles of scrambled speech material are superposed therefore they have the appearanse shown in the photograph figure 46 The superposed traces show a definite pattern with regions of high and regions of low anpli tude and also sharp indentations These lat tsr are the crossover points of the ode wave will not be as satisfactory for matohing at pat teWns of straight speech There iw also a marked tendency for the peaks to occur alternately above and below the center line but the amplitudes of the peaks are not al alike jince the speech amplitudes tend to average out over a nunber of cycles the ampli tudes of the superposed peaks reflect pretty accurately the amplitudes of the coding wave at those points The probable shape of the coding wave based on this evidence has been partly traced in 04 The twodimensional scramble can be handled by natching spectrograms if a repeited code is used Experiments along these liaes are described in Preliminar• Report No 22 If the code ts nonropeated however it would be ox ceedingly difficult and time consuming to htinl It would help dle by unaided matchiMn considerably if the original frequency location of each element in the scramble could be deoter mined This night be accomplished by the methods described in Section 5 HI aeteroining the code for multiplica tion or phase reversal systems can be accom plished quite readily if the code is repeated at sufficiently short intervals In the one systen which was Met in Project 04 3 Preliminary Report Mlo 18 the code wave was repeated 100 times per second In this case the soranbled signal could be applied to the vertical plates of az cecil losoope with the horizontal sweep synchronized with the code cycle It ic obvious that every time he coding wave passes throuEgh zero the soranbled signal also passes through zero re It has been found experisentally that if only the crossovers of the coding wave are reproduced the speech will be intelligibly de coded The decoding wave need not be the reci prooal of the coding wave It can be like the one drawn in at the right in the photograph It is only necessary therefore to generate a wave having its crossovers at the indicated points and reverse the phase of the scranbled signal of these points H2 HS Level modulations by theaselves are but with they other might systems very well used not in private combination in be an attempt to foil the matching of speech patterns The level modulations themselves however need not be solved cryptographically 31 There appears to be no method either cryptographic or noacryptogrehio for breakins the noise masking abod if the noise is predis torted random and suffiocietly high in level to really ask the speech Thee requirements however make the technical difficulties for system operation very great and it is unlikely that this method can be used over radio channels Cracking this system therefore becomes a matter of solving the noise distorting systea Project C43 has had no experience along these lines 1l K2 18 14 Scraebled veeoder chan nels can theoretically be solved by matching osoillograns Actually u mentioaed in Section 4 above this procedure is very difficult because the channels look so much alike 7 Deternination of the Message ply needs better execution than it received in the first attempt TVe objective of deecodin work is us ually not to determine the codes used but to lean the intelligence which was tranmitted un der these codes In the case of repeated code systems the procedure for obtaiting intelligence is obvious once the ode has been determined by the methods outlined above It is only neces sory to set this code into a mAchine similar to that used at the receiving end of the system being eonitored and listen directly to the transmitted speech If the naterial has been recorded while the code was being deternined the recorded material con in general be decoded in the a ne way __ In the once of nonrepeate od e eye te is the determination of the code sequenceegaaino leaves us in general a long way from the deter nination of the message Obviously all the material must be recorded in scrambled form It is aeoossary during the process to establish time reference points in the scramble perhaps by superposing clicks or spurts of tone during the recordine process and referring the code sequences to these points A doeoding aaohine must be available such as the one described in Preliminary Report INo 18 which is adaptable to a variety of coding system The code sequence 54 _ here are some alternative possibili ties which uy apply in special gues In the case of nonrepeated code TDS for instance the process of matching variable area patterns has actually restored the speech in reproducible form Variable area patterns can be played back just like the sound trackhs used with motion pie tures A playback neine of this type is describ4d in Preliminary Report fo 18 The re aranged elements are moutned on a strip of ad hesive and scanned With a light slit and photocell Considerable noise is caused by the Joints between the separate elements but this could be largely elizinated by a specially de signed squelch circuit perhaps controlled by a separate light bean and photocell to out off the output wherever a joint is passing under the scanning been The first attempt to use this decoding method was unsuccessful as discussed in Preliminary Report go 28 However there is nothing basically wrong with the method it asi Ll L2 1 Channel mixing systems woald be exceedingly difficult to handle cryptograph ically if a sufficient number of channels Were involved so that noncrytographic methods were inapplicable The only possible method of at tack appears to be matchinj spectrograms Since however about 25 percent of normal speech con sists of pauses many of the switch pointA will occur in these pauses and it will therefore be difficult to establish continuity by matching 1Žk 7_T7 mudt be set into this machine perhaps in the form of a punched tape The scrambled aster al must then be re produced and fed into the enahane maintaining proper sanchronism between the reproducing and decoding systems Obviously this is a very formidable job If the solution of the coding system requires speotrogran rather than variable area patterns it is still theoretically possible to play back the rearranged pieces A playback ma chine for spectrograms is described in Prelia inary Report No 17 This first model requires a negative transparency of the spectrograms to be scanned by a light slit and photocell with a multifrequenoy light chopper interposed ahead of the photocell Again the method is basically sound The experimental machine described in the report needs considerable improvement before it will yield adequate quality for the purpose described above in order to overcome the de ult aue ytejitb gradation of quality caused by the joints by slight nisplaoements of the elements by epill over at the boundaries etc Jurthermore in order to get good patterns for matching the signal must be subjected to a very high degree of onspression which distorts both the time and the frequency distribution of energy It may be necessary to make one kind of pattern for match iu and another kind for playing back as wae done with the variable area patterns described in Preliminary Report No 26 As a final alternative it is possible to learn to read speech speatrograu by visual inspection Theoretically therefore the rear ranged speotrograns might yield the mesage di reotly Fere acain however the boundary distortion will increase the difficulty of read ing the ptterns It has also been found that the beat patterns for matching are not the best for reading and it aax be necessary to make two ' 55 sets of patterns However since spectrograas have been continually improving the possibility of visually deterziningth3 intelligence froa rearranged spectrograzs zust be listed as a adi tinot poseibilityý ana one which ifit is feaal ble is the zoat general of all methods since the basic procedure is the same for all of the sorambling methods which can be handled in this manner PAC TCAL EVAiUATION OF PRIVACY SYSTEMS The material in the foregoing chapters is intended to be useful not only for possible interception and decoding of scrambled meseesges but also to aid in the production of new privacy systems and to estimate the degree of security which we might expect to obtain from these or other systems Uxperience has 1hown that there is a strong tendency to underestimate the secu rity or military value of a given privacy system as soon as laboratory studies have indicated that the system can be cracked In this chapter therefore an attempt will be made to point out the great difference between what might be termed theoretical or laboratory evaluation and practical or field evaluation In order to balance the effect of the previous chapters this one is written from the standpoint not of the man interested in decoding a system but of the man interested in getting a practical privacy system into use in the field 1 Orackina Time The objective of a laboratory study of a privacy system is to obtain some kind of qusa titative measure of the time or effort required to decode the system The questions are How long does it take to determine the code and how much equipment and how many people are required In general the procedure is to acquire a pair of actual models of the system under scrutiny ThI coding and decoding processes are studied in de tail possibly with the aid of nathematical anal ysls to determine whether there are anyweakmesm or any oharacteristios of the coding process of which advantuge miGht be taken to assist in the oraohina process Possibly a noncryptographio method will be found to apply In this case the oraoking time reduces substantially to zero If nonoryptographio methods are not applicable awilable cryptographic 'ools and methods are brought to bear Usually a new scrambling aye toe will require modifications or ohbn es in the existing tools or techniques Possibly the basic methods can be improved for use against this particular system or possibly new methods can be devised Presumably after all this develop ment work the project personnel will have become skilled in the art of decoding this particular system The oraoking time can then be determined quantitatively perhaps with estimates as to how far this nay be reduced by further skill In the oase of repeated code systems the craoking time determined in the above way substantially represents the total decoding time because as mentioned previously this code oan be set into a receiving machimn and the message obtained directly Some additional time might be added however for determining what was said during the time that the oode was being deter mined The procedure outlined above is very well illustrated in the series of Preliminary Reports covering the development of cracking methods for the repeated cori TDS system They include rnathematical analysis Fos 3 and 6 the 1 and development of a new decoding tool eos 7 and the reduction of the docodiag technique 14 In the case of the maLl to a routine 11o tiplication system the ohronoloical steps are all listed in one report No 18 Too often the cracking time as de termined above is quoted without qualification to describe the security of a system It is of course usually understood that the use of this figure involves the following usumptions 1 that the enemy kows all about the eoding system 2 that he is equipped with an adequate supply of the machines our own models may still that he be far from the production stage 3 has developed the same decoding tools and tech niques that we have some of our tools may be entirely new and secret 4 he is equipped with an adequate supply of the decoding tools 5 he has trained men in their use and 6 he is in a position to receive a good signal free of inter forense Such aesuaptions certainly represent an extreme possibility Experience has shown that there is a strong tendency to forget just how extreme a condition such assumptions repre sent Bren if the assumptions are valid there are still other factors which affect the mili tary value of a privacy system as will be die cussed in a subsequent section 2 Nonreneated Code Systems If the code is hanged periodically it may be necessary to have several decoding teams working in parallel in order to keep up with the transmitted material as was mentioned in the previous chapter The number of teams whioh will be required depends on the relation I j between the intervals of the code changes and the cracking tine No particular difficulty presents itself in expressing the decodin of fort under these conditions in term o ___ hours The evaluation is complicated however by the necessity for additional equipment not only for decoding but for recoriing In the cue of nonrepeaGed code s97 teow the cracking time for any given portion of a message will in general be Iona compared to the duration of that portion of the message Zvery portion of the message must be cracked in dividually ad the decoding effort can be am pressed as a ratio of decoding time to message time This ratio may be 1 0Q0 or 100 000 to 1 that is each second of nmssage will take 1 000 or 100 000 seconds to decode Gonversely it would require 1 000 or 100 000 teams to keep up with the messags as they are spoken This kind of evaluation is somewhat unsatisfactory because the length of tine it will take the enemy to determine the intellipece in a particular sentence which might carry mill tar information will depend on whether or not he im at the moment working on this sentence or whether he is wasting his tivA decoding previous material which might contain no information of value to his It ha in fact been proposed that the seaurity of such highprivacy systems could be materimlly enhanced by keeping the circuit 100 per cent busy with all kinds of material possibly even from recordings making certain that the eaey has no way of determining when the circuit is being used for passing important information As in the ease of noa repeated code systems it seems a bit unrealistic in evaluating such a system to assume that the ineny will seBise upon the few seconds of message time which are important and to compute the length of time it will take him to decode that portion of the s 8 Code Analysis 1 any schemes have been proposed for generating everohanging codes by a combination of short cycles geared together in such a way that the number of elements in the cycle is the product of the number of elements in the indivi dual cycles One scheme is to use odd ratios such as 99 to 100 so that the code cycle will not repeat until the smaller wheel has %ad 100 revolutions In other words there are 9 900 steps in the code cycle before it repeats Another scheme is the cycloneter type in which one wheel rotates one step for each revolution of another wheel Again the total cycle is the product of the number of steps on the vidual wheels indi Such schees should be dofstinuished from truly nonrepeating codes because wherever cyclic processes are used they are subject to analysis This is a matter pertinent to the field of cryptanalysia and will not be discussed here In general it may be said that the dif ficulty of solving such long cycles is not de termined by the total length but rather by the length of the individual suboycles Systems designed to produce a long code sequence usually contain provision for read justing or realigning certain elements period ically or from day to day Assuwing that we kncw all about the system except the momentary settings estimates can usually be made of the length of tim and the number of people it would take to determine the unknown settings by ans lysing a given sample of the cod sequence The analyst requires a kncwledge of the code for a long sequence of soraubled speeoh before he can begin the work aimed at determining the unknown setting Es must obtain and solve a suffi ciently long sample of the scramble by the methods outlined in the previous chapters and then analyzs this sequence to obtain the set tinge Too frequently the evaluation of a coding system is based on the analysing tim alone whereas tho time required for solving or unscrambling a long sequence of scrambled speech nay be overwhelsingly greater than the analyting time In fact if there is no way of solving the code sequence from the soraable alone then the analyst can contribute nothing and the system is still secret regardless of any inherent weak ness of the cyclic coding system EvaIuation The continuously changing military sit uations of modern warfare require rapid means of e communication in order that the required mili tary actions can be taken Obviously a perfectly seouoe speech privacy system is of no military value if it requires so much time for encoding and decoding that it slows up the conunaeation system to the point where appropriate steps can not be taken when needed Obviously also a oraoking system is of no value if it is too slow to permit counter measures to be taken according to the intercepted intelligence or certain purposes 15 minutes or even 5 minutes cracking time is much too slow Where this is true a privacy system giving 15 minutes or 5 minutes privaoy is just as good as one with an hour's security This is important because systems which afford a few xinuzea of privacy uan be produced in portable form whereas those pro vidin loner privacy at the present cannot itigAdvantae iht als be taken of te Avnaemgtas etkno h element of surprise consider also the equipment and train ed personnel required for decoding intercepted 0oounicationa As a specific exaaple the small TD unit required about U• minutes for de coding but it required a vanload of highly speoialized equipment as described in Prolimimary Report No 24 Suppose the small portable TDS unit were used in mazy planes and tanks and other mobile equipment that required som privacy Suppose also that different codes were used within different groups of units and that the codes were changed at some reasonable interval Would it be worth the enemy's while to provide enough decoding equipnent and enough traid Suppose we suddenly introduce in the field a lowgrade privacy sye tem in large quantities 11ow long would it take before the enei diagnosed the systea developed a decodin method manufactured receiving sets of the proper type and also deooding equipment distributed these where neeaed and organised end trained personnel to use then Until he has done theoe things the units provide complete seorec7y different kind of systen night then be introduced which would again prorido aeorecy for a tim m It ta intended in the foreoin sidrto to point out that there are other onsiderations in the evaluaticn of privaey systeas than the tire it takes a highly speoialised group such as the personnel of C43 to decode the system under the ideal conditions of a laboratory The decoding time alone is so often quoted because it is the only element which tan be described quantitatively While there is always theoreti cal agreezent about the existence of the other considerations apparently they cannot be pointed out too often or too stronaly personnel to follov these units around and de code their neseages If it is not worth his while then units rated as low in privacy W provide high grade privacy under such conditions Obviously the foregoing does not apply if the unite are used to convey wessegs between the hiaher echelons of command In such cases the meseages have a longer term significance to the enemy and he can u'ford to devote consider ___________ __ able time aza equipment to intercepting and de oodina them C 4 y K 4 K S I I I liii II it I I fjfjjj f liii iIiiHhI i i II a I I 4' V I I Iii I' ii4 Ii A' iyi ii 1' ij liii j A'I I 2 _______ WC¾ r t I flrzrc I 'p ______ ' 17 It 1 k V A A 1 L ti It u I iii 41 l j IA U3 R'I fjJf j r 11 j '4 ii I 1I I Eg 5 I j d I IfII III S a Ii Ill IfIa tiIg a 63 1 I 2 Embr- 8% 14 1 2 2 21 22 121 22 2 1 1 2 141 F may T 2 2 92 322 1 1 31 22IA2 2 p 2 and are nun-or no nPdunIn u2nuao uoaahaal vi Innuv215nnhalw wanna- win nunpiunalw Ipullaau ha a unvln pain I 7 Inn 0 oval IHHBI 5R a I IHI son pad-noon as nonuwnn nu on an aunt than H are Hal-H nulnoualyll unanllvu nun aapu plo oaanwnuaau In unuawiun new who trial tuba-I alu vl nHIInHH nuIalnulno pal cubic HE nameluno Annd naai an taupe-an Quin-nah Hula n ua o lunlnulnl Inning Ian Had-Hols annuilnnl na Hahn app-Bonn nanub a gunman-092 Ilwuu I Mac-Hols In Il awuun planar-12Iuho i aunt unini- nxal nu pic Hun cinnarunn Hun-uduul In and-nuwb c I pr no Hounakphna nee-n15 Iapvh 93 an nailnon rd pawl naidapb penal coupons Inmuv are nunwlu menad w I bHvau lno Hnionuwou 4 z m - r 5-4 4m u unBojn uma can I 8 gm gagig gi n iggug agn nagging sang-gonna a o gh gu iig galggo af gag nsiaad gag 2331 50 09 38 angugu ggsidlioa ggo 33 2333 530353983 V0 3398831333 3 Ei 1 1 L E a -7 3L fars p qu 5707 72 52 32 3 - Qwigi g g T Erl mf w 7- u W71 313 431 335Taxi a a az r 341 11 3- 1 Ulrggugilgon ga p rt 7 gain palladi- 5E 3 3 goioglil glganagi await-E3355 announce gon lglgul nowp sni uuvg gigs l e 0 i gov-igeginrogguir mag on as an In lg-084 na Ed annual-I I Sharia-adv 1 Ilnllo A InJail l u l I I h nomakob H drawn-am I Dunn h gaaoggi'ia ggaauoslglagga IIEO 8ngigg-gulauaggua i gagu ae 9 83 3 u3holua 43 33003 jga liaho Ail gunk sagoaaon i oi an 3335 - 31030053 ng ga asg gunk-bun lggigg ggsgi iagigglg a Sumac-gag iugx xggoaub v r111 a gr-rno-nr-r-w litmus Pm ands pug - as un t Hal s-109000 13mm Inc-unm- mzmomudopmm mqmw mm-minumo 'p and mama-om ma a 'Iunodz Hutton-Adam 'onx luau-m- In mm queasy-mu 'm mum an uncut 10 pm In unm- mnouw noun no to tutu-nu our in non wm I u lI qumu muum'mu-m oqzuqmqotm nnoxo mtanmnI-nmn I'm-Dianna 'opoo a Q1 lwh 1343 usable a 2 an 1 HI ung 1 rmt r r n 7 I 1 f '1 j #1111a ii11111' I dI I 'B A I I i iltiiiIi iJ II j _ K 'iy K t2Z I 4 1 LThLTJV I I I I A ii 2 4 U A I £ 4 w C I 1 13 1 I iLl 'Ii' jF 1 IiII A U S 1111111 iljJj 11111 1 77 I ii i U 4 4 K i ji f i 1 if' If 3 t4 I 111 1 J I U jlj 4 I'j i I ii jIII I lit II iji 4i jJ a i A ftp El-7r This 1 tho no Ill systo- no tho 11-011mm illustration Int a bond or non hon boon added to inure-no tho Irina Ono mm- of tho highest of tho tour from hands into which tho channel no dividod h1- hom r111 with thorn-1 nolno In tho opp-r npootmmn this noioo no study and in tho lower one it go - torn-d onond off about 4 t1 - por UQ HOto that although tho min on introduood into only on Johnna 1t nppourn 1n nah of tho four nidohndn 1n the than pttornn m1 nhovn that 111 ml 0h snob-n11 con- ulna oompononta tro- onoh Inb bnnd - a Figure 57 - Time Division ultiplox lith Hoiaa Channel - 4A 1 A Eb w n em noma a can mm 8 93 il I 314i l jirllirl 25 r1 1 - - rm-4 v55 4 7 7 13 uL mx 2n ommos pm and amounts m an 0 titanium-m - 09 can Winn-I103 Jog-amompuS-mmw taunt-Wm lap-panama 'tmmouuoddnuqau onnunttuddom ozu I zv mm Manama an u1mluoupJquvIumon f1 140 A U 4O O h ajA 144 13 0 4 Oh 11141 i 0a 47 i 1 4 g IN o l 3 4 nun-ulna a lggi gghgiaaggd gaggaga LIA-IF gng aguahnig 3 35092558333053ng33 155th 9 35 333 g a i naaS-n s lli QilarI I I 01 111 1 ltrl liq hlwuchI 11ltti 17 3 73m - 91 1 AWL WW I 5 In ws 9 53 1 2m 8 me gniiq au ia g zuo 19 422hi 1i in fit E 1 filL ii n I jsiX I I j t 1 ii IiiI ii 1K it I j hi Er 1 95 I J i'a n wn I Iii j 1Ri Ix u h ' A ii P11 Ai' The upper'apeotrogran shone a continuouo noiae nith aeteral eorde or eyllahlee shoeing through counting the harmonica aha-e that tho fundamental of the nniee in aboutpl cyclee Examination of the aignal eith an oncilloecope ahoee that the noise onneiete of abort pulaeo about 10 1111- aeconde apart These can he removed hy'a blanking circuit - the loner spectrogram shone a aanple not the aane as the one ahoee eithont tho pnleeo the outstanding charecteriatic an in the aanplo ahoee in an alnoet coaplete lack of tho etructnre of unreal apeech alao tho energy in dintrihnted nnre or lone evenly oeer tho ireqpenoy range for each enmd or ayllahlo Ihewe are no characteristic recon-non arena There are no regular boundaries either vertical or horizontal the aeqnonce or eordo and epaceo looko normal in the opectrogran and has the annual oadonce of speech to the ear Those cherecteriotico are to he empeoted ehen the scrambling qyetee operatee on the none can directly In this particular system the apeeoh none nnltipliod hr'a codina eeoe the latter'eaa repented 100 tinoo per aocnnd nith pulse bet-eon eech cycle It in obeiono thot a of required to remove the cod- ing nave at the receiving end nhich actonnta for the of tho pol-en It any he noted that phone at a irregular rate to aohiove privacy in aloo eoaentially a multiplication process extent thot the hue no ealnoe othor then pane and ninna unity Spectrograne or ouch ayltenleonld be expected to look the Iigurn 65 - Multiplication - ex r 7 a u mt w 14 mm-h -- I why mung 9 3 H3555 Santa II I we a oiogo ulguong gig gaggaaiaglagigag ni5 eg NH Auggili oB Egn ego In I lilli lurk I1n1 irt h Err II I a lid TT Zgh ___ ___________________ _______________ I I 1 ilJf if I I I I I I It 1 44J Ii ' j i a 0j I El kl i1 mi hi 1A a 9 1111 III 11111 a ' _____ __ 101 a Ln wiloll- 531Iotll-oln 9 run lull-ii5 llun 1 1-- Engatgagiigg uvga yegg bagsggig Ito-3 I magnum Eva nrll I 4 @555 Luz 7 61 7' fun - 1 A 14 4 14 1 4 1 N A w 4 I NiI' 44 4 4 If I 4 ui Ow 44i 14 14 0 a A 10 a 9 po 4 4 A 4 103 A PC Q 04a NONCRYPTOGRAPHIC Dl•CODI•D I•TBODS Used by Zisousaion 27 yes 33 38 28 and 29 'es Yes 30 ABsC 31 AB and 32 33 Yes yes 34 Yen No 35 yes 37 ITO 38 Diagram 1 Captured set or junctional Equivalent a find bytyen lze into stop Program got Simp•ale Condtion 2 Cozproxalse Deood iz Uethods a Interadiate Condition b Listen to Portion of Frequenoy Band o Listen Part Time d Limlter Peak Chopper Compressor e Reotifir f superposition a Approxiatm CMe by Trial h SPoi l Uood Code by Reooding is Direotional Discriminator 3 Automatio Deooding a Total Energy b 0 Energy Yerqumnoy Dietri utioe PitOh Change Corrector do Wobble Corrector o Code Wave Generator tTO f Parallel Autozaria as Inharmonio Detector U and 37 34 34 34 35 35 Se 38 37 No ITO 37 38 No 39 3o 39 ITo 2105 AT1 r ser TITLE Final Report Part I Speech Privacy Systems Interception Diagnosis Decod ing Evaluation u AUTHOR Sj Koenig W ORIGINATING AGENCY Bell Telephone Labs Inc New York N Y PUBLISHED BY Office of Scientific Research and Development NDRC Div 13 SATIl Oct '44 ABSTRACT ICC 0 5 Secr L AN4JA U S Eng ASU I rIStUSTIASFON None 0410 AoSS NO 4 None 4573A '1'1 1 photos tables diagrs The results of three years' experience in diagnosing decoding and evaluating speech privacy systems are summarized Speechprivacy systems may be used in connection with radio telephone systems or wire systems but radio interception problems only are discussed The decoding techniques described apply to wire as well as to radio communi cations The sound spectrograph is described including its history method of operation and capabilities It analyzes speech in terms of its three basic dimensions frequency amplitude and time and portrays the analysis in the form of spectrograms Basic speech scrambling methods are also explained in which the original speech is transmitted with its parts modified displaced or Interchanged Cryptanalysis and cryptography which apply to telegraph types of communication are also described DIVISION Electronics t SECTION Communications A CSIWOY 29345 $ ' J SUBJECT HEADINGS Communication systems Secret _ 23992 87 Decoders 28877 A 2 0 CAL INDEX WrightPateron Air Force Has DOy on OhN s
OCR of the Document
View the Document >>