THE CODEBREAKERS

by DAVID KAHN

  Another simple technique is the band-shift. This is a kind of telephonic Caesar substitution, in which all the frequencies are shoved upwards or downwards a certain distance, with the portion pushed out of the frequency band reentering at the bottom or the top. For example, a factor of 1,000 might be added to all frequencies in the 300-to-3,300 band, so that a tone of 500 c.p.s. would be shifted to 1,500. One of 2,800 c.p.s. would then be enciphered to 800.

  Band-splitting splits the frequency band into several smaller bands and interchanges these. Filters can divide a 250-to-3,000 band into five subbands of 550 cycles each: subband A of 250 to 800, subband B of 800 to 1,350, sub-band C of 1,350 to 1,900, subband D of 1,900 to 2,450, and subband E of 2,450 to 3,000. Then the scrambler’s switches and circuits may replace A by C, B by D, C by E, D by A, and E by B, thus jumbling the normal tones. The better band-splitters shift these substitutions every few seconds or milliseconds. The result sounds something like a recording of a Mah-Jongg game played too fast.

  Masking systems bury the voice signal in noise. The music from a phonograph record can be electrically superimposed on the voice, drowning it out. The descrambler, which must have an identical disk precisely synchronized with that of the scrambler, subtracts the phonograph signal out, leaving the voice. These systems resemble null ciphers, which interlard the true message within a welter of spurious symbols. Another system is wave-form modification. A fluctuating electrical current operates upon the voice current to produce rapid and extreme variations in the amplitude of the transmitted speech. This sounds rather like a radio whose volume control is being turned up to full blast one instant and then down to a whisper the next. In the de-scrambler, an identical synchronized current reverses these effects.

  All these encipherments transform the speech only in the frequency dimension, along the vertical axis, as it were. None extends horizontally along the time axis. Systems that encipher by changing the temporal relationships of speech’s continuous flow must preserve it momentarily to permit the transposition. Usually they have used magnetic tape.

  Time-division scramble, or T.D.S., chops the stream of speech into split-second portions and shuffles them. It does so by tape-recording the voice and then picking off segments in jumbled order, using, say, five pickup heads that a mechanism activates in mixed sequence. The result is a literal hash of sounds. The descrambler uses five recording heads to lay the sounds back on a. moving tape in their proper order. Another tape-based scramble, the wobble, slides a pickup head back and forth along the length of the tape as the tape passes beneath it. As the head moves opposite to the tape’s direction, it will read off the signals faster than they were recorded, and these will sound higher than normal. As the head moves with the tape, it will read off the signals more slowly than they were recorded, and these will sound lower than normal. The result will be an alternation of squeaks and growls, sounding exactly as if a phonograph record were alternately raced and almost stopped.

  Most of the basic scrambler systems were invented during the 1920s and 1930s by engineers for the growing radio and telephone companies. A need for them first became apparent when the radio hams began listening in to the conversations of erring husbands and their wives and on stockbrokers giving tips on the first public radiotelephone service, offered after World War I by the Pacific Telephone Company between Los Angeles and nearby Catalina Island. The American Telephone & Telegraph Company installed an inverter. While it prevented casual eavesdropping, it would not keep a determined amateur from inverting the inversion. And several did just that on the East Coast in the latter 1920s when the telephone company was setting up its radiotelephone link to Europe. Among them was a young man of 20, William Roberts of Trenton, who even sold some of his “De-Scramblers” to Latin American countries.

  Growing realization of the insecurity of the inverter caused its replacement by band-splitters on both the A. T. & T. transatlantic radiotelephone circuit and the Radio Corporation of America’s circuit between San Francisco, Honolulu, and Tokyo. Called the A-3, this Bell Telephone device not only switched the substitution assignments for its five subbands but inverted them as well. However, of the 3,840 possible combinations, only 11 were considered suitable for privacy, and of these only 6 were actually used. They were brought into play in a cycle of 36 steps, each of which remained for 20 seconds, giving the A-3 an overall period of 12 minutes. It began operating between the R.C.A. post in San Francisco and the Mutual Telephone Company post in Honolulu in December, 1937—and a few days later the Tokyo post, which was still using the old inverters, asked what kind of system was in use on the other leg of the circuit, since they could not understand it. The military took the query as proof that Japan was monitoring the mainland communications.

  It was the A-3 that brought news of World War II to President Roosevelt, who was awakened early on the morning of September 1, 1939, by a call from the American ambassador in Paris, William C. Bullitt. As the United States was drawn closer and closer to war, the President Conferred with his emissaries abroad more and more by scrambler radiotelephone. During the Battle of France he sometimes spoke with Bullitt several times a day. Characteristically, Roosevelt liked the telephone because it cut through the red tape of diplomatic routine and the delays of coding and cabling and because it gave him personal contact with the speaker. Occasionally he spoke with Premier Paul Reynaud, and frequently and increasingly with Churchill.

  The President’s words sped from the White House to the overseas switchboard in an A. T. & T. building at 47 Walker Street, New York. In common with all other transatlantic conversations, the nasal Roosevelt drawl then entered a special locked room, barred to all except government-licensed employees, where the A-3 equipment mangled it. Here engineers watched dials and listened to the sound to make sure that the speech was properly scrambled. At the transmitter, channel mixers continually shifted the transmission from one frequency to another, so that anyone listening on one circuit would hear it go suddenly blank.

  And someone was indeed listening. The Deutsche Reichspost—which, like other European post offices, handled telephone and telegraph traffic as well as mail—realized that the only telephone link between England and the United States was the radio circuit, and it reported “The special national political importance of this communication connection has caused the D.R.P. to try with all available scientific means to decipher the conversation carried on this connection.” A task force under Postal Counselor Graduate Engineer Vetterlein of the D.R.P.’s Forschungsanstalt (“Research Bureau”) set to work on the problem. The engineers soon learned the nature of the A-3 system and found that they had to wire circuits for only the six different combinations of subband substitutions. Naturally, they had to experiment to find the exact subband divisions and the sequence in which the six combinations were used, but from start to finish the solution took only a few months. They completed it by September, 1941. Within a few more months the D.R.P. had established an intercept and voice-cryptanalysis station on the Dutch coast. Its elaborate equipment instantaneously unscrambled the conversations, losing only a syllable or two after a key change until the proper one was found. When this was in operation, the German Postal Minister, Wilhelm Ohnesorge, notified Adolf Hitler:

  THE REICHSPOST MINISTER BERLIN W 66, 6 March 1942

  LEIPZIGER STR. 15

  U5342-1/1 Bfb Nr. 23 gRs SECRET REICH MATTER

  Decipherment of the U.S.A.-England telephone connection

  Mein Führer!

  The Forschungsanstalt of the Deutsche Reichspost has completed as the latest of its enterprises an intercept installation for the telephone traffic between the U.S.A. and England, which has been made unintelligible using all present knowledge of communications technology. Thanks to the devoted work of its scientists, it [the D.R.P.] is the only place in Germany that has succeeded in making the scramble, which had been made unintelligible with the best methods, again understandable at the instant of its reception.

  I will give the result
s of our interceptions to the Reich Leader of the S.S., Party Comrade Himmler, who will submit them on March 22.

  I will limit the circulation of this communication pending higher decision in view of the fact that if this success were to come to the knowledge of the English, they would further complicate the problem of telephone traffic and cause it to be sent on the telegraph cable.

  Heil mein Führer!

  (signed) Ohnesorge

  To the

  Leader and Reich Chancellor

  of the Greater German Reich

  Berlin W8

  Dr. Ohnesorge appended a concrete example of the intercept station’s success: a cryptanalyzed and translated conversation plucked from the ether at 7:45 p.m. September 7, 1941. A Briton who had just arrived in the United States was talking with a colleague back in England about the need for a man named Campbell to have an assistant and about their propaganda bureau.

  The group continued to send transcripts to Hitler’s desk, including a 1942 chat between Churchill (at Whitehall 4433) and a Mr. Butcher in New York, and one between Major General Mark Clark and the Inspector General’s office in Washington. At 1:00 a.m. July 29, 1943, they hit the jackpot: a radiotelephone conversation between Roosevelt and Churchill. They were discussing the coup in Italy that had just ousted Mussolini’s government:

  “We do not want proposals for an armistice to be made before we have been definitely approached,” said Churchill.

  “That is right,” agreed Roosevelt.

  “We can also wait quietly for one or two days.”

  “That is right,” said Roosevelt again.

  Churchill said that he would contact the king of Italy, and Roosevelt replied that he too would get in touch with “Emmanuel.” “I do not know quite how I shall do this,” he admitted. The Germans took the conversation as evidence of the treachery and complicity of the Italians: “This is complete proof that secret negotiations between the Anglo-Americans and Italy are under way,” the war diary of the O.K.W. noted. This does not seem to have been the case; in any event, the Allies were cool to the coup.

  Later the Forschungsanstalt again picked up a Roosevelt-Churchill conversation—Churchill was practically addicted to the telephone, calling Roosevelt at all hours from his bombproof shelter in Whitehall, and placing great faith in the scrambler. This conversation, early in 1944, “lasted almost five minutes,” wrote Walter Schellenberg, the Himmler aide who studied it, “and disclosed a crescendo of military activity in Britain, thereby corroborating the many reports of impending invasion.” Soon thereafter the A-3 was replaced by a more secure system, and English became Greek to the listening Germans.

  Transcript of a German descrambling of an intercepted Churchill transatlantic conversation

  Similar activities had, rather surprisingly, been started in the United States even before the D.R.P. project began. Early in October, 1940, the communications division of the National Defense Research Committee set up a group on speech secrecy. It would concentrate on cryptanalysis, partly to eavesdrop on enemy radiotelephone talk, partly to evaluate proposed Allied scramblers. The N.D.R.C. contracted with A. T. & T.’s Bell Telephone Laboratories for these studies, and during World War II the laboratories handled most of the American speech-scrambling work. It was done in two small workshops in the vast pile of stone at 463 West Street, New York, that housed Bell Labs. Though the shops faced on an inside courtyard, both had their windows painted black because of the secret nature of their work. In charge was Walter Koenig, Jr., a short, reticent engineer then turning 40 who had gone to work for Bell right after being graduated from Harvard with an A.B. in chemistry and physics. Much of his work dealt with acoustics, in which the telephone company had a natural interest, and he slid into descrambling because he had helped develop an instrument that was to play an important role in scrambler cryptanalysis.

  Before his device came into widespread use, however, a much easier-to-operate and more common instrument proved to have unsuspected capabilities as a tool for solving ciphony. This was the human ear. Anyone who has managed to converse at a noisy cocktail party should not be surprised at the ear’s ability to pick out speech—and the right speech—from amidst a babble of noise. Nevertheless, wrote Koenig:

  Beginners in the study of privacy systems never fail to be amazed at the difficulty of scrambling speech sufficiently to destroy the intelligence. The ear can tolerate or even ignore surprising amounts of noise, nonlinearity, frequency distortion, misplaced components, superpositions, and other forms of interference. We can therefore very often obtain partial or even complete intelligence from a privacy system by partial or imperfect decoding…. These non-cryptographic methods are very important, because they may reduce the delay in obtaining the intelligence substantially to zero…. Some of them, of course, result in poor quality, but the saving of time, labor and equipment may be very great.

  With some experience in hearing how words sound when scrambled, with some practice in trying to make out scrambled speech, and with repeated listenings to a scrambled conversation, one can understand a goodly portion of what has been said even without electrically cryptanalyzing the scramble. As a not-at-all extreme example, some Bell Telephone Laboratories engineers recovered an average of 47 per cent of the words scrambled by the A-3 simply by listening to it several times. This means that almost half the intelligence leaked through. In one test, intelligibility rose to 76 per cent, or three quarters of what was said. This is enough to give an eavesdropper the gist of a conversation.

  This weakness results from the large safety factor with which the system of oral communication is invested. Speech contains many more elements than it actually needs to be understood. Psychologists and communication engineers have demonstrated this in a great variety of tests—which, interestingly, employ scramblerlike equipment. One test, for example, eliminated (by electrical means) all sounds below 100 c.p.s. in a series of nonsense syllables. Subjects missed less than 10 per cent of the syllables, and in running speech probably would have lost nothing. Why, then, does speech include these low-frequency sounds? Because low frequencies get around the objects and corners of everyday life much better than high frequencies; without them, oral communication would be reduced much more to the line of sight and would lose many of its present advantages. The excess of detail defends speech against the noise and accidental distortion of ordinary activities by ensuring that, even if one component is eroded, the others will sustain the message. This superabundance resists the deliberate deformations of scrambler systems just as strongly. Thus, only 1,000 c.p.s. of the full speech-band of about 7,000 c.p.s. will allow a listener to hear 45 per cent of a series of nonsense syllables. This helps explain why the Bell engineers could understand 47 per cent of the A-3’s scrambled speech even without cryptanalysis.

  “The fact that the ear is such a good decoding tool,” wrote Koenig, “makes the production of privacy systems very difficult. Scrambling systems which look very effective on paper sometimes turn out on trial to degrade the intelligibility very little, although the scrambled speech usually sounds unpleasant. Most methods if they are pushed to the point where they do succeed in hiding the intelligibility are impossible to restore with good quality. There are in fact very few speech privacy systems which achieve a high degree of privacy with acceptable quality.”

  The ear, however, cannot reconstruct the specific nature of the scrambler. This calls for precise differentiation of frequencies and for a kind of total recall of the order of many minute speech segments. These matters were best handled with the sound spectrograph, an instrument that portrayed sound pictorially. Ralph K. Potter of Bell Labs had devised it—with some later help from Koenig—purely for research, but its applications to cryptanalyzing scramblers soon became evident. They were even demonstrated to Dr. Van-nevar Bush, the highly respected head of the Office of Scientific Research and Development, N.D.R.C.’s parent body.

  Bush saw how the spectrograph, by laying down a permanent visualization of s
crambled speech, enabled the scientists to compare this with normal speech, to deduce from the difference the type of scrambling employed, and so to crack it. The device records the voice sounds on paper as a series of horizontal lines representing the main frequencies. In normal speech, these lines appear and disappear, rise and fall in flowing patterns as the frequencies do. Sounds loaded with low frequencies, like /fül/, show up in the spectrograph record as a heavy concentration of lines near the bottom of the paper. The lines for /fēl/ are much higher. In scrambled speech, the normal patterns are distorted. Inversion has the dark lines of the more powerful middle-to-low frequencies near the top of the spectrogram. Inflections which normally climb slope. A band-splitter shows the long horizontal divisions of the subbands. T.D.S. consists of disjointed segments divided by sharp vertical boundaries.

  Mere examination of the spectrogram will thus disclose the type of scramble; solution then becomes the jigsaw-puzzlelike task of cutting apart the spectrogram along the boundaries of the scramble and reassembling it to re-create the flow-pattern of normal speech. This reconstruction will suggest the key used in the scramble, and the cryptanalyst will set up his own apparatus to this key and run the recording of the scrambled message through. Koenig and his associates perfected spectrographic cryptanalysis to the point where, in field tests at Camp Coles in 1943, four people working in a laboratory set up in an Army van truck solved T.D.S. test-scrambles within 15 to 18 minutes of the time of transmission. As a result, eight spectrographs suitable for field use were built and delivered—three to the Army, three to the Navy, and two to the British—between January and May 1944. Some of these were apparently used in cracking a new Japanese scrambler intercepted by the Army at Point Reyes and later at Two Rock Ranch, California. The results sometimes yielded valuable information about forthcoming Japanese moves.