The Emperor's Codes

by Michael Smith

  On one occasion Fabian actually burned a signals intelligence document in front of Willoughby to demonstrate that he was not allowed to see it. After the move to Brisbane FRUMEL transmitted a daily radio intelligence summary to the naval staff officer at MacArthur's headquarters and he conducted the daily briefing. But despite his security concerns, Fabian freely admitted that part of his reason for not co-operating with Central Bureau was that, in his opinion, it had nothing to offer FRUMEL since it was less advanced and had entirely different interests. ‘FRUMEL was concerned solely with information on Japanese naval circuits,’ he said. ‘The Central Bureau was not.’

  This was simply not true. Few wars had seen more need for complete co-operation between the army and navy. The Japanese Army was forced by the very nature of the campaigns it was fighting, cut off from its home bases by thousands of miles of ocean, to pass messages on naval communications circuits, often in naval codes and ciphers. Messages would be translated from one system to the other providing a wealth of potential ‘cribs’, if only they could have been followed through the system. There were also a number of joint navy–army codes and ciphers providing potential ‘cribs’.

  Naval systems such as JN25, JN40, the merchant-shipping code broken at Kilindini, and JN11, the auxiliary fleet system, all carried large amounts of important military intelligence, particularly on movements of troops. The raw JN25 messages intercepted by FRUMEL's mainly Royal Australian Navy operators, and by FRUPAC (Fleet Radio Unit Pacific) in Hawaii, offered an important insight into Japanese attempts to reinforce Guadalcanal and New Guinea in the face of the Allied advances. While they were available to MacArthur, they were not seen by the Central Bureau code-breakers, who needed any assistance they could get to ‘cheat’ their way into the Japanese Army codes.

  Fabian's refusal to co-operate also affected Central Bureau's attempts to keep track of air-to-ground messages and inform the Australian air defences of enemy aircraft movements, since virtually all of the Japanese aircraft operating north-east of Australia were naval air force.

  ‘Brisbane was in the position of needing material and traffic analysis results from Melbourne for the sake of the naval air problem, whereas Melbourne thought that it had nothing that it wanted from Brisbane and was disinclined to co-operate,’ said Nigel de Grey, a deputy director of GC&CS and one of those attempting to mediate to solve the problem. ‘Fortunately, Brisbane established a “black market” with 7th Fleet which, however gratifying at the moment, was entirely reprehensible in principle.’

  Sandford had initially passed material produced by Central Bureau to Fabian but after failing to receive anything in return had set up a private link with the US 7th Fleet while at the same time asking Bletchley Park to intervene in order to get the material through the proper channels. The British, who were concerned both at the lack of co-operation and the insecure procedures it was encouraging, spent a good deal of time battling to get OP-20-G to persuade Fabian to pass information to both Central Bureau and Kilindini. His attitude represented a microcosm both of the sometimes obstructive attitude Bletchley Park itself faced in co-operating with OP-20-G on Japanese codes and ciphers and of the even more bitterly competitive relationship between the US Navy codebreakers in Washington and their US Army counterparts now based at Arlington Hall, a former girls’ college at Arlington, Virginia.

  There was at least one happy fallout for Central Bureau from all the rivalry and rancour. If Fabian did not want Nave, the US Army codebreakers were very happy to have him, said Joe Richard.

  Fabian's dislike of Eric Nave was very fortunate for us. Nave became an indispensable person. Early in the war, Central Bureau found that reading air-to-ground messages containing the weather gave away the intended target for that day. The warnings of Japanese air raids gained from reading these air-to-ground messages saved the Allied air forces in New Guinea and Australia early in the war and made possible the destruction of the Japanese air forces in that area when Allied air force got better aircraft and became stronger. Since General George Kenney, commander 5th Air Force, used these messages to gain command of the air, just reading them would have made possible the eventual winning of the war in MacArthur's area.

  The diplomatic section of Nave's Special Intelligence Bureau remained in operation at Park Orchards in Melbourne under the control of the AIF's Director of Military Intelligence. The operation was run by Hubert Graves, the former British Consul-General in Kobe, Japan, and another former British Consul in Japan, Henry Archer, with the assistance of Professor Trendall and two Australian Army linguists. The majority of the small unit of intercept operators working for the SIB were AWAS.

  By early 1943, the Australian Special Wireless Group had deployed 51 Wireless Section to Darwin where it was picking up a good deal of Japanese military traffic, particularly from Borneo and the Philippines, and 55 Wireless Section to Port Moresby, on New Guinea. At one point the Japanese were only thirty miles away from Port Moresby and Bob Edwards, one of 55 Section's operators, recalled seeing RAAF Spitfires and USAAF Thunderbolts doing victory rolls as they flew back from the front line.

  There were occasions when bombers returned with gaping holes in wings or tail assembly, sure evidence of the activity they had been involved in. There were two sad occasions. Firstly when a Flying Fortress, returning at night from a raid, hit the top of the mountain behind the camp and blew up with the loss of all its crew. The second tragedy occurred one night when a bomber crashed on take-off. As its bomb-load went up, it sounded like all hell had been let loose.

  Despite their closeness to the front line, the Allied forces based at Port Moresby were well looked after, Bob Edwards recalled.

  For entertainment, deck-tennis tournaments were a feature of camp life and the inevitable picture shows and USO shows, including one in which Bob Hope appeared. There was a lot of interest displayed when Australian girls made their appearance with an Australian concert party. Poker, five hundred and chess were also popular pastimes. On more than one occasion, Keith Walshe [a professional entertainer] organized good concerts with talent from our own section and some imported from our counterparts in the RAAF at Port Moresby. Those so minded joined a church parade of a Sunday evening and attended a service at St John's on the Hill at Port Moresby where Padre Hammond and a US Army padre officiated. The cup of tea and a yarn after the service were happy and interesting interludes in the routine. For a fortunate few, Padre Hammond organized a cruise round Port Moresby Harbour on a New Guinea pearling lugger. Fishing with hand grenades proved a dangerous and frustrating occupation.

  Each wireless section now had an Australian Special Intelligence Personnel Section attached to it, providing a basic codebreaking, traffic analysis and intelligence-reporting facility. There was a wealth of information to be had from reading the operator logs. Procedural codes, designed more to help operators discern what was being said than for security reasons, often produced important material, frequently confirming DF fixes of locations. The procedural code CHI A, which stood for ‘this station is located at’, was normally followed by a simple kana equivalent of the name, while the times of transmission of HI KE, the code for an air-raid alarm, could be compared with times of Allied raids to provide locations for the stations concerned. Low-grade messages sent without the use of codes or ciphers included casualty lists and preparations for Allied attacks. Plain-language operator chat often gave names that could be tied to specific units.

  The Japanese system of recruitment also provided the traffic analysts with an easy method of identifying individual units. Each infantry regiment was strongly identified with a local district. It recruited from that district through a home depot in its garrison town. The division it was subordinate to also had a home depot covering the area in which the district lay and the neighbouring districts from which the division's other infantry regiments came. If a member of the regiment, from the lowliest soldier to its commanding officer, did anything of note – was promoted; decorated; captured by the e
nemy; wounded; or killed – this had to be reported back to both the regimental and divisional home depots. So it was relatively easy for the traffic analysts to identify new units from the home depots they were in contact with.

  Although the Japanese Army codes were still frustratingly difficult to break, traffic analysis of 55 Wireless Section intercepts provided an early success for Central Bureau. It very quickly built up a detailed picture of the Japanese troops attempting to cross the Owen Stanley Range, north of Port Moresby, allowing the Australian 7th Division and the US 32nd Division to defeat them at Kokoda and force them back to the northern coast.

  It was JN25 messages picked up by FRUMEL's intercept stations in the southern Melbourne suburb of Moorabbin and at Adelaide River, two hours south of Darwin, which led MacArthur to deploy the Australian 18th Infantry Brigade to Milne Bay in eastern New Guinea in September 1942 to repulse an attempted Japanese amphibious landing. But traffic analysis provided by Stan Clark's E Branch provided the backbone of MacArthur's intelligence about Japanese military activity on the ground.

  A detachment of the RAAF's No. 1 Wireless Unit was sent to Port Moresby in January 1943 to assist 55 Section in its coverage of airto-ground radio networks. The Japanese followed set procedures that allowed the intelligence analysts attached to the intercept units to predict precisely when and where they would strike. Direct telephone links between the intercept sites and the fighter controllers ensured that Allied aircraft could be in the air to intercept the enemy bombers.

  Enemy weather reconnaissance flights provided the first clues. Naval air-to-ground traffic always contained a high percentage of weather messages. The codebreakers soon realized that those broadcast by stations at Truk, Rabaul and Tokyo were always of weather over territory held by the Allies. They invariably mentioned the name of the target, which was sent using a simple and easily broken substitution cipher. The difference between the time the weather reports were issued and the time of the raids was so standard that it was a simple task to predict when the Japanese aircraft would be overhead.

  Japanese aircraft maintained strict radio silence when about to take off for a raid but the tuning-up of radio sets located by the interceptors’ DF units provided forewarning that the enemy mission was about to start. In addition, the Japanese usually dispersed their aircraft to rear bases, only moving them to a frontline airfield shortly before a raid. The staging movement was normally accompanied by much slacker radio security than the mission itself and Allied aircraft could frequently be dispatched to mount a pre-emptive attack against the aircraft on the ground at the forward base.

  The Guadalcanal campaign provided the Allied cryptographers with two useful bonuses in the form of ‘pinches’ of Japanese code-books. The first came when US Marines captured a JN25c codebook shortly after landing, although this was superseded a week later by the JN25d book. The second was a major haul of various different codebooks and signals instructions recovered in early 1943 from a Japanese submarine. ‘The New Zealand ships HMNZS Kiwi and HMNZS Moa surprised the I-1 at Kamimbo and sank her by ramming on 29 January,’ said Phillip Jacobsen, who was now serving with Station AL, a small US Navy radio intelligence unit based on the island.

  While fifty or so Japanese survivors buried some documents, including codebooks, a later salvage operation by the submarine rescue vessel Ortlan recovered many other codebooks, communications and other secret documents from the I-1 and the surrounding waters. These were brought to Station AL and we carefully dried them out page by page, as we knew their value. Although the codebooks were mostly superseded, it was very useful to have these complete codes and fleet vocabularies to aid further code recoveries.

  Australian troops captured a list of Japanese naval area designators and a book of Japanese call signs at Milne Bay. But by early February 1943, when the Japanese finally gave up their fight for Guadalcanal, the Allied codebreakers were still struggling to make the break into the high-level Japanese Army codes. A solution was, however, close at hand. The Japanese Army's Water Transport Code, or senpaku angosho 2, a four-figure superenciphered code known as 2468 since this appeared as the first group, or ‘discriminant’, of each message, was under attack at a number of centres. The lead appears to have been taken by Wilfrid Noyce, a classicist from King's College, Cambridge, and a well-known mountaineer, who together with Maurice Allen, an Oxford don, realized that the first letter of the third group of each message was not random. The third group in the 2468 received at Delhi was the repeater group or indicator. As in the main naval systems, this gave the starting point of the additive which was added to the encoded message.

  The lack of randomness was the entry point into the code that the Allied codebreakers had been seeking. Accounts of who subsequently broke it differ, with the earliest date being ascribed to Noyce and Allen in March 1943. But Arlington Hall also claimed to have broken it first, in April, and, according to John MacInnes, Kilindini was receiving enough 2468 messages for Brian Townend to break it ‘about the same time as solutions were reached by Delhi and Washington’.

  Since Delhi was already sending decoded material back to Bletchley Park in April 1943, it seems likely that it was at the fore-front of the first break into high-level Japanese military codes. But it appears to have been joined there by Central Bureau, where Joe Richard had begun working almost obsessively on the 2468 code to take his mind off the death of a friend:

  I decided to do double work after the accidental shooting of my friend John Bartlet. After the traffic analysts had extracted all the data they needed from the four-figure traffic, it was simply filed unsorted. So I volunteered to Colonel Sinkov to come back evenings and sort it by system. Each day's intercept was contained in a manila folder and the day was based on Greenwich Meantime so that all the traffic wherever intercepted could be collated together.

  Colonel Sinkov agreed and sent me to Major Norman Webb, an English officer who escaped from the FECB at Singapore with some records and twelve intercept operators. Webb put an Australian soldier to work pulling the traffic and, being a careful man, had him number each message consecutively so that if necessary it would be returned exactly as stored.

  Working on the basis of Tiltman's previous break into a Japanese military system, the Allied codebreakers had made the assumption that mainline army systems were still based on a codebook and a cipher additive. So Richard needed to find a way to strip off the additives. To recover additives, a depth of messages must be available for each set of additives used and the messages superimposed in proper relative order, as determined by the indicator. This might appear anywhere in the message but somewhere else it would be repeated, so it was relatively easy to find. But the problem with this type of system was that the indicator was disguised by the use of a second, separate additive system. It was this system that Richard had to uncover to have any hope of breaking the messages. ‘We decided to start with traffic for 19 December 1942 in order to catch any changes that might occur on 1 January 1943. So on an evening in January 1943 in a second-storey room at 21 Henry Street with the blackout curtains drawn and working on a big table under a dim drop light, I started sorting the traffic.’

  On the table in front of him, Richard had three separate types of message: the Water Transport Code; the Army Air Force General Purpose Code, koku angoo-sho 3, known to the Allied codebreakers by its discriminant group 3366; and the General Army Administrative Code, or rikugunangoo-sho 4, which they referred to as the 7890. He first sorted the messages by system. ‘Then I sorted each system by the repeated text group – all four-figure systems repeated a group from near the beginning as the last group. I used the repeated group because I thought it would enable me to place fragmentary messages together. In the first few evenings, I made two discoveries.’

  The first was in the 3366 where Richard found sixteen messages passing between Rangoon and Saigon in which the first twenty-one textual groups were identical. This seemed to indicate that they all used the same additive and contained large amou
nts of identical plain text. It should by rights provide a way into the system. But it came to nothing when in March 1943 the 3366 system stopped for ten days and returned in a changed form, Richard said.

  The other discovery was in the Water Transport Code, the 2468. This system repeated the third group – the other two systems 7890 and 3366 repeated the fourth group. I noticed that when I sorted the messages by the first figure of the repeated group in 2468 the piles of messages stayed the same from day to day. For example, the daily piles for the figure three would have lots of messages in them while all the nine piles would have only one or two or none. This was non-random behaviour and differed from the way the 7890 and the 3366 sorted. I reported this non-random behaviour to Colonel Sinkov some time late in February 1943 and I think he brought me a typed three-page letter from WEC also noting that the first figure of the repeated group in 2468 was non-random.

  A few days later I saw a message from Arlington Hall to all centres saying they would make an IBM listing of 2468 messages to look for this non-random figure. About two weeks later Arlington Hall sent another message saying their listing did not show any non-randomness. This puzzled me until I remembered that the non-randomness changed in character about every four weeks and I decided Arlington Hall must have mixed up two or three months’ traffic across these changes. I remembered that I had not told Colonel Sinkov that the non-randomness changed every few weeks. So I went to him and told him – he was not very happy.

  That evening I came back determined to find the exact day and hour when the first change occurred. I pulled the 2468 folders for about three days before the day that I thought the first change had occurred and for about three days after and started to copy down the date and message number. I found three groups of every message. I soon found this was a long job and was about resigned to staying most of the night when I noticed that there was a relationship between the non-random figure and the first figure of the group that preceded it. For any given figure in this preceding group only three figures and always one of the same three figures appeared as the non-random figure. So I prepared a three-column table for the editors. If a message fitted on the table it would belong in that particular period. After I had constructed this ten-row, three-column table I left the office.