We worked on a particular code called JNA10 and a code called Kuibyshev. This code was very similar to JNA10 but was not on the same codebook and because we could never find any depth to it, Mr Flintham gave it to me to play around with. That was the sort of term they used. It didn't mean play, of course. It meant jolly well try and get something done with it, and I was successful in that I did get a depth of five which was good, because the greater the depth we had the more likely we were to be able to confirm the groups.
We recorded the frequency of groups on what became known as a Dottery. Each time a particular group was used it would be recorded by a dot against it. We used to record the number of times that a particular group had come up and what we thought it meant. When it had appeared twenty-five times with the same meaning, we would draw a little square round it. That was our way of knowing it was a confirmed group. Most times, we didn't get all the message, only parts of it. Then it was sent off to America to see if they could add bits to it. Everything we did was sent to Washington, which was codenamed Susan.
As with all the Wrens at Bletchley Park, they were quartered at nearby country houses. Edith Bennett and her friends were deposited at Woburn Abbey.
We had huts in the courtyard because there were so many of us there. There was one pipe that went through the huts providing hot water for the showers at the end of the block of huts and that was our only heating. We used to go to bed dressed in more clothing than we went out in and the grass used to grow through the floorboards of the huts. I shared a room with one other girl and before we could clean our room out we had to go along with a pair of nail scissors, clipping the grass that was between the floorboards, and all the time we were doing it we were absolutely roaring with laughter.
During our off-duty hours we used to do a fair bit of hitchhiking. Wherever a lorry was going, that's where we would go. It sounds awful to do it now but in those days it was completely safe. We wouldn't go on our own, maybe three or four of us if it was a nice day. Or we even might just walk down the drive to Woburn itself where there used to be a little tea shop. That was sort of high days and holidays, if we could afford to do it. You actually had a tablecloth on the table and a cup and saucer that matched.
A couple of the girls had boyfriends, but I already had a young man. He was in the army, in the Rifle Brigade. We occasionally were invited to the American bases, but I only went once. They used to come and fetch you, but it was such a rough-house. I saw so many goings-on that night, I can tell you. It was all right while the dance was going on, but it was afterwards. It was dark outside and we had GIs trying to stow away on the way home. So I didn't go again. I thought, Golly, that's enough for me.
With D-Day only a few months away, the information that the Allies now obtained from the messages sent back to Tokyo by the Japanese representatives in Berlin was making a major contribution to the planning for Operation Overlord, the invasion of Europe. Most Japanese diplomatic messages were now sent by radio. They were intercepted in Britain by the Foreign Office interception stations at Brora, Sandridge and Whitchurch; at British embassies in the neutral countries; and at various points in Asia and Australasia, including Canberra, where a number of RAANs worked, and at the Abbottabad and Bangalore stations in India. The Royal Canadian Navy also intercepted diplomatic material at Winnipeg and Point Grey, Vancouver.
BREAKING THE JAPANESE SUPERENCIPHERED CODES
The process used to break the Japanese superenciphered codes like JN25 can perhaps best be demonstrated by taking an example and working through it to ‘strip’ off the additive cipher in the same way that the codebreakers did. Remember, in a superenciphered code, the originator encoded the message in a series of five-figure code groups using the codebook and then enciphered the encoded message by adding to the encoded groups a series of five-figure groups taken from an additive cipher book. It is this additive cipher that the codebreakers needed to strip off in order to reveal the encoded message underneath.
The first step was to go through all the available Japanese messages in the system the codebreakers were trying to crack, sorting them according to the ‘indicator’ groups. These were the groups inserted into each message, sometimes at the beginning, or more often in a previously predetermined position, which told the Japanese receiving station the starting point for deciphering the encoded message.
The codebreakers looked for a number of messages with the same starting points, or with starting points that were close to each other in the additive cipher book. For the sake of example, let us say we are lucky to find five messages with similar starting points. The first twelve groups of each are shown in the following table:
Table 1.
For the sake of clarity, in the example shown in Table 1, the starting points for the five-figure messages the codebreaker is trying to decipher have been sent ‘in clear’ at the beginning of the message. Each is using different but overlapping parts of page 3 of the additive cipher book, indicated by the fact that they all commence with the figure 003. The ‘indicator’ for the first message is 00300. This means that the additive sequence in use begins on page 3, first line, first column—i.e., the first additive group at the top of that page, since the Japanese numbered the columns and lines from 0 to 9 rather than from 1 to 9. The second message starts with page 3, line 0, column 3, i.e., the fourth group on the page.
The codebreakers would then write the messages out on a large sheet of graph paper, known as a worksheet. The messages were aligned so that each group in any one column had been enciphered using the same additive group. On our worksheet we can line up the first group of message two (30962) underneath the fourth group of message one (49073), knowing that both these groups were enciphered using the fourth additive group on page 3 of the cipher book. Every subsequent group in message two will now appear underneath a counterpart in message one that was enciphered using the same additive. The third message begins with line 0, column 1—i.e., the second group on the page, so its first group is lined up under the second group of message one—and so on. If we now line up our messages so that all the groups in each column have been enciphered using the same additive group, they look like Table 2 below.
From this point on, it will probably be easier to understand the process if you copy out this example and work on top of it, stripping the additive off as we go along, just as the codebreakers did. You will need to leave a space at the top of the page for the additive groups we recover plus two lines between each row of figures so you can fill in the code groups and any original text we recover.
Table 2.
The number of groups using the same additive cipher group that can be aligned in the vertical columns is referred to as the ‘depth’. The more depth there is on the worksheet, the easier it should be to break the messages. In Table 2, there is a depth of two groups from the second column onwards, the second group in message one (78389) and the first group in message three (27755) having been enciphered using the same additive cipher group (namely, the second group in line 1 of page 3 of the additive cipher book). The example grows to a more useful depth of three by the fourth column; a depth of four by the fifth column; and a much more healthy depth of five from the seventh column onwards.
Sometimes the codebreakers would be lucky enough to know from previously broken messages what the sequence of additive cipher groups for the given starting point was. But if they were starting from scratch, which was frequently the case, the first task was to hypothesise whether any individual enciphered groups matched up to known code groups. By subtracting the known code group from the enciphered group, they produced an additive cipher group, which could then be subtracted from all the enciphered groups in the same column to yield potential code groups. If other known code groups then began appearing in their correct positions, the code breakers would know they were on the right track. As we continue working through the example, this will be demonstrated.
Codebreakers would look for key code groups. These involved words that mi
ght be expected to appear more frequently than others, the first being hatsu, the Japanese word for ‘from’, which appeared at the start of most messages to indicate who was originating the message. Depending on the station sending the message, the likely originator might also be known. The next code group after the one giving the originator would be shuushifu, the Japanese word for ‘full stop’.
We can begin to try some of these entry devices, starting with hatsu. It would be best to begin with the first enciphered group of message four, in the seventh column, since that column has a depth of five. The advantage of starting with a depth of five is that if either of the already recovered code groups for hatsu works then we can begin stripping off the additive from all five code groups in the column, with the possibility that some of those code groups may already have been recovered. There would be a number of five-figure JN25 code groups for a common word like hatsu. But let us say there are just two—58743 and 78225—and we have already recovered both from breaking previous messages. Note that the sums of all the figures in each of these code groups, as with every five-figure JN25 code group, are divisible by three. This was the ‘garble check’ for the Japanese operators. If after deciphering the message they found that any of the code groups left were not divisible by three, they knew that either the message had been sent or taken down incorrectly, or that their sums were wrong and they had to start again. But this garble check also turned out to be an invaluable tool for the codebreakers. If, after stripping off the additive groups, they were left with any codegroups that were not divisible by three, they, too, knew they had gotten something wrong.
So let us try the first of our hatsu code groups: 58743. We subtract it from the first enciphered group of message four: 87033, using the Fibonacci non-carrying system (see page 55), which for ease of use was common to all additive systems. Each digit is treated separately, so if we subtract 8 from 7 we get 9 and do not carry any figures over to the next digit. Similarly, if we were to add 8 and 7 we would get 5 rather than 15. Subtracting the code group 58743 from the enciphered group 87033, we come up with a possible additive group for column four of 39390. By subtracting 39390 from the respective enciphered groups, working from top to bottom, we arrive at the sums 57699; 90561; 04578; our hypothesised code group 58743; and 50418. The arithmetic that produced these results, using the Fibonacci system, looks like this:
Table 3.
The last group of the seventh column is in the third position of the fifth message, precisely where we might expect to see shuushifu. Because this is a code group that we see all the time, it has already been recovered from breaking of previous messages as 50418, giving us the confirmation we need that our additive group is correct. Our worksheet now looks like this:
Table 4.
We could now try one of the two hatsu code groups on the other messages, but that would give us less chance of a breakthrough since it involves columns with depths of less than five. So put that option to one side for the moment. If we try the shuushifu code group 50418 instead, we only have one group to worry about and therefore may have more luck. Hypothesising it for the third group of message two, 75834—remember you subtract the potential code group from your enciphered group—produces an additive group for that column of 25426, while in the third group of message four, 18443, it gives us an additive group of 68035.
Now we subtract these new additive groups from the enciphered groups to come up with the other code groups in their respective columns. Some of these will not have been previously recovered but equally some well-known code groups can now be filled in: in this case, we know from previously broken messages that 76883 is ‘maru’ (the suffix used after the name of each ship); that 84717 is ‘supplies’; that 45435 is ‘begin’; that 34131 is ‘good’; that 41595 is ‘Commander-in-Chief’; and that 66201 is ‘radio silence’. Our work-sheet now looks like this:
Table 5.
We are not going to be able to strip the first four groups of message one and the first two groups of message three at this stage so it is easier to move along the worksheet to give us more of the potentially breakable body of the messages. The following table moves along the worksheet so that the fourth column of the previous table is now at the extreme left:
Table 6.
What we might do now is look at the sequence of additive groups we have recovered to see if it matches any known sequences. If it does, this will make our job much easier. But if it doesn't, we still have a number of options open to us.
Our first option is to return to the two possible code groups for hatsu, in the first enciphered groups of messages two and five that we have not yet tried. If we start with the code group for hatsu that we have already used, 58743, and subtract it from the first enciphered group in message five, 57508, we get an additive group of 09865. When this is subtracted from the groups in this column, it produces code groups of 39617, 40752, and 04794 respectively. There is a problem here. The Japanese operators’ garble check ensures that the sum of the figures in all code groups is divisible by three. Code groups 40752 and 04794 conform to this rule, but 39617 does not. Either it has been taken down incorrectly by the intercept operator or, more likely, 58743 is not the code group in use here. If we try the other hatsu code group, 78225, instead, and subtract it from the first enciphered group in message five we get an additive group of 89383. When this is subtracted from the other enciphered groups in this column, it produces code groups of 59199, 60234, and 24276 respectively, all of which fit the garble check, which has now proved very useful. If we try both hatsu code groups on the first enciphered group of message two, only the 58743 code group passes the garble check. So if we fill in those two columns our example now looks like this:
Table 7.
Unfortunately, none of the resultant code groups has been recovered from previously broken messages. But we are beginning to make headway with the last of the messages on our worksheet, message number five. We know it is a short message, and unlike any of the others, the whole message is shown in this portion of the work-sheet. We know it is from the commander-in-chief (C-in-C), and we know it is talking about radio silence. The most likely code groups to be used with radio silence in this code are those for ‘begin’ and for ‘end’. Since we know that the stations concerned had not previously been maintaining radio silence, they cannot be ending it. So ‘begin’ seems the most likely option. If we try the code group for ‘begin’—45435—with that enciphered group, we get an additive group of 28057, and when we subtract that from the other enciphered groups in the same column, every one fits the garble check. We also get another recovery in column five of message three where 38664 is known to be the code group for ‘until further notice’.
Table 8.
With all these code groups in place, an interesting possibility opens up. It appears that the text in message five is being repeated in message three. The most likely reason for this would be that the commander-in-chief's message is being passed on to a station further on down the line. We can test our theory on message five, providing us with two more potential additive groups and allowing us to extend the fifth message, and the stream of additive groups, still further. When we do so we find that the garble checks are working out, and yet again we are on the right track,
Table 9.
With luck, we will be able to continue stripping off the additive cipher by ‘piggy-backing’ two groups at a time in each message. As we move along the worksheet, a number of already recovered code groups are appearing in the messages. None is anywhere near as complete as message five. But by good fortune, some previously recovered code groups have turned up in the repetition sequence in message three. The code groups 83302 (‘your’) and 02633 (‘mission’) fill in the major remaining gaps and, although it cannot be confirmed, established Japanese practice leads us to guess that the unrecovered code group 70635 is probably ‘luck’.
Not only have we successfully stripped additive cipher to uncover the code groups underneath, but we have also decode
d an important message indicating that a major operation is imminent. While that information will be sent out in the form of an intelligence report, the worksheet. which should look like table 10, will be handed on to other codebreakers who will attempt to use other messages and knowledge of Japanese procedure to break into the as yet unrecovered code groups, building up the codebook for future use. The string of additive groups will also be recorded so that when the same indicator groups appear again, the codebreakers dealing with those messages will have a much easier time.
Table 10.
This is of course only a demonstration exercise. It was rarely this easy. Although, even so, note how few code groups are recovered. We have availed ourselves of only a limited number of the possible means used to break into the Japanese superenciphered codes.
Among the more important diplomatic intercepts in the run-up to D-Day was a detailed report by Oshima Hiroshi on a tour he made in November 1943 of the German defences in northern France. One of the US codebreakers working on the Purple messages later recalled the excitement of working through the night and into the next day on Oshima's detailed rundown of the Atlantic Wall:
When I picked up the first intercept I was not sure what I had because it was not part one. But within a few hours the magnitude of what was at hand was apparent. I remained on duty throughout much of the day, continuing to translate along with colleagues who had pitched in to complete the work. I was too electrified to sleep. In the end we produced what was veritably a pamphlet, an on-the-ground description of the north French defences of ‘Festung Europa’, composed dictu mirabile by a general.
The gaps in Oshima's report were more than filled in by Colonel Ito Seiichi, the Japanese Military Attaché who had made his own tour of the entire German coastal defences, sending a massive thirty-two-part report back to Tokyo. The reports reassured the Allies that Hitler remained convinced that the main thrust of the invasion was to be along the Pas de Calais.
The Emperor's Codes Page 27
