The Dragon was a machine built in America in 1944. It was one of a few machines given the title of Rapid Analytic Machinery (RAM) built by the Signal Security Agency (SSA) at Arlington Hall, Virginia. It was created to assist in the breaking of "Tunny", the British name for the Lorenz SZ40/42 cipher attachment.
SSA photo of Dragon, National Cryptologic Museum/National Security Agency.
Dragon was designed by Captain George at the SSA, an expert in telephone switching systems, was completed on 24th August 1944 and then shipped over to Bletchley Park to arrive on 14th October 1944. It was placed into Rooms 22, 23 and 24 of Block F which were made into one larger room to accommodate it.
Dragon was a crib dragging machine which would test a guess at a section of text which was believed was to be found within the enciphered message being tested. Should such a guess be successful, and a match found, this would give the cryptographers in the Testery the foothold required to begin to break into the message and work backwards to the beginning of the message. Once the start point of the message was recovered, this would give the start positions of the PSI wheels for that message and therefore enable a full deciphering of the message to be generated on one of the British Tunny machines.
A lot of the messages could be deciphered manually by the code breakers in the Testery, but on low dottage days (days where the number of dots on the Motor(Mu) wheels were few), it was much trickier to find a break, and this is where Dragon would be brought into play.
To understand how the dragon worked, it is first best to understand a little about the machine it was built to help break. This machine was the Lorenz SZ40/42 - a formidable cipher machine used by the High Command in Germany to send top secret enciphered teleprinter messages to and from the top generals on the battle fronts.
The Lorenz had twelve wheels which were used to generate two seperate key letters for each letter of the message being transmitted. The wheels were split into three sections, the first five being called the CHI wheels (pronounced with a hard C like a K and rhymes with try), the next two the MU (or motor) wheels and the final five called the PSI (pronouced sigh) wheels. The two generated letters were "added" together with the letter being set using a method invented by Gilbert Vernam in 1918. This would then give a final enciphered letter which would be sent by wire or radio transmitter to another Lorenz machine at the receiving end. The receiving Lorenz, having it's wheels being set to the same start positions as the transmitting one, would add back on the same two letters which would then reveal the original message.
If you want to know more about the Lorenz and how it encoded messages, please take a look at my other site which has a simulator of the Lorenz and gives more detail on how it worked Virtual Lorenz
As an example, let us assume we're going to send the letter M. The Lorenz will generate two letters - for our example, the CHI wheel will give the letter X and the PSI wheels will generate the letter R. The machine then adds these two letters together to give a further letter, in our example the letter Q (X + R = Q). The result of these is added to our letter to be encoded M which results in the letter J (M + Q = J). The important thing to note here is that it doesn't matter which order we do these additions in, the result will be the same. We can add M + X then add that to R, the result is still J
When deciphering the message we received (the letter J), the same two letters from the PSI and CHI wheels are added to the received character. Again, it doesn't matter which order the letters are added, the final result will be the original letter. J + X + R = M
The following is an excerpt from The Newmanry History - Part of the General Report on Tunny (1945), 55A Page 363.
The purpose is to set a common crib P, of up to 10 letters, in a given de-chi D, i.e. to find a stretch of D (if there is one) where the underlying plain text is P, so that P + D = PSI' , which when the extensions are removed, yields PSI.
Dragon adds P to a stretch of D in all positions in turn: in each position it contracts P + D, i.e. omits repeated letters, and then compares each impulse of the result, independently, with the corresponding PSI wheel : if all five can be fitted the machine stops and displays the settings of D and all PSI's at the last letter of the crib.
Over the next couple of pages, I will try to explain what this means and how Dragon did it's job.
A specific part of the General Report on Tunny which covers the detail on the machines that were used at Bletchley Park (Colossus, Robinson, Dragon and others) was scanned and made available by the late Tony Sale on his website if you wish to know a little more. www.codesandciphers.co.uk
The point where the Dragon was able to do it's work was not on the original enciphered message, but rather on a message which had one part of the message stripped off, specifically, that which was generated by the CHI wheels. The input for the Dragon was therefore called a de-chi.
Each time the pin settings of the Lorenz were changed (this varied throughout the war and some wheels were adjusted more regularly than others), the cryptographers in the Testery at Bletchley Park, people like Captain Jerry Roberts, would work out the pin settings of the wheels by hand methods
The de-chi was then usually generated by Colossus which began breaking into the message using Bill Tutte's "1+2 break in" to find the start positions that had been set for that specific message. See Virtual Colossus for more details on how this was achieved. Colossus was able to continue on to break the MU wheels and PSI wheels, but it appears that is was a better use of time for Colossus to move onto another message and let other machines and Testery manual decrypting to do the final message decrypt.
This initial break gave all the information required to generate the same CHI wheel pattern of letters generated by the Lorenz and therefore, if they added these letters to the original message, you would end up with a de-chi. Remember, the PLAIN + CHI + PSI = CIPHER and similarly, the CIPHER + CHI + PSI = PLAIN, which means that adding just the CHI to the cipher message leaves us with CIPHER + [CHI + CHI cancel out] + PSI = PLAIN. We're left with just CIPHER + PSI which is our DE-CHI.
This resulting de-chi would then be made into a punched hole tape ready to be read into Dragon using a standard Teletype Model 14TD tape reader. It is likely that the reader was wired to read the five impulses in parallel straight in rather than converting to a set of serial pulses (the normal output method).
A Teletype 14TD showing how this model can be started and stopped to show the current de-chi letter being read. This was likely a wired connection from Dragon to stop when a result was found.