The transcript of a Lecture given at the IEEE 18th February 1999 by Tony Sale Colossus I first need to give you an introduction to the problem of which Colossus was the solution. This is all about the German Lorenz cipher machine, and there it is in all its glory and Colossus was the machine which in WW II assisted greatly in breaking that cipher. Now to understand a bit about how Colossus was used and what it actually did; first of all pre-war teleprinter traffic, a 5 bit Baudot code and paper tape, which I am sure some of you will remember intimately. Standard 5 bit baudot code like that, and that was a 32 element code 5 bits and that was a well established pre-war technology. Now the encipherement of teleprinter traffic actually started with Gilbert Vernam at AT&T in 1918. He realised that by using modulo 2 arithmetic you could actually make a very simple enciphering and deciphering system. Its enciphering by adding. Its quite different from Enigma. Enigma is a substitution cipher . When I press the A key the Q lamp lights and I write down Q as the substitution for A for that particular configuration of the Enigma machine. This is different, when the operator presses an A key on his teleprinter the electrical signal goes into the machine and the machine adds an obscuring character to it and adds it modulo 2 bit by bit. The result of that is then another character, in this case F, and the magic of modulo 2 arithmetic is that if you transmit F to the other end and add back exactly the same obscuring character, then it cancels out and leaves you with the plain text. So Vernam's system was a very simple, in principle, system where you had plain language add obscuring characters and at the other end add back exactly the same obscuring characters and the they cancel out and leave you with the plain language. He actually proposed this in 1918 and in 1919 the Americans actually built one of these. Vernam proposed that the added characters be pre punched onto a roll of paper tape as random letters. Of course if they are completely random then its unbreakable. The difficulty is in a hot war situation to make certain that the same tape is at both ends of the link and they are synchronised. That was a difficulty and caused the system not to be used at that time although it was revived much later. Now when the German high command wanted a machine which Hitler could use to talk to his Generals, they went to the Lorenz company and the Lorenz Company decided to build a machine based on the Vernam principle and they designed this machine, the Lorenz to produce a sequence of obscuring characters, but because it is a machine it generates a pseudo random sequence, unfortunately for the Germans it was more "pseudo" than random and that was how it was broken. Now the first messages sent using this system actually used a Hellschreiber, an early form of fax, which prints out letters on a strip of paper. Here is a Hellschreiber message of about 1940. They set up a section in BP to try and find how to break it, I won't go into the details of the break, although it is a fascinating story. Basically what happened was that the Germans made a mistake, they sent virtually the same message twice one after the other with the Lorenz machines back to the same start position. So the obscuring character set was the same on both transmissions. But the operator, who keyed all this in by hand started to make differences in the second entry of the message. You can see he substituted NR for NUMMER and NR means the same as NUMMER, whats the difference? The difference is that from that point on the entry characters are relatively different even though the obscuring characters generated were the same on both transmissions. This pair was worked out by John Tiltman in BP and he recovered the actual obscuring character set being generated by this cipher machine, not knowing what it was, but that was what it was generating. Then a mathematician, Bill Tutte, came on the scene and John Tiltman gave him this long stretch of pure key , pure obscuring characters, and asked Bill to find out what he could about the structure of this machine. Bill started to write this out at various periodicity's, remember this was BC, Before Computers, so he had to write it out by hand but when he wrote it out on a repetition rate of 41, various patterns began to emerge so 41 had some significance in the structure of the machine. Then over the next 2 to 3 months Bill Tutte and the research section worked out the whole logical structure of the Lorenz cipher machine from that one mistake that the Germans made, never having seen the machine, a remarkable tour-de-force. So here is the logical structure of the cipher machine. What they found was that it added not one but two obscuring characters one after the other sequentially to the input character the result being transmitted. At the other end another Lorenz machine set to exactly the same configuration regenerates the two obscuring characters adds them back and by the magic of modulo 2 arithmetic they cancel out and leave you with the plain text. So that was the problem the Germans set us in trying to break the messages. It was found that the key setting was actually in two parts. Here's the actual machine again, it has twelve wheels broken up into two groups of five which give the first and second obscuring characters and the two wheels with intermittent motion in the centre. Bill Tutte being a mathematician called these the Chi, Psi and Mu wheels, lesser mortals call them the K, S and M wheels. The key was found to be in two parts, first of all round the periphery of each wheel are tiny mechanical lugs which can be set either left or right and those are set by the operator. There are 501 of them and the operator had to get them all exactly right! These were known as the wheel patterns and remained set over a number of messages. Here are the wheel patterns for ZMUG, the German mistake. In modern terms these are 0s and 1s in BP terms they were dot and cross. The lengths of these patterns are all different and mutually prime to give the maximum length before they repeat. Now what Bill Tutte found was that if you took the enciphered text , ie the result of the plain language plus two obscuring characters. if you took any one stream of the five streams on the paper tape, and look at it then the Germans had been very careful in their choice of wheel patterns so that together with the German language component the distribution of crosses and dots was flat random, ie 0.5 dead, and that was the same for all five streams. There was therefore no chance of trying to do a match of the pattern to try to find the position of the wheels. The second part of the key was the turning of the wheels, using the finger notches, to different start positions for every message. This was a 10^19 combination, non trivial! So the problem facing the code breakers was first of all to find the patterns and then to find for each message the start positions of those patterns so that they could regenerate the same set of obscuring characters and decipher the message. But Bill Tutte also found that although individual streams were 0.5 dead, streams taken in pairs were 0.47 but this meant going through all possible start positions of both tracks together, 1,271 for k1 and k2. This led to the double delta method for finding the wheel start positions. Because Bill Tutte also found that more leverage could be obtained by using the delta of a stream rather than the actual. When they had worked out the logic of the machine they then went to Dollis Hill and Dollis Hill engineers built them these machines called Tunny and these were the logical equivalent of the Lorenz machine but built using uniselectors and relays rather than wheels. So when the code breakers had laboriously worked out by hand the settings that the German operator had used for a particular message then they could plug those settings up, the patterns and the start positions, on the front panel of Tunny and then they could read in the enciphered text and as if by magic out came plain German on the teleprinter if they hadgot it right. But it was taking 6 to 8 weeks to work out the settings on any one message. Max Newman came on the scene in 1942 and he was a mathematician with an engineering background and he believed that it was possible to automate the finding of the wheel start positions. He first commissioned from Wynn-Williams at TRE a machine called Heath Robinson, and it was a "Heath Robinson" machine. What was postulated was that if you took the enciphered message tape and you punched the patterns that the code breakers had worked out onto another paper tape and the ran these two tapes in synchronism at high speed and did a sophisticated cross correlation measurement, the double delta measurement, between these two tapes and then slipped the relative positions of the tapes by one character and repeated the cross correlation measurement, by comparing the scores you should be able to find the position when the pattern tape was in synchronism with the pattern position which the German operator had used for that particular message tape. Now Heath Robinson was a difficult machine because of the difficulties in keeping two paper tapes in synchronism at high speed. The sprockets tore the tape and it was very difficult. But it worked well enough to show that the basic idea was correct. Alan Turing then suggested that Max Newman go to Dollis Hill and there he was put in touch with Tommy Flowers, and Tommy Flowers was the brilliant Post Office engineer who had the idea of designing Colossus to meet Max Newman's requirements for a machine to speed up the breaking of the Lorenz cipher. Tommy started in March 1943, a blank sheet of paper, never been done before. The reader had already been developed for Heath Robinson, and Arnold Lynch is here who designed that, but Tommy Flowers wanted to speed everything up to break the cipher in a reasonable time span and whereas Heath Robinson ran at 1,000 ch per sec he wanted Colossus to go up to 5,000 cps. So Colossus was the machine he started designing in March 1943 proposing a machine with 1,500 valves in it, the biggest machine at that time had 150 valves in it so this was an enormous leap into the dark but Tommy Flowers was convinced it could be done, nobody else was, so he got Dollis Hill to do it more or less as a private venture, because nobody else believed it would work. They got it working at Dollis Hill by December 1943 it was then shipped to BP reassembled over Christmas 1943 and operational January 1944 just in time for D Day. It really was an absolutely crucial thing to have happened because they were able to decipher the messages from Hitler to his Generals in time, because it reduced the time of breaking from weeks to hours and so suddenly just in time before D Day they deciphered these messages which confirmed that Hitler had swallowed the deception campaigns, the phantom army in the south of England, the phantom convoys and was convinced that the attack was going to be across the Pas-de-Calais and that confirmation gave Eisenhower and Montgomery the confidence to go ahead with D Day even though the weather wasn't particularly brilliant. After D Day the French resistance and the British and American Air Forces bombed and straffed all the landlines in Northern France forcing the Germans to use radio communications so suddenly the volume of messages intercepted went up enormously and they rapidly built ten Colossi to handle the vast increase in traffic. By the end of the war they had decrypted 63 Million characters of high grade German text between the Generals and Hitler an absolutely remarkable penetration of an enemy communication system. What I want to do now is to take a leap in time to 1948. Now in 1948 I was a young lad of 17 and by that time I had done everything in electronics, built my own TV transmitter and radio transmitters and receivers and done all that stuff. I came across a newspaper article which said that digits had been stored on the face of a cathode ray tube. So with the arrogance of youth I spent the whole of the summer holiday trying to store digits on the front of my cathode ray oscilloscope, built out of a wartime Loran indicator and I failed and I was absolutely disgusted at this. I could not find any way of storing digits on the front of a cathode ray tube. I eventually forgot about it and went on to other things. But when I went to the Science Museum in 1989 to join Doron Swade in the Information Age Project I spent six months researching the history of all the early computers because I realised I had to get up to speed to be able to talk to the pioneers of computing about their achievements. In doing this I came across the papers of FreddieWilliams and Tom Kilburn, his PHD thesis in fact, which explained exactly how he had done it, and of course what he had done was to actually switched on and off the beam, and I never thought of that, I was just trying to deflect it from side to side and of course it was the switching on and off which gave the anticipation pulse which enabled you then to build a store. So I then fixed up a demonstration of this using some microelectronics, a few amplifiers and there is an example of the storing of digits on a cathode ray tube using exactly the same principles as originally used and I demonstrated that at the inaugural meeting of the Computer Conservation Society in 1989. I then asked Tom Kilburn whether I could come up and show him this in Manchester. At that time Tom Kilburn was very much a recluse and he eventually said "O yes I suppose you can come and show me it". So I took it up to Manchester and showed it to him and he said "didn't work like that" so I said "well yes tell me how it did work". He said "It wasn't the anticipation pulse, it was the switch on pulse which was very much larger" but this is not clearly documented in any of the literature so I rapidly changed my circuitry around and used the switching on pulse and it worked much better as Tom Kilburn had said. That was an interesting insight into a revelation of something which had not been clearly documented before. That exercise convinced me that it was possible to rebuild early computers and I had formed the Computer Conservation Society in 1989 at the Science Museum as part of the gathering together of all the information about early computers and myself and Chris Burton had restored a Pegasus computer to full working order and we also had an Elliott 803. Chris and I spent a long time talking about early machines and eventually we tossed a coin for it, he did the Manchester machine and I did Colossus but we both put our heads together on how we could do this. I now want to tell you how we actually did it. The next part of the story starts in 1991 when the owners of Bletchley Park decided they were going to leave the site. The owners were then British Telecom and the Government and they applied for planning permission to demolish all the buildings in the Park and sell it for development for houses and supermarkets which Milton Keynes needs like a hole in the head. But Milton Keynes has not got much history and myself and a number of colleagues who knew about the importance of Bletchley Park got together and started the campaign to save the Park from being demolished. This gave me the excuse for rebuilding Colossus. This is the building, F Block, where the first Colossus computer was assembled, this has been knocked down and is just a piece of grass. There were 4 Colossi in F Block but there were 6 of them in H Block and this still stands and the Colossus rebuild is in the room where no 9 stood in 1945. The Colossus computer. I gathered together all the information there was about it and that amounted to eight of these wartime photographs plus a few close-ups of some bits. So this was the complete set of photographs but what was even more surprising was that I looked at this one and I thought "someone has printed this the wrong way round" because all the others are the other way round, so I went and talked to Harry Fensom and some of the other people who actually assembled these machines and they said "Oh well we wanted the paper tape reader at the far end of the room so we just assembled the racks in the reverse order". That didn't improve the ability to decipher what was on the racks. So we had these photographs which showed the overall views, the power supplies and the rear rack containing an enormous amount of circuitry and the thyratrons. So I now had some information about the machine and a simplistic logic diagram which Tommy Flowers gave in one of his reports. The paper tape was read with the optical paper tape reader and then the internal bit streams and this was a major contribution which Tommy Flowers made. He realised that it was possible to use valve circuits to generate these patterns of bits corresponding to the patterns of lugs set round the wheels of the Lorenz machine, rather than read them from another paper tape as on Heath Robinson. This does away with the synchronisation problem of two tapes but of course it puts up the number of valves enormously because you have got to have at least one for every bit around the wheels. So the basic logic of Colossus is that it reads paper tape, it generates the bit streams corresponding to the wheel patterns it then does the double delta algorithm, if that's the one your using, and counts the results and then prints out the resulting count. Tommy Flowers had also given a slightly more techy diagram of the machine which shows more of the parts of it. So all I had to go on at the start was these 8-10 photographs and some fragments of circuit diagrams and these reports which Tommy Flowers and Allen Coombs and other people had written after the war. So I decided to have a go at it. It seemed a good thing to do and like Everest it was there so might just as well go and climb it. I first of all found out what valves were used and these are EF36s GT1C thyratrons. These GT1C Thyratrons, I have deep in my psyche a hate of GT1C thyratrons. My first job when I went to Marconis after leaving the Air Force was to help in the design of a switched power supply using GT1C thyratrons and I have hated the bastards ever since. Other valves on Colossus, 807s 6J5 and the photocells, more about theselater. We were obviously going to need an enormous number of valve holders and that raised a problem because there was no way to get valve holders of the right sort but what I found was that in fact the valve holder used in Colossus were all surface mounting, actually this little chap here. This is a surface mounting octal valve base and that is a type 21A. Now those are as rare as rocking horse droppings but there were lots and lots of chassis being chucked out of various exchanges in which were the type 21B. These are through the chassis valve holders to lower the valve height and so lower the height of the equipment. What I found in a flash of inspiration was that I could actually take these 21B holders, put them in my lathe and part them off like this, recover the disc from the bottom and then reassemble them in the reverse direction and hey presto here is a 21A holder so that is what we did. That was an enormously important leap forward because we needed thousands of these valve bases. We then activated various underground links, found out when these units were being thrown out and made sure we were there with a van when they were thrown out of the door. So we collected all these valve bases and its quite remarkable that Gil Hayward, the designer of the MK 2 Tunny, lives in deepest Wales with his own workshop so we would send to him boxes of the 21B bases and he would then machine them and send us back 21As and he's done nearly 2,000 of these over the last 4 or 5 years, a remarkable effort. So that is how we got over the valve base problem. The next problem was could we ever make any of the circuits work. The first thing I built was this counter because the counters were absolutely crucial to the whole thing. Harry Fensom had told me that they had a lot of trouble with the counters getting them finally to work properly, getting the tolerances correct. It is a bi-quinary counter, a divide by two circuit followed by a ring of five based on a pre-war design. So here it is, the first chassis, it uses breadboard construction, a piece of metal kitchen foil over a piece of wood, and it worked. As an interesting sideline when I got it working, and it actually works up to 12.5 Kc/s, it only has to work at 5 kc/s, but this one ran up to 12.5 kc/s without blinking too much. So when I got it working I rang up Tommy Flowers who I hadn't really met very much till then and said "I've made this decade counter can I come and show it to you?" and he rather reluctantly agreed. So I took it down and set it up on his kitchen table with power supplies and my oscilloscope and as soon as we got the thing working and got some waveforms coming out of it he got really interested and then we got talking about Colossus and he suddenly stopped and said " I never realised I remembered so much about Colossus" and it was just the triggering of memories by seeing the waveforms, seeing the components which actually brought back an enormous amount of information about Colossus so that was a very rewarding visit. So we now had a working decade counter so we had shown the technology could be reproduced at least at that level. The next question was what was the height and width of the racks. Of course I asked the WRENS who operated Colossus and the engineers and they waved their hands in the air saying oh about that, which is not exactly an engineering definition. Because I had realised that these were standard Post Office components I took three starting points, an octal valve base, because I had some of those, I took a 20 way jack socket strip, which hadn't changed for 50 years, and then the toggle switches shown on the front of the machine, I got the pitch of those and then I used a very sophisticated 3D system, actually it was De Lux Paint III a remarkable program running on an Amiga home computer, so I constructed a wire frame model from the photographs and then used the program to rotate the view point to normal position and then used that to calculate the height of the rack starting from those three components, and they converged on 90 inches, plus or minus about 2 inches which was pretty good for that sort of measurement. So I was quite happy that the height of the racks was 90 inches and that's what it is. Next question; what was the width of the racks. Well they obviously weren't standard racks but what I found was that in fact the width of the rack was dependant on the number of valves you got across a chassis. So if you wanted 14 valves across a chassis for electrical reasons that was what you wanted, that was the width of the rack. so we used that method to work out the widths of the racks and then I drew a CAD drawing of the complete machine with full CAD measurements of all the components. So now we had the racks and sizes of racks. Next problem: the paper tape reader. Now we could draw the bedstead and again by using third angle projection we could work out the height and widths of the bedstead on which the paper tape ran. Then I hit a problem because none of the photographs show the optical reader system. But luckily I made contact with Arnold Lynch, who is here tonight, he designed the optical system for Tommy Flowers in 1942 and he came up to my house and that was 22nd April 1994 and we spent a very happy afternoon reverse engineering the optical system on my CAD system and there it is. It is determined by the size of the photocells. These photocells, luckily again engineers had kept some as souvenirs quite illegally so I had eight of those, just enough to do the machine and they have to be spaced apart because they are fairly big objects and so that determines the optical magnification required to go from the small size of the paper tape to the width of a row of photocells. So we worked all that out and agreed that was what it looked like, we had some idea of the stepped nature of the chassis that went inhere. Then in the best engineering tradition I built a prototype of it and there it is. I learnt very early on in my life that it is very important to only engineer a test piece sufficiently to be able to prove what you want to prove. You can spend an awful lot of time over engineering parts of something you are going to test. So this is possibly under engineered. A car head lamp bulb from my 1937 TA MG, a wartime set of collimator lenses, a piece of tape round a drum, a cameral lens and the mask and the photocells. So there it is and it worked, it gave signals out of the photocells so then I assembled the whole of the frame to hold the photoelectric reader set. In actual fact when we finally came to get it working I had underestimated the width of this by about an inch which wasn't bad considering the lack of data to work on. So then we built a test rig consisting of a vacuum cleaner motor, some pulleys off a little trolley and that ran tape round and gave signals out from the photocells. Here is rather a nice photograph showing the spots of light projected through the tape and going past the mask, which I will describe in more detail in a moment, that's the original mask as designed by Arnold Lynch. Now what I found with the mask was, this was the design which Arnold Lynch produced to meet Tommy Flower's requirement for a flat topped optical pulse falling on the photocells. The transparent portion is the differential of a circle in order to give a rectangular optical pulse as the spot of light went past the mask. But what I found was that if you took a series of ones and noughts on the tape then this particular mask was causing heaving of the bottom level by the fact that it didn't actually allow the complete darkness as you went from one element to the next. So I changed the design to this design which is the one we have got now where I concentrated on getting an absolutely clear dark patch and to hell with the shape of the waveform coming out of the optical reader there and that is actually much better. There is no longer any heaving of the base line. I spoke to Don Horwood and Harry Fensom about this and they seemed to remember that they had had problems with that and they thought they had ended up with a mask rather like mine actually in BP. So there is the picture as it was in 1994. Now the next problem was the pulleys and that was an interesting exercise. Chris Burton who was fully involved with me in doing a lot of this early work, he said oh I've got a big lathe so I can turn up pulleys. We thought we were going to have to buy slices off of an aluminium billet and turn them from the solid, this would have cost a very large arm and a leg! Then Chris Burton found a founder in Telford, who actually made horse brasses but he also cast aluminium so he went along to him and asked if he could cast these pulleys for us, particularly the ones with perforations in and he said yes he could if Chris made him a mould. So Chris asked him how much would it cost, and he said well quite a lot of aluminium, I'd probably have to charge you Ï5 each for them and Chris said "SIX I'LL HAVE SIX!". So Chris made the mould and got them cast. When they were being cast he said "What sort of aluminium is it?" and the chap said "don't ask", it was cylinder heads and sumps and all sorts thrown in to the pot. But Chris managed to get a working set of wheels out of that and they have been obviously invaluable. There are the solid wheels mounted on the bedstead. Then here are the perforated wheels on the bedstead. Now at about that time, after the wheels had been made, on 15th June 1994 Harry Fensom and I went down to Plymouth to meet Allen Coombs. Allen Coombs had had a stroke and he was a very ill man but he desperately wanted to see us and talk to us. So we went down there and spent a day with him. He had great difficulty communicating because of his stroke but at the end of the day he said "well I've got all these drawings, documents and notes that I have kept, quite illegally, here they are you have them". That put me in a bit of a quandary because although I had got myself security cleared in order to do this.. There is an interesting sidelight on this, when I first said to GCHQ I want to rebuild Colossus, they said "No way" I said I really do want to do it have you got a problem and they said well knowing you probably yes. So I said right give me my full security clearance back again, because I was in MI5 for many years, and take me behind the wire and tell me where the cow pats are so that I don't put my feet in them. So they did that and that was very fortunate because it gave me a sort of quasi authority for taking these documents from Allen Coombs but I have refrained from bringing them fully into the public gaze because they were held by him quite illegally. But thank goodness he did. Now what they amounted to was about 10 fragments of circuit diagrams and these really are fragments. So what you have got is Panels R1-R5, not what they do nothing about where they are on the racks just scraps of circuitry and about 10 fragments of these with all these various bits of circuitry on them. But no indication where they are on the machine, no machine drawings just the actual pieces of circuit diagram. Like for instance there, another set of Panels, so now we start the real detective work. First of all what are these panels, what do they do and secondly where are they on the machine. One of the key elements which did come from Allen Coombs was a hand written rate book and that has been absolutely invaluable because this was just a hand written document which lists all the components in a Colossus. It also gives this nomenclature along the top and we worked out that this must be what they called the various racks. One list is of the standard Post Office components another is the non Post Office items which had to be bought in. So that gave us a bit of a clue as to what the layout might be. Then we had to start trying to work out where everything went on the racks. This involved trying to locate the racks actually on the Colossus. Now what I found out by studying these fragments was that the draughtsmen who had done this, obviously working under extreme pressure, they had used an absolutely standard notation of V1, V2 etc V21, V31 down here and what I worked out with a little bit of thought was this was actually a standard layout on a chassis. So starting at the top left hand corner was V1 so here is a counter control circuit so V1, V2 , V3 across the top then V21 starting the next row and then V31 and V41. So now if you look at the circuit diagram you can identify the valve types and you could now lay out valves by type on the panels. Then you could begin to get an optical impression of what the panel looked like because you knew if a valve was an 807 it was a very large one, a 6J5 was a smaller one and EF36s had top caps. So here for instance are the actual original sketches I did for a panel, EF36s all along there then a 6J5 then 36s there 6J5 and a central tag strip. So now I go round the photographs looking for a chassis that had that optical characteristic, and I found them. So that is the way we worked out where the panels went on the racks and we also managed to work out the rack layout for the whole machine, eventually, after weeks and weeks of work. So we now had the letters of the racks, the J rack, the K rack, the relays and the counters, the C rack, R rack, W for the thyratrons, I'm not sure why thyratrons were W, and the power supply units. So that meant we now had the racks in position and were able to carry on reading the diagrams to get the information out. So now we could go round the machine finding out where all these various panels were. So now we could tie down that this was the counter rack here and these were the counter control circuits here and this was the J rack and the K rack and so on. But there was still one major problem. In the fragments that had come from Allen Coombs there was no master control circuit, which is pretty fundamental. But there was a master control circuit for Super Robinson which was a later machine. So I reasoned in my engineering manner, nobody is going to reinvent the wheel particularly in wartime so its almost certain that the later one was a spit copy of the Colossus one. So I based the Colossus one on the Super Robinson master control circuit. Now the master control circuit controls the operation of the whole machine. The machine actually has two basic cycles first of all there is the long cycle which is right the way through the message tape which you are trying to find the settings on. The tape is joined into a loop with a sticky bit in the middle and special holes are punched offset between the data tracks to give a unique mark for the end of data and the start of data. The various things that have to go on in this interval of about 100 millisec, not a long time, is that first you have to unlatch the holding relays and then you have to latch the count standing on the electronic counters onto relays you then have to zero the counters and the rings, and then set the rings to their strike start position and then open up the pulses from the sprocket holes that precess the rings at the start of data. So that is the basic logic of the control circuit so I then built the control circuit and it worked. But one of the difficulties in the whole machine was these dreaded thyratrons. Now Tommy Flowers had decided that from the point of view of economy it was better to use gas filled triodes because they can hold two states in one envelope permanently. So you have gas filled triodes, 501 of them, and they are cut up into rings the lengths corresponding to the ring lengths round the peripheral of the wheels on the Lorenz machine. And then you have to precess this round by alternately firing and locking and unfiring the thyratrons, because the only way as you probably know to quench a thyratron is to drive its anode negative with respect to the cathode once its struck. So you have this bistable thing around each ring but unfortunately the Germans had been most inconsiderate in making most of the ring sizes odd and so you have another thyratron down here and you have another complete set of logic here just to handle the odd case there and then get it back into synchronism again and that control circuit was absolutely diabolical to get working but I will say that once it is working it is remarkably stable, but it was a real devil to get working. So you have the two cycles in the machine, the long cycle which is the cycle round the whole length of the tape to get the count of the coincidence and then you have the clock cycle and the clock of course is the sprocket hole signals off of the tape and whatever speed the tape is running at, that is the speed of the machine and it is a very elegant and simple system. Everything was strobed, the data and the output of the rings, was strobed on the back end of the clock pulse. Those then propagated through the all the logic circuits to give you the double delta function and then the result of that was then sampled on the front end of the clock pulse to be counted as to whether it was nought or one at that point. So that was the minor cycle on the machine, the major cycle was the tape right the way round. Then of course you also had to get some output from the machine that required a typewriter. There is an intermediate stage of building the machine. There we have some of the circuits in place and here are the first bits of the counter control circuits there then I managed to get the whole machine working at 2 bit level all the way round the machine. This was a bit of a Sale lashup in order to do this because we didn't even know whether the machine actually would work at all so it was lashed up in a fair hurry and then His Royal Highness the Duke of Kent came and officially switched on the basic working Colossus on the 6th June 1996. There is Tommy Flowers in his wheel chair there and that was a marvellous occasion. I said to His Royal Highness, "I think we should switch on Colossus now, that switch there" "That switch there?" and this was a 1940s light switch which switches on the whole of Colossus exactly as it was and it worked I'm pleased to say. But there was some trepidation about this because in the best Sale tradition I had decided the previous day that the master control circuit looked a bit tatty, we had actually rewired a new master control circuit so I started about the morning of the day before and changed the master control panel. I got it working at 12.15 on the morning of the day when His Royal Highness was coming to switch Colossus on. Not an exercise to be repeated. So there is the machine getting more and more of it together. And then we had a stroke of luck. At that time we had no official reports, well at least, because I was security cleared, I had sight of a couple of the histories of the Newmanry but they were not released to me permanently, but then the Americans were forced to put into the public domain in '96, about 5,000 documents from WW II. They went into the National Archive and I immediately downloaded a listing off the Internet and my eyes came out on stalks because one of the documents in that list was a "Report On the British attack on FISH" and that was in the public domain in Washington so I pulled all sorts of strings and got copies of that back to the UK tout de suite and I found that it was a complete exposition of the way Colossus was used including all the later techniques of which I had no real knowledge. This report was written by Walter Jacobs on 7th May 1945. These were American service personnel seconded to Bletchley Park and they then wrote reports back to America on the work here. The other important report was written by Albert Small unfortunately it is a negative photographic copy so its very difficult to show but here is the introduction written out and it is a marvellous tribute from Small to the UK. "Daily solutions of Fish messages at GC&CS reflect a background of British mathematical genius, superb engineering ability and solid common senes. Each of these has been a necessary factor. Each could have been overemphasised or underemphasised to the detriment of the solutions; a remarkable fact is that the fusion of the elements has been apparently in perfect proportion. The result is an outstanding contribution to cryptanalytic science" That's a very nice tribute from an American. So these reports then enabled us to work out a lot more of the circuits because obviously there were many circuits on the machine where we had no idea what they did and what they were used for but these reports enabled us to work that out and so we have progressed from there to the machine that we have now got which is almost complete. I would now like to show you a brief video of Colossus working, actually in 1997, but at the end is a section on the machine as it is now. The typewriter was a great difficulty because they actually used an IBM typewriter. I haven't got one of those but I've got a Flexowriter which has the same mechanical movement and is just as diabolical to get working as the original typewriter was. Then the video was shown 2nd part of video Now this is where we took a step back and decided that we were now going to rewire the machine to proper Post Office standards, and this is David Stanley an ex Post Office engineer who thought he had retired who is actually busy rewiring Colossus with all the proper looms and tag blocks. We still have the basic machine working but the extra circuits we are putting in are being built in this way to the full Post Office standard and then we will re-engineer the first circuits so the whole machine is actually working to full Post Office standards. You can see the increase we now have in the number of circuit panels on the machine and the size of the thyratron rings and then we have put in all the counters now so we have five, four decade counters so that is the machine almost complete now. It will take about another six months to complete. That is the end of the presentation I have had to truncate it a little bit but I hope that it has given you a flavour of the problems that we faced in trying to recreate a machine with absolutely minimal information and it has been laborious detective work of almost archaeological dimensions to try and get all the information and tease it out of fragmentary data. Thank you.