Rebuilding Colossus

The Colossus Rebuild Project

by the late Tony Sale who led the project and became co-founder of TNMOC.

1993, Sources, Detective Work and Feasibility

In 1993 I gathered together all the information available. This consisted of eight wartime photographs taken of Colossus in 1945 plus some fragments of circuit diagrams which some engineers had kept quite illegally, as engineers nearly always do.

The original drawings of the mechanical assemblies had been deliberately burnt in 1960. The first stage was to produce accurate drawings of the frames. The project began to acquire a more public face. The Bletchley Park Trust, which had been established in 1992, obtained permission to set up and open to visitors an Exhibition of wartime code breaking in the CAA assembly hall in A Block in Bletchley Park. Part of this Exhibition was about the German Lorenz cipher machine and the role of Colossus in breaking the Lorenz ciphers. The exhibition opened in February 1994.

The ambition to rebuild a working Colossus to be as authentic a standard as available information and analysis permitted was publicly stated.

The visit to Dr Allen Coombs

In June 1994, Harry Fensom, one of the original Colossus engineers, and I visited Dr Allen Coombs in Plymouth. Coombs had engineered the Mk 2 Colossus from Tommy Flowers' Mk 1. As they left he gave me all his wartime notes, including some circuit diagrams of Colossus. This was a marvelous gift and really made the rebuild possible.

The site for the Rebuild

We would rebuild Colossus in its historic environment -- in one room in Block H where Colossus No 9 had stood during the war.

Getting Under Way

The Colossus Rebuild Project's bank account was opened on 6th July 1994 with startup money from my wife and myself.

Before we could make serious progress we needed the frames that supported the machine. I located, in Bletchley a supplier of steel, Charles Head (Blacksmiths) Ltd. They were very keen to help with such a prestigious project and have, over the years, supplied me with large amounts of cut steel at very reasonable prices.

The Royal inauguration of the Rebuild Project

On 18th July 1994 HRH The Duke of Kent formally opened the Bletchley Park Museums and inaugurated the Colossus Rebuild Project. At this stage there was just a heap of steel on the floor of the room in H Block.

Christmas 1994, the bedstead built

The steel frames for the bedstead, the paper tape reader, were built and the illuminating lamp and lenses were installed together with the amplifier chassis, the mask and the photo cells and electrical signals emerged from the amplifiers for the first time. This was in late 1994.

I then built the amplifier chassis and by Christmas 1994 we had a working bedstead and thoughts could turn to the other racks.

Sourcing components

Colossus was originally built at the Post Office Research Laboratories in North London and used a large number of standard issue components. Luckily a large number of old Post Office telephone exchanges were being decommissioned and changed to digital. We were able, through Gerald Palmer, to collect large amounts of equipment which otherwise would just have gone for scrap.

The next problem was obtaining the large number of valves required. Mainly Mullard EF36 pentodes but also 6J5 triodes, 6V6 and 807 tetrodes and the GT1C thyratrons. I had myself acquired some of these over the years but not enough for a complete Colossus. I started going round various vintage radio dealers and found that the valves did exist, but at a price, which meant I would have to get some serious sponsorship. This came, but, inevitably, only later.

Gathering a Team and Next Steps

In early 1995 I was joined, full time, by Cliff Horrocks who had recently retired from the Diplomatic Wireless Service at Hanslope Park. Cliff eventually became the Manager of the Colossus Rebuild Project, and looked after the documentation and the tasking of the many volunteers who offered help. Another retired engineer, Paul Bruton, joined us part time and with John Pether, who had a large workshop, coming in at weekends, we now had the nucleus of a team.

We decided to attack bits 4 and 5 of the 5 bit cipher text characters. These bits were enciphered on the Lorenz machine using mechanical wheels known as K4 and K5. These wheels had the shortest lengths of 26 and 23 bits, thus requiring the minimum number of thyratrons and valve bases in the vacuum tube rings used in Colossus to simulate them. These rings were also known as K4 and K5. Finally, we would need a master control panel to hold everything together. Later we would need the whole typewriter logic so that we could print out results.

With the optical reader, the bedstead, and the power supplies, operational it now became clear that the Master Control panel was crucial to further progress. It was this panel which held the tubes and circuitry for handling the two timing cycles used in Colossus. The sprocket hole cycle and the tape join cycle.

So I built a panel to test these ideas and verify our understanding. I installed it above the jack field panel on the front rack. Eventually it worked.

Meanwhile we had constructed the massive W rack which went in the centre of the back racks of Colossus so now I had to consider the thyratron rings which went on the W rack.

So I wired up two thyratron ring panels and two ring controller panels and eventually got them working together with the uniselectors which selected the start thyratron to be set before the ring started precessing round and the patch panels used to select the bits to make up a wheel pattern. So now we had K4 and K5 ring pattern streams coming from the W rack.

Figuring Out the Physical Layout

Next we needed the two staticisor panels for the K streams. We found the circuits for these and I wired up two panels and put them on the front rack under the staticisors for the Z streams.

We were now fairly confident that we were going to be able to make the circuits work. So far, for testing and verification, we had just stuck panels any old where on the single front rack built so far, but before duplicating the panels we had to consider where the panels would have gone in the original.

We had identified the R rack as holding all the staticisor panels so now moved the first test panels from the J rack onto it. Also on these panels were the one bit stores which enabled the delta to be produced. When these were working and installed on the R rack we had DeltaZ4 and DeltaZ5 the first components of the double delta algorithm soon followed by DeltaK4 and DeltaK5.

Meanwhile Cliff Horrocks had located the circuits of the XOR logic gates and had laid out and built the logic panel containing 5 gates. This went onto the top of J rack and when we connected in the Delta Z and Delta K signals from the R rack we had the result of DeltaZ4 XOR DeltaK4 XOR Delta Z5 XOR K5 which is the double delta calculation needed to be counted.

The decade counter panels were moved from the back of the J rack onto the C rack together with the metal box to hold the set total switches and a counter control panel also installed. When the counter control panel was wired we could connect across the output from the XOR logic panel and start counting results.

Printout and Display

With the signal processing circuits shaping up we now had to tackle the apparatus for count printout and display. I knew that originally an IBM electric typewriter had been used as the printer. I could not find one of these but did find a pair of Flexowriters from the 1950s.

John Pether had installed a large lamp panel on the S rack used on Colossus to display counter outputs and ring start positions, together with the master switch panel used by the Colossus operator to initiate runs and the uniselectors used for setting ring start positions.

I connected rows of lamps on John's lamp panel to the latching relays and we had a lamp display. Not in the original format but at least giving a display of the counts.

So by early 1996 we had a basic Colossus working, only on two tracks, but it could show Lorenz wheel setting on a real cipher text.

Making it Look Right

We now concentrated on making Colossus look like the period photographs. John Pether made the large switch panel on the K rack, I cut and fitted the sloping plug panel in front of it. Charles Head (Blacksmiths) made up the tubular stand for the typewriter, we added the relay boxes on the S rack and Colossus began at last to look like the photographs.

By now we had achieved considerable media coverage, on TV, Radio and in Newspapers and Magazines. I was asked to give the Royal Institution Friday evening lecture on 23rd February 1996. My wife and I dined with Sir Peter Day, President of the Royal Institution and HRH the Duke of Kent. before the lecture.

Publicity attracted our first sponsorship --, from Mr Frank Morrell, an ex Post Office engineer and designer of the first decoding machine for the Lorenz cipher, and parts of Heath Robinson. At Easter 1996 David Stanley joined the team and what he had learnt in his Post Office apprenticeship helped us to lay out and wire racks to exacting Post Office standards.

A Royal switch-on

At this stage we decided that we had enough of the basic machine working to have a large re-union of the wartime people involved, and to ask HRH the Duke of Kent to come back and formally switch on Colossus as it then stood, on 6th June 1996, the anniversary of D-Day. On his arrival the Duke first went into the Colossus Rebuild room to formally switch on Colossus in the presence of Tommy Flowers, the Colossus team and some of the original engineers. After the Duke had left, the guests went over to Block H and filed through the Colossus room to see Colossus working and talk to Tommy Flowers.

As a result of all this publicity I was invited to give a lecture on Colossus at the Institute of Electrical Engineers in London. This in turn resulted in more sponsorship money, this time from Mr Keith Thrower and from "The Mrs L D Rope Third Charitable Settlement" via Crispin Rope.

Going for a Mk 2

We now made the decision to go for a Mk 2 Colossus. This, with 2,500 tubes, was far more powerful than the Mk 1, the prototype and there were ten Mk 2s in Bletchley Park by the end of the War and it was a Mk 2 that stood in the room in Block H where we were rebuilding Colossus.

But in order to decide what we had to build, we needed more information about how Colossus was used to break the Lorenz codes, and this came from America.

In 1995 the American National Security Agency (NSA) had been forced, by application of the Freedom of Information Act, to put 5,000 World War II documents into the US National Archive (NARA). I eventually obtained a listing of these and was amazed to see a whole series of documents relating to Colossus and the breaking of the German Lorenz cipher. One of the most important documents ones was written by Albert Small, an American seconded to Bletchley Park who worked with Colossus The title of the paper was "The Cryptographic Attack on Fish". In late 1996 I obtained a copy of this and other related documents. This material was enormously important. Unlike the very sparse UK documents here was a complete and detailed account of how Colossus was used to break the Lorenz cipher

We now needed more sponsorship and Quantel Ltd. came to our rescue. They gave us the funds to buy enough GT1C Thyratrons to make up all the rings. Luckily various dealers had kept these from just after the War.

So now Cliff laid out the ring controller plates and the thyratron plates and various volunteers wired them up.

Now I turned to the paper tape reader and got that working on all five tracks, but not very reliably. So by 1999 we had repeated the 1996 results but with signals now correctly routed through the switches on the J rack.

2000 and the big K rack switch panel

This was the major, and very obvious, difference between a Mk 1 and a Mk 2. It was a vast improvement over the complex jack plugging needed to configure an algorithm on the J rack jack plug socket panel on the Mk 1.

After a lot of detective work we managed to identify the chassis plates needed for the logic behind the switches on the K2 panel. Cliff laid these out and more volunteers assembled them at home.

In 2000, I made a wrong decision as to which logic plates were associated with which sets of switches on the K2 switch panel. David Stanley wired up the whole of the back of the K2 panel, based on this decision, and connected up the logic chassis plates.

David also wired up the counter rack and the set total switches.

March 2002, and a disaster!

During 2001/2 I had built in my garage a replica of Heath Robinson, the precursor of Colossus. This required two paper tape readers and I had built two amplifier chassis for the photo cell outputs. These worked much better than the amplifier chassis on Colossus, so the first thing I did was to install a Heath Robinson chassis on Colossus. This immediately improved the Colossus paper tape signals and Cliff then rewired the original Colossus chassis to the Heath Robinson specs.

So now with good reliable paper tape signals we could start commissioning the K2 switch panel. Disaster struck - it became obvious that we had allocated the wrong logic panels to the groups of switches on the K2 panel.

After a lot of discussion David Stanley said that the only way to get out of this was to re-wire the whole of the back of the K2 switch panel and he said that he would do just that. It took him nearly six months!

2002, getting the K wheels working, K2 switch panel back!

Bob Alexander had divided up the thyratron panels into rings of the size required for the various wheel lengths on the Lorenz machine so I now started to commission the thyratron rings. This was an extremely difficult task. To get a ring working both the ring controller chassis panel and the ring panels had to be coaxed into operation. K4 and K5 were working, now K1, K2 and K3 had to be tackled.

We needed to connect the thyratron rings to the patch panels used to set the ring patterns and make up the complex wiring looms from the patch panels to the pointer uniselectors and up to the top tag blocks on the W rack ready to go over to the ring start jack panel on the S rack.

David finished his massive task of rewiring the K2 switch panel and with due ceremony we re-installed it on Colossus. After correctly connecting up the relevant logic panels the K2 switch panel began to work, much relief all round. So now we had signals coming onto the "Q" bus on the K2 rack, 5 wide from the paper tape reader plus rings K4 and K5.

At this stage Charles Coultas had joined us and we commissioned the delay or "remembering" circuits. This part of Colossus provided the five-fold increase in processing speed which was such an impressive part of the Mk 1 to Mk 2 upgrade. This gave us the "R" signals on the K2 switch panel.

Bob Alexander and Peter Merriman designed the system for setting and incrementing the ring pointer uniselectors on the W rack and the chaser uniselectors on the S rack. Rob Dickson designed the typewriter drive logic but was taken ill before he could complete it. Phil Hayes took up Rob's design and implemented it on the S rack.

All Ks working, 2003

By early 2003 I had beaten all five K rings into submission and we could now begin some real testing of the Colossus Mk 2.

We re-installed the original tape amplifier chassis which Cliff Horrocks had rewired and at last got good reliable signals from all five data tracks and good start/stop signals at the beginning and end of the cipher text on the tape.

I had, in the meantime, upgraded Virtual Colossus on my web site to a Mk 2 complete with all 12 rings and patch panels. This, together with the Walter Fried weekly reports that I found in the American National Archives, enabled me to work out what algorithms we needed to break the K wheel settings on a real cipher text. So now we could try setting up these algorithms on the Colossus Rebuild. At first results were not very good. We just could not get reliable and consistent counts. But gradually by changing valves and getting rid of dry joints and bad connections it began to settle down.

Meanwhile David was installing the massive inter-rack wiring between the W and S racks and making up and installing the looms for the chaser uniselectors. It was these uniselectors which "followed" the ring pointer uniselectors on the W rack and drove the ring position lamps on the lamp panel.

John Whetter joined us, helping to wire up the pattern patch panels.

By December 2003 we were getting reliable counts on the (1+2)= . algorithm for setting K1 and K2, the 4=5 given 1=2 algorithm for setting K4 and K5 and the slash (/) count for setting K3. This was on a cipher text of a re-encipher German decrypt using the K2 one back limitation and BREAM patterns from the Walter Fried reports.

The wheel start positions were set by using a croc clip lead feeding the strike pulse onto the ring pattern patch panel. The pointer uniselectors were not yet working.

The home run to 1st June 2004!

At a team meeting in January 2004 we decided to go all out for a working Colossus Mk 2 on 1st June 2004, the 60th anniversary of the first running of a Colossus Mk 2 in Bletchley Park in 1944.

The critical path was the inter-rack and chaser wiring looms to implement wheel starts from the wheel start position jack field on the front of the S rack. We decided to do just the K wheels and we thought that this was achievable.

Phil Hayes thought that he could get some if not all, of the typewriter driver system working. At least giving wheel positions and counts. So we all set to.

The result was that on Thursday 20th May I filmed Colossus Mk 2 setting all five K wheels on the BREAM cipher text and after editing, this video was shown to 120 people at our commemoration event at the Science Museum in London on 1st June 2004. This was a marvelous occasion with over 30 of the original wartime people there, WRNS, maintenance men and Knockholt interceptors.

What next?

We now have to get the automatic stepping of the pointer uniselectors working and get the set totals and the typewriter output correct.

Then we need to demonstrate the speed up obtained by using the "remembering" circuits. We think that about 4 months is required in which to evaluate, video and document the Colossus Mk 2 breaking wartime Lorenz cipher texts.

The Colossus Mk 2 rebuild

The rebuild can be seen, in The national Museum of Computing in Block H located on Bletchley Park, in the original room where Colossus No 9 stood in World War II. It is a marvelous tribute to Tommy Flowers, Allen Coombs and all the engineers at Dollis Hill and a great tribute to Bill Tutte, Max Newman, Ralph Tester and all the code breakers involved at Bletchley Park ... not forgetting all the WRNS who operated and supported Colossus and the radio interceptors at Knockholt without whom there would have been no messages to break.

Lastly I would like to thank my wife Margaret for agreeing to the use of our own money to start up the project and for her continuing support and encouragement.

The financial sponsors: A E & M D Sale Mr Frank Morrell The Mrs L D Rope Third Charitable Settlement Mr Keith Thrower OBE Quantel Ltd The Computer Conservation Society

Contributions by special low prices: Charles Head (Blacksmiths) Billington Exports Ltd Claude Lyons Ltd

The Colossus Mk 2 Team Manager, Cliff Horrocks (deceased)
Bob Alexander Charles Coultas Phil Hayes Peter Merriman John Pether David Stanley John Whetter Rob Dickson (deceased).

Thanks also to: The many hundreds of individuals who have searched their garages and lofts and sent valves for Colossus.

References

Randell, B.: 'The Colossus', University of Newcastle upon Tyne, Technical Report Ser. 90, June 1976

Good, I. J.: 'Early Work on Computers at Bletchley', Annals of the History of Computing, Vol 1 (1979). p. 38

Flowers, T. H.: 'The design of Colossus', Annals of the History of Computing, 1983, 5, (3), p. 239

Good, I. J.: 'Enigma and Fish' in Hinsley, F. H., and Stripp, A. (Eds.): 'Code breakers' (OXford University Press 1993) p. 149

Sale, A. E.: 'Colossus 1943-1996' M & M Baldwin, (1998), 3rd imp 2004

Archive material

The American National Archive (NARA), College Campus, Washington DC.

NR 4628 SPECIAL FISH REPORT (BOX 1417)