By the year 2000 I had spent seven years rebuilding Colossus, which although still
not complete, had demonstrated the enormous code breaking power of a dedicated
electronic computer. In 2000, whilst I was banned from Bletchley Park
and could no longer work on my rebuild of Colossus, my thoughts
turned to Heath Robinson. I had always been intrigued by its rather esoteric logic
design and decided to try to recreate it to see how it really worked.
The first problem encountered was lack of any photographs of wartime Heath
Robinson. It was briefly described in a paper by Allen Coombs in the 1983 edition of
the Annals of the History of Computing with a fragment of a circuit diagram but no
layout or visual description. This paper described the 25Kc/s phase modulated
method used in the logic circuits.
In 1996 Harry Fensom and I had visited Allen Coombs at his house just outside
Plymouth. Harry, who was a GPO engineer at Dollis Hill, had worked on Heath
Robinson and then on the Mk 1 and Mk 2 Colossi. Allen Coombs had had a slight
stoke and could not communicate very well but he gave me all his wartime notes
which he had kept, quite illegally, after the war. I promised to be discreet in my use
of these notes. We did not discuss Heath Robinson with him. I was then deeply
involved in setting up and running the Colossus Rebuild Project for which his notes
and circuit drawing fragments were a great help.
In 2000, on looking back again through Allen Coombs's papers I suddenly realised
that one A3 sized sheet was in fact circuits of Heath Robinson, recognisable by the
ring modulator circuits. On checking back to Allen Coombs's paper of 1983, it was
clear that the parts of circuitry in that paper had been copied from this original sheet.
There was also, next to it, a machine drawing of a 19 inch plate with a rectangular
aperture in it. This machine drawing had the title "Apparatus Telegraph
Transmitting", case number 11951. I had already identified this title, confirmed by
Arnold Lynch, as being used for the Bedstead design. When I was building the
Colossus bedstead in 1995 I had searched the BT (Ex Post Office) document
archives. Cases 11950 and 11952 existed. No trace of 11951. Confirmation of
secrecy?
In 2000 I first put together in best birds nest style a lashup of the 25kc/s master
oscillator and two bridge logic circuits. I had to design the transformers for these
circuits from scratch. No such details on the surviving circuit diagram. However I
reasoned that if the circuits were wired up at Dollis Hill they probably used standard
GPO components and the transformers in Amp 32 units seemed likely candidates. I
had collected lots of these to obtain surface mounting valve bases for Colossus so I
stripped out some transformers from the Amp 32's and then took the transformers to
pieces, removed the wiring and rewound them to work at 25kc/s. I hadn't lost my
transformer design skills and the lashup worked well showing the 180 degree phase
shift as the bridge bias was altered. So I had shown that it was possible to recreate
the Heath Robinson circuitry.
There the matter rested until in May 2001, still being prevented by the Bletchley Park
Trust from completing my Colossus, I decided to completely rebuild Heath Robinson.
I first studied in great detail the A3 sized circuit diagram of Heath Robinson. A
considerable amount of information could be deduced from it. Firstly the number of
Post Office jack strips referenced on the circuit page corresponded exactly to the
aperture in the drawing of the 19 inch plate. So this was a jack field of 20 way jack
strips. But this implied a 19 inch rack whereas up until then I had assumed that the
only rack was the wide not very tall main Heath Robinson rack built at TRE. However
talking to Harry Fensom, he confirmed that the "Valve Rack" had been built at Dollis
Hill and was a tall 19 inch rack. Furthermore the words for chunks of circuitry on the
circuit diagram were "plates" and "panels" confirming Dollis Hill origins. (TRE would
have called them "chassis"). So now there were 13 plates of which by inference the
jack field was number 6.
Next what size were the plates. I was pretty certain, and Harry Fensom confirmed,
that the standard GPO transformer cases were used as in Amp 32's. This allowed 6
transformers side by side long ways across a 19 inch plate. So the 12 transformers
required for the six logic circuits per plate implied two rows of six transformers with
the valves centrally in a row between each transformer pair. This gave a plate width
of 6 inches. Now with a slightly narrower plate for the master oscillator, all the plates
fitted onto a Colossus rack height of 90 inches.(Harry confirmed that they only used
one rack height for all machines.)
There were two transformer types on an Amp 32, one with a large bobbin and
laminated core, the other much smaller. Which to use? I noticed that on the circuit
for the detector transformer labelled T1 contained two cores. That confirmed that
the smaller cores were used since two small cores would fit into one transformer
can.
So now I revisited the design of the transformers based on the small bobbin and
laminated core. Again my old design skills didn't fail me and the resultant transformer
at 25kc/s had zero phase shift input to output, almost zero leakage inductance and
only loaded 600 ohms by 10% with open secondary.
I decide to use plywood for the plates to avoid cutting steel until I was sure of the
layouts. When I mentioned this to Harry Fensom I was pleased and relieved when he
said that that was exactly what they had done at Dollis Hill. So I built a plate for the
master oscillator, two plates each containing four logic gates and a detector board.
All using the rewound transformers and with extra components like the bridges
inside the transformer cans.
A test of eight logic circuits in series showed that nearly zero unwanted phase shift
could be achieved with the addition of a capacitor and a damping resistor across the
anode coil of the output coupling transformer in each logic circuit.
Next problem, how to simulate the tape reader bedstead before actually building
one. I realised that I could use the line printer parallel port on a PC as a signal
source. Eight data bits would give me the signals as if coming from eight photo cells,
i.e. previous and current bits from cipher and Chi tapes for two tracks out of the five
on the tapes. The strobe bit on the LP port could simulate the sprocket pulse. A
quickly written Qbasic program gave the right information written to the LP port but
the output lines were switched between 0 and +5volts. A rapidly wired interface box
using Op Amp chips produced the +- 10 volts required to drive the ring modulator
logic circuits.
A few delay loops in the Qbasic program allowed the data rate to be set and ensured
that the simulated sprocket pulse arrived correctly in relation to the changing logic
drive signals. I had also wired up parts of the jack field panel so now the whole
circuitry could be tested from paper tape signals to detected output from the double
delta algorithm and it worked extremely well up to a simulated data rate of 2kc/s.
Because I was using in my Qbasic tape simulation program the same cipher text and
Chi patterns as in my Virtual Colossus, I could do a direct comparison of results from
the double delta algorithm. They were exactly the same on Heath Robinson as on
Colossus.
Now came the counter circuits. No circuits for these but Harry had told me that a ring
of ten thyratrons had been used. I knew Wynn-Williams had worked before the war
on counting circuits for nuclear experiments. I looked up his papers in the
Proceedings of the Royal Society and found the circuit for a thyratron ring counter. I
showed this to Harry who agreed that the Heath Robinson first stages of the
counters were like that. So I built them and after some slight experimenting they
worked. Now I had to make the relay second and third stages of the counters.
It now remains to build a Bedstead paper tape reader unit. I looked up my files for 1994
on Colossus and reordered the steel. The Heath Robinson bedstead was later used on Colossus
so they were the same dimensions and construction.
So now, very shortly, there will be a complete, working, rebuild of Heath Robinson,
achieved from one A3 sheet of circuit diagram, and a lot of detective work.
