Harwell Dekatron / WITCH Restoration - 2010
Post this page to popular social media
Each TNMOC project has either a working group or project team assigned to do the work. Working groups are either managed in association with the CCS (Computer Conservation Society) or solely within the Museum.
Below are updates on the progress of the Harwell Computer restoration working group during 2010.
For details on the progress made during 2009.
For more detailed information on this system please visit the CCS Harwell Dekatron / WITCH page.
17/08/2010 update from Tony Frazer
It is a while since I did a progress update, as we have slowed up during the hot summer period.
I gave in to temptation and started work on the arithmetic rack - starting with the pulse generator unit.
First of all, I vaccumed and wiped down the unit. All valves were removed, tested (where practical) and re-seated in their holders. Several showed a low mutual conductance and were replaced with known good examples. A loose wire (for V1 anode supply) was soldered back in place. Some in-circuit testing of passive components and diodes was also carried out.
Yesterday (16th Aug) Dick Barnes, one of the designers visited us. Together with Eddie and Ros, orders were successfully entered using the manual keys on the control panel. The trick was to enter a space character before keying the order. It was noticed that the indicator lamps and relays of the sending and receiving stores were operating, more or less as expected.
Dick suggested that we could punch up a tape to check that all send/receive relays, store selection relays and indicator lamps operate correctly - so that is the plan!
27/06/2010 update from Tony Frazer
Eddie has almost completed testing the rack interconnect cables (see picture below) The thin white cables running above the thick brown cables are temporarily bridging known intermittent or disconnected wires. When we obtain the correct tools for maintaining the Plessey connectors, we can re-terminate affected cables.
During the Vintage Computer Festival the talks on the Harwell Dekatron Computer were well attended and we were able to demonstrate an instruction being executed by the relay racks. Our initial hope was to have two small loops of tape, mounted on tape readers 1 & 2 which performed as follows:
(Bootstrap searches for block 1 on reader 1 and executes the next instruction on reader 1)
Tape Reader 1: Block marker 1; Search for block 1 on reader 2; Transfer control to reader 2; (loop)
Tape Reader 2: Block marker 1; Search for block 1 on reader 1; Transfer control to reader 1; (loop)
The above sequence would repeat indefinitely, however we encountered a problem as the transfer instruction on reader 1 was not being read in. Today I discovered the reason for this was that I had omitted to include a space character preceding the transfer instructions!
The following was then tried to prove my "hunch":
Tape Reader 1: Block marker 1; Search for block 1 on reader 2; space; Transfer control to reader 2;
Tape Reader 2: Block marker 1; Search for block 1 on reader 1; space; Transfer control to reader 1;
This worked perfectly!
I then punched up 4 tapes as follows, with code in ()'s repeated 8 times:
Tape Reader 1: (Block marker 1; Search for block 1 on reader 2; space; Transfer control to reader 2; Block marker 1; Search for block 1 on reader 3; space; Transfer control to reader 3; Block marker 1; Search for block 1 on reader 6; space; Transfer control to reader 6;) finish;
Tape Reader 2: (Block marker 1; Search for block 1 on reader 3; space; Transfer control to reader 3; Block marker 1; Search for block 1 on reader 6; space; Transfer control to reader 6; Block marker 1; Search for block 1 on reader 1; space; Transfer control to reader 1;)
Tape Reader 3: (Block marker 1; Search for block 1 on reader 6; space; Transfer control to reader 6; Block marker 1; Search for block 1 on reader 1; space; Transfer control to reader 1; Block marker 1; Search for block 1 on reader 2; space; Transfer control to reader 2;)
Tape Reader 6: (Block marker 1; Search for block 1 on reader 1; space; Transfer control to reader 1; Block marker 1; Search for block 1 on reader 2; space; Transfer control to reader 2; Block marker 1; Search for block 1 on reader 3; space; Transfer control to reader 3;)
Tape readers 4 + 5 were not included as 4 has a weak spring on the pawl and the spring is missing on reader 5.
This also worked perfectly! (Actually, an intermediate version including tape 4 worked intermittently due to the weak spring.)
The above sequence keeps the machine busy for 10 minutes and I would like to acknowledge all the effort Eddie has put into getting the relays working so reliably. Well done!
Another bonus of the above program is that it almost completely tests the Input Table. All that needs to be tested is tape reader station 5, which can be done by temporarily swapping a tape reader from another station.
We are now investigating undocumented modifications to the computer made during the course of its life - most probably at Wolverhampton, and resuming work on the store exerciser.
See Rack rear inter-connecting cables image
24/05/2010 update from Tony Frazer
Work on the store exerciser continues. Testing of the Dekatron selector circuit comprising thyristors and uni-selectors revealed problems with spurious triggering of the thyristors. This was hardly surprising as the circuit was constructed on veroboard, alongside other 'noisy' circuits. I have almost completed rebuilding the thyristor board with star earthing, a lower gate-cathode impedance, dedicated voltage regulators, protection diodes and R-C snubbers on gate-cathode and anode-cathode. Hopefully this will suppress the false triggering - if not, I can make minor software changes to prevent simultaneous operation of the uni-selectors - but at the expense of selection speed.
Eddie is taking a well earned break, and Johan is also getting his own project started - restoration of an ICL System 25.
John P has examined the Creed reperforator which appears to be in generally good order - although it was noted that the centrifugal governor gets quite hot in operation - removing the cover indicates sparking, so the next step is to test the spark quench circuit components.
Many of the historic photographs of the computer show an attendant oscilloscope, used for monitoring waveforms from various units on the arithmetic rack and assisting with their calibration as part of routine maintenance procedures.
John P and I found a Cossor 1049 Mk III oscilloscope (mfd c.1953) which is in good cosmetic condition, and a Kaymar 'scope trolley which I have sanded down and repainted with Hammerite hammered dk green. I hope to augment our display with a contemporary oscilloscope such as this one, although we have several other similar Cossor oscilloscopes which may prove better candidates - or sources of parts!
Stop press - I have got both of the Cossor 1049 MkIII's working - I'm calibrating one of them at the moment...
28/04/2010 update from Tony Frazer
I mentioned last time that development of the store exerciser is under way - here are a few more details:
The selection of a Dekatron for testing is accomplished using 3 uni-selectors (which are essentially electrically operated multi-bank rotary switches). A pulse generator which will be triggered by a positive-edge of a pulse comprises 3 mono-stables to create a 1.5mS pulse for the 1st guide electrodes. A separate delayed 1.5mS pulse is also produced (just under 1.5mS after the 1st guide pulse) for the 2nd guilde electrodes. This ensures a slight overlap of the guide pulses. In order to perform marginal testing, the timing resistor for each monostable can be shunted by a second resistance. thus lowering the pulse timings by 0.5mS.
I plan on using opto-isolators to couple the TTL voltage levels to the +/-60v required by the computer.
A pre-settable threshold detector comprising an op-amp will be used to detect which cathode is glowing (it should be +20v when lit instead of -10v). Again, an opto-isolator will couple this to TTL voltage levels.
I have found the Velleman K8055 USB interface cards very useful for this type of control/monitoring application elsewhere in the museum (visitor counters and NPL network simulator), so no surprise that there is one interfacing the above circuits to a laptop!
I have also started developing a VB6 application which will select Dekatrons, perform test sequences and monitor the cathode. Success/failure will be logged for further examination. Typical failure modes I have coded so far (some are unlikely to occur in reality): no cathode lit; multiple cathodes lit; failed to advance cathode x; failed to advance cathode x under marginal conditions. A visual representation of the stores will give a quick guide to the progress of a test programme: Dekatrons failing test will become more red; Dekatrons passing tests will become more green; the more a Dekatron is tested, the brighter its displayed icon will be.
Eddie has extended the tape reader test stand to include lamps, so that the tape reader contacts can be tested by stepping a test tape through the reader (see picture below).
John Pether has cleaned and examined the Creed Page Printer, which we have started testing under power. The motor runs very well, with just a hint of that smell that electric motors emit when they have not been run for 35 years! I built a simple switch panel to set up 5-line character codes and operate the print solenoid. This appears to work OK, but it has revealed a problem with the coupling of one of the solenoids to the print mechanism. Ironically, the one which is most difficult to get at! A peculiarity of this printer is that it has been modified to receive parallel data. It is therefore not necessary to be able to adjust the speed of the motor to synchronise it with incoming serial data. As this is the only printer of this type that we have, John suggested that I build an interface which will enable us to use the more common Creed serial printers with the computer for ongoing demonstration purposes.
John is now looking at the Creed Reperforator - the first observation is that it is missing the chad box (a hopper and chute which collects all the little perforations). We already have a chad box on a reperforator elsewhere in the museum, so we could use this as a template to make a temporary one until an authentic replacement can be found.
Further to our initial test of the relay racks last week, Johan discovered how to single step through the sequence we observed, so this will be useful when we have completed the basic tests of the machine first.
Ros is analysing the paper tape software (up to tape number 50!) and developing some simulator code. Many of the programs are simple student exercises, although the calculation of Sin(x) is reported to be impressive!
Here is a short video of the relays in action for the first time in over 30 years!
See Tape reader test rig now with lights! image
21/04/2010 update from Tony Frazer
Quite a lot to report this time, as considerable progress has been made in several areas...
Johan has been working on the installation of a contactor and "red button" and extension of the mains supply conduit under the computer.
Construction of the store exerciser is progressing, with the uni-selector unit almost completed and the electronics unit construction under way. I will describe this more fully next time.
Eddie has made a second pass of the machine to adjust the high-speed type relays and check the 6v and 50v lamps, many of which were faulty. In preparation for testing of racks 1+2, the fuses have been temporarily replaced with a slightly more modern version [image 1] which is easier to remove and rewire if necessary during our testing.
Eddie conducted the first partial power-on test of the relay units (racks 1+2) today. After checking the power cables from the PSU rack to racks 1+2 for continuity and absence of shorts, these were fitted in place. This only required the 50v and 6v supplies, so all fuses for these supplies were removed from the Power Distribution Unit [image2]. After switching on the power supply, the fuses for the 6v lamps were gradually replaced without incident. This was repeated for the 50v relay fuses, again without incident. Various relays started operating as a result of the 30 second timing pulse.
Various switch operations were attempted from the control panel and interactions between various relay sets were noted and various supervisory lamps lit during the sequence [image3]. Although we are unable to verify that the operation is correct at this time, it was encouraging that when we operated the Cancel and Start keys, the same pattern of lights seemed to result from each operation [image4]
Obviously too early to comment on whether this part of the machine is operating correctly - but the initial power-on test of the relay units went better than we anticipated, with no blown fuses or wisps of smoke.
We can now continue examining the sequence of relay operation and interfacing with the tape readers.
See Image 1 - fuses
See Image 2 - fuse rack
See Image 3 - relays (covered)
See Image 4 - control panel
06/04/2010 update from Tony Frazer
I have now completed the partial rewiring of racks 4 and 5 and cleaning/replacement of U points. All store units have been replaced on the racks and repopulated with Dekatrons.
In the case of two store units, the top 5 stores in each unit have been populated with GC10B Dekatrons, rather than the original GC10A type. This has been done for two reasons: Firstly, we know that we do not have sufficient good GC10A tubes to completely populate the machine and rather than leaving the machine partially populated, we can draw from the supply of GC10B Dekatrons, some of which were stored with the computer and others that have been kindly donated to us. Secondly, the original GC10A Dekatrons are very feint and difficult for visitors to see, whereas the GC10B is much brighter and will help us demonstrate the machine. I also intend to fit GC10B Dekatrons to the accumulator in due course.
My plans for the store exerciser are sufficiently concrete to begin construction of the unit... The main purpose of the store exerciser is to provide a means of repetitively testing racks 4 + 5, connections, relays, lamps and Dekatrons and identifying faults (including intermittent!) for further investigation. It also occurred to us that this will enable us to demonstrate the stores in operation to our visitors while we continue work on other parts of the computer.
The store exerciser will be a standalone unit, built on our portable test rack which already has a HT power supply fitted which will provide the Dekatron anode supply. A detachable harness will connect the store exerciser to racks 4 + 5 while it is in use. The basic operation of the store exerciser will be to choose send or receive lines, select a Dekatron within a store within a store unit. This will be achieved by means of 3 uni-selectors for the store unit, store and Dekatron selections respectively. A series of actions and tests to that Dekatron will be applied under the local control of a PIC processor and associated circuits which will generate authentic guide pulses and sense whether the cathode voltage goes positive. A possible test sequence may be to send 10 pulse pairs to read a Dekatron's current state, then apply a pulse pair to advance one position, then send a further 10 pulse pairs to read the Dekatron's new state. The store exerciser will be controlled by a laptop computer (also mounted on the rack) by means of a USB-to-parallel interface card and associated circuits to drive the uni-selectors and return data from the PIC. An option will be provided to perform marginal tests by reducing the guide pulse amplitude by a pre-set amount. Test results will be displayed and logged for further analysis.
Eddie has completed adjustment of all relays on the racks, and is now servicing the tape reader feed mechanisms. An authentic stepper circuit has been assembled on a piece of plywood (see picture) which charges a capacitor through a resistor. Operating the switch dumps the charge from the capacitor into the solenoid of the tape reader, causing the pawl and ratchet wheel to advance the tape one character. Some pawls were found to be hanging downwards rather than engaged with the ratchet wheel - this was easily adjusted, however one reader has a missing spring and another a broken spring, for which we will need to find or fabricate replacements.
Johan has been fitting a spur to the ring main in the project area, with additional sockets fixed to the floor underneath the computer. We also have an isolator and distribution panel fitted at the bottom rear of the PSU rack, and red warning lamps visible from the front and rear of the computer which are lit whenever the isolator is live. Johan plans to fit a contactor and emergency power cut-out to the spur in the near future.
Ros Mennie (daughter of Dick Barnes, one of the original designers of the computer) is joining the project team and has started examining the software we had recovered from paper tapes that were stored with the machine (and has already found a bug in my 5-hole code to character translation routine!)
See Image of Eddie trying out the authentic reader stepper circuit
22/03/2010 update from Tony Frazer
After a few weeks break and implementation of improved working environment and safety measures, progress continues on the restoration...
Eddie continues with the cleaning and adjusting of relays in the 9 store boxes and Johan has been working on display lighting and cabling.
All store boxes have now been replaced on rack 5, ready for testing.
Rack 4 partial rewiring is now complete and 2 broken U-points (see picture below) have been replaced (same problems as with PVC wiring on rack 5).
Eddie has completed cleaning and adjustment of relays on all store boxes and is now testing the high speed relays on the Translator Unit.
The next steps are to examine components and wiring of the remaining 4 store boxes, repopulate them with tested Dekatrons and fit them to rack 4. An exerciser will then be built to test/demonstrate all store selection and gating circuits under authentic conditions.
See image of 2 broken U-points on rack 4
02/02/2010 update from Tony Frazer
I have now replaced 4 store boxes on Rack 5. Johan straightened the internal framework of one of these and I used the same technique on the remaining store and replaced a diode which had been crushed by the bent frame with a similar one from the spares that came with the machine.
Eddie has started work on the relays in the Rack 5 stores.
I am looking at the possibility of using the more readily available Russian OG-4 Dekatrons as a direct replacement for the GC10A. OG-4's seem more similar to GC10A than GC10B (low strike voltage) although they have a different pin-out. No problem - I adapted the base of an OG-4 with epoxy and an octal base from some unknown broken valve. An advantageouas side-effect to this is that the OG-4 is now slightly longer (but still not as long as a GC10A!)
27/01/2010 update from Tony Frazer
I may have mentioned in a previous blog that there was considerable damage to the wiring on racks 4 and 5 (used for the Dekatrons stores). The send/receive guide pulse and power connections run the length of the racks to all stores and across to the extension on rack 5. On both racks, these wires span exposed areas between the bays like the strings of a harp, hence many had been broken off and others had suffered weakening near the solder joints. All such wiring on rack 5 has been replaced with single strand PVC insulated plate wire using the same insulation colour scheme. Fortunately, the laced loom running up the centre U-points appeared undamaged and has been left in place.
In addition, the send/receive guide pulse connections (a total of 36 wires) linking racks 4 and 5 were at some point hard-wired between the two racks. All these wires had been cut in order to separate racks 4 and 5. This wiring has been replaced on the rack 5 side and 2 18-way Jones connectors have been attached using existing threaded holes on rack 4 (clearly shown in the picture.)
The restored store unit has been replaced on rack 5 as shown in the picture. I am ready to test a second store which has been cleaned and populated with GC10B Dekatrons (much brighter than the original GC10A's). This has almost exhausted our supply of spares of GC10B, so we will be looking out for more of these. 3 more stores and the rack will be fully populated. We have sufficient GC10B to populate the accumulator and one option may be to populate half the stores in each of 2 store boxes with GC10B to facilitate demonstration of the machine to visitors (i.e. bright Dekatrons in both a send and receive store.)
Eddie continues to clean and adjust relays - of particular note are relays where spring sets have been deliberately kinked rather than adjusting the travel to make contact.
Johan has made a start on straightening the casing and internal frame of one of the stores (with all Dekatrons removed!). It had evidently been impacted causing crumpling of the interior metalwork and possible damage to components.
Next step - continue work on stores with the aim of fully populating rack 5.
See image of New wiring at the back of rack 5
See image of Rack 5 fitted with 2 stores - Dekatron GC10B top, GC10A bottom
07/01/2010 update from Tony Frazer
Restoration work on the first store proceeded faster than I expected.
A total of 3 diodes required replacement as the brittle casing had broken.
All Dekatrons were removed and tested. I assembled a simple tester by connecting the variac up to one of the spinners. By increasing the voltage to 130V (remember the spinner includes a voltage multiplier), the tube starts to spin. Now gradually reducing the voltage to 120V it becomes apparent if the tube is sticky as the glow keeps snagging on a particular cathode, or just refuses to spin. Such tubes (12 out of 80) were set aside for further examination and possible reconditioning as a future activity.
Some tubes which tested OK were found to have loose bases (7 out of 90) - these have been repaired with clear laquer (nail varnish) and held together with a rubber band until hard.
Two tubes were found to be dead (possibly due to wiring problems in bases which were both loose - repair may be possible.)
The full complement of 90 Dekatron tubes were then replaced in the store.
A short circuit of the 370v supply rail to ground was discovered, localised with the aid of an ESR (low ohms) meter and was due to a component lead shorting against an adjacent valveholder pin.
I connected my prototype relief power supply to the store unit (see the pictures below). The unit draws 17mA with all 90 dekatrons glowing, which is much lower than I expected.
Encouraged by the lack of problems so far, I will continue work on the store units and repair the wiring of Rack 5.
See several images of the powered dekatron store
I'm really pleased so many Dekatrons actually worked - this was unknown and could have been a major problem.
However, the glow from the original GC10A's is so feint (as seen in the images above), and the failure rate such that I will probably populate two stores with the later GC10B's which have a much brighter glow and a Dekatron expert tells me that they fare better over time. Perhaps eventually we can move over to using GC10B's if we can find a source for them.
05/01/2010 update from Tony Frazer
This week I examined the wiring and components of the stabiliser unit. The only apparent problems were a loose chassis socket and missing MES 6v lamps in the anode circuit of two EL360's, which were replaced with new lamps of similar rating.
I replaced the stabiliser unit in the power supply rack and connected it to the recently rebuilt rectifier unit. Using the variac, I increased the voltage gradually to 160V at which point it became apparent that one of the EL360's (V15 stabilising the -200v supply rail) was running hot and the anode had a dull glow. The power was disconnected and that circuit block is being examined to determine the cause.
Before Christmas I mentioned that I had prototyped a relief power supply for rack 5. When final assembly is complete, I hope to exhibit at least one store unit powered up on static display. This will be more useful than our Dekatron spinners in explaining to visitors how numbers are held in the stores.
I have made a start on this by removing all 90 Dekatron tubes from one of the store units and have tested the matrix of selenium rectifiers used for row/column selection, all but one of which is OK (one has been mechanically damaged by compression against the store box frame - fortunately, we have spares!).