The University of Iowa's DEC PDP-8Restoration Log
Part of
the UI-8 pages
|
The University of Iowa's DEC PDP-8Restoration Log
Part of
the UI-8 pages
|
This is a chronological log of the progress restoring the University of Iowa's PDP-8 computer. Entries are added at the end as work progresses. Click on any thumbnail image to see full-sized image.
Bug 67: Continuing the work from Dec 16, 2024, we found a second G209 board that was equally bad. As with the first bad board, all of the bad diodes were on the bottom (B-side) of the board. Some diodes were shorted, some were pretty good but showed some reverse leakage. We replaced all dioes that tested less than perfect.
|
| Melted ground clips |
|---|
Bug 68: Evidently, one of the clips, which was supposed to be on Pin C (ground) of the backplane socket inadvertently came into contact with pin A (+10). The current flowed through the scope probes to the other ground clip, melting both ground clips (but fortunately, doing no damage to the probes or scope).
We ordered replacements, and extras, and the parts required to build new
ground clips that will end not in alligator clips but in insulated sockets that
push down over pins on the backplane, eliminating this risk.
|
|
| Layer 3 | Layer 4 |
|---|
The mechanism parts that we must move from one punch to another are explained in the diagram on page 12 of High Speed Tape Punch (BRPE Type) Description and Principles of Operation, Teletype Corp. Section 592-802-100 Issue 4, July 1969. The exploded diagrams in the parts list are a helpful guide for assembly and disassembly, particularly High Speed Tape Punch (BRPE 6 and up) Parts, Bell System Practices Section 592-802-800TC Issue 2, June 1966, Figure 2, magnet and toggle mechanism. Unfortunately, these manuals do not suggest the order in which to remove parts to mine one punch for parts to install on another.
So, we have worked slowly inward from the front, removing parts one small layer
at a time, and placing each layer of parts in a small plastic bag along with
a tag indicating what punch it came from and the order of removal. The
photos here show the two punches, one with 3 layers removed, one with 4 removed.
|
| Wire-wrap ground clips |
|---|
Note that grounding clips are commercially available that include push-on connectors for "modern" wire wrap pins that are 0.025" square, designed for 30 AWG wire. Our old pdp-8 backplane, however, uses larger pins designed for 24 AWG wire, so we had to make our own.
Molex 1501811020 connector sockets are desined go push onto .062" round pins that are intended for use in a broad line of Molex multi-pin connectors. These pins, as luck would have it, slide snugly onto the wire-wrap pins of the PDP-8 backplane.
We used these Molex .062" sockets to replace the alligator clips on two of the scope probe ground clips, with 1/8" heat-shrink tubing to protect the outside of the socket and provide strain relief. These should be safe to push onto backplane pins even when the power is on with no risk of shorting adjacent pins.
We also made two similar sockets that clip to the scope-probe tip to allow
secure and relatively risk-free cliping of the probe to wire-wrap pins and
we made a shorting jumper. We made the latter because the memory tuning manual
told us to short some backplane pins to ground for some tests and this seems
safer than using alligator clips.
|
| Removing pawls |
|---|
|
| Linkages to move |
|---|
During extraction of the parts, we noticed that the drag link for the sprocket track on the tape is a different part than the drag links (part number 143080) for the data tracks. This is explained in note 3 on page 3 of the parts list, where it says that a different drag link (part number 146678) is used for punches that have an advanced feed hole. Our 6-bit punches have this feature, but the 8-bit punch we want does not. So, we also extracted the extra drag link shown in the photo.
While going over the parts list, we discovered that both of our punches are missing the felt wick (part number 142849) that sits on top of the drag link pivot bracket (part number 142846). The clip that is supposed to hold this felt is present.
Having extracted these parts, we began reassembling the punch from which they
were extracted, in part so that we will have some practice on this complex
job before reassembling the punch we intend to widen to 8-bits.
|
| Side view of blocks |
|---|
|
| Top and end views |
We noted some interesting differences between our 6-bit punch block and
Bob's 8-bit block. The most notable is that the cast-iron body of our
6-bit block shows the pebble texture of sand casting on the hollow of the block,
while the hollow of Bob's block has a machined surface. Evidently, the 8-bit
block was machined from a solid billet of cast iron.
|
|
|
Measuring the block was frustrating. It was very easy to make clerical errors, and locating centers of screw holes is particularly difficult. What seemed to work best was to measure to the screw head and then add half the screw head diameter. This only worked because Teletype Corp. used cheese-head screws with heads that are very concentric and very uniform in diameter.
To find the distance between two screw holes, the easiest solution was to measure the distance across both screws, including both screw heads, and then subtract the head diameter.
Note that any time you add or subtract measurements, the error in the result is the sum of the errors in the individual measurements. We found that we had to repeat measurements many times before we were confident that we had a useful value.
The drawings at the left are the result of this effort.
|
| All 8 drag links |
|---|
The first step in widening our BRPE punch to 8 bits was to remove the shaft on which the drag links pivot (part number 124847) and then replace the slightly over-length drag link for the advanced-feed hole punch on the 6-bit punch block and add the two missing drag links.
With the pivot pin removed, the springs on the long toggle arms pushed
the drag links left and right, and many of the lower short toggle arms
(part number 143079) fell out. They all fell out in the next step.
|
| Installing toggle arms |
|---|
|
| Toggle arms installed |
The problem was then to install or reinstall all 8 toggle long toggle arms
(part 143078)
plus the feed punch link (part 124279) that punches the sprocket holes
along with the felt washers (part 124244) that help space and oil these links. Because the shaft was inserted from the front, this involved working from
front to back. The first few parts were easy to install, but as work
progressed, the space became more and more congested. Needle nosed pliers
were essential, and installing the very last felt washer was frustratingly
difficult.
|
| All the links assembled |
|---|
The next step was to swing the lower toggle arm up to join with the upper toggle arm. Particularly at the back, this was difficult, as the lower arm had to be slid up between the teeth of the metal comb that aligns the linkages. We found that using jeweler's screwdrivers to prod things into alignment worked.
The comb teeth serve as spacers (part 124282) between the toggle joints, and the teeth are separated by smaller spacers (part 124300). All of these in front of the feed-punch-link were removed, and this linking the upper and lower toggle arms much simpler because the joint was exposed until the next comb tooth and spacer were slid into place.
Finally, with pieces assembled, it was time to tighten the nuts on the ends
of the posts supporting the comb. With these loose, the comb structure
sags noticably. Snugging everything tight was done in steps, making sure that
all the toggle and drag links moved freely at each step. The final photo
here shows the assembled result, although some of the felt wicks are out of
place.
|
| Expanded to 8 bits |
|---|
|
| Robbed for parts |
After reassembling the eccentric linkages that tie the punch bail to the main shaft, and after assuring that the whole mechanism moved freely, we set to work robbing two solenoid assemblies from our parts machine and mounting them on the widened 8-bit machine. Fortunately, when Teletype Corp. built our machine, they included all the necessary wires in the wiring harness, with the ends taped and neatly bundled where it was hard to find them. As built, the ends were nicely tinned, and the wires were cut to exactly the right length.
One tricky detail is the selection of the correct armatures for each solenoid. The parts list drawings show 3 variants, and all 3 were present in our punches. I turns out that the outermost punch on the 8-bit block nees the same broadly eccentric armature as the armature that controls the tape advance, part 142871.
The assembled result is shown to the left, along with the machine we robbed
for parts. You can see which solenoids we moved because the originals in the
widened machines have flat metal steel brackets while the moved solenoid
brackets are cadmium plated and have coined surfaces (a waffle texture).
We made one mistake in our reassembly. The solder terminals
on the solenoids need to be oriented down, not toward the front. We will
need to remove and reattach the solenoids to fix this.
|
| Assembled Correctly |
|---|
With all of the lubrication done, we finished reassembling the punch (save for
the 8-bit punch block we need to make).
|
| The motor running |
|---|
We will not run the mechanism under power until we have done some of the
preliminary adjustments.
| |
| Noise on MA0 |
|---|
The ringing disappeared when we put the G209 boards back in. The loading
provided by the boards appears to come quite close to providing ideal
termination for these lines. The noise was also damped when these lines
were loaded, and in any case, the noise does not approach the
switching threshold of the R-series logic.
| |
| MA11(0) on Dec. 9 | |
|---|---|
MA11(0) now |
|
Note that the two traces shown here were recorded using different horizontal
scales. The red trace in both images is taken from the same pin.
|
| Advancing tape |
|---|
Looking at how the linkages worked, we noticed that the feed blocking
pawl was not completely disengaging from the toggle extension of the feed
toggle linkage. As a result, the knee joint in the linkage always buckled
so the stroke in the feed pawl was insufficient to turn the feed ratchet.
(All of these mechanisms are described in Figure 9 of
High
Speed Tape Punch Unit (BRPE type) Description and Principles of Operation
Teletype Corp., Section 592-802-100, Issue 4, July 1969.
Moving the stop plate that limits the movement of the blocking pawls was sufficient to fix the problem. With this plate moved, pressing the feed-out lever caused the tape advance sprocket to advance one step per revolution of the main shaft.
At this point, we decided to take a risk and mounted 10 feet
of pre-punched tape in the mechanism (without a punch block) and turned
on the motor. The machine gives a constant 110 cycle-per-second buzz when
the motor is spinning, but it is well enough balanced that the chassis does not
vibrate too much. To our great satisfaction, the tape advanced at 11 inches
per second whenever we held down the feed-out lever and stopped promptly when
we let it up. The photo here shows the tape in motion, with the moving
parts and moving tape quite blurred.
Bug 64 and Bug 67: Having replaced all of the G209 boards and verified that the noise on the memory address lines appears to be acceptable on Apr. 7, it was time to test the memory. For a preliminary test, we toggled zeros into memory locations 0 to 7, and on reading, we always got zero back. Then, we toggled ones into locations 0 to 7 and got the following back; note that we repeated the experiment twice, toggling zeros into memory between attempts to store ones:
| address | trial 1 | trial 2 |
|---|---|---|
| 0 | 011100111011 | 011100111011 |
| 1 | 111110111011 | 111110111011 |
| 2 | 111010011111 | 111010011111 |
| 3 | 111010011011 | 111010011011 |
| 4 | 111110111111 | 111110111111 |
| 5 | 011110111110 | 111110111111 |
| 6 | 110100000010 | 111110010011 |
| 7 | 110010000010 | 110010000011 |
Of course, we had hoped for all ones, but we have never gotten any 1's back from memory before, so this represents tremendous progress. It looks like bit 5 never works, and bit 1 always works. The other bits are marginal, working sometimes.
Repeatedly reading the same locaiton slowly shifted the value toward 000000000000 because of the read-refresh cycle of the memory. Any bit that read as zero on any trial was written back as zero and stayed zero from then on.
Bug 66:
The G007 sense amplifier in MB27 handles bit 5 of every word. On
Nov. 20, 2024
we were unable to balance this sense amplifier. This problem may explain
why bit 5 never worked in our first marginally successful test of memory.
|
| BRPE wiring diagram |
|---|
The only difference that is possibly significant is that Figure 6-10 specifies 1N91 diodes, while our machine has 1N1693 diodes. The 1N91 is a germanium diode rated at 65V and 150mA continuous. The 1N1693 is a silicon diode rated at 200V and 0.6A continuous. The 1N1693 is a better choice in this circuit, but nowdays, the 1N4004 rated at 400V and 1A continuous would be a reasonable substitute.
Teletype's schematic is awkward to follow, and our drawing also gives the layout
of the face of the Amphenol socket (or the back of the plug that fits that
socket). This is a key detail needed to make a cable to connect here.
|
| 110 pulses per second |
|---|
|
| 250 μsec peak-to-peak |
Our measurements confirmed that our machine actually runs at 110 revolutions
per second, or about 9 ms per revolution. The pulse amplitude given in the
above-cited bulletin is for an older style of magnetic pickup.
Our pickup matches the description in paragraph 3.03
High Speed Receiver Sets (BRPE) Description,
Teletype Corp. Section 592-831-100TC, Issue 1, Sept. 1969.
(This is part of a later edition of Bulletin 215B.)
This states that the pulse amplitude at this speed should be 40 V peak to peak
with a maximum pulse width of 250 μs using the same test circuit.
Our measurements show that the 250 μs pulse width is measured
between the two peaks of the waveform, exactly as specified in the figure
just cited in the 1965 version of Bulletin 215B. The entire pulse is
closer to 1.2 ms, including the rise to the first peak and the fall from the
second. Our machine has a peak to peak amplitude just under 50V.