VCF Straight 8 repair

[djg]

You are making me rethink the idea of adding the A/D option to the Straight-8 at the RICM.

Much easier for the second person.

 

[czunit]

You could always just go with an AF01. Negibus device should work great.

 

[djg]

I sent Rick Lewis with VCF museum a procedure to try to get the machine working. After removing the two cards I added the machine was functional. He then reloaded all the demos using the teletype and they ran. Hopefully it will stay running for a while.

I'll ground unused signals to the R111 on next visit and bring something to use to demo the A/D. We are thinking about what would make a good demo.

 

[DougIngraham]

I'll ground unused signals to the R111 on next visit and bring something to use to demo the A/D. We are thinking about what would make a good demo.

When originally in service my machine had a storage scope. For DEMO days (open house) I wrote an etch a sketch program that read two pots (connected to the AF01) and displayed the points on the storage scope. With a non storage scope you could remember XY coordinates and the machine would be very busy displaying those points. A joystick would be a more intuitive entry device than a pair of pots.

You need two channels for this but it makes a good demo. I suppose if you have two channels and a plotter you could do a paper etch a sketch. Use a switch on the front panel to raise/lower the pen.

 

[djg]

Time for the next installment of As The Bits Flip...

Visited VCF again. Original plan was to help someone with their computer but they had to cancel at the last minute so worked on the Straight 8 again. I had brought stuff for it but hadn't done any prep work to make it go faster.

Got started late in the day after helping pack up the flea market. Was studying the data line interface schematic to determine what additional signals need grounding to fix the problem previously reported with the Teletype interface not working with cards needed for A/D installed. The machine had been running demos all day. Suddenly the Teletype starts printing for a little bit. Found the memory was now always reading 0 like two workshops ago. For once failed at about the best time it could.

Probed various signals until I found the read signal from R650 MC16 instead of going to ground when high was -1.2V instead. Turning off machine for a bit would make it would work for a little while then fail again. I could see the high level dropping until it failed.

1 Yellow is memory read strobe, 2 cyan is sense signal from core, 3 magenta is read signal, 4 blue is base of pullup transistor.

Probed card with DVM and nothing obvious. Put it on extender and tried freeze spray and no change. This card is odd. Normal diode transistor logic used in this machine is transistor to pull up to ground and pulldown resistor to -15V with diode clamp to -3V to limit low level. This card uses push pull stage running non saturating to drive the signal. Probing signals was inconclusive on where fault was. Tried replacing the pull up transistor. Ran long enough to give me hope but then failed again. That left two diodes and one resistor attached to the base that could be causing the problem. Quickly pulled the card after it failed and saw for a short time low forward voltage readings on one of the diodes. Should have measured that on operating failed board. Replaced it and fixed the problem. Now I know one more way the diodes can go bad. Reloaded all the demos from Teletype and tested. This finished the first day.

Second day I added ground to PE6 J&K (line(0) & line(1)) and PC5 T to prevent floating inputs from causing unwanted logic to activate. PC5 T is expander input to gate. Seemed like should be OK to ground but not normal signal input. May have been better to create PC5 card with just clamped load on P so didn't have the extra logic. All the demos still worked.

Had brought some stuff to try to measure the ADC calibration before doing the calibration procedure. Thought my Rigol signal generator could generate -5V signal but when set to DC it could only generate -2.5V. ADC full range is -10V. I had done some controlling of my test equipment but since replaced that laptop. Was able to find libraries for communication and udev rules so could access devices etc and then wrote some code to set voltage on Rigol 5074, measure it on HP3457A, and have PDP-8 send out serial port ADC reading. Yes I'm playing with Rube Goldberg setup and took bigger chunk of time than may have been worth. May have been able to get to -5V if I told it to generate a pulse with a long on time but didn't try.



10V range and 12 bit/4096 counts gives 0.0024V LSB so about 4 LSB error. Not too bad for 50 year old electronics. The 10V reference card was outputting 9.9891V. HP lines are difference between commanded voltage to Rigol and measured voltage by HP. Adc lines are difference between HP measured voltage and ADC output value. The 1 and 2 are I did the measurement twice to look for changes. Pattern of differences between two reads for ADC is suspicious that something may have not been quite right in measurement.
I had also brought a DC power supply and manually turned the knobs to sample the rest of the range. Supply had fine and coarse voltage knobs so much twiddling.

That doesn't look so good. Didn't have commanded voltage so used average of the two HP measurements. Since HP lines shows the two measurements varied significantly looks like we are having issues with output varying likely due to pot noise. Should have gone down to 0V so we can see if less than 2.5 is really less error or it was the different measurement methods. Will need to get/make a commandable voltage source that can do the full range if I want good data.

That finished my time so didn't actually get to adjusting the converter.

Occasionally the A/D conversion hung. Wouldn't repeat enough for me to see why so not quite perfect. Will poke again some other time. We will see if the rest of the machine operates properly. Seems good enough to create a demo for it.

 

[djg]

Should put the last work up before the next.
Back in March I visited again to work on the Straight 8. It was acting up again.

When I turned on the machine the Teletype chattered but didn't print anything.
Checked voltages - ok.
Got sidetracked and when got back to it a little later problem was gone.

Ran my normal set of diagnostics
Loaded maindec-801-1-pm. AC bit 2 was on at first halt. Went away while getting
out schematics to debug. Cleaned contacts in case.
Ran maindec-08-d01a-b-pb. After 5 passes halted at 4612. RAR of 6611 didn't
leave link set. While investigating read 4613 which should be 2002. MB
was 2002 but AC was 3002. Bit 2 went to zero when writing up this description.

Checked diodes on PA2 card and PAB9. Ok.

Put JMP 4602 at 4613 so it kept repeating test didn't fail. Wiggled AC cards
didn't fail.

Restarted and it ran 30 times without failure.

Started maindec-08-d04b-pb.bin. Ran 4 passes then endless loop. location
34 was 0 instead of 7430.
Reloaded and let it run 25 passes without error.

maindec-08-d05b-pb.bin, 25 passes without error
maindec-08-d07b-pb.bin, 25 passes without error
maindec-08-d1b1-pm.rim 25 passes without error
maindec-08-d1l2-pm.rim. Passed. Doesn't print so not sure how many loops

After cooling off AC2 was one. Both PAB9 AF and AP -3V. After a little
bit AP goes to GND. Tried swapping PAB8 and 9. Didn't fail. Swapped back &
didn't fail.

Next morning at 60 degrees machine worked fine.

While waiting for it to fail again ran test of ADC with full 0 to -10V range.
Had 10 mV offset and gain .018% low. Can't complain about that for its age.

This is difference between consecutive reading as the test voltage was stepped. The reddish line is the heoretical step of 2.4 mv. Peak error is about 15 LSB. Looks somewhat repetitive. There are adjustments to the DAC cards to adjust the step when each new bit turns on. Didn't take the time to see if I can improve it.


I had modified a grip strength meter to use as input to the ADC. Wrote some code to capture the maximum and maximum average over a couple seconds and print it on the Teletype.


It acted up a few times but wouldn't stay failed long enough to track down problem. Swapped the AC card for the bit that was acting funny before I left.
Machine has now failed hard so hopefully I can fix the problem.

 

[rittwage]

It doesn't sound like you will ever find ALL the problems on this one, without old problems reappearing.
Did it live a hard long life in bad storage conditions? Those pins do not look great.

 

[djg]

Did it live a hard long life in bad storage conditions? Those pins do not look great.

It was in a barn for its entire life.
https://www.pdp8online.com/shows/vcfe15/restoration.shtml
https://spectrum.ieee.org/teenage-hackers
https://www.resistors.org/index.php/Main_Page

I was hoping it would settle down but it may not. Would have to replace a lot of possibly suspect components and try to replate back plane connectors etc to get better reliability. These weren't reliable like modern computers when new. Haven't seen service log for this era machines to know new failure rate. Have log for 8/E era machine and DEC was out a couple times a year for a machine that was used at least 5 days a week for school administration. We have had some pretty long periods without failure.

I hate intermittent problems. Normally its been nice enough so I can figure out a way to get it to reproduce long enough to fix.

 

[DougIngraham]

I was hoping it would settle down but it may not. Would have to replace a lot of possibly suspect components and try to replate back plane connectors etc to get better reliability. These weren't reliable like modern computers when new. Haven't seen service log for this era machines to know new failure rate. Have log for 8/E era machine and DEC was out a couple times a year for a machine that was used at least 5 days a week for school administration. We have had some pretty long periods without failure.

My Straight 8 would only see service about once every 6 months or when something happened. It was quite reliable. Much more so than the CDC 3400 mainframe in the next room over which saw weekly service. The only problem I had direct experience with was one of the DF-32 drives was becoming increasingly unreliable. I was there when the service guy showed up. He ran some tests and flipped the platter over. The back side was worse so he installed a new platter.

The Straight 8's are really old machines now. I had good access from 1975 through 1978 and then I bought it in the early 1980's. My boss bought the CDC 3400 and I spent too much time putting the 3400 back together. Someone stole the memory power supply (cut the cables) and pulled wires out of the backplane. I was able to repair the backplane enough to punch instructions into the IR and make it execute but we ended up scrapping it out as the cost in time and materials exceeded the perceived value. I would make a different call today.

Keep plugging away at it. Mine got flakey until I reinforced the -15V power supply with a modern switcher. There was excessive ripple and no cause ever found. I expect that the filter caps are not what they once were and these are not actively regulated supplies. I watched the rail and turned up the switcher until the ripple went away. Most of the load is still on the original PS.

 

[pbirkel]

Keep plugging away at it. Mine got flakey until I reinforced the -15V power supply with a modern switcher. There was excessive ripple and no cause ever found. I expect that the filter caps are not what they once were and these are not actively regulated supplies. I watched the rail and turned up the switcher until the ripple went away. Most of the load is still on the original PS.

You connected the SMPS directly across the original PS rail (fingers crossed), or perhaps isolated it with a diode first? Tell us more about this approach :-}.

 

[Radix]

I take it you caught this one as well:


Robin

 

[DougIngraham]

You connected the SMPS directly across the original PS rail (fingers crossed), or perhaps isolated it with a diode first? Tell us more about this approach :-}.

I ran two twelve gauge wires for the -15 and two for the GND connection. If I remember correctly, I also ran a separate pair of 24 gauge for the remote voltage sense of the SMPS power supply. The AC input is paralleled across the input of the mains transformer so it turns on with the normal front panel switch. It is bolted to the inside of the door just below the mercury switch assembly. The -15 is the big supply on a Straight 8. With this addition I was able to eliminate the ripple and get at least some power factor correction. I did consider completely replacing both -15 and +10 supplies with SMPS for the advantages that would give but chose to keep it as original as possible.

 

[djg]

Keep plugging away at it. Mine got flakey until I reinforced the -15V power supply with a modern switcher. There was excessive ripple and no cause ever found.

Do you remember how much ripple you had? Looked at the pages for mine and it had 190mV ripple peak to peak.

 

[djg]

I take it you caught this one as well:

Not sure. Will check end of the month when I'm next there.

 

[DougIngraham]

Do you remember how much ripple you had? Looked at the pages for mine and it had 190mV ripple peak to peak.

Sorry David, I don't remember. There is an argument for writing this stuff down on a piece of paper kept with the computer (a physical maintenance logbook). Instead my notes are on a machine that is not easy to bring up right now. Seemed like a good idea at the time. And I do have a backup on a 400gb tape but the drive for those tapes is kaput.

190mv does not seem like much. That is 1.3% right? And the marginal tests would never find anything even bringing down the -15 to -12. It seems unlikely that 190mv would be a problem.

The ripple was enough that it was triggering the power fail interrupt when that diagnostic was run. I suppose I could have used that as a 120hz Real Time Clock if it hadn't also affected other stuff.

 

[pbirkel]

I ran two twelve gauge wires for the -15 and two for the GND connection. If I remember correctly, I also ran a separate pair of 24 gauge for the remote voltage sense of the SMPS power supply. The AC input is paralleled across the input of the mains transformer so it turns on with the normal front panel switch. It is bolted to the inside of the door just below the mercury switch assembly. The -15 is the big supply on a Straight 8. With this addition I was able to eliminate the ripple and get at least some power factor correction. I did consider completely replacing both -15 and +10 supplies with SMPS for the advantages that would give but chose to keep it as original as possible.

I now understand you to be describing a split-load situation rather than a shared-load. I was originally thinking that you had a shared-load situation with no changes to the rail distribution downstream of the original PS, hence the question about connecting two PS outputs to the same power rail.

With a split load I'm guessing that the side of the split fed by the original PS still have a reduced ripple because the reduced load leads to less "droop" on the output capacitors each half-cycle (assuming full-wave bridge rectification), correct?

In splitting the power distribution did you specifically put the memory subsystem on the original PS, or instead on the SMPS, or adopt some other basis for splitting up the load on that rail?

 

[djg]

The ripple was enough that it was triggering the power fail interrupt when that diagnostic was run.

That's what I wanted to know. You had RIPPLE which if I saw would say that's not good vs my ripple which I said looks fine. Did find in an 8/I the ESR was high on the bus bar before I took off the screw terminal caps and did some cleaning of the terminals. I assume the same problem that aluminum wiring had when they tried it for general use.

 

[DougIngraham]

I now understand you to be describing a split-load situation rather than a shared-load. I was originally thinking that you had a shared-load situation with no changes to the rail distribution downstream of the original PS, hence the question about connecting two PS outputs to the same power rail.

No, shared load. SMPS is connected to the unused recessed spade lugs which are basically across the -15V filter caps. The SMPS has active feedback and the original unregulated linear supply is nothing more than a half wave bridge and some big caps. No regulation unless you consider the ferro resonant cap regulation. The SMPS fills in. If it was two regulated supplies there could be a problem. To the SMPS the 120 HZ ripple just looks like low frequency variations in load. The SMPS I used is a 15V 40A supply (600 watts). I just looked at it and I misremembered, I didn't bother to hook up the remote sense wires as it was not necessary.

One thing I did try was to replace the big stud diodes thinking they might be the problem. The original parts were still available. That made no difference at all. Then I tried replacing those with Schottky equivalents and this did help just enough that the power fail circuit liked it better, but still not enough to pass the diag. I left those in the machine because they produce less heat. That is my story and I am sticking with it. No, it was not because it wasn't any fun to change out and twice was enough. The caps are probably part of the problem. I wish I had tested them and will probably do that this fall when the basement cools off and the additional heat the machine generates will be welcome again.

You can PM me if you want to discuss this some more. I feel a bit guilty hijacking David's thread.

 

[pbirkel]

You can PM me if you want to discuss this some more. I feel a bit guilty hijacking David's thread.

Thank you Doug for the education :-}. Got it now; this was a novel, to me, approach. End hijack!

 

[djg]

I take it you caught this one as well:

Looks like I didn't. I think that was the one I put the wirewrap to. Added another wrap to the next pin which was fine. Thanks for pointing it out.