Project Time! #1

I’m going to start this with my latest toy – HP 6920B Meter Calibrator. I got it because I need a 1kV DC voltage source. What for? Well I actually need to calibrate a meter but that’s something I will explain at a later time. Suffice to say this device is not enough on it’s own.

A good few years ago I tried building my own 2kV DC PSU for an oscilloscope CRT. That didn’t go very well though I’ve learned a lot. Biggest issue was backfeed to the control circuit due to parasitic gate capacitance of the switching MOSFET – I was driving a lot of current at 12V to get 1kV which was then doubled. These days I’d probably try less step-up at the transformer level and more stages of the multiplier, possibly also NPN instead of N-MOSFET or maybe a driver chip in between PWM controller and the power transistor. Point here is, making these voltages is not easy (or safe) so I decided to just look for something to buy.

There were a few candidates but then I found the 6920B on ebay for about 200EUR shipped. It can do much more than just output 1kV DC – and it’s a bit of a restoration project too so I was happy to pay that price. And it works, although it was pretty filthy inside – but being manufactured in 1976 and having vent holes on all sides will do that. Here’s a photo from the inside before I started cleaning it:

A not-so-short list of what was wrong with it:

  • pretty dirty both inside and outside
  • all the protective loom tubes are sticky
  • the main 10-turn potentiometer gets stuck
  • pilot light (neon bulb) lost it front panel cover
  • output indicator light bulb burned out
  • output switch contacts oxidated
  • selector switches contacts dirty or oxidated

I really didn’t want to do that but in the end I just showered the guts with hot water. Took several days (and that’s outside in the sun) to dry, mostly because the water that got into the wire loom tubes couldn’t evaporate easily. I did desolder and remove the potentiometer before applying water. It’s usually a poor idea to let the transformers get soaked in water but a) there was no other choice unless I desoldered each an every wire to both PCBs and b) the cores seem to be coated with some protective lacquer. The windings will dry out eventually – you just need to be patient.

The wire loom tubes being all sticky (and dirty) – that’s new to me. Could be cigarette smoke residue, though usually the PCBs would be affected too. It could also be this particular polymer is slowly breaking down over time and releasing this goo on it’s own. Water didn’t do much to it, I had to just go over each piece of tubing I could reach with a cloth and rub it clean. Again I really didn’t want to desolder everything after I noticed (removing the 10-turn pot) the wire strands are prone to breaking right next to solder points and there isn’t much slack to just strip more of the wire.

Potentiometer is pretty much sealed but there is a screw that can be removed and the hole is just big enough to inject some fluids into it. So in went 100% IPA and some 3M contact cleaner (it’s a wire pot), that got it moving and improved the wiper contact. But after a few days of drying and playing with it by hand I noticed some signs of the wiper getting stuck again so I injected a few drops of pot cleaner/lubricant. That’s some good stuff and it wors nicely now, no more issues.

Nothing can be done about the pilot light right now, but at least it works. I will have to come up with some replacement cover – the whole light assembly is one part and with the front broken off (that’s not typical) the neon bulb is exposed and not held by anything in place.

Output indicator is also a part that is supposed to be replaced as a whole, and trying to get it out of the front panel and apart left me with several plastic pieces that were meant to be one. The good news is, after some creative use of glue and drill bits, I now have the orange cover firmly back in the front panel and the internal bits await a new bulb that should just slide into the front cover once I solder the wires. The locking tabs are gone now but the wires are rigid enough to prevent the bulb from sliding out. Finding a small enough 2.7V bulb is a problem but I got a reasonably good replacement candidate already and a few more options on their way to me – I just need to pick the best one.

Rotary selectors were treated with 3M cleaner, work perfectly now. Then I’ve made the mistake of spraying some of that into the output switch and got a nasty brown mess of old, dirty grease oozing out – it had to be desoldered, opened up and properly cleaned. It can be done and gives a nice access to the contact points – those are silver and were already completly black. I washed the whole thing in 100% IPA and used silicone grease when putting it back together. Seems to work just fine.

So here’s how it looks now:

Does it work? Sure does, I got it re-calibrated following the manual and the DC output is pretty much perfect. So was AC for a time, but after about 6 hours of use something went bad – and now AC is too high. So much in fact that it cannot be compensated by recalibration. Many hours of looking at the schematic and head-scratching followed. This part is bit more technical and you might want to download the manuals for full schematics and some brief explanations of how this device works (it was designed around 1965, back then manuals came with schematics and even theory of operation). The block diagram:

After poking around a bit I concluded my issue is with AC reference, which actually is explained a bit in the manual, here’s a simplified version of it:

It regulates the AC voltage coming from transformer down to 1Vrms using a resistor divider, and one part of that divider is a photo-resistor coupled to a light bulb. That bulb is driven by an amplifier that senses the 1V value via a feedback. There’s actually two parts to that amp – first an AC stage that converts the AC do DC and then the second one compares the resulting DC with a known voltage from DC reference circuit, then drives the photo-resistor. Well that’s all nice on paper but the actual implementation is a tad bit more complicated and here’s the actual schematic of that part. I put come colors on it to make it easier to explain though I’m not sure if it helps or not. Well, lets get to it.

First I checked the ripple on the light blue line – it’s not insignificant but the whole device is a bit of a mess when it comes to ground reference – called -S on the schematic. And BTW my eyes are not that good anymore, at first I thought it says -5 and this confused me a lot. You’ll notice the AC REFERENCE CKT. is referenced to ground but powered by 3 different sources:

  • +12.6V from DC REFERENCE for the low power stages
  • +6.2V as the reference point through R69 calibration pot
  • some unspecified voltage from CR24/CR25 diodes (about 50V actually)

It’s that CR24/CR25 line that has ripple, but that PSU branch is not tied to ground. Instead it’s connected in various ways to other parts of the device. Ripple is most likely caused by the CR3 thyristor that discharges C17 – indeed it’s C17 that has highest ripple on it, and it’s sawtooth-shaped. So while all this is kinda suspicious, I think it’s actually meant to be like that.

Well then, there are 2 stages here, just 6 transistors, how difficult can that be to analyze? Voltage from sensing point is routed to base of Q19 which provides main amplification. It’s 100k collector resistor is too much impedance so there is Q20 which is just a follower. C19 is just to prevent high frequency noise from being passed through and it’s not shorted. Might be open but as far as I can tell isn’t, and that would not cause the problems I have. Then there is Q21 which drives the CR30/CR31 rectifier via C20.
Q21 gets pretty hot due to high collector voltage (about 30V) but the way it shares it’s emitter current with Q19 should offset any temperature-related changes in operating point. There is something interesting here, the diodes are not in parallel but there is the R66 resistor in there too. Which means the rectifier is not balanced, it has easier time pulling the output lower (via CR31) than pushing it higher (via CR30 and R66). Output from this stage is at R67.

The second stage is a differential amplifier made out of Q22 and Q23, driving the bulb-regulating Q24. Lightbulb itself is powered by 15V regulated by VR7 – and it’s also this regulated voltage that provides current to CR34. Voltage drop of CR34, about 800mV, is feeding the collector of Q22. Any temperature rise in this diode will lower it’s forward voltage, which in turn is going to limit the collector voltage of Q22 and in turn help compensate it’s temperature-related Vbe drop. Then there is the negative feedback path of C21 and R74 – this serves 2 purposes, to reduce amplification of any AC still present on the input and provide a bit of a low-pass filter to slow down any rapid changes in the output. That being said some AC does make it through to the collector of Q24 but the lightbulb has a decent “inertia” of its own not to care too much.
One thing I’ve noticed about this design is any wires connected to base of Q22, even through C21 and R74, will act as antenna and send this amp into multi-megahertz oscillations. Perhaps it doesn’t do that on it’s own but I will mod this part by adding about 100nF capacitor between Q22 base and ground to prevent that. Or maybe its collector instead of ground if that works well – would be easier.

When you look at the Q22/Q23 pair you realize the base of Q23 is held at 0V via R72. Then it stands to reason that the pair would be in balance if base of Q22 is also at 0V. And sure enough, it pretty much is. This also explains why there is an imbalance in the first stage rectifier – notice the DC reference is +6.2V via a 10k resistor. And first stage DC comes through 2k resistor. Then, to get 0V at the tie point the first stage must produce mostly negative voltage, at -6.2/10*2=-1,24V. Well, mostly, there is a helper feedback path here via CR32/CR33 that provides some negative voltage from the AC ouput of the reference.

And this is how this block keeps the balance. Except it doesn’t, but it still regulates, just imperfectly. So why? Well the “0V” at base of Q22 is actually -17mV or so (depends on temperature). If you calculate just how much (or should I say, how little) voltage regulation the R69 calibration pot provides, it’s pretty much at the limit already. Q22 is already biased to negative voltage on the base and yet the output of the whole circuit is too high. Shorting one of the CR32/CR33 diodes, to provide more negative drive from the output, helps but not a lot. Adding external negative DC bias does cancel out the error. But then it looks like this:

So what is wrong here? I’m not very good with analog stuff but after ruling out DC leakage via C21 I’m inclined to belive one of the Q22/Q23 transistors is faulty. Well, not precisely that, both still work but at least one of them has a widely different amplification than the other and the pair is now badly mismatched. This causes rather significant input offset which is outside the specs of the calibration. So I need to match two NPN transistors and replace these ones. Unless someone has a better idea? To sum this up:

  • it worked for few hours before AC went too high (about 3% high now)
  • AC is regulated but outside spec, R69 doesn’t have the range to bring it back
  • DC part of the instrument is still spot-on
  • seems temperature dependent (much more than it was when output was correct)

INB4 “replace all the caps” – those are still good and will stay until they fail.

UPDATE: Replacing Q22/Q23 pair with modern NPN parts matched to 300uV didn’t help, it actually made things worse! Also, I tested the original transistors (once removed from PCB) and those are matched to less than 2mV – so good enough. The fault must be in the AC stage then, but how could that be?

Then I realized I didn’t check C20 for DC leakage, just for capacity and ESR. And sure enough that bugger has well over 60uA of leakage at rated voltage, and never fully reforms. To make sure I wired a temporary 10uF/50V replacement and the AC reference is properly regulating 1V again. Turns out that current is excessive enough to cause voltage drop on R65 that throws DC stage off balance.

So, one cap has failed after all – but only in a very specific and limited way 🙂

Computer Week-end, contd.

Phoebes are a bit delayed but I will have some in 2-3 weeks. In the meantime I will open DocBrown orders this Saturday, as usual at noon CEST.

As a reminder, I still can’t ship to New Zealand, Malaysia and pretty much most of Africa and Middle East, except Israel. And since some countries might enter another lockdown due to 2nd COVID wave, this could change at any time.

I’ve been busy with some neglected personal projects that have nothing to do with ODEs or even computers. I don’t think those fit on this blog but there is no other content right now, and dire lack of photos lately, so I could maybe show some of those. I know some people who would be interested but at the same time I don’t want to confuse the general public. What do you think?