HP 6289A power supply not outputting full current

[iroc86]

I ran a few new tests. I removed the panel meter from the circuit and applied an external voltage to see if the drift was coming from the meter movement. It wasn't. The needle remained dead-even for the duration of my test (about 30 min, plenty long enough to see the drift on the power supply meter circuit).

I then measured the voltages of TP46 and TP47, referenced to +S, on a cold unit over the course of 60 minutes. The panel meter is connected across these test points, so the difference between TP46 and TP47 is essentially the same voltage that the meter sees (through some dropping resistors, of course). To eliminate any interference from the mechanical meter movement, I installed a 100 ohm resistor in place of the meter. I also measured the meter voltage across the 100 ohm resistor for comparison to the test points.

The data is plotted below. The ~1 mV rise in meter voltage at startup is the same as before, but note the unusual drops around 24 minutes and 49 minutes. This behavior correlates directly with two little humps on the TP46 graph. The power supply sat on my bench, undisturbed and with no load, for the duration of the test.

I then swapped out Q11 (the TO-39 dual BJT) and replaced it with two 2N2222s, just to see if there'd be any difference in response. The absolute numbers have changed a bit, but the weird fluctuations remain... and only after the unit has been powered on for upwards of an hour. The power supply output voltage (30 V in my tests) remained rock solid the whole time, so I'm pretty certain this is drift from the meter circuit only.

Any ideas on what might be causing this? Could it be from a passive component (e.g., a bad resistor), or would this be more likely a semiconductor issue? The only original transistor left is Q15, which functions as a current source according to the service manual. It's a 2N2907.

[xavier60]

What is Q15 doing? Is it operating as a constant current source?
I can't figure out how the whole circuit can work properly with constant current fed to the Emitters of the output transistors.
Is VR6 fitted as shown?
What voltages are across VR6 and R62?
What is the Colector voltage of Q15?

[iroc86]

Per the manual excerpt below, Section 4-38 states that Q15 is a current source to set the bias current for the amplifier. I'm not sure how it works.

Yep, VR6 is there. Earlier revisions of this supply had a slightly different arrangement with resistor R40 in its place. Interestingly, the schematic has the polarity of the diode incorrect. My PCB matches the proper orientation, though--cathode to +12.4 V. It's supposed to be a 4.22 V Zener.

The voltage across VR6 is 4.00 V. Across R62, it's 3.30 V. The collector of Q15 to ground (+S) is about 8.65 V. Vce measures 0.024 V.

[xavier60]

With Q15 saturated, it isn't operating as a current source which I don't think it should be for now. Check for changes in operating state of Q15.

[iroc86]

With Q15 saturated, it isn't operating as a current source which I don't think it should be for now. Check for changes in operating state of Q15.

Maybe not... but an LTspice simulation seems to show that Q15 is saturated with Vce right around where I measured at ~20 mV. Hmm.

[duak]

Here's some thoughts:

As per section 4-39 in the clip from the manual above, the purpose of the circuit is to prevent damage to the meter movement during overload by limiting the current to it.  Q15 is a current source that can provide a certain maximum current of about 5.5 mA.  Assuming D1, R62 & Q15 are good, then when Q15 is saturated, it's an indication that it is not limiting current.  Is the meter pointer significantly above maximum?  If not, what happens to Q15's Vce when the meter reads above maximum?

When I look at the 'meter' curve in the above graph I think that a fan is coming on at 30 minute intervals.  Could the HVAC be starting and stirring the air up?  Less likely is that the HVAC or some other repetitive load such as a pump is affecting line voltage.

When I think about it, Q12 & Q14 are within a closed feedback loop so their temperature differences should be insignificant.  OTOH, the VBEs of the transistors in the first pair are terrifically important and should track for minimal drift.   My copy of the manual shows the first pair is in separate packages and are designated Q11 & Q13 whereas IROC's schematic shows a dual part Q11 and the above picture shows they are in a common package.  hp probably made a production change to improve the circuit by going to a dual part in one package.

I wonder if the transistors in Q11 are actually matched.  What is the part number on the package and in the manual?

[iroc86]

Interesting thoughts, duak. (FYI, I forgot to label D1 correctly in my LTspice schematic; it's representing VR6 with a ~4.7 V Zener, which is probably close enough to the real 4.2 V part for simulation purposes.)

No, the meter pointer is in the middle of its range. I'm doing all my testing at 30 V. The meter graduations go up to 50 V. When S2 is in the lower range, it's divide-by-ten for 0-5 V.

I think I see how Q15 is supposed to work based on your description of the current limiter. But what does the manual mean about Q15 setting the bias current for the amplifier per section 4-38? If you disconnect the Q15 collector in the LTspice simulation, the meter voltage drops to essentially zero, so Q15 is still doing something even when in saturation.

I measured Vce of Q15 and selected the 5 V range while driving the output voltage higher. At about 10 V, Vce starts to increase, but even when the supply is set to 50 V (ten times the meter range), it only reaches about 70 mV... so it's still basically in saturation. Now, if I try the same thing on the lowest amperage range (180 mA), Q15 will come out of saturation when the output current reaches 0.5 A or so. Is this behavior inherent to the difference between sampling the output voltage vs. the current sense resistor voltage?

Regarding the dips on the meter curve, I'll pay attention to the airflow in my room. We're using heat now, so there wouldn't be a heavy electrical load from AC, but temperature could play a difference. I can try putting a thermocouple in there and running the test again. Even so, I'd be surprised if it's the furnace, as my lab bench is about 15 feet away from the nearest vent and I had the lid installed on the power supply case. This thing would be super sensitive if a 1-2 degC ambient temperature change caused the panel meter to deflect so significantly.

That being said: for Q12 and Q14, component temperature really does seem to make a difference. I can squeeze one of the BJTs with my finger and the meter input will move by about 200 µV. If I squeeze the other one, it'll shift by the same amount in the other direction. This isn't really enough to see on the panel meter, but when I hit the transistors with freeze spray, the meter input will move by 2-3 mV, which is upwards of 2 V on the meter on the 50 V range... quite visible.

I took a closer look at the 6289A errata (attached) and HP does actually call out replacing Q11 with a dual part in later revisions. Under Change 4, they're saying it's a 2N4045, which appears to be a matched pair with "tight Vbe tracking" per most of the data sheets I've found. That's probably what's in my unit, or at least something close.

While playing around with a few things as I was writing my reply above, I noticed that I can get the get the panel meter to move significantly by touching the components in the meter amplifier circuit. I know the human body is going to have some conductivity, but I didn't expect to visibly see such a change. This leads back to cbutlera's post in Reply #46 about PCB leakage. Might this be what we're seeing with the meter amplifier circuit and any adjacent components?

[duak]

I goofed in my calculation of the current in my previous reply; it should be 5.5 mA.  (I corrected the reply).  A more common name for the circuit Q15 is a part of is a current source.  A current source can always provide less current than it's designed for - it's just never supposed to provide more, hence it also limits current.  An analogy is a valve that's fully open ie., saturated - it can't do anything more to get what's downstream to accept more flow.

It's not incorrect to say Q15 provides a bias current.  IMHO it would be more clear to say that Q15 provides the emitter current for the output stage of a differential amplifier.  This is the lion's share of the current provided to the circuit; R64 and R65 provide a smaller amount.  Most flows to -6.2V through R71 & R70 and the rest through R74.

Q15 could be bad in that its gain has fallen off or it developed big leakage but it's really unlikely.

Curiouser & curiouser...

I wouldn't expect operation to change between Voltage and Current modes, unless there was something wrong.

When touching the parts with a finger, are you making electrical contact with any of the actual circuitry?  If not, and the circuit responds when touched and then recovers instantly when not, the circuit could be oscillating.  Maybe look at some of the signals with a scope to see if you get something weird to happen.

It could be leakage from something or other on the PCB but the picture doesn't indicate any contamination.  OTOH, there could be a bad part somewhere - capacitors get leaky and so can transistors.  The meter circuit doesn't have any capacitors though.

I can't remember if you've looked at the 12.4, 6.2 and -6.2 V right on the parts in this circuit while the event is happening.

I have an hp 6284A.  If the meter circuit is similar enough and if I can scare up some space & time I'll see how it responds.

[xavier60]

If Q15 is supposed to operate in an unsaturated state, it would be odd that the designers would have added this extra source of possible variance to the meter circuit.
Try adding an extra 0.6V of diode drop to the zener.

[duak]

I pulled the hp 6284A out of retirement and managed to clear a space for it on the bench.  It was made in '79, has a dual transistor for the meter amp and has a zener diode VR6 instead of a resistor.

Using a Fluke 8062A DMM, I measure 3.804 V across VR6 with the meter reading anywhere between zero and full output voltage.  I see that Q15 is saturated with a Vce of about 50 mV.  I expect that the voltage across VR6 is reduced by Q15 drawing more base current because it is in forced Beta mode, ie., the collector current is limited by the rest of the meter circuit so the voltage across R62 is reduced and the base current must rise to maintain equilibrium.  If I select low voltage range and pin the meter, Q15 comes out of saturation with a Vce of about 500 mV and I measure 3.822 V across VR6.   At this time Q15 draws less base current because it is working as a transistor and its collector current is contributing to the voltage drop across R62.  Repeating what I said above I think Q15 is part of the circuit to protect the meter and it works by limiting the current for out of range values.

Assuming VR6 is a 4.22 V zener as per the update instructions, this tells me that VR6 might be running at a low current.  BTW, hp's original design used a resistor.  I suspect the resistor was replaced by a zener not necessarily for stability but rather if the +12.4 V supply failed with an overly high voltage, the meter movement would be damaged because the meter current limit would also be increased.

I don't have a logging meter so I have not tried to replicate IROC's test of meter circuit stability over time. I did not find that touching Q12 or Q14 cases made a difference in the meter reading or in the voltage between TP45 & TP47.

[xavier60]

I pulled the hp 6284A out of retirement and managed to clear a space for it on the bench.  It was made in '79 and has a dual transistor for the meter amp and has a zener diode VR6 instead of a resistor.

Using a Fluke 8062A DMM, I measure 3.804 V across VR6 with the meter reading anywhere between zero and full output voltage.  Q15 is saturated with a Vceo of about 50 mV.  I expect then that the voltage across VR6 is dragged down by Q15 drawing more base current because it is in forced Beta mode, ie., the collector current is limited by the rest of the meter circuit so the voltage across R62 is reduced and the base current must rise to maintain equilibrium.  If I select low voltage range and pin the meter, Q15 comes out of saturation with a Vce of about 500 mV and I measure 3.822 V across VR6.   At this time Q15 draws less base current because it is working as a transistor and its collector current is contributing to the voltage drop across R62.

Assuming VR6 is a 4.22 V zener as per the update instructions, this tells me that VR6 might be running at a low current.  OTOH, hp's original design used a resistor.  I suspect the resistor was replaced by a zener not necessarily for stability but rather if the +12.4 V supply failed with an overly high voltage, the meter movement would be damaged because the meter current limit would also be increased.

I have not tried to replicate IROC's test of meter circuit stability over time.  I did not find that touching Q12 or Q14 made a difference in the meter reading or in the voltage between TP45 & TP47.

Is it possible for you to decrease the temperature of Q15 to see if it drops out of saturation?

[xavier60]

Anyway, I'm hoping that iroc86's  meter drift problem is being caused by Q15 drifting in and out of saturation because the it's operating too close to the edge.
Another way of covering up this possible cause is by decreasing R62 by a small amount.

[duak]

xavier, I don't have any cool spray to try on Q15.  I had some dry ice packs here the other day but they're long gone now.

I also edited my previous reply to hopefully clarify it but it was about the same time you submitted your reply.  As I said, I think it's to protect the meter.

One way to increase the current limit is to put a resistor in parallel with R62, say 1K0 and see what happens to the drift.

[iroc86]

Thanks for testing out your 6284A, duak. I'm really glad we have something to compare to. I have a 6236B triple output that I thought was similar, but the meter circuit is much simpler... no amplifiers, just a resistor network. I'm not sure how/if HP is limiting the meter current on that model (FYI, it has zero drift from what I can tell!).

Out of curiosity, did you let your supply warm up before taking your measurements, or were you going from a cold unit?

That's interesting about Q12 and Q14 not responding to human touch. I'd also like to extend xavier's request about Q15's temperature--I'd be curious if either Q12 or Q14 would pull one way or the other when doused with freeze spray (on my unit, if I spray them at the same time, the voltage across the panel meter more or less stays the same). I saw that you don't have any spray, but would you be willing to try heat gun?

It looks like you're getting Q15 to drop out of saturation when switched to voltage display, unlike my unit. Can you also try it on the current range?

To your earlier questions about the voltage rails, I did do some cursory measurements on the 12.4, 6.2, and -6.2 V rails. They're all rock solid throughout the entire warmup cycle. The 12.4 V rail is a little low at 12.0 V, but it's still within HP's +/- 1.0 V tolerance.

I took a few more measurements over the weekend and will try to post up some new graphs tomorrow. FYI, I'm not doing this with a logging meter, either... just a pair of Fluke 27s, my HP 3466A, and a stopwatch. I'll put on one of Dave's videos and just burn through the data points while I watch on the side.

More to come. Thanks, guys!

[xavier60]

The input selection possibly alters the amplifier's input Common Mode voltage which will also alter the current draw through Q15.

[duak]

I suppose I was a professional logger back when I was paid to write down numbers.

The final numbers were after a 30 minute warm up.  They were a few mV lower when it was cold.

When the meter reads full scale on either voltage or current, the voltage between TP46 & TP47 is almost exactly 1.000 V.  Because hp used wirewound pots even the fine adjustment is somewhat granular and is obvious with a DMM and damn near impossible to set to an exact XX0.0 V.  By heating Q12's case with a soldering iron I can reduce the above voltage by about 10 mV and by heating Q14's case I can increase the above voltage by about 10 mV.  It's hard to say how much their temperatures changed but I'd guess it's more than 10 deg C.

Pinning the meter in both voltage and current modes cause Q15 to come out of saturation.  For all in-scale readings Q15 is saturated.

The diff amp has a gain of X10.  With an output voltage of 1 V for full scale, the input voltage is 100 mV.   R74 is the common emitter resistor that long-tails the input stage of the diff-amp and It'll have about 5.5 V across it.  Since both the current sense resistors and the output voltage connect to +R, the + output terminal, there isn't really a common-mode voltage to consider.  Anyway, the input to the diff amp will cause a variation of +/-100 mV across R74 which is probably not significant vis-a-vis Q15.  However, if something is on the edge this variation may be significant.  Note that the sensed voltages into the diff-amp are of opposite polarities for the voltage and current modes so there is a distinction between them.  hp goes over this in the manual.

I'm going throw out something here - the current calibration adjustment R56 is wirewound.  This is where the resistance element is a thin wire wrapped around an insulator in the shape of an arc.  A moveable contact on the wiper touches one or more turns of the wire and makes a variable resistance.  I mentioned above that the voltage and current pots are wirewound and that the adjustments are granular.  Suppose R56 is just on the edge of contacting the next turn on the resistance wire.  Slight temperature changes and/or extraterrestrial ethereal forces may be just enough to tip it over and make the meter jump.  Maybe a cleaning and slight back and forth working of the wiper will make a difference.

[iroc86]

I too am seeing about ~1.000 V between TP46 and TP47 when the meter is at full scale. The linearity seems to change, too. I don't have any data to back up that assertion, but it seems like the meter reading will fluctuate as I adjust the output voltage. It usually reads within a "needle's width" when going to/from specific output voltages, but occasionally it'll end up a half-graduation higher or lower. That might have something to do with the wirewound pots, but I'd expect consistency because I can always get the output voltage to return to its original value. The meter circuit is just sampling off of that, so I figure the issue is somewhere in the amplifier.

Regarding the wirewound adjustment pots, that's a very good thought about the adjustment being on the hairy edge of the contact. Early on in my troubleshooting, I removed R56 and R72, ran them through their full ranges, and cleaned the contacts with CAIG DeoxIT F5 Fader lubricant. I've also turned the pots to a different setting just to move the wiper to another area, but it didn't make any difference. I can try tossing in some modern trimmers to see if that changes anything.

Thanks for heating up Q12 and Q14 with your soldering iron. Those fluctuations are about the same as I'm seeing.

Here's a new set of charts. This time I also logged the temperature in the vicinity around Q12 and Q14, just in the open air. The lid of the case was closed as best as possible with all of the test leads coming out. I let the unit warm up for 35 minutes before switching the furnace on (just to rule out that possibility). As expected, the panel meter responded slightly, but nowhere near as significant as the first set of data I posted. I shut off the furnace at 52 minutes in, and the temperature measurement stabilized. Interestingly, the meter voltage was on a rising trend the whole way, even after the temperature mostly stabilized around the 50 minute mark. The meter just started to level off at the very end, but I had to shut things down and go do something else.

After a a few hours, I tried again on a cold supply (though only over 20 minutes and no furnace cycling). The response was almost identical, including the odd blip at the ~12 minute mark. I have no idea what caused the two big dips in my data from a few days ago.

duak and xavier, I'm going to probe around some more based on your responses and will report back...

[iroc86]

I did some more testing with Q15 and meter current limiting. The maximum current through the meter is about 1.7 mA when pegged in either voltage or current modes. (FYI, when I'm referring to voltage and current "modes," I mean the setting of the panel meter switch, not the output voltage or current limiting modes of the supply itself.)

duak, can you confirm that you saw Q15 drop out of saturation when your meter was reading voltage only? If so, this is different behavior than what I observe. My supply will only limit meter current via Q15 when the meter is in current mode. This seems odd to me, since the resistor networks in front of switch S2 provide the same input voltage to Q11 regardless of mode--the only difference is the polarity, as you mentioned.

Here's what I mean: In the high range modes (50 V and 1.8 A), the maximum input voltage to Q11 is 100 mV based on the resistor networks, which corresponds to a full-scale reading on the meter. This is based on 50 V across R59-R61 for voltage mode and 1.2 V across R56-58 in current mode (i.e., a 1.2 V drop across 0.66 ohm current sense resistor R54 with a 50 V input and 27 ohm load = ~1.8 A). Note that in this scenario, R56 is adjusted to 100 ohms, which correlates to the actual pot adjustment on my unit.

In the low range modes (5 V and 0.18 A), the 100 mV input again corresponds to full-scale on the meter, except now it's at the 5 V (voltage) and 120 mV (current) input levels due to the divide-by-ten resistor network. Anything exceeding a 100 mV input to Q11 should presumably cause Q15 to start limiting the meter current. At full output voltage or current in the low range mode, the input to Q11 will be about 1 V. So, anything from 200 mV to 1 V ought to provoke some response from Q15.

I'm not sure why Q15 is dropping out of saturation in only the current mode. That's when the base of Q11B is grounded and Q11A is positive. Might this have something to do with the nature of the differential amplifier and the way the voltage references are set up? Or perhaps Q15 doesn't need to fully drop out of saturation to limit the current in voltage mode? (In voltage mode, Vce of Q15 will increase slightly, to maybe 60 mV or so, but it's vastly different than the ~500 mV in current mode.)

I know this isn't directly attacking the meter drift problem, but if this part of the circuit isn't working properly, it might be a clue.

[xavier60]

RF could explain the odd behavior like being over sensitive to being touched by hand.

[xavier60]

It is actually normal for the amplifier to draw a little more current when one input has a + input and the other is zero,
 compared to when one input has a - input while the other is zero.

[duak]

Iroc, I measure about 1.56 V between TP46 and TP47 when pinned in voltage mode and about 1.58 V when pinned in current mode.  Likewise Q15's Vce is 485 mV and 5.10 V respectively.  Xavier points out that because the two modes are slightly different the common mode voltage into the diff amp  will affect the current drawn from Q15.

I'll bet if you reduce the voltage across VR6 to about half of what it is now, it should limit maximum meter readings to the upper third or so of its range.  You should also definitely see Q15 come out of saturation.

These things were introduced in 1967 when it was cheaper to use discrete parts than an op-amp.  I'd bet the original circuit didn't have Q15 but during a design review or testing an internal supply failure pointed out how the meter could be damaged.  VR6 replaced the resistor as a running change after some field failures surfaced.  These days, I'd use an op-amp and a band-gap reference diode and be able to limit the meter current to 125% of full scale +/- a couple of percent.

I've encountered one discrete circuit where one of the transisors developed a strange leakage current that manifested as an intermittant drift and replacement solved the problem.

This next part is clutching at straws:  Since this is a precision circuit it might benefit from using transistors that weren't optimized for switching.  I used 2N5087 & 2N5089 low noise transistors years ago when I designed some low noise stuff but there are others that may be more available.  I believe the ubiquitous 2N2222, 2N2907, 2N3904 & 2N3906 were all gold doped to reduce switching times.  This makes them slightly leakier and may contribute to popcorn noise.  That being said, I've used these transistors for various low level iinear circuits and they've been just fine.  I see in the manual that hp specified a Sprague 2N2907A but if there was a problem with any of the above, I'm sure they would have changed parts.

[iroc86]

All very interesting stuff, guys. xavier, thanks for pointing out the current draw difference vs. polarity on the amplifier circuit. I think I can see how that might affect Q15 downstream.

I didn't know about the doping on the jellybean switching transistors--those things are so ubiquitous that I never really considered how a switching transistor might behave differently than an amplifying transistor. It's hard to say what's actually in these units, too, with HP's internal numbering scheme. Over the course of refurbishing a variety of HP equipment, it's amazing to see how often their parts lists have conflicting cross-reference numbers. I certainly give them credit for improving the designs over the years, and it's testament that a design from fifty years ago is still useful and usable, despite the issues that crop up with age. (My 6289A appears to be from around 1986.)

Re: VR6 and its effect on Q15 saturation, I did notice some of that in one of the LTspice simulations I ran. I reverted back to the original design with a resistor and it changed where the "knee" of Vce appeared in relation to the amplifier input voltage at Q11. I think I will try adding another diode in parallel with VR6, as xavier suggested, and see how that affects the saturation point. I might have a few other Zeners I could swap in, too.

So, where are things right now? I replaced R63 and R72 with regular resistors to eliminate any potential temperature effects with the wirewound pots. From the little bit of preliminary testing I did, it doesn't seem like that's the issue. Even over just 10 minutes of warm-up, the meter drifts by about three widths of the needle. It's tough to systematically diagnose this issue because I want to start from "dead cold" every time to ensure I'm not chasing my tail. That, and the intermittent times when the needle is a half-tick low for no apparent reason at all. It just takes time...

Next, I'll probably swap out Q15 and see what that does. I don't have any spare 2N4045 TO-78 packages for Q11, so that'd be the only original semiconductor left.

At some point, I probably need to ask myself how much of this is "good enough" and within my expectations for a 30-year-old piece of hardware based on a 50-year-old design. As a side experiment, I'm thinking of spinning up a PCB for the meter circuit with modern hardware, in the same topology, to see if the drift is any better. There are a few matched pair SOT-23 transistors out there that might be interesting to experiment with.

[xavier60]

The easiest way of causing Q15 to stay in saturation for slightly higher current is to  put a resistor across R62 to reduce its value by about 10%.
Also check for possible RF interference or oscillations. Although unlikely in this case, that configuration of transistors can oscillate.

[duak]

Well, the US put men on the moon (and returned them safely to the Earth) using technology contemporaneous with these supplies.  The initial temperature drift problem is probably inherent in the design.  I have an hp 6002A and it uses op-amps in the same basic topology but it was far more difficult to get working right and it's got a noisy fan that runs all the time.

The instability you see is another thing.  There's got to be something blinky in there.  Keep puttering along with it and it'll be found.  BTW, If you bypass Q15 to 12.4 V with 470R, it may give more clues.

My supply was missing a cover over the power resistors in the back so I tried to find what it looked like.  In the process I found this link showing someone replacing the analog meter with a digital one:
 

The digital panel meters I worked with would accept either +/- 200 mV or +/-2 V so at first glance it wouldn't need a meter amplifier/limiter circuit.  The meter accuracy could be improved - just sayin'

I think I saw a posting on EEVBLOG recently where someone had replaced the LED/LCD with Nixies in an hp something or other.  Personally, I like the look of LEDs and they're from about the same time.

[iroc86]

Yeah, the random instability is really the kicker here. After you all helped me diagnose the output current issue (back in May, gosh!), I started using the power supply semi-regularly. It became really annoying when the needle would sometimes read a half-tick lower at turn on. The supply held its output voltage just fine of course, but I'd have to pull out a DMM and check it just to confirm. Kinda frustrating on a benchtop lab supply that's inherently going to be powered off and on over the course of a day.

I'll add both of your suggestions to my list and report back. Should have some time this weekend to play around and let the unit go through a few warm-up cycles.

The digital panel meter is a neat idea. I had actually considered this a while ago and picked up a bunch of original HP 5082-7611 7-segment displays, the classic red digits they used in all of their older test gear. I'm a bit of a purist when it comes to refurbs and restorations, so I thought this'd be a cool way to upgrade the old 62xx supplies without resorting to modern parts that ruin the vintage look. My plan was to use an ICL7107 and eliminate the original meter amplifier altogether. (FYI, this is essentially what HP does in their E36xx supplies. I took apart an E3630 triple output supply we had at work and found a TC14433 ADC in quite nearly the exact datasheet application circuit for a 3.5-digit voltmeter. )

This was the post that gave me the idea, though he was swapping LCD to LED in an HP system multimeter: LED display for HP 3457A multimeter - I did it :-). Perhaps it's the one you were thinking of? Not Nixies, though.