Author Topic: HP 3456A voltage offset  (Read 3930 times)

0 Members and 1 Guest are viewing this topic.

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #25 on: October 15, 2024, 08:15:24 am »
The leakage from flux residue usually does not change that much with temperature (may even go down from a humidity effect). It is still possibly to have leakagef from residue. AFAIK some of the critical nodes use extra teflon isolation posts, not just the PCB diretly.

Some spike during the AZ switching is normal (specially with non zero voltage measured), as the compensation circuit for the switching charge can't be perfect with nonlinear capacitance of the FETs.

Q105 should not be less of an issue (at least not cause a negative offset) as it is connected also for the low side reading and thus with compensation of the input resistance.

U307 is originally a LF351. There should be better matches than an OP07, that could well be too slow. So more like TL071, AD711, LF411 or the LF351 as in the original. I don't see how a fast OP-amp there should cause much porblems. It is more that too slow an amplifier could cause instability / ringing. A faster amplifier may need better decoupling  (e.g. maybe add an extra capacitor directly on top of the chip). However the +30 V supply may not like capacitive loading - so there could be an issue in the design. The TLE2071 has the oddity that the supply current goes up a low when saturated in one direction - this could be an issue more than the higher speed.

Maybe also check if the +5 V is clean near the amplifier - it is used as a auxiliary reference and interference there could end up at the output.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #26 on: October 15, 2024, 01:07:24 pm »
Yes, most of the sensitive nodes are on Teflon insulators.

It isn't the autozero itself that's causing the spikes, it's the ADC switching between the S+4 and S-4 slopes during run-up (which happens very frequently with a 0 V input). I'll grab some scope captures when I can. Those spike are much more apparent with the input amp in the x10 and especially x100 gain configs.

I'll freely admit that my understanding of things like circuit poles and zeroes aren't nearly as well-developed as I'd like. Why would a slower part oscillate where a faster one wouldn't? All things being equal I'd figure more gain at frequency would tend to reduce stability. Would a slower part necessarily have less phase margin at frequency?
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #27 on: October 15, 2024, 04:37:53 pm »
U307 works with local feedback for a gain of ~25 and thus quite some gain. The main loop for the amplifier has separate compensation (set by C303 and R324-R327 depending on the gain). A slow amplifier for U307 would additional phase shift and could destabilize to amplifier. The LF351 is already not that fast for the rather high gain.
A problem with a fast amplifier for U307 could be that the +30 V is just from an OP-amp output and no good decoupling (and also not easy to add) at this supply.

It is rather unusual to get effects from the S+4 / S-4 switching at the amplifier. Normally the amplifier should not see much, as the signal goes to the virtual ground at the integrator and only to tiny residual signal there (may be a few 10 mV) and some 100 K to the amplfier. So interference would more be via the supplies. Especially the +5 V may effect the amplifier output - so maybe check decoupling caps at the +5 V (at the ADC, maybe near the amplifier). An effect from the +5 V should still be about the same for all the gain settings - so likely not the culprit.

Having some signal from the ref. switching at the amplifier or supplies may lead to nonlinear errors.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #28 on: October 15, 2024, 07:03:16 pm »
I tried replacing Q110 with a new PN4392 and... it had no effect.

I think Q101 can be ruled out because there is no change in the displayed offset when the analog filter is on or off. Q112 was changed previously, so unless the new part is unacceptably leaky it can be ruled out. That leaves Q105 and Q116. I haven't checked them yet.

I tried some more investigation with warm air and everything I heated resulted in the offset very quickly moving in the positive direction--in fact it went comfortably positive (5-10 μV) before slowly falling back to -5 μV after heat was removed.

@Kleinstein, your reply reminded me to get the scope captures--I'll do that now.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #29 on: October 15, 2024, 07:47:22 pm »
Here are the captures. They were all made with the Volts input shorted, and the instrument in the 100 mV DC range with autozero on. The only difference is that some were made with the analog filter off and some were made with the filter on, as the file names indicate.

To reiterate, the only time this appears to affect the measurement is when the 3456A is set to 100 cycles integration and autozero is on and the analog filter is on. Change any of those and the problem goes away.

And this is all secondary to the offset anyway :P
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #30 on: October 15, 2024, 09:49:17 pm »
There is quite some high frequency background at TP303. Also for the case with the filter off it does not look like a clean signal, but more like something oscillating.  An oscillation can also cause an offset. At least it is not a good sign to have that much background.
A point worth noting is that the  background changes between the normal input reading and the zero reading in the AZ cycle. This makes me think of some unwanted coupling to the input / switching part or filter circuit and not so much the supply to the amplifier.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #31 on: October 15, 2024, 11:01:42 pm »
I had my scope in peak detect mode as I normally do, which has the side effect of accentuating noise. But your point stands: the noise is greater than I'd like. What's most interesting to me is that in the filter off case the waveform at TP303 is identical for both the input measurement and the zero measurement, and for some reason turning on the analog filter, which should remove this kind of noise, instead adds it in.

In hindsight, I think I may have made an error when I replaced Q120 and U105 (gate bias amplifier #1) quite a while ago. I made the same mistake I almost made with U307: substituted a slower part (OP07 again) with the thought that this would improve stability and performance. I think I'm going to replace Q120 with an LSK389D (it is currently an LSK489, which has a transconductance about an order of magnitude lower than LSK389) and the op amp with a TLE2071.
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #32 on: October 16, 2024, 06:37:06 am »
For Q120 it is not about low noise, but reasonable low low offset and low input current. Especially a positive offset ( > 20 mV)  could lead to addition input current. One could consider replacing Q120 and U105 with a low input bias MOSFET - possibly just a selected TL071H / TL031 or OPA145. So only a low bias OP-amp as a buffer (these were rare when the 3456 was designed and Thus the dual JFET plus OP-amp as the standard solution at that time).

For U105 a slow OP-amp could be OK, as here the OP-amp speed sets the overall BW of the amplifier. The FETs just work as source followers and thus little effect of the transconductance on the BW. The limited slew rate of the OP07 may be more an issue than the GBW - not so much with the shorted input, but possibly with a higher voltage.
 

Offline Wallace Gasiewicz

  • Super Contributor
  • ***
  • Posts: 1379
  • Country: us
Re: HP 3456A voltage offset
« Reply #33 on: October 16, 2024, 12:21:17 pm »
I fixed an episode of buzzing on the amp. I think this is what you are talking about..... I do not know if this will help but here is a rendition of my notes from earlier this year:   

Had buzzing on the square wave output from amplifier. tried increasing the mica cap and this helped but noticed more square wave distortion on low range-----putting my finger on Q303 stopped buzzing, replaced Q 303 with 2N4117     
This fixed the buzzing but did not help the offset. I was still not able to adjust out the 100mV range to zero, it remained 120 uV low.   

I know my notes are not really complete but maybe check Q 303.that was not helping.     

EDIT:  I also had trouble with a noisy 33 Volt Supply It was fixed by replacing caps.


« Last Edit: October 16, 2024, 12:51:06 pm by Wallace Gasiewicz »
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #34 on: October 16, 2024, 03:37:15 pm »
I GOT THE OTHER ONE!!

Removing Q105 rewarded me with an input offset voltage of 0.0 μV!!! I replaced it with my very last PN4392, and with the instrument cold/just powered on I had an offset of about -2.0 μV, which is very similar to that for the 3456A that's on my bench at work and which also has a current cal. I'm letting mine warm up now to see if the offset settles out to zero like it does on the working instrument.

(Edited to correct the part changed)
« Last Edit: October 16, 2024, 03:58:30 pm by stevopedia »
 

Offline Wallace Gasiewicz

  • Super Contributor
  • ***
  • Posts: 1379
  • Country: us
Re: HP 3456A voltage offset
« Reply #35 on: October 16, 2024, 03:53:19 pm »
I thought you had already replaced Q 110???
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #36 on: October 16, 2024, 03:56:41 pm »
I did. In my excitement I got it wrong--the one I actually did just now was Q105.

Also, the new part produces the same offset as the old one. Still, this is a powerful clue, and if nothing else I'm fine with sacrificing 4-wire ohms for an otherwise working instrument.

Edit: I can't even say that much. The offset quickly increases (decreases?) to about -5.0 μV over the course of a minute or so after power-on even with Q105 removed. So much for that.
« Last Edit: October 16, 2024, 04:12:07 pm by stevopedia »
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #37 on: October 16, 2024, 06:25:05 pm »
So I replaced the PN4392 as Q105, and I think my pessimism over that not working was a bit excessive: the offset is definitely reduced with the new Q105 in place vs. either the old part or no part at all.

I think the noise on TP303 (input amplifier output to ADC) is responsible for a lot of what I'm seeing now, both the offset and the noise in the measurement; the least significant (100 nV in the 100 mV range) digit is quite unstable/noisy even with the input shorted and generally always has been. I checked all the power jumpers by the ADC section (JMPR403 and JMPR503-507) and the +12 V and -12 V rails had the same noise that I saw on TP303. That certainly won't help, especially given the -12 V rail sets up the constant-current tail on the input amplifier's input JFETs Q310 and Q311.

Looking at the schematics (mine has the LM399-style reference of schematic 5B), there's shockingly little in the way of filtering on those rails. I count exactly one cap on -12 Va, a 0.1 μF wet tantalum (C401) which won't do much to reduce high-frequency noise. -12 Vb has a 2.2 μF wet tantalum, C404. But the +12 V reference has no filtering at all!
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #38 on: October 16, 2024, 09:45:25 pm »
Not much decoupling on some of the voltages is not good. However the problem is that some voltages, like the +12 V,-12VB or the +30 V (at the amplifier) are directly driven from OP-amps and thus can not have much capacitive load. So there is no easy way to improve on that situation. Part if the signal on the supply is likely normal and that way by design.  Instead of a normal decoupling capacitor one may have to include a series resistance to get at least some dampening.

At  leat the +12 V part seens very little load and should not see transients - so here capacitance is not needed.
The problem is more the -12 V.

I agree that the "noise" / high frequency background on the amplier signal can be a problem. It may be enough to get some rectification at the JFETs and additional bias this way. If all supplies and the signal seem to have the same noise, maybe check the ground that is used for the scope. It could be just a bad ground point.

Not having reference filtering is a common point in many DMM designs. The LM399 ref. has quite some white noise that can effect the ADC. This part is missed in other DMMs too.  Adding some ref. filtering could be possible, e.g. at the input of U501 and between pin 2 and 6 of U505. It could be worth the relatively moderate effort.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #39 on: October 17, 2024, 01:02:38 pm »
For the moment I'm focusing on the slope-switching noise/hash I'm seeing on the +/- 12 V references--it doesn't show up on any of the other power rails. Of these, -12 Va is probably the most important, since the others are generated from it, as you noted. I think the noise is coupling in there and U504 and U505 are faithfully duplicating that noise on their outputs.

Yesterday I did a little further investigation. I found that the switching hash was present on pins 2 (- input) and 6 (output) of U501, but not at pin 3 (+ input). I connected a 470 pF ceramic cap directly across U501 pins 2 and 7 (ground, conveniently) on the bottom of the board and this quieted U501's output beautifully. The noise was still present on the collector and emitter of Q501. (It was much stronger at the collector--I think Q501 is acting as a common-base amplifier for it in this case. Not especially important, but I thought it was an interesting observation.)

U402 looks like it could be the source of the switching noise. I'll try putting a 470 pF cap across its power inputs (not coincidentally -12 Va and ground) and see what happens.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #40 on: October 17, 2024, 03:43:16 pm »
It didn't work. Nothing changed.  |O |O |O |O

I am so close to giving up on this thing again. I don't know why I'm suddenly so demoralized--perhaps because I thought I'd finally cracked it after ten years--but the fact remains.
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #41 on: October 17, 2024, 07:02:08 pm »
The collector of Q501 should not have much AC signal and there is already a capacitor to ground (C506). One could try adding extra capacitance there. Different from the amplifier output there is no problem adding capacitance there. With not enough capacitance there the transistor can not drive the -12 V fast / strong enough as it should. Maybe a problem with the ground connection ?

There is already a 100 nF capacitor (C401 on the ADC page) at the -12 Va. So 470 pF more does not make a big change.
From the scope pictures it looks a bit like ringing as the interference signal. Some RC combination e.g. 22 or 39 ohms and 100 nF could help in dampening the oscillation more than added capacitance.
Also just a little more load (e.g. 5 K to GND) to the -12Va could help, as the output resistance of the transistor gets lower with more current.


The loading of the reference switches is a principle tricky part for this switch configuration. So don't expect the -12 V to be perfect, there is some good enough. After all the HP3456 normally works OK and is usually considered a good meter for it's time, depite the weakness with the -12 V ref. / supply.
The question is only how much AC part is normal acceptable for the -12 Va.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #42 on: October 18, 2024, 01:33:31 pm »
The loading of the reference switches is a principle tricky part for this switch configuration. So don't expect the -12 V to be perfect, there is some good enough. After all the HP3456 normally works OK and is usually considered a good meter for it's time, depite the weakness with the -12 V ref. / supply.
The question is only how much AC part is normal acceptable for the -12 Va.

You're absolutely right, of course. I've been letting myself get carried away. I think it's time to take a deep breath and a step back. I also have the luxury of access to a working 3456A I can take reference measurements from--a capability I haven't taken advantage of.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #43 on: October 23, 2024, 12:46:40 am »
I took the opportunity today to measure the gate leakage current of all the switching JFETs I've recently extracted from the 3456A with a Keithley 485 picoammeter. For most of the FETs it was too low to measure (less than 100 fA!). But for the one that was on the analog filter circuit, it measured about 50 pA with 18 V reverse bias on the gate-channel junction. On a whim I replaced Q102, Q103, and Q104 with some of the verified-good original 1855-0298s, but as expected this made no difference.

With the Volts inputs shorted, the meter's offset is about 6.5 μV. I tried shorting the lower input node (Q102, Q103, Q104, Q105) to the low Volts terminal with a probe lead and the displayed offset went to zero, as expected. But when I shorted the upper input node (Q109, Q116, Q115 etc) to the low volts terminal, the offset shot all the way to -1.1 mV, a full three orders of magnitude higher! I don't know what to make of that, though it would seem to me that if the instrument was working perfectly then shorting the upper input node to ground should also result in a 0 V reading... though I'll test that on a known-good unit tomorrow.
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #44 on: October 23, 2024, 12:57:54 pm »
Soemthing like 50 pA at 18 V is about what is expected from a somewhat bad JFET. Less than 100 fA is surprisingly good for the good ones.

Getting the zero readign with the lower input not shorted is no surprise, as both reading of the AZ cycle would see the same voltage and thus zero difference. There much be something badly wrong to get a different result.

Connecting the upper input node (Q109,Q115,Q116,...) to the input should not make that much difference. It could make a difference for some self test / gain check, that tries to read the internal -12 V reference. To protect Q112 one may want some series resistance (e.g. 10K) and not a hard link to the input terminals. A wrong ref. reading would result in totally off readings - maybe the factor 1000 to high, though it should trigger an error message too.
The test with the upper terminal ground could still show a small offset, e.g. from amplifier input bias acting on R104 and the offset from R149/R150 that would still be active for the zero reading.

A point to check could be if the offset changes between 1 PLC and 10 PLC setting (100 mV range). A difference can come from addition input current from the switching / compensation ciruit around R147/R148 / Q102.
Too much offset of the Q120/ U105 combination could still be an issue here, though I would expect only a small error. If needed one could adjust R126 or R127 to trim the offset. This may be easier than exchanging the dual FET.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #45 on: October 24, 2024, 03:14:53 pm »
To be clear, my experiments with directly connecting the input nodes to the low Volts input with the Volts inputs also shorted together were only temporary, using a regular "pokey" multimeter probe, just to see what happened. The meter never does a check of the -12 V reference (via Q112) unless it's in the self-test mode, and I had it in the normal DC V mode. I'd have to repeat it to be sure, but I don't think AZ made a difference.

The effect of shorting the high node to the Volts input terminal is to bypass the section from R103 through Q116. What doesn't make sense to me is why that should result in the meter's displayed offset being almost a thousand times greater than without the short, and AZ having no effect.

GB1 (Q120/U105) is working fine; offset is about -6 mV.

Another thing I noticed: with the Volts inputs shorted, I connected one end of a meter probe to the low Volts terminal and touched the other to GND at TP101, which should have literally no effect (the schematic shows the terminal shorted to ground). The displayed offset went from about 7.0 uV to about 4.0 uV and returned to 7.0 uV when the probe was removed. I think that indicates at least part of what I'm dealing with is that I need to implement the one service note HP put out for the older 3456As that involves drilling out the rivets originally used for grounding the outguard case parts and replacing them with screws. The purist/preservationist in me is sad that I can't (easily) get the prescribed Pozidriv hardware, but such is life.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #46 on: December 09, 2024, 03:16:08 pm »
It's been a while since I last posted an update here. I've tried a bunch of things, and now I know a lot of things that don't affect the offset issue!

  • Implemented the service note where the grounding rivets are drilled out and replaced with screws: No effect.
  • Replaced Q101: No effect.
  • Replaced some of my replacement FETs with verified-good original parts: No effect.
  • Carefully and scrupulously cleaned the top and bottom of the A20 board in the vicinity of the input switching circuit, separately using 99% isopropyl alcohol, acetone, and soapy (Triton X-100, a nonionic surfactant, in the form of Photo-Flo) water: No effect, or maybe a few single-digit counts on the 100 mV range (tenths of a microvolt).

I was messing around with it last night, probing with my oscilloscope and another DMM, when I noticed that the offset had dropped to -1.5 μV or so, or well less than half its normal value of about -5.0 μV, and I have no idea why. I also found that turning on my desk lamp, which as an LED bulb and is connected to the same outlet (an isolation transformer) as the 3456A, consistently reduced the magnitude of the displayed offset by about 1.0 μV (10 counts on the display). I turned it off for the night, and this morning I turned it back on and the offset was back to its usual value, though I was able to reproduce the lamp's effects again. I wonder if I'm dealing with common-mode noise on the power input generating this offset somehow. If so, that raises the question of why the autozero doesn't cancel it out--that's the whole purpose of having an AZ system!

I dunno, man. This thing is stubborn.  :-BROKE
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #47 on: December 09, 2024, 04:20:31 pm »
An EMI effect is a real possibility. EMI (e.g. from a noisy LED lamp or SMPS) can cause some extra offset, that the AZ mode is not allways able to correct. The AZ mode is not made to correct EMI effects.
When the 3456 was build there were a low less SMPS and cell phones around and no Wifi. I don't really think it is common mode noise, as the power supply is still via a classic mains transformer and not a separate DCDC converter. A CM filter at the mains side could still have an effect (could be part of the line filter already).

It can still be tricky to improve the EMI suppression, maybe some ferrites at the input side. In many other aspects at least the circuit diagram looks good - could still be a layout thing for the really high frequencies. For a quick check one could test click on ferrites with the mains cable or a the analog inputs.
 

Offline stevopediaTopic starter

  • Contributor
  • Posts: 36
  • Country: us
Re: HP 3456A voltage offset
« Reply #48 on: December 09, 2024, 10:44:03 pm »
I agree. I don't doubt that's at least a significant portion of what I'm seeing, especially with respect to noise. After all, I live within line of sight of an FM broadcast transmitter tower!

It gets weirder, though. I left the 3456A on during the day with its volts inputs shorted and the outer upper cover sitting on top of the instrument. When I got home from work the display read about -7.0 μV. I took the cover off and set it aside. A few minutes later, the display was bouncing around +/- 0.2 μV! I rested the top cover back on the DVM and the indicated offset slowly began to rise again.

With the cover on, I tried to reproduce the effect my desk lamp had, and--nothing! I knew I saw what I'd seen then, so I took the upper cover back off and bingo--instantly reproducible. I realized that I had the desk lamp pointed directly at the instrument so I tried blocking the light, and the effect was the same as if I'd turned the light off and with immediate effect. So either I'm seeing something thermal that's ridiculously sensitive, or there's something photosensitive in the input switching area.

As it happens, C104 on my instrument appears to be some kind of ceramic capacitor in a transparent glass or epoxy housing (not sure which; haven't touched it to find out). The parts list says it's just a 470 pF, 100 V DC capacitor, but for some reason HP saw fit to use a different component here instead of the identically-rated polypropylene caps specified for A20C101-103. Of course the part number and mfg code point to an HP custom part (probably a custom spec) so it's impossible to know for sure what the difference is, but it's worth at least trying a replacement.

Edit: I just tried moving my desk lamp right up to C104, and I could make the offset go as strongly positive as it went negative! I wonder if these caps were always light-sensitive, or they became more sensitive with time.
Second edit: It's not C104, but there's definitely something either directly photosensitive or something phenomenally sensitive to temperature on that board somewhere.
« Last Edit: December 09, 2024, 11:05:36 pm by stevopedia »
 

Offline Kleinstein

  • Super Contributor
  • ***
  • Posts: 15149
  • Country: de
Re: HP 3456A voltage offset
« Reply #49 on: December 09, 2024, 11:10:16 pm »
Some parts are indeed light sensitive. This was however less an issue in the old time with THT parts, but more with small SMD parts like  TSSOP and similar with not much plastics. In the old days it were diode (and transistors) in a glass case. For some odd reason they still make low leakage diodes in a glass case.
Thermal EMF can produce an offset of a few µV with some parts.

For the FM band it can make a difference of the cover just rests on the meter or it is screwed down. For me RF effects are still a bit unpredictable.
 


Share me

Digg  Facebook  SlashDot  Delicious  Technorati  Twitter  Google  Yahoo
Smf