HP 3456A voltage offset

[stevopedia]

Hello all,

I've got an early (serial 2015A04414; has a fan) HP 3456A DVM that's been beguiling me for a while now. Ultimately the fault is that it displays a DC offset of more than 100 counts with the input shorted when cold, increasing to 400 counts or more as the instrument warms up. It passes all the self-tests and I've followed the procedure for input switching troubleshooting (8-C-26 in the op & service manual) at least twice now. I got to the final step of "replace Q102 and Q112" both times, and replacing them (once) with new PN4392s (by Linear Systems, sourced from Digikey) had no effect. The offset is real, in that it follows range adjustments and can be measured with another hi-Z DVM.

Here's a list of all the tests, changes, or other work I've done thus far (all parts are on A20 unless otherwise specified):

• Replaced all the electrolytic caps in the inguard and outguard power supplies
• An error in the procedure under 8-B-10 in the OSM (steps c. and d. should be swapped) lead me to replace U405 with TLE2071 and Q408 with PN4392. As it happens these were unnecessary because autozero has always worked in this instrument, but as the comp changes haven't seemed to cause any problems they have been left in place
• Following the corrected procedure under 8-B-10 shows the ADC is working perfectly (checked with Valhalla 2701C and HP 3458A at work)
• Found input amplifier oscillation; cured by replacing U307 with TLE2071 (fixed self-test failures on 4-6)
• GB2 has an offset of about 17 mV--OSM says to replace parts if offset exceeds 5 mV, but I doubt this is really an issue
• Tried replacing Q120 with LSK489 and U105 with OP07 (C108 was removed) but this made no change. GB1 output offset is about 10 mV now. This was unfortunately done before I started taking meticulous notes, so I have no record of how it performed before the change, but it having made no difference is enough for me in this case
• Noticed C701 (on the unregulated +33 V line) is bulged, but don't yet have a proper replacement
• Jumpering across K103 at R09 & R110 has no effect on the displayed offset

At this point I'm completely out of ideas. Admittedly I'm far from an expert in these instruments, no matter how thoroughly I've read (parts of) the OSM. What am I missing?!

[Kleinstein]

Is the offset a fixed 100/400 count, or a fixed voltage (e.g. 10 µV) ?
Where is the offset measurable with another meter ?
A point to check would be if the problem is really an offset voltage or maybe way too much bias current.

One could trim the offset of the GB2 buffer via the resistors R320/R321. It is not ideal (may have more temperature drift) but easier than changing a dual JFET. I still doubt it would make a big difference for the overall offset.

There is an offset adjustment via R150 ( resistor choosen to compensate an offset) - so maybe check how much offset this really is.

[stevopedia]

The offset is a variable voltage. The numbers given are approximate.

By memory, the offset appears somewhere between the switching input at R102 and the output at J19, and can be observed at all points downstream (TP303, TP401, etc).

Bias current is a new one for me. How do I go about checking that?

The OSM has a procedure for determining the correct value and placement of R150 as part of 8-F-18. The values it gives are for offsets of 2.0 μV or less. But the offsets I'm seeing are 100 counts or more on the 100 mV range, equating to 10.0 μV or more. The OSM simply says: "If the offset is >2 μV, calibrate and/or repair the instrument."

[Kleinstein]

Ideally the meter should have a very low input current (e.g. < 20 pA or so) with zero input voltage and maybe a little more from the input resistance with high voltage. When dirty or with some bad parts the current can be higher and the input resistance lower.

A simple test is comparing the reading with a short and some 10 M across the input (may be the meter internal divider and thus 10 M input resistance mode). A test also for other voltages can be done with a low absorbtion (e.g. C0G, PP or PS type capacitor of some 10 nF) at the input and abserving the rate of voltage drift. 10 pA of input current would than cause a drift of some 1 mV per second, that can be observed at the meter. With the capacitor one can start at different voltages and this way check the input current at different voltages, nor just near zero.

It is normal to have some offset at TP303, TP401 and so on. The question is if the offset is also there in the auto zero mode.

Is there also an offset in the 4 wire resistance mode ?  There may be an extra zero compensation in software, that could make this test useless. However ideally there not be an extra constant.

For some odd reason old LM339 comparators have a tendency to fail. So it may be worth checking of the gate signals for the input switching are all working, especially the gates of Q104 and Q102.

[stevopedia]

Remember, I'm seeing this offset with the inputs shorted (my shorting "bar" in this case being some 2.5 mm desoldering braid, saturated with solder and just long enough to be securely clamped in the terminals). So while I don't think input current is the culprit, I'll try the procedure you described when I get home. (It's both clever and effective--I like it!)

I should have said sooner that I can bring at least the 10 V range into cal, using the Valhalla 2701C and 3458A I mentioned in my first post.

I'm aware of the offsets generated by the input amp and the correction for it that AZ provides. If you're asking whether the 3456A displays the same offset if AZ is on, the answer is yes.

Now that you mention it, I haven't checked any of the ohms ranges yet, being so focused on DCV. I'll test them and see what happens.

I also haven't actually checked the gate signals on the input switching FETs either--another good catch! I'll probe those too.

Thanks again for the pointers!    I'll report back with what I find. (That will be at least five hours from now!)

[stevopedia]

I'm back as promised, with results in hand (or, at least, what few results I have).

The input current test with a capacitor was inconclusive. I don't think any of the film caps I have on hand are suitable (I believe they're all PET) and I couldn't find any C0G parts that were in the range we're after (all the ones I have are tens of pF or less). If PET is good enough I'll do the test again, but I had a 100 nF cap and it drained rather faster than 1 mV/s.

I checked all the switch FETs in the input switching that are controlled by logic (i.e. not the diode connected ones) and all the driving signals were OK.

I put together a little harness to test 4-wire ohms (Volts inputs shorted, Ohms sense inputs shorted, volts low shorted to ohms sense low) and the display shows  0.00095-0.00105 Ω (it's a little noisy). Is that measurement what you meant by "an offset in 4-wire resistance mode"?

[Kleinstein]

A resistance offset of 1 mohm would be some 1 µV with a 1 mA test current. So it looks like less offset.

A polyester cap can be a bit tricky. They have quite some dielectric absorbtion (some 10 x more than PP) and one could be fooled from that.
Getting more than 1 mV/s with a 100 nF capacitor is still a bit fast, even for a PET capacitor. The dielectric absorbtion effect would change with more waiting time (e.g. compare with 1 min and 10 min of soak time).

What is the 10 M resistor case reading - that may be the easier, more direct way ?

The resistors in the 3456 are still about balanced (100 K for zero and 4x27K for the input). So it would need some 1.2 nA of input current to get 10 µV of offset.
How is the offset in the 100 V range compared to the 1 V range ? With the input divider the resistance matching should be a bit better, so input bias would have less of an effect.

For a test one could also add some 10 K in parallel to R104 to intentionally get more imbalance and thus get more sensitive to input current. So if the offset gets significant larger it would point to a current problem. Going through the settings (4 W res, 100 V, 1 V with AF on/off) could also give a hint if one of the switching fets is the source: the one that is switched on usually no longer leaks as much.

A possible test to find leaky semiconductors is by local heating. The leakage of most semiconductors goes up quite fast (about doubling for 10 K) and this often also applies to parts that are leaking more due to a defect. It does not need much heat to see a change, so a non touching hot resistor / soldering iron can be enough.

[stevopedia]

I put in an order with Digikey for some nice polypropylene film caps. They should be here in a few days.

Do I understand you correctly that you want me to measure a 10 megohm resistor with the 3456 and report what it says? I can do that, but I don't know how well calibrated the ohms ranges are.

The offset follows DCV range adjustments as if it were a real input voltage: I see three digits in the 100 mV range, two in the 1 V range, and one in the 10 V range. To my memory there is no offset in the 100 V or 1000 V ranges--but I'll verify that after work.

I like your ideas re: paralleling R104 and searching for faulty parts with heat. I'll try those too and report back.

[Kleinstein]

The idea with the 10 M resistor (could also be the internal divider) is to have this at the input with the meter in DC voltage mode. So 1 nA would result in some 10 mV which are easy to read with the meter.

The sign of the offset could also help to check some leakage paths (e.g. the gate leakage would usually be negative).

If the offset is due to leakage current it could be Q109 or the matiching between R104 and the 100 K from the divider that causes a low offset for the 100 V range.

[stevopedia]

Ohhh!!! I'm glad I asked for clarification regarding the resistor. I'd completely misunderstood you!

With a 10 MΩ resistor across the Volts input terminals, the instrument shows an offset of about -2.78 mV and slowly increasing (moving away from 0). I also have some proper numbers with the inputs shorted after the 3456A had been powered on for 10 minutes or so. Values given are exactly as the display showed.

• 100 mV range: -00.230 (the last digit was noisy)
• 1 V range: -.000022
• 10 V range: -0.00002
• 100 V range: -00.0008
• 1000 V range: -000.001

I tried connecting a 10 kΩ resistor across R104, but didn't see anything notable in the displayed values as I went through the various functions as you described. It could very well be that I simply didn't know what I was looking for or got some part of the procedure wrong.

To test with heat, I used my hot air rework station. I set the temperature to 120 °C and turned down the airflow--enough to gently heat a targeted transistor but (hopefully) nothing more. That lead me to be suspicious of Q115, as it seemed to react the most to the air (the offset becoming more negative as Q115 was heated). I didn't remember if the portion of the input switching troubleshooting concerning the switch FETs (8-C-26 g. in the OSM) checked that part and I found that it did, so I decided to repeat the whole procedure in 8-C-26 g. That immediately identified Q104 as suspect: reading was -00.0245 mV at first and changed to +00.1440 mV when the gate was shorted to the Low terminal! I don't know how I missed that previously.

Anyway, I replaced Q104 with a new PN4392, but this had no effect. In fact repeating the Q104 check given in the service manual has the same result: huge positive swing in the reported offset, with the manual prescribing replacement. I continued with the switch FET test procedure from there, but all the other parts tested good. I'll bet the original Q104 is also actually good, and the behavior when Q104 is turned on is a symptom of whatever the true cause is and not a bad Q104.

[Kleinstein]

Some -2.8 mV at 10 M indicate an input current of 280 pA. So too high in the input current. The specs should be like  < 20 pA.
If the leakage is in the pos side part the effective resistance are some 104 K and 280 pA would result in some -28 µV offset.
Balancing with R104 would only fully work if the leakage is more at the amplifier input or other side of Q103. So the input current could be the culprit for the offset.

Reducing R104 would increase the imbalance in the resistance. With the extra switch Q103 the resistor balance may not be that important anyway. So not such a good idea.

Due to the heating effect I would suspect Q115. For a test one could see if a voltage (e.g. 1.5 V battery) applied between the sense H and Voltage Lo terminal would effect the offset reading. If there is quite some effect this clearly points to a bad Q115. The test procedure for Q115 in the service manual should also work, but needs to open the case. For a first test one could check the function even with Q115 removed (obviously no 4 wire ohm then).

The test procedure for Q104 looks a bit odd to me. Chances are it would show a change even if a different FET is leaky. It adds a 2nd path to ground reducing the resistance to about half and thus expect half the effect of input current.

The 100 V range shows relatively little offset (-0.8 mV compared to -2.2 mV expected from 100 x the 1 V range offset). This make Q109 a little suspect - maybe worth retesting with some heat.

[stevopedia]

After reading your reply I took a closer look at the input switching schematic. I admit I half-glossed over it before, but something about what you said made it click for me--thank you!

I'll try your idea to apply a voltage to the 4-wire sense high terminal as soon as I can.

I should mention that in the course of my testing last night I found that there was a substantial voltage across the switching output resistor R103. The main DMM I've been using in the course of this diagnosis is a Keithley 196, which is rated to > 1 GΩ input impedance in the 3 V and 300 mV ranges. Using that meter with its low input connected to the 3456A's low Volts terminal, I found that the input switching side of R103 was at around +1 mV and the J19 side of R104 was at around +0.02 mV. That condition persisted even after disconnecting the input amplifier from J19. That would imply 100 nA flowing through R103 via some parallel path. That seems substantial to me now--I should have investigated that more closely. I'll do that too, when I can.

I'm fairly certain Q104 is a major part in the autozero system, so in hindsight I completely agree that with your assessment that the manual's procedure for Q104 is flawed. If the two input nodes are for some reason not at ground even with the inputs shorted, then deliberately turning on Q104 will force a change in the measurement whose direction depends entirely on the polarity of the charge on the upper and lower input nodes (as the OSM refers to them). That's not the first problem I've found in the manual either, unfortunately.

By the way, are discussing drain-to-source leakage or gate leakage with these FETs? Looking at the schematic now, it seems to me that shorting the gate of Q115 to the low volts terminal (which I did try, and per the manual it tested good) with the high reference terminal open would serve only to test Q115 for gate leakage or perhaps gate leakage of diode-connected Q117 since the fail condition is the displayed offset moving in a positive direction or to zero. Then again, since the high ref terminal is normally not connected, a Q115 that was leaky drain-to-source wouldn't cause any problems if Q108 and Q117 were both good.

[stevopedia]

Applying +10 V to the high ratio ref/sense terminal with the Volts terminals shorted had no effect. Same with applying -10 V. 

[Wallace Gasiewicz]

I know you already replaced Q 102    
I had a very similar problem where I could not get the 100 mV range to adjust high enough on my 3456.   
After switching multiple FETs and doing lots of things that did not help, I eventually replaced Q 102, the pre charge FET.   Everything works fine now.   
I believe that that FET was the problem although I changed so many FETs that I cannot prove it.  Anyway that was the last change I made and before that the 3456 would not adjust high enough on 100 mV range and after that replacement it now adjusts easily..   

I do not recall what the shorted value was I was using a newly calibrated volt source.

I do think that I had problems on the other ranges, but I was able to adjust them out, so I was fixated with the 100 mV problem.   
The alleged troublesome FET tested OK when removed.   
As I remember this my repair went on for far too long.   Good Luck!!

[Kleinstein]

The main suspect is gate leakage. With open ohms sense inputs there should be little voltage on most of the other JFETs that may have drain - source leakage.

As the 4W resistance case shows little offset (1 µV range) and with the strongest heat effect at Q115, I would still suspect Q115. This is even though the external voltage did not have an effect. The test for Q115 in the service manual should work better and should also work with the meter in AZ mode (less noise and drift). One could also repeat the offset test in the 4 Wire mode (100 ohm range) with an added resistor (1 M or 10 M) in series to the sense H input. This would measure the input current for the sense H input. If Q115 is the culprit the sense H input should have less input bias current than the main input.

There is a chance that the leakage can be from more than 1 souce, as the 100 V range showed less (like 1/3) offset than expected, though in this case there is still a chance for some thermal EMF at the divider and trimmer.

[FrodeM]

What is the reading at TP104?

I had a 3456A with a front-end switching problem once, and when measuring TP104 with another known good 3456A, it mirrored the exact same offset as my unit had. From that I was able to realize that the unit was measuring the gate bias and not the shorted input terminals. Turned out a previous owner had applied some sort of oil(!) to the connector between the two inboard PCBs, and in turn random switches were stuck.

[stevopedia]

Thanks for your inputs, everyone.

FrodeM: With the input shorted, TP104 is steady at about -17.5 mV (yes, I know that's technically out of spec, but I don't think that's my issue). It does not move with the 3456A's displayed offset.
Wallace: Thanks for the reminder to adjust the charge balancing circuit--I've probably done enough under the hood to warrant that. This particular instrument has been beguiling me for about ten years now, on and off. I started thinking of it as "the basket case" a while ago for a reason!

Since I couldn't shake the feeling of going in circles chasing my tail in this troubleshooting, I've done some sanity checks on my 3456A using the HP 3458A and Valhalla 2701C I'm lucky enough to have access to at work. I've verified that the ADC and input amplifier are both working extremely well, so the problem absolutely must lie within the input switching. I also verified with the 3458A that there is in fact no difference in voltage across R103, so I'm not dealing with stray current paths--the low-range DCV input impedance on the Keithley 196 is simply low enough to have a significant impact on the circuit.

I also tried recalibrating the DC V ranges just to see what happened. I was able to bring all the ranges into cal at full scale, but they all read substantially low (some 20 counts with a 6-digit display) at 10% scale--for example, with 1 V input on the 10 V range the instrument showed about 0.99980 V. Since I'm quite certain the ADC and input amp are both good, this means my apparent linearity error is actually an offset arising within the input switching that is proportional to the voltage at the high Volts input. That would align nicely with a leaky gate somewhere, I think. And I just realized that gate leakage implies we're looking for FETs that are off in the DC V ranges. Now I understand even better why Q115 is particularly suspect.

I tried a test for 4-wire ohms: I shorted the Volts input terminals together and to the Ohms Sense Low terminal, and connected the Valhalla 2701C across the Sense terminals. I found that the offset was still present in that configuration, if I was reading the results correctly. So given a FET with a leaky gate must be turned off, I think I can rule out Q115. I have ideas for more things to try, but I do have to get some Real Work™ done today!

[Kleinstein]

A FET with a leaky gate can still turn off OK. There can be separate leakage from the gate to source and drain (no need to be symmetric).

I point that hints to a bad Q115 is that the 4 wire ohm case showed very little offset.  0.8 mohm correspond to 0.8 µV and thus way less than some 20 µV seen in the DCV case.

In 4 Wire mode the input path switches between sense H and sense L. So Q115 is still off half the time, so an external meter could still detect leakage current at sense H, though not necessary the same as to the meter input side.

Current through R103 would be the input current of the amplifier. Even if there would be some input current there this would see a resistance of some 100 K in all paths and thus a rather constant offset for both readings in the AZ cycle and thus less (like 1/10) offset from the same current.

The likely culprit is gate leakage at one of the FETs (Q115,Q109,Q112,Q114,Q120,Q110, Q118) that than causes a voltage drop at the 4x27 K for the protection or R106 or the 100 K from the divider. For the other reading in the AZ cycle Q103 is off, and thus the current not reaching R104 for the zero reading. Leakage from Q104 would be off when the FET is on. So it could also cause and offset voltage and measurable input current.
Gate leakage from Q103 would cause a positive offset as the zero reading would be negative.

[stevopedia]

I found one of the culprits!

Earlier today I found that turning on the analog filter caused the input offset to drop substantially--it seemed to have been the primary source of offset when the instrument was fully warmed up. So putting that together my realizations I replaced Q114 with a new PN4392. And--it worked! The offset with the filter off is now the same as with the filter on! I'm going to let it warm up again and see how the offset changes then. I've got four new 4392s left--hopefully that will be enough.
 
I hadn't considered that Q115 was actually switched off during half a 4-wire ohms measurement cycle. In that case, it definitely could still be a problem.