HP 3455A last digit jitter in Hi-Res Auto-Cal mode

[orin]

I have had two 3455As through my hands and on both, the last digit bounced around when in Hi-Res (i.e. 6.5 digit mode) with Auto-Cal on.  See the blue line in the graph.

If you listen carefully, you can hear the auto-cal relays clicking.  The last digit jitter is in time with the relays... hmmm.  Now the way auto-cal works is that the meter reads an auto-cal constant for every real reading it takes.  There are 13 auto-cal constants.  I set up a Labview program to read a value from the meter every second and the graph is the result.  The spikes in the graph are every 13 readings, pretty much confirming that the bouncing is related to auto-cal; but how?

I consulted with cncjerry who has seen/fixed several 3455As and we went through various testing, up to his very generous loan of one of his working AtoD cards.  Thanks again Jerry!  His AtoD card was more stable, so I was pretty sure it was something on the AtoD card.

I found that the zero-crossing output was unstable.  I made it more stable by increasing the positive feedback (2Mohm to 1Mohm resistor), but no change.

I found the resistor in the integrator capacitor reset circuit was 39Kohm but 24Kohm on the schematic.  I reduced it to 11Kohm eventually with no change.   I measured the residual voltage on the integrator capacitor as the meter stepped through the auto-cal constants, finding something like 0.2mV max difference - once count is about 0.8mV on the integrator capacitor, so that was a dead end.

I found lots of noise looking at the input to the zero-crossing detector, so replaced the opamps that drove it - again, no change (and I'm not sure that the noise wasn't common mode noise, it being difficult to get a decent ground on the AtoD board).

I then monitored the counts being returned from the inguard module using a Digilent Analog Discovery - the counts can be read using the Discovery's SPI decoding and it is happy to accept the 5V levels.  I found the pattern in the counts I read.  I could see the auto-cal constant counts interleaved with the counts for the meter input.  I could see the pattern in the counts for the meter input and in all cases, where there was a spike, the previous auto-cal count + or - 10V.  +10V and the next input reading spiked up; -10V and the next input reading spiked down.  The AtoD was either showing a memory effect, or there must have been leakage of the previous input onto the integration capacitor.

Leakage was eliminated by monitoring the signal that held the integrator in reset and the signal that gated the input to the AtoD board.  They change simultaneously.  That pretty much left the integration capacitor as the only thing that could have memory.

So I considered dielectric absorption in the integration capacitor.  I noted that during the integration of a full scale input, the voltage on the integration capacitor averages +/- 9V depending on the sign of the input.  See the scope screenshot (yes, the ramp on the peaks of the triangular waveform is normal and appears on one of the screenshots in the service manual).  It is interesting to note that the 3456A service manual, in section 8-60 comments: "The average charge before rundown is kept low to minimize integrator capacitor dielectric absorption".  Presumably they had noticed the effect on the 3455A when they designed the 3456A.

The integration capacitor is a polypropylene capacitor which should have very low dielectric absorption.  It's 0.082 uF 200V.  I found some on ebay and tried one.  No change.  I read Bob Pease's article on Dielectric Absorption as well as the mentions in "The Art of Electronics".  Time to try for a Teflon capacitor.  Nothing to be found other than surplus Russian capacitors on ebay.  So, I ordered one.  The thing is huge compared to the original (see picture).  I installed it and... problem fixed.  The last digit is much stabler, bouncing one digit occasionally, similar to what my other meters do at 10PLC integration period (the 3455A actually uses 8PLC).  I let the Labview program run overnight with the result in the final attachment.  I think that is as good as I can expect from a 3455A.

So, if the last digit on your 3455A bounces around in time to the relay clicking with auto-cal on, consider replacing the AtoD integrator capacitor... there are no more Russian 0.082uF teflon capacitors on ebay, but 2x0.039 200V 5% in parallel would do the trick given the tolerance on the original 0.082 capacitor was 10% (and they might even fit in the original space with one on each side of the board).  I used a carefully sited dab of hot glue to hold my monster capacitor in place.

[Edit - additional information]

The hot glue failed.  I need to find a better way of securing the capacitor.

It's two counts out in low-resolution mode with a 10V input.  I don't know why yet.  It's one count out with the old capacitor, but flashing down to 0 counts out once per auto-cal cycle.

US Patent #US4357600 A which was for the 3456A (http://www.google.com/patents/US4357600) describes the 3455A's failings well in its "Background of the Invention" section which is basically a description of how the 3455A works.  They specifically mention dielectric absorption as a problem.

[orin]

Followup on dielectric absorption.

I built the Bob Pease test circuit* and here are a couple of results.  The capacitor is being charged to 10V for about 160ms, discharged to 0V for about 5ms, then left essentially open circuit as the hold capacitor in a sample and hold (LF398A).

First the original capacitor and second, the monster Russian teflon capacitor.  The traces are 100 * voltage on the capacitor, so about 1.6mV appears on the original capacitor and it's down in the noise for the teflon capacitor.  1.6mV is enough to make a couple of counts difference on the 3455A.

*A reference to Bob Pease's article: http://www.datasheetarchive.com/files/national/htm/nsc03883.htm  This has the best resolution schematic - Fig 5.  Yes, the timing circuit could be easily be replaced with a microcontroller.

[pelule]

I guess this is a normal behaviour during measurements with Auto-Cal.
That's is indicated by the performance tests, allowing a +/-1 digit jumping for high resolution modes.
I bought some time ago a original service manual. Former owner was Siemens Technical Engineering and Development.
Inside was a by classic type writer written Siemens instruction, dated 1982.
- how to reduce +/-1-bit jumping in hiigh resolution mode and increase at the same time the measuring speed.
=> dont't use Auto Cal continously during measurements, only switch Auto-Cal on for 2 cycles every 10 minutes to correct drifts.

BR
PeLuLi

[orin]

I guess this is a normal behaviour during measurements with Auto-Cal.
That's is indicated by the performance tests, allowing a +/-1 digit jumping for high resolution modes.
I bought some time ago a original service manual. Former owner was Siemens Technical Engineering and Development.
Inside was a by classic type writer written Siemens instruction, dated 1982.
- how to reduce +/-1-bit jumping in hiigh resolution mode and increase at the same time the measuring speed.
=> dont't use Auto Cal continously during measurements, only switch Auto-Cal on for 2 cycles every 10 minutes to correct drifts.

BR
PeLuLi

I had heard that instruction to not use Auto-Cal continuously.  My unit was/is particularly bad with +/- 2 digit jumping.

It sure does drift if you turn Auto-Cal off.  It happened by accident once with my Labview program and I thought something was very wrong with the recorded values overnight drifting high by 20ppm.

[pelule]

A +/2 digit jumping I never had on my three 3455A (I have sold 2 of them meanwhile).

Auto-Cal shall compensate Warm-Up drift, Zero drift of Om-Amps and Aging drift.
If the DMM is warmed up well the "short term" drift should small, or the DMM has an issue.
Warm-up need at least 3 hours needed in my experiance (I do warm-up before I test or do precise measurements).

Now do the selftest (press TEST button) - does it show an error?
If not, the following tests may help.

Short the inputs and switch Auto Cal off - observe Zero Reading - what is the zero drift over time?
If there is a "large" drift, there seem's something wrong.

Next connect to stable ~10V and measure the drift.
If there is a "large" drift, there seem's something wrong.

May just require the adjust of the offset or has hugh leakage problem in the input stage, or a bad opamp.
First do the leakage test, How to do, is described in the service manual in detail.

PeLuLe

[orin]

A +/2 digit jumping I never had on my three 3455A (I have sold 2 of them meanwhile).

Auto-Cal shall compensate Warm-Up drift, Zero drift of Om-Amps and Aging drift.
If the DMM is warmed up well the "short term" drift should small, or the DMM has an issue.
Warm-up need at least 3 hours needed in my experiance (I do warm-up before I test or do precise measurements).

Now do the selftest (press TEST button) - does it show an error?
If not, the following tests may help.

Short the inputs and switch Auto Cal off - observe Zero Reading - what is the zero drift over time?
If there is a "large" drift, there seem's something wrong.

Next connect to stable ~10V and measure the drift.
If there is a "large" drift, there seem's something wrong.

May just require the adjust of the offset or has hugh leakage problem in the input stage, or a bad opamp.
First do the leakage test, How to do, is described in the service manual in detail.

PeLuLe

It passes the self test and leakage tests fine.  Everything on the input multiplex node tests out fine.  Leakage IMO wouldn't produce a reading that depended on the previous conversion.  As I mentioned in the original post, I monitored the counts produced by the AtoD converter (including the Auto-Cal counts) and every reading depended slightly on the previous count.  The AtoD converter auto-zero circuit is a possible culprit, but I checked it out and the residual voltage on the integration capacitor couldn't have produced a single count error.  The only other component with 'memory' of the previous count would be the integration capacitor, more specifically, dielectric absorption in the integration capacitor.  My post today was to show that indeed, my integration capacitor exhibits sufficient dielectric absorption to cause my symptoms and that the teflon capacitor which fixed the problem shows negligible dielectric absorption.

The large drift was probably with only an hour or two warmup... and the lab temperature is far from constant.

My 10V source is a Fluke 731B which is stable - at least when monitored by a 34461A.

I've not tested drift with the input shorted and Auto-Cal off.

The service manual mentions the integration capacitor as a possible suspect for 'AtoD noise' and the fact that the teflon capacitor fixes the problem points to the integration capacitor as being my problem.

I have seen some comments that absorption can increase with age - so maybe these capacitors were a lot better when new.  Now, as far as absorption is concerned, my capacitor is about the same as a good regular polypropylene capacitor.  I did't post the picture, but a pair of WIMA MKP 4 capacitors (33 and 47nF in parallel, http://www.wima.com/EN/mkp4.htm) gave a similar curve to the original capacitor and an 80nF polystyrene capacitor was a little worse.  I have tried both in the 3455A and get similar jitter in the last digit.  I'm going to post the rest of the 'scope captures for the capacitors I tested on a separate thread.

Given the comments in the 3456A manual and related patent about dielectric absorption in the integration capacitor, it was a known problem that they fixed in the 3456A by reducing the average voltage on the integration capacitor (not to mention abandoning 'auto-cal' and using another method of cancelling out the drifts).