HP 3455A zero stability

[Vacuuminded]

Hi All,
Have a 3455A 6.5 digit multimeter that I have finally gotten up and running after some hiccups.  While trying to go through some of the cal procedures, I have found the unit does not stay zero'ed.  The "Auto Cal" function seems to do its job well, that is not the issue here.  With Auto Cal OFF and a nice shiny copper strap placed between the hi/low terminals, the unit reads .000000V in hi-accuracy mode.  HOWEVER, that reading slowly creeps DOWNWARD in time, and have seen it as low as .000050V.  It takes only a minute or so for the first couple uV drop.  This also seems to happen in 2 wire resistance mode, but to a lesser extent.  Not sure about 4 wire mode, and resolution is such that it doesn't register in ACV modes. 

HERE'S where it gets weird though...  If I hit Auto-Cal (or change mode/range which automatically enables Auto-Cal again) the unit reads 0 after a couple of counts, as it should, being corrected.  When turning Auto-Cal back off again, the unit stays at 0, and starts creeping downward again. 
Does this sound like something is faulty with the input DC amplifier, something funky with the A/D converter, or perhaps something else??  I'm a little dumbfounded by this and at a loss where to even begin.   
OR is this perfectly normal behavior for these meters???  It has been on 24/7 for about a week now, so this is certainly not a matter of it not having enough time to stabilize. 

Incidentally, this exercise has made me realize that the Auto-Cal apparently does not function in my 3456A, that Simco just cal'd earlier this year.  Ugh. 

Any input at all with this?? 
Regards,
Chris

[Kleinstein]

It looks like the 3455 uses a JFET pair as the DC input amplifier. So there is a chance so see some drift (e.g. thermal). However 50µV of drift looks like quite a lot. So there might be still a problem, like a not so stable supply, or a not so stable or wrong current with the FETs.

Besides voltage drift, there is a chance to have some leakage currents causing extra offsets. Have you checked the input bias.

[macboy]

100% absolutely perfectly normal.

The input signal passes through many components: wires, connectors, FETs, relays, resistors, any many solder joints and other metal-metal junctions. Thermal EMFs  are everywhere, and the only solution is the auto-zero, or "auto-cal" as HP called it back then.

All high resolution meters experience this problem, not just yours. But the auto-zero is doing its job as designed. If you disable it, then the meter doesn't measure "zero" for each reading, so you can get a higher reading rate, but you will also see drift over time. Some meters have an optimized auto zero mode that does one auto zero per several measurements; frequently enough to reduce zero drift, but also allowing higher reading rate. If you use GPIB with your 3455 you can simulate this by turning off auto-zero but sending a command to set the mode every so often, to trigger auto-zero on command.

[Kleinstein]

It is normal to see some drift, a couple of µV per minute is already rather high for a meter that is well thermally stabilized. Not sure how good the JFET based amps were. So the error might be normal, but could also be higher than normal. The manual seems to be rather detailed so there might even be some drift adjustment.
It is also possible to have drift not only with the input amplifier, but also with the ADC itself. So if the drift rate different in different ranges (the lower ones).

Thermal EMF plays a small role, but this is more like fractions of an µV and to a large part also with the auto zero (ACAL mode). 

[Vacuuminded]

Thanks for the replies, finally had a chance to play with this some more.

As a sanity check, I re-cleaned the relevant areas of the inguard board with a soft paint brush and Freon-TF, as I could see some small spots of residue from flux, or lord knows what.  This seems to have helped quite a bit, easily cutting the "creep" to half or less of what was seen before.  Naturally, condensation caused by the cold cleaner caused issues until it had COMPLETELY dissipated, easily taking over an hour.

This made me wonder, has anyone used a conformal coating on circuits like this, post manufacture?  It appears that some of the devices themselves have/had some sort of coating on the body itself, but not on the leads nor board.  Curious as to your thoughts or recommendations here.  This would seemingly help instability due to humidity in a big way.

As a sidestep relating to this, what is considered the "best", non-conductive coating out there to use now??  Years ago, clear Krylon enamel was supposed to be very good, but I can't imagine that would be the ideal stuff to use on critical, sensitive circuits like this.

Regards,
Chris

[Kleinstein]

Usually humidity is an issue in rather humid areas only. The inside of such a rather high power meter is usually warm and thus lower relative humidity. It would be a different thing one a battery powered meter.

Dirt is usually more a problem with bias currents than voltage drift. It takes a lot of leakage to cause a significant drift for reading a low impedance source.

[orin]

I turned my 3455A on last night and put a Pomona shorting plug on the input.  I let it warm up for half an hour or so and let Auto-Cal zero it.  Then I left it with Auto-Cal off.

24 hours later, the reading was pretty steady at -5uV.  Ambient temperatures were similar at about 17 deg C.

So a 50uV drift does seem rather large for a well warmed up instrument.

If I get chance, I'll fire up Labview and log a days worth of readings.  I wrote a program to log the '55A when I was investigating the last digit jitter.

[Vacuuminded]

Thank you all for the comments, and thank you Orin for doing a check on your instrument!  It's times like this that I wish I had cobbled up some way to interface HPIB with a logging program to see just how much it drifts, and see if it tracks with temperature, humidity, or what.  Probably time to invest some time and energy into some sensors, and figure out what is needed to utilize it. 

It makes sense that these super high impedance circuits would not be a candidate for any sort of coating.  This being said, I'm still wondering what is considered to be the best materials to use for conformal coating in a more general sense. 

Since cleaning the board and letting the unit run (it's been on nearly 24-7 since this begun), the "creep" seems to have mostly stabilized.  I haven't seen it go beyond -15uV yet. 

What really has me irked now is that the 3456A "zero" isn't zero.  The auto-zero function DOES work, but it seems to "zero" at about 2uV instead of 0.  I'd like to tweak this, but afraid of how it may affect the rest of the calibration..  Looks like I've gots me some readin' to do!   

Regards,
Chris

[Kleinstein]

After cleaning the sensitive input part it is a good idea to check the input bias current - dirt like flux residue and similar can cause a significant increase in the bias currents. Sometimes cleaning is making things worse. The test is relatively easy:
Either a high value resistor (e.g. 10 M) and measure the voltage. To also get the bias current at a voltage different from 0, one can watch how fast a good low leakage low DA capacitor in the 1-10 nF range is charging / discharging. Because of DA it may need some soaking time: so keep it at the voltage of interest for quite some time (e.g. minutes to 1 hour) before the actual test.

A high bias in combination with the protective resistance at the input can cause offsets in the µV range that AZ might not correct for. Also thermal EMF might cause this, especially if relays are used.

If an instrument was not used for a long time it can take quite a while to stabilize, as the higher temperatures when on drive out humidity and this can be a very slow process.