Testing "working" 3458As

[dkozel]

I'm eagerly looking forward to the arrival of the precision resistors and other components to build three of branadic/DrFranks' LTZ1000 boards. I've been living entirely in the precision frequency realm for the last years and this is my first incursion into volt-nuttery.

I'm in the very fortunate position of having a few labs worth of closets to dig around in and have found a pair of Agilent era 3458A DMMs. No calibration records of course and the lab manager said "Oh those, no one's used those in a few years". Cosmetically they are in good condition and they have electrical test stickers from 2016 so major power supply issues are unlikely. After a quick check of the fuses and line voltage selector I turned them on and both passed self-test. I recently bought one of the DMMCheck Plus boards so decided to use it for a sanity check.



+5.0000 VDC, yep, fairly happy with that. The cables are mismatched and the DMMs not warmed up so the difference in measurement didn't disturb me too much.

The trouble is, I don't really know where to go from here. My goal is to verify that both of these DMMs don't have gross issues like drifty A3 boards, hopefully get them checked against something calibrated, and setup a measurement system to monitor the three LTZs over the next year (or until someone realizes that they actually want to use these!). I've setup GPIB access using a Raspberry Pi  so the setup can be pretty standalone.

My next step is to run some tests over time. TiN has a spreadsheet that I've seen in a few threads now. It lists meter temperature, CAL? 72, CAL? 1,1, and CAL? 2,1 for each day. I believe that this does not rely on any external source as it is a relative internal measurement. I'll setup the logging to capture that every few hours over the next week or two and see what can be learned from the results.
https://www.eevblog.com/forum/metrology/(3458a)-yet-another-3458-repair-thread/msg1885664/#msg1885664

Next up will be making some sets of cables and trying to figure out what my test setup will be with two multimeters and three DUTs, sadly no Keithley 7168 scanner cards around. 

Anyone who has 3458A knowledge, I'd be very interested in any other tests you'd recommend I run. Short of sending it to Keysight (Ha $$) I'd like to verify that it's operational as well as possible.

[lukier]

I've recently expanded my lab with two 3458A as well and verifying these instruments is not easy.

Obviously I did  CAL? 72 logging. Pro tip, if the meters were not powered up for a while I'd let them warm up for a week and only use CAL? 72 samples from the days after that. SN18 also mentions that LTZs powered off for too long might need 6 weeks aging to bring them back to stability. Nice thing about CAL? 72 logging is that you don't need any extra equipment. DMMCheck or your new LTZ boards (not aged) are not stable enough.

Another tests I did is just every function range, having two meters helps there (even uncalibrated) as a sanity check. I've used new Keithley DMM6500 that has very recent calibration as well for this purpose (but again, not precise absolute checks, DMM6500 is 10x noisier than 3458A).

Tests done so far:

• DCV - various values generated by Fluke 5440B from microvolts to 1000V
• ACV and FREQ - various values generated by Fluke 5200A, from mV to 100V, 10s Hz to 1 MHz
• OHM/OHMF- a box of mostly Vishay VHP resistors, 1 Ohm, 10, 100, 1k, 10k, 100k, 1M and some Caddock/Ohmite 10M/100M/1G
• DCI/ACI - like V but via Wavetek 4600 trans-conductance amplifier, from 1mA to 1A, lower currents from Keithley 2400

Another useful test IMHO is INL test. Normally one would need JJA, but you have two meters, so if they are both in a good condition and similar temperature the INL difference between them should be below 0.1ppm.
For more info see here: https://www.eevblog.com/forum/metrology/dmm-linearity-comparison/msg1909922/#msg1909922

Edit:
Another useful test is the shorted input noise on various ranges and NPLCs, see TiN's dmm_noise thread.

[Kleinstein]

The CAL72 test is not sensitive to reference drift very much. It tests the ADC's gain by measuring the 7.2xx reference level, that is also used for the ADC reference.

After long time of not using such an instrument there might be a similar (e.g. humidity related) effect with the ADC gain too. So the values for the fist few days are less accurate, as there is like humidity going out and the circuit getting used to higher temperature again.

Another sanity test is noise, e.g. in the 100 mV range to test the amplifier and 10 V range to test the ADC.

[ap]

CAL72 test essentially refers everything to the internal reference, so any reference drift is not detected by checking CAL72. So that test detects drifts in the A/D (and related components on the input board). Any drift beyond limits detected I have seen (and unfortunatelly it were many units) has always been related to the A/D board, not any other parts of the meter. So beware of buying expensive 3458As with unknown state other than 'passes selftest'.
The issue with verifying precise meters in general is that you need precise references. If you do a ball park test with some precision resistors bought or something like DMMcheck what have you won? This reaches 4 to 6 digit accuracy usually. Or by using uncalibrated calibrators. This does not tell you anything other that the meter is operational in these ranges. You can as well buy a lower uncertainty meter new with cal cert cheap instead. Having highest accuracy gear (and for non-professional voltnuts use from time to time at least) requires accuracy verification with known uncertainties to make sense. That does not mean one allways needs to go to the highest level cal cert.

[lukier]

This does not tell you anything other that the meter is operational in these ranges.

But that is what's needed to test if the meter is working - the question of this thread. How far it is from SI Volt & Ohm is another story. The second 3458A I got was very cheap, but in a bad state, was a "donor unit", some parts were needed etc and of course NVRAM was gone. So I just calibrated it off 5440B 10V for now, as my 5440B has very stable and well aged dual SZA263 reference. Therefore I could do all the tests in respect to 5440B and see if the meter is working. For the functionality test it doesn't matter that both 3458A and 5440B are 100ppm away from SI volt (they aren't that far off my case).

Once I get better volt & ohm home I'll recalibrate both 3458As and 5440B, but that's another story.

One more test comes to mind is checking the input bias (< 20pA).

[MiDi]

One more test comes to mind is checking the input bias (< 20pA).

What would be best way to check that without electrometer?
External 10Meg at input and perhaps 1nF MKP in parallel?

[lukier]

One more test comes to mind is checking the input bias (< 20pA).

What would be best way to check that without electrometer?
External 10Meg at input and perhaps 1nF MKP in parallel?

Something like that, or maybe even simpler using internal 10M (FIXEDZ ON command), as described here:
http://gpete.blogs.keysight.com/2013/09/measuring-injected-current-and-input.html

[Kleinstein]

Checking the input bias would be either with a resistor, like the 10 M internal resistance, or with a capacitor.

The capacitor would be more like 10 nF, so that the DMM internal capacitance of a few 100 pF would not alter much. The current is than from I = C*dU/Dt.    With 10 nF and 10 pA bias this would be some 1 mV/second, which is relatively easy to read from a set of data.

The capacitor method has the advantage, that one can also measure with an applied voltage.


If one has an old instrument some of those crude test would be a good idea before sending it in for calibration. Having the unit run for some week before the calibration is likely a good idea too, to get out the moisture and thus keep drift low. Also some possible defects that might have developed over time might show up.

[Henrik_V]

Another sanity check I learned from a now retired (but still active ) fellow, who really knows 3458s :
Take a 9V block battery and a 10M or 100M resistor. Measure the voltage with and without that resistor in series, in both polarities.
If one of the protection 'diodes' got leaky (lower input impedance) you will see it.

[dkozel]

It's really awesome that I can post a question and wake up to such advice and discussion. Thank you all.

Today I'll setup the two DMMs and get them running. I'll monitor the temp and various CAL? values out of curiosity to see how they change. I'll use the warm up/re-burn-in/drying time over the next few weeks to figure out cables, switching, and refine the control and plotting software. I'm hoping to produce vaguely similar plots to TiN's comparing three or four references measured by both meters. Chances are that this will require making a custom switching setup and if so I plan on basing it on the various designs in the DIY Low Thermal EMF Switch Scanner thread.

For cables I think I'll end up with some Cat 5/6 based ones for the moment as I don't plan on doing any high voltage measurements. That seems to be a reasonable low cost route and spend a bit of the money on decent spades and crimping.

[tggzzz]

The trouble is, I don't really know where to go from here.

I have a few toys that might indicate a direction...





Given that we are a mere 40 miles apart, it might be worth having a mini cal-fest in the new year. That would give you a chance to "dry out" your 3458s

I mention "dry out" since my Solartron 7081 manual suggests that, if it has been stored in non-lab environment, turning it on and wrapping it in a blanket for 24 hours (so it bakes at 40C) can beneficially evaporate stuff on the boards. It certainly improve the stability of mine. You should not infer I am suggesting going that with any 3458.

[lukier]

Here are my measurements so far. #1 is the good one, #2 is the one with the more troubled past.
#1 drifts 0.05 ppm/day while #2 is currently at -0.18 ppm/day, still within limits (<0.43 ppm/day) but not great. Maybe next year I'll buy new ADC board, we'll see how it goes.

These figures are just raw drift rates and for less than 7 days they can be quite jumpy (less averaging). Also the temperature is not taken into account and especially the sick ADCs can exhibit higher tempco I think. Once I get more days of data I'll pick samples at the same temperature and this could improve these figures a bit.

[Kleinstein]

The large excursion on the meter #2 seems to be only at the beginning. So likely this is not a problem. After that it looks fine.

[lukier]

The large excursion on the meter #2 seems to be only at the beginning. So likely this is not a problem. After that it looks fine.

This is not absolute ppm drift, but running ppm/day, so unfortunately it is steady at ~0.2 ppm/day

[martinr33]

Sounds like you are in for a new ADC board. It is worth doing, as the updated units are very stable over time.

You might want to rummage through those closets again, and look for a Fluke 732a. I feel like you need something like this to stay spot on. Even a 2ppm/year drift on these meters is 200 counts.

[lukier]

Sounds like you are in for a new ADC board. It is worth doing, as the updated units are very stable over time.

Yes and no, 0.2 ppm/day is still bellow 0.43 ppm/day specified by SN18, but I agree it is not ideal. Hence why I'll think about it next year as the new ADC board is more than I've paid for the second meter.

You might want to rummage through those closets again, and look for a Fluke 732a. I feel like you need something like this to stay spot on. Even a 2ppm/year drift on these meters is 200 counts.

I wish I had closets with Fluke 732A. Even second hand beaten up drifty ones go on eBay for pretty penny, hence why I'm building my own 4 boxes with LTZ1000 and 7-10V stage and some other features.

[dkozel]

Had a bit of a distraction with getting the Pi onto the wifi here and then dealing with the linux-gpib driver, but the logging is setup and is now taking measurements every 12 hours. Hopefully I'm correct in assuming that ACAL needs to be run before each measurement to update the DCV 10 Gain value (CAL72). The code I'm using is here:
https://github.com/dkozel/voltage_reference/blob/master/warm_up_logging.py

Here is the first 12 hours of data. The DMMs were turned on about four hours before the start of logging so definitely way to early to care much about the numbers. I'll leave this logging running until at least the middle of January as I have some other work coming up that will be distracting.

DMM "22"

Time	T Celsius	10V Gain	40k Ref	7V Ref	10V Gain ppm
2018-12-04T18:04:59Z	46.1	0.985288471	40001.864	7.08983721	0
2018-12-05T06:04:59Z	44.3	0.985288206	40001.864	7.08983721	-0.269

DMM "23"

Time	T Celsius	10V Gain	40k Ref	7V Ref	10V Gain ppm
2018-12-04T18:05:00Z	45.9	0.987604634	39999.1402	7.09641088	0
2018-12-05T06:04:59Z	43.6	0.987606166	39999.1402	7.09641088	1.551

I have the DMMCheck board which I briefly used to check a few of the ranges, but I'll do so more methodically when I get the chance and share the data. I have a pretty precise SMU that I can use to generate test currents and voltages down to the ~50uV and ~100pA accuracy.

For the INL test (Reference Thread, yes sadly the closets don't have a JJA (I did check just to be sure), but doing a relative comparison to each other does seem interesting. I'm guessing that I can use the above SMU as it is the stability rather than accuracy that is important and the noise levels are quite low on it.

For the Noise Test (Reference Thread) That seems quite straightforward and I can run it during the 12 hour downtimes. Though I was already looking at getting a low EMF short and found that it's already been done for the 3458A by texaspyro (Thread). But yah, this project is going to be expensive enough so I'll use basic copper wire and not loose a month waiting for PCBs!

The input bias current link is interesting, thanks for linking that lukier! Without spending too much time thinking about it, it seems like we'd want to shield the jumper cable as it runs the length from the HI terminal to the ground point, maybe some scrap coax and ground both center and shield at the back and just the center at the HI?  The capacitor option is interesting as it could be a through hole component mounted just between the HI and LO terminals(?) keeping stray fields from being a potential issue.

tggzzz! Yes, definitely I'd enjoy meeting up sometime next year for a comparison and chat. That's a fun collection of items for sure! I'm a bit behind on the LTZ build due to some BOM confusion (so many details!), but February/March I should have them built and warmed up.