Precise Offset Compensation Ohm Measurements and Validation of DMMs

[Dr. Frank]

Ohm Measurement Problem

I came across a very uncommon error on ohm measurements, when I checked a brand new 34465A against a 3458A, using my old precision ‘reference’ resistors.

Setting on both DMMs was: 4W Ohm for values <= 10k, Offset Compensation for values <= 100k, and NPLC 100. So the highest accuracy of < 1ppm could be expected.

For the 100k resistor, there was a gross reading difference of about 15ppm between both instruments.  This is no ppm nit-picking, and no volt-nuttery any more, that is a severe measurement problem, obviously.

In contrast to that, a 10.0000k Vishay Bulk Metal Foil resistor, which had been accurately measured to 1ppm independently by Third Party, gives identical readings on both DMMs to 1ppm.

This resistor serves as a baseline calibration standard for the 3458A, which conducts by its AUTOCAL function all necessary internal precision ratio transfers, calibrating all ranges from 10.00 Ohm to 10MOhm, to about 2ppm uncertainty.

As the 34465A was brand new, I concluded, that the 3458A has a failure in its Ohm measurement circuitry, maybe a defect of one of the FETs. This would also discredit the complete AUTOCAL functionality for Ohm, and for the current ranges.

At that point, my confidence in the 3458A was completely shattered.

From a metrological point of view, that’s the classical ‘man-with-two-clocks’ problem: As the absolute value of the 100k resistor had not been independently measured, it is not possible to make a decision about, which of both instrument is failing.

Even worse: As I before had relied on only one precision instrument, i.e. the 3458A with its AUTOCAL function, I had absolutely been ‘blind’ for any possible failures.

As a conclusion, an independent verification of the functionality and calibration of any instrument is always needed.

[Dr. Frank]

Part 2: Resistors

These resistors I have salvaged from old HP instruments. These are precision wire wound type, from 1969, <= 0.01%, T.C. 1...5 ppm/K.

I used them as shunts and as resistance standards for the last 25 years, also to calibrate my DMMs, like the 34401A.

I had determined their absolute values several times with 3 different 3458As, in 1990, 1996, and since 2009 with my own 3458A, every time using that same setting. They all showed a decent drift over the years.
The same error occurs on other resistors, like the 5450A (37ppm) and a decade resistor box (18ppm).

The 5450A allows an independent 100k calibration from the 10k Vishay resistor, by using a precise 10:1 ratio transfer.
Description is here:
https://www.eevblog.com/forum/blog/eevblog-544-fluke-5450a-resistance-calibrator-teardown/msg823545/#msg823545

So in the end, it has been decided, that the 3458A displays an erroneous value.
And maybe, that I also did not detect that error, for the last 25 years.

[Dr. Frank]

Part 3: Root Cause

I started to measure deeply inside the guts of the 3458A ohm circuitry, especially the timing and found … a totally different explanation of this effect. The 3458A has no defect, and its AUTOCAL is working correctly.

When I measured another PWW 100k resistor, both instruments showed exactly the same value. Therefore, this error obviously depends on the DUT.
At first I found out, that the timing of the 34465A and the 3458A in OCOMP mode is different on the identical settings.

The 3458A always makes NPLC 10 measurements which were averaged to give higher NPLC numbers. So a NPLC 100 ohm measurement consists of 10 x 400ms wide On/Off cycles.
The 34465A only makes measurements of 2s on and 2s off.

Therefore, this has something to do with charging effects, obviously.
Using NPLC 10 lets the 34465A also display wrong values, like the 3458A.

[Dr. Frank]

Part 4: DMMs with OCOMP

So each instrument having an OFFSET COMPENSATION function may be affected, see table.

I went through all the user manuals but did not find any clear explanation about this failure, unlike 4W and OCOMP functions, which were extensively explained in measurement tutorials.
Also in the specifications, there is no parameter characterizing this problem.

The 34420 and the 3458A only had very hidden, and not very obvious hints about that problem. They propose either to switch off OCOMP, or to experimentally introduce a delay.

Both measures actually helped, but w/o OCOMP, big errors from thermal voltages can be observed; especially on ranges, where the test voltage is 100mV or 1V only, i.e. on the 10 Ohm.. 100k ranges.

Normally all instruments have delays on the order of several 10ms only, the 3458A increases that to 50ms during 10k calibration.
The FLUKE 8508A has a very long 5 seconds delay, which makes this instrument extremely slow, but FLUKE probably knew about that problem, and also limited their offset compensation to max. 20k range.
The 7 ½ digit Keithley 7510 directly adds 10ppm error for OCOMP on the 100k range, for no obvious reason.

Using an appropriate delay of about 5 sec for 100k, and 1 sec for 10k ranges, all the measurements now were correct, and especially the verified 100k value was in accordance to 3458A and 34465A in each case.

[Dr. Frank]

Part 5: Explanation of the effect

Anyhow, to find an explanation for this error, I checked the impedance of these resistors for parasitic capacitance and inductance.

I measured the impedance over frequency, and there is a remarkable difference between both 100k resistors, (1) the black, naked one, and (2) the red one in the box.

When I measured the red HP pww resistor (2) directly, w/o the case, the error vanished, and also its frequency response was smooth.

The capacitance of the case is about 40pF. This gives a time constant on the order of µsec, which cannot affect the reading.

When I measured the leakage current of the jacks, I finally found a relaxation effect of the expected order of magnitude and time scale.

This current decreases slowly, over several seconds from about 20nA to several hundred pA, where it stabilizes.
That happens each time, when reversing the voltage.


Therefore, all precision resistor measurements with OCOMP functionality,  on values of 1k and above have to be carefully analyzed for parasitic capacitance (order of nF and more), but also for Polarisation effects, coming from the plastic isolation in jacks and cables.

This presumably also affects standard resistors like the ESI SR104 or the WEKOMM type.

Choosing an appropriate delay of 1 .. 10sec, OCOMP, and 4W connection then may allow measurement uncertainties of < 1ppm.

[jpb]

Interesting. Thank you for sharing your investigation. It shows that even something as apparently simple as a DMM needs to be understood in detail when it comes to making measurements down to ppm.

Off topic but a similar issue, I used to work on I(V) characterisation of FETs and other devices. The measurement rate there can have a huge effect on the results because of charge getting trapped in surface states. Basically curve-tracers gave sets of measurement curves that were useless for predicting the circuit behaviour under operating conditions.

[HighVoltage]

Very interesting indeed.
You are always making such perfect explanations, thank you.

When I made myself an inductance "reference" of 1 mH I came across this effect of different values when the inductor was inside a box or outside a box. Even when I used closed metal cores around the inductor, with and without airgap. Only when I switched to a toroidal inductor core, did all problems vanished and the values stayed stable inside and outside the enclosure.

What instrument did you use to get the impedance graphs over frequency?

[dr.diesel]

Very interesting indeed.

I look forward to study of this with my own instruments capable of OCOMP.

Thanks Dr.Frank

[Alex Nikitin]

Thank you, a very useful investigation and a reminder that the dielectric absorption/polarisation/leakage can be a problem even at medium impedances if you look for the best accuracy (1ppm error for 100K is produced by 100G in parallel). Teflon is good as an insulator as long as you don't move the wire!

Cheers

Alex

[deadlylover]

Very enlightening, thanks Dr. Frank!

The R6581 can do OCOMP from 10Ohms-1MOhm, the manual says add 2ppm gain error for the 100k range, and 5ppm for 1Meg.

[Dr. Frank]

What instrument did you use to get the impedance graphs over frequency?

I used an HP4149A impedance/gain-phase analyzer.

[Vgkid]

Thank you for making these post Dr.  Frank, they have been very informative.
Of the 2 bench meters I own only one of them is capable of ocOhms. My 3456 does this, but when coupled with 100nplc it is a very slow process, only working as a 4-wire keasurement until several 10's of Kohm.

[Kleinstein]

@HighVoltage: The change in inductance is a different thing: its just the not fully closed flield that is still incluenced by metal in some distance. With a core that is not alsolutely free from magentistriction you also have to care a lot about mechnical force on the core - this can also change a lot - so may need to keep it leveled to keep gravitational effects at least constant.

I was allways wondering how they can measure high impedances (not just 100 K) so fast - so we see that we have to be carefull and it does not really work well so fast.  Interesting that with insulators it's not just leakage currents and the capacitance but also dielectric absorbtion. So we can't just rely on the RC time constant beeing  short - capacitance can still get us.

[robrenz]

Excellent post.

I wonder what aspects of this relate to offset compensated low ohms measurements on the Lom510A which uses 1A drive current for 166ms out of each 1 sec measurement interval. Obviously a problem on resistors with any significant inductance.

[Kleinstein]

The inductance with low ohms resistors might in principle have similar problems, but usually you avoid a ferromagentic core in a parasitice inductor, not normally no relaxation to cause problems. Only the resitor material might be just on the border of getting ferromagnetic - so some magneto resitance might happen.

The supprising thing with the parasitic capacitance is that the time constant is notjust  given by RC, as one normally learns in school. The tricky part here was that there is also a much slower dielectric relaxation that can cause trouble. The simple RC current decays exponetially, so one can wait for something like 12 time constants and we are OK to the PPM level. But with an internal relaxation, a small fraction of the capacative current come much later, with time constant set by the material, not the RC combinaton.

Even worse, the worst case relaxation current goes down quite slow. So just waiting may not be a solution to the ppm level.

[babysitter]

Very interesting.  Looking at the ocomp measurement waveforms with an oscilloscope before the "valid" measurement might hint at troubles if the voltage over the DUT is not stable. Also, the classic "component tester" setup with very slow excitation frequency (arbgen or programmable PSU) might indicate dielectric relaxation.

I wonder if the "ham radio standard microwave oven test" might be used for qualifying isolation materials - put your plastic in a microwave oven together with a glass of water, turn it on and see if your material under test gets hot - then, something is trying to follow the field.

[lukier]

Thank you for this detailed investigation.

Recently I got 3457A that has OCOMP and I've read the manual, journal, schematics, but I wouldn't suspect parasitics affecting the measurements. Very interesting indeed.

BTW I'm not 100% sure, but I think 3457A has true 100 NPLC (unlike 3458A 10 NPLC + digital). With AZ and OCOMP it takes ages to get a reading

[Kleinstein]

The microwave ofen test will test dielectric absorbtion in the GHz range. The problem here is dielectric absorbtion in the Hz region. So 9 decades slower. You would have to do the microwave test at a much higher temperature to speed things up (e.g. 200 K higher) - thus to hot for most plastics.

The more sensibel way would be to measure the voltage over the resistor, after diskonnecting a voltage (e.g. 1.5 V or more if self heating and resultiing thermal EMF of the resistor is not a big problem). If there is significant current from dielectric absorbtion this should show up as it's rather close to what happens in the ohms measurement, in the compensation phase.

So ideally the meter should not just wait a predefined time (even something like 5 s might not be enough) but do a measuremt over time and warn if there is to much "drift"  / residual current from relaxations.

Unless you use very long or shielded cables there should not be so much capacitance to get in the 100s of pF range.
As shielding might be a good idea one might have to think about driven shields when doing very accurate higher ohms measurements - not just GOhms resistors but allready if Gohms in parallel might be a problem.

[Vgkid]

BTW I'm not 100% sure, but I think 3457A has true 100 NPLC (unlike 3458A 10 NPLC + digital). With AZ and OCOMP it takes ages to get a reading

Sounds similiar to my 3456A, so that is most likely correct, I havent checked the output for that.

[Andreas]

Hello Frank,

that is probably also the right explanation for the problems that I had with the AC-Multiplexer on my T.C. resistor measurements.
Above 1kOhm (e.g. 12k)  I got unreliable results (but I also have EMI capacitors in the range of some nF which have to be charged every time).
The self adjustment time after switching the MUX of 80ms before the measurement actually starts was not sufficient to give reliable measurements.
Although more than 20 Tau are within those 80 ms.

With best regards

Andreas

[TimFox]

I believe that I have seen a similar problem with 1 Megohm wire-wound resistors in IET (originally GR) SR1 resistor standards.
I have two 1 M units that give unstable short-term readings on 6.5 digit DMMs, while a normal RN65 1 M metal film gives stable short-term readings.

[TiN]

Hi Frank,

That's for thorough investigation, this is interesting. Good trap for young players, to check and triple check everything, not just believe "magic features" to improve performance.

K2001 have max 200k range if OCOMP enabled, while 2002 is 2M max autorange. I can check timings/capture some data to add for your measurements.
There is not much needed to upset 100K resistance reading as even 10GOhm in parallel with 100K gives 10ppm error.

What was your current leakage setup? 32V PSU in either polarity to case and floating terminal of resistor?

Also to fully complete this experiment would be cool to have guarded measurements test.
I'd expect those should resolve issues with OCOMP even without delay.

[Dr. Frank]

Hi Frank,

That's for thorough investigation, this is interesting. Good trap for young players, to check and triple check everything, not just believe "magic features" to improve performance.

Hi TiN,
I'm sorry, but I'm already an old-player..
Everybody will fall into that trap, as this is not explained in the documentation.

K2001 have max 200k range if OCOMP enabled, while 2002 is 2M max autorange. I can check timings/capture some data to add for your measurements.
There is not much needed to upset 100K resistance reading as even 10GOhm in parallel with 100K gives 10ppm error.

Well, please do so. I expect the same effects. Keithley instruments prefer shorter NPLC numbers, obviously. So NPLC 1 will even worsen the measurement.
Isolation resistance is another subject, which is not covered here, and which will provide a constant decrease of the reference resistance, but which will not be affected or mitigated by the delay.

What was your current leakage setup? 32V PSU in either polarity to case and floating terminal of resistor?

This setup was for demonstration purposes only..not for quantitative results.

Yes, I applied +/-32V between the case and one jack.
The resistor was applied, so you would probably measure the sum of all three jacks..

I also measured one jack w/o a resistor connected, like on the picture of the interior of the box.
Anyhow, I found loading time & current of the same order of magnitude.
You also have to scale with the 1V or 5V which are normally applied

Also to fully complete this experiment would be cool to have guarded measurements test.
I'd expect those should resolve issues with OCOMP even without delay.

I did that, but the passive guard of the 3458A did not change anything.

An actively driven Guard only may help, I assume.
That can be found on some Keithley electrometers only, using Triax cables.

[Vgkid]

I believe that I have seen a similar problem with 1 Megohm wire-wound resistors in IET (originally GR) SR1 resistor standards.
I have two 1 M units that give unstable short-term readings on 6.5 digit DMMs, while a normal RN65 1 M metal film gives stable short-term readings.

The ESI SR1 resistors should be flat mica card wound, not much inductance.

[TimFox]

Yes, they are mica-card-wound resistors.  Perhaps the problem is due only to the shunt capacitance in the aluminum box?