Correct setup for resistance measurement for performing calibration adjustments

[Rax]

Hi all,
Did a search through the forum but didn't quite find something covering this methodologically.

I may start on adjusting the ohmmeter section of a meter and have standard resistors available, but not sure if I'm doing this correctly. This one type of measurement and calibration seems by far more finicky and hard to set up correctly than anything else (I'm willing to stand corrected if that's not quite so). May explain why one single measurement (I can't personally feel quite right even calling it a "calibration," as it's pretty much one single measurement and allows for no adjustment whatsoever anyway) is as expensive as it is. Almost - and in some cases exceeding - as high as the equivalent service for an entire instrument with multiple features and modes etc.

So what's the correct mode to do this? I've typically just hooked up two sets of probes - after zeroing them, if the meter needs it, with a short I cobbled together by connecting with heavy gauge wire four banana jacks - and connected them to the resistor. Depending on the resistor, this may be two sets of banana-to-banana cables, maybe one banana-to-banana and one banana-to-witches hat, etc.

BTW - are the large "Kelvin clip" four wire probes an absolute requirement? I feel they don't do much differently from a good quality set of four regular probes.

[TimFox]

Kelvin clips are easier to use.
If you use four generic clips, they can work on a leaded resistor if you are place the "current" clips outside of the "voltage clips".

[alm]

I may start on adjusting the ohmmeter section of a meter and have standard resistors available, but not sure if I'm doing this correctly. This one type of measurement and calibration seems by far more finicky and hard to set up correctly than anything else (I'm willing to stand corrected if that's not quite so). May explain why one single measurement (I can't personally feel quite right even calling it a "calibration," as it's pretty much one single measurement and allows for no adjustment whatsoever anyway) is as expensive as it is. Almost - and in some cases exceeding - as high as the equivalent service for an entire instrument with multiple features and modes etc.

If you look up the BIPM definition of calibration, that is basically a documented measurement. So calling it calibration would be appropriate. As for pricing, my guess is that it's because the labor is pretty much the same, since a DMM calibration is usually done these days with both a calibrator and DMM under computer control stepping through all the points. Although of course a DMM will take more time. it could also be a volume thing: the average cal lab will get a lot more DMMs to calibrate than resistors.

So what's the correct mode to do this? I've typically just hooked up two sets of probes - after zeroing them, if the meter needs it, with a short I cobbled together by connecting with heavy gauge wire four banana jacks - and connected them to the resistor. Depending on the resistor, this may be two sets of banana-to-banana cables, maybe one banana-to-banana and one banana-to-witches hat, etc.

What kind of device are you using to measure the resistor? At one end of the spectrum, a 3.5 digit handheld DMM, you would just use whatever leads. On the other end of the spectrum, a direct current comparator, you might use shielded four-wire connections, all copper connections, a very stable ambient temperature and an air bath for the resistor.

BTW - are the large "Kelvin clip" four wire probes an absolute requirement? I feel they don't do much differently from a good quality set of four regular probes.

No, not at all. Kelvin clips are just more convenient to connect than separate clips, and require less space to grip on to.

[Rax]

What kind of device are you using to measure the resistor? At one end of the spectrum, a 3.5 digit handheld DMM, you would just use whatever leads. On the other end of the spectrum, a direct current comparator, you might use shielded four-wire connections, all copper connections, a very stable ambient temperature and an air bath for the resistor.

To measure the resistors (my "standards" are ESI SR1s) I'd use a calibrated Fluke 8502A. I'd then use the "standards" (I put these in quotation marks because they're not the highest of standards, and I have not really "calibrated" them either) to adjust a Fluke 8505A I repaired/am repairing and calibrating.

[alm]

Here's my $0.10: 2 cents per paragraph

Both meters have a similar accuracy of about 30 ppm/90 days in the middle resistance ranges, so you won't be able to ensure the 8505A is in spec, but should be able to get it pretty close. I would definitely four-wire connections for everything under 1 MOhm. Thermal EMF could be a factor. The voltage while measuring resistance is about 1 V, and a copper-nickel junction produces about 10 uV/°C. So thermal EMF errors could easily be in the order of 10 ppm. See also Fluke metrology 101: Watch Out for Those Thermoelectric Voltages!.

To combat this, you can either (or both) use offset compensation if the meters support this, or use materials (copper) that minimize thermal EMF. Offset compensation involves the meter measuring the voltage with the current source enabled (as a normal resistance measurement), and then another measurement with the current source disabled (should measure 0 V in a perfect world), and subtract the latter voltage from the first. As you see in the tables in the documents I linked, thermal EMF depends on the type of metal, so you can minimize them by using all copper connections, but make sure the copper is not oxidized. A simple solution is using UTP (cat5/6/7) network cable, I like solid strands best. I connect the stripped strands directly to binding posts on both the DMM and device under test (e.g. resistor). If the DMM only has banana jacks (boo Keithley!), then you can insert the copper wires and then insert Q-tips to press them against the jack. I strip the outer sheath and extract individual pairs, but you could also leave the cable in tact especially if it's shielded. In that case, use the shield for guard connection (more about that later). Leaving connections for a while after touching them to wait for thermal equilibrium also helps with thermal EMF.

I believe the SR1 resistors are manganin, so they should have a somewhat substantial temperature coefficient in the order of 10 ppm/°C. And of course the meters will also have a temperature coefficient. So ideally try to match the ambient temperature to the one it was calibrated at as closely as possible, and definitely register the ambient temperature you do the "calibration" at. Some resistors, like the NBS/Rosa type (the L&N resistors you often see on eBay for ~$100), have a special hole for inserting a temperature probe. Here it also helps to let the resistor sit for a bit after handling, since your hands will heat it up.

If you are using a shielded cable, then connect the shield to the guard terminal at the meter end, make sure to enable external guarding (aka disconnect guard from LO) on the meter, and connect the shield to the lo force/current terminal at the resistor end. If the resistor has a shield, then connect this to the same point. If you are not connecting guard at the resistor end, then make sure the meter is set to internal guarding / connect guard to lo. The meters manual might give more information about this topic, but I think this application note gives a clear and quite deep treatment of the subject of guarding.

Then take a number of resistance readings, say 16, and check if there is an upward or downward trend (suggests something isn't fully stabilized yet) and check if the sample standard deviation is well below the target uncertainty (say << 30 ppm). For very high and low resistance values you expect a larger uncertainty. You will also need this standard deviation (well, the standard error of the mean, which is standard deviation divided by the square root of the number of samples) to calculate the uncertainty of your calibration according to the standard Guide to the expression of uncertainty in measurement if you want.

[Kleinstein]

The calibration normally does not use the kelvin clips, but 4 separate wires. The standard resistors usually have seprate connectors for drive and sense. The Kelvin clips are for convenient testing of components that don't have 4 wire connections, like normal resistors.

For the test with individual clips, it does not really matter if the current contacts are inside or outside. The reading should be the same, no matter which sequence.

The short for the zeroing should be an zero because of the sequence the 4 cables are connected, not because of thick wire. With the right sequence one gets a zero reading even with thin connections.
With 4 wire resistance more there is no need for heavy gauge wires - so 4 wires from an UTP cable are thick enough. The more limiting factor can be the isolation.
High resistance like 1 M and up may need care with the cable isolation (e.g. maybe usewell isolated posts to support the cable if needed).

[iMo]

.. You will also need this standard deviation (well, the standard error of the mean, which is standard deviation divided by the square root of the number of samples) to calculate the uncertainty of your calibration according to the standard Guide to the expression of uncertainty in measurement if you want.

After looking into the Guide I know now why we need the upcoming new quantum computers so urgently.. 

[mzzj]

With 4 wire resistance more there is no need for heavy gauge wires - so 4 wires from an UTP cable are thick enough. The more limiting factor can be the isolation.
High resistance like 1 M and up may need care with the cable isolation (e.g. maybe usewell isolated posts to support the cable if needed).

UTP cable insulation resistance is probably over 10 teraohms.
Brand name USB3 cables have also variety of wire sizes with screening and good insulation.

I’d be extra vary of any ready made ”laboratory” banana cables. These often have PVC insulation that easily measures below 1 gigaohm and can ruin your day even at moderately low accuracy levels like handheld dmm measuring 40 megaohms.

[Rax]

Been reading all this and doing tests and such, and I'm still not sure I'm able to acquire consistent resistance readings with my setup. Maybe there's a reason it costs so much.

I don't really have all-copper cables for all hookups for this, though I'm not convinced that is the reason for my conundrums. That said, I'm working on eliminating thermocouple effects. I feel I'm in pretty good shape for voltage measurements, but not so much this stuff.

Not sure how everyone else is doing (outside of the gurus) but maybe pics and relatively detailed lab setups would help? Hopefully, I'm not the only one facing similar roadblocks.

[RandallMcRee]

I briefly looked at the fluke manual.  Are you using four wire? If not, you should for normal high quality measurements.  You should also be using the sampling available on the meter. I would think 128 samples would be a starting point.

Also are there any differences between the front and rear jacks on the meter? Sometimes the rear jacks have more hookup options.

[Rax]

I briefly looked at the fluke manual.  Are you using four wire? If not, you should for normal high quality measurements.  You should also be using the sampling available on the meter. I would think 128 samples would be a starting point.

Also are there any differences between the front and rear jacks on the meter? Sometimes the rear jacks have more hookup options.

Thank you, Randall. I have not used the rear jacks at all.

I've always been using four wire, regardless of the actual resistance being measured. I've not tinkered with the sampling, just stayed with whatever default my meter thought is appropriate.

I feel one of my main issues is the zero-ing. The 8502A is not the best meter there is at this. I mean, certainly, this is the dragon of all metrology, says my noob anxiety mode. Next to ground loops and a few other gremlins.

[guenthert]

[..]
I don't really have all-copper cables for all hookups for this, though I'm not convinced that is the reason for my conundrums. That said, I'm working on eliminating thermocouple effects. I feel I'm in pretty good shape for voltage measurements, but not so much this stuff.
[..]

There *will* be thermal EMF.  With oxidized copper on copper pretty high even.  One can use either

i) use clean pure copper contacts everywhere (not always a realistic option),

ii) exercise patience (if ambient temperature is stable and little heat is generated locally, e.g. due to self-heating, then temperature difference will over time reduce -- this can take hours though)

iii) measure thermal EMF before/after measuring voltage across a resistor and subtract it later (some DMMs offer this as 'offset compensation') or

iv) measure multiple times and reverse polarity of the driving current, averaging the result and thereby cancelling the thermal EMF across the resistor (won't do anything about thermal EMF across contacts which aren't reversed -- if in doubt, reverse them all, not all at once of course), or

v) use (low frequency) AC to begin with, e.g. in a bridge using an AC null detector (e.g. a lock-in amplifier)

[Rax]

ii) exercise patience (if ambient temperature is stable and little heat is generated locally, e.g. due to self-heating, then temperature difference will over time reduce -- this can take hours though)

On this aspect - I think air baths were discussed as a stabilization/control idea. I'd appreciate some thoughts and suggestion on options with that.