DC 1 A current measurement linearity at ppm level?

[maxwell3e10]

Has anyone looked at linearity of high current ranges on precision DMMs (3458A, 34470A, DMM7510)? I need to do tests where the current varies between 0.1 and 1 Amp over a few minutes and am wondering how much non-linearity one can expect from shunt self-heating?

HP3458A specifies only 100 ppm short-term accuracy on 1 A range. The shunt resistor is 0.1 Ohm and has TC of 5 ppm. Why is the specification so poor? Is it just because of the internal resolution in the ACAL measurement of the shunt resistor value? So maybe the absolute accuracy is poor but linearity is much better?

[TiN]

Interesting questions. If you define step delay time, I can help you test linearity in 100mA to 1A range. Or -1A to 1A in 10mA steps maybe?

[Dr. Frank]

 I also was always wondering about  these bad specifications of the 100mA, 1A ranges.

In the CLIP, these range shunts consist of R212, a 1 Ohm, T.C. 50 ppm/K, and R213, a 0.1 Ohm, T.C. 5 ppm/K resistors.

The T.C. w/o ACAL gives the apparent T.C. of these ranges, and that's about 28 ppm/K.

That's already an obvious contradiction between the specification of these resistors and the ranges.

During ACAL Ohm, that big T.C. might affect the 10:1 transfer process, which itself is obviously precise to < 10 ppm, as with the lower ranges, but it will add up into the 24h accuracy specification for 100mA and 1A.

The strange DCI specifications might also be caused by the internal thermal voltages, which can't be auto-zeroed during the measurement.. So HP included these effects generously into the T.C. spec.

If you look closely into the schematics, you'll see these strange solder joints/bridges JM 201..208. These might be used to compensate for equivalent solder junctions of the shunt resistors.

That magic stuff kept me away from replacing these shunts by better ones.

In the end, to verify the real behavior of your specific instrument, it would require a calibrated and variable current source being at least 10 times more uncertain than these two ranges, i.e. well below 10 ppm.. even the 57xxA calibrator @ 40..60ppm won't do the job, I fear.

Frank

[TiN]

Yes, I actually tried replacing shunts before, and it actually made TC much worse for medium current ranges, even though replaced resistors itself are way more stable.



57xx specifications are rather conservative, so there could be some good data to see.  And since we can use multiple DMMs , it's possible to estimate difference between different units on same source. But self-heating and PCR of the involved shunts/source parts is important, that's why I want hear from maxwell3e10 on specific delay times and such.

I also have few of these DCCT heads, which specified for current linearity <1ppm.



But for 10-2000mA that would need some more work. I already have CSNG 2.5 ohm shunts for them though.

[e61_phil]

I saw up to 35ppm deviation between ACALs on the 3458A depending on what is happend closely before ACAL. If you run the meter with 1A and than run ACAL it is far away from an ACAL run without current before.

The spec of the 8508A is worse, but the 1A (2A) range seems to be much better than the 3458A. We will see after the next calibration of the 8508A...

You don't need to have a super accurated 1A source to test the meter. You only need a super stable one. A 57xxA should be good enough for such a test. One would short the input of the 3458A current input (use a resistor in series, to lower the residual current through the 3458A even further) and than let the calibrator and the monitoring shunt/DMM stabilize. After the stabilization open your short and monitor the behaviour of the meter under test. This can also be done with a jump from 100mA to 1A.

I played with some Burster 1240-1 resistors, a Fluke 8508A and two 3458As. The 3458A had the best specs, but the worst behaviour. (Some 1240 were also bad)

[maxwell3e10]

Thanks for the feedback. We have done measurements with about 10 steps from 0.1 to 1 A with about 2 minutes on each step. The obvious test is to compare step-up and step-down measurements and they give opposite sign of non-linearity.  Also tried random steps from low current to a particular point and then back or alternating back and forth between a high and a low value while keeping the average roughly constant. All these measurements give different results!  The question is if it is from our DUT which includes magnetic materials that can have hysteresis, or from the meter. So far the meter we used is DMM7510 (but if HP3458A or another meter is clearly better, its a good excuse to get one ).

It would be interesting to find the time constant of whatever drift is happening. Perhaps it would have better linearity if one only measures for a couple of seconds at each point, we haven't tried that yet.

Fundamentally, what is the source of the drift and large TC? Could be thermocouple effects, I suppose. For 1 A on 0.1 Ohm shunt, 1 ppm corresponds to 100 nV. With Seebeck coefficient on the order of 1 uV/K it wouldn't take much temperature gradient in the meter to mess things up. A shunt might have a power dissipation coefficient of about 10K/W, so one can see about 1 K changes. Also I am wondering if whatever alloy is used for shunt construction would have a particularly large Seebeck coefficient. Perhaps there is a compromise for a material with higher TC but lower thermal EMF. Also mounting and connection symmetry would be  crucial.

[e61_phil]

I think 2min are too short for stabilization. I would try to make one bigger jump instead of the smaller ones to learn a bit more over the system.

If your system is fast enough you can turn the polarity to overcome thermoelectric effects. Even if you can't use this in the end, it will help to learn were the sources of error are.


Another approach could be to build 10x 100mA current sources. Every source with a switch to switch the current in your DUT or around. You can measure every source with very high relative precision (because it is always almost the same value) and the sum of the currents should be the "real" current. If you are really interested in ppm current linearity I wouldn't rely on any meter with specs which doesn't fullfill your needs.

[maxwell3e10]

If your system is fast enough you can turn the polarity to overcome thermoelectric effects. Even if you can't use this in the end, it will help to learn were the sources of error are.

Maybe one way to do it is to setup a make-before-break reversal switch across the DMM, so one can switch the current from positive to negative through the DMM without disturbing the system too much.

[TiN]

I guess this is a "teaser". Data still collecting, 3458A NPLC50 input direct to AUX output on calibrator, locked on 220mA range. 
Standard deviation on points about 40-50 nA.



-110 to 110 mADC sweep first, then will do -1A to 1A sweep. Just one meter for now.

[e61_phil]

Especially for the -1A to 1A sweep I would always step between positive and negative for better stabilization.

For example: 10mA, -10mA, 20mA, -20mA...

If negative is interesting at all.

[maxwell3e10]

Yes, its a question if one tries to minimize the thermal effects or just measure how big they can be. Also its not obvious which effects will be even with current and which will be odd. If its a thermocouple-type shift, it maybe odd.

[e61_phil]

I wouldn't sweep two unkown devices against each other. I think one can learn more with a current step on the DUT.

[Kleinstein]

For high accuracy current measurement at hight currents, I would consider an external shunt, that can be higher power and if needed could use power compensation. So have a heater coupled to the shunt and use it to keep the power / temperature constant.

I know of 3 effects that contribute to shunt nonlinearity:
1) Self heating and the TCR
2) Self heating and thermal EMF
3) magnetic effect from the magnetic field the test current produces itself. This tends to push the currents to go a path more to the outside of the resistor and thus cause an increasing resistance with higher current. I don't think this would be relevant at 1 A already. There could be magnetic effects in some alloys too.

Shunt alloys are usually low thermal EMF, but of cause nothing is perfect.

[maxwell3e10]

What is a good shunt resistor to use? I've seen some people mention Burster 1240. Also on Newark site there are these Alpha resistors which have TC of only 2.5ppm/K, does anyone have experience with them?
https://www.newark.com/alpha-electronics/pcyr1000b/res-metal-foil-0r1-0-1-750v-axial/dp/96Y7893

[e61_phil]

A couple of weeks ago, I played with some Burster 1240. Most of them were far better than specified. But not all of them were very good.

Thermal compensation can be quite difficult. One have to make sure, that the resitive element itself stays at a constant temperature. But there will be always thermal resistance between the sensor and the shunt resistance. The thermal flux through this thermal resistance will change with applied current. Therefore, the shunt resistance temperature will change, even if your heater is perfectly stable.

[TiN]

Well, even with zero effort, -1A to +1A sweep looks not too awful. 

This time I connected more DMMs, in a single loop (source Iout => 3458A I -> 3458A LO -> 3458B I -> 3458B LO -> FLK I -> FLK LO -> Source return).

Here's slower run with NPLC 50 on HP meters:



Rerun with NPLC 10 instead, and Fluke configured in 10A range.

[maxwell3e10]

Interesting that the two HP3458 behave so different. Are they from a different era or modified in some way?

The third order non-linearity suggests a thermocouple effect.

[TiN]

Yes, one is modified, as outlined above  . I will let you guess which one for now, sorry 

[HighVoltage]

I made a similar Test a few weeks ago to test a 1 Ohm Burster 1240

Source: Keithley 2460
Manual setting from +100 mA to 1A then again from -100mA to -1A

Loop:
A) Burster 1240-1 Ohm on 34470 (1)
B) 34470A (2)
C) DMM7510
D) 3458A

Here are the raw data.
I am traveling, may be someone can put the data in to a graph

[Kleinstein]

Interesting that the two HP3458 behave so different. Are they from a different era or modified in some way?

The third order non-linearity suggests a thermocouple effect.

Thermal EMF would be more like a second power INL contribution, as the error would scale with the power and thus square of current.
Self-heating and TCR would cause a 3rd power contribution, as there is  the current as an extra factor.

[e61_phil]

I took HighVoltages data and tried to plot the INL. I choosed the 1240-1 as reference, because I think it is the best device in this list. The results looks wrong. There is a big outlier on the 3458A and the 7510.

@HighVoltage: Which 1240-1 do you use? And why are all values positive?

[maxwell3e10]

Great data, its exactly what I would have taken if I had all the equipment!

I threw away outliers at 0.1A (maybe different range was used) and also one point each at 0.8 and 0.9 A. Here is the plot. It looks like  34470A is pretty bad and DMM7510 is perhaps slightly better than HP3458A.

[HighVoltage]

I took HighVoltages data and tried to plot the INL. I choosed the 1240-1 as reference, because I think it is the best device in this list. The results looks wrong. There is a big outlier on the 3458A and the 7510.

@HighVoltage: Which 1240-1 do you use? And why are all values positive?

The two outliers are corrected, they were a typo. Enclosed is the corrected file.
The Burster 1240 was my "1.0-A" SN: 187251

I am back in the lab tomorrow (and we have a holiday), so I will repeat this test
with a second Burster 1240-1.

If I remember right, the 100 mA was measured in the 100 mA range
In my new test I will take all measurements in the 1A range.


 

[e61_phil]

The Burster 1240 was my "1.0-A" SN: 187251

Thanks! If this is the case, I expect even better results with another 1240-1.

[TiN]

HighVoltage's data with polynom fit: