DC 1 A current measurement linearity at ppm level?

[e61_phil]

HighVoltage's data with polynom fit:

Polynom? But at first order?


Now we have five unknown devices. Is there a mathematical way to find the best one?

[TiN]

Yes, code for line 3:

Code: [Select]

p3 = np.polyfit(real3,ideal,1)
pv3 = np.polyval(p3,real3)
diff3 = ideal-pv3
diff_ppm3 = (diff3/1.0)*1000000.0

Best unknown is still uknown. Without the calibrated reference, you can only compare unknowns, but which one is closer to truth? 

[e61_phil]

Best unknown is still uknown. Without the calibrated reference, you can only compare unknowns, but which one is closer to truth? 

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

Replace "two" by n

As one can see the graphs depend on what is fitted to what and that gives completely different results. Would be interesting to find the "best" of the options.

I don't agree with the calibrated reference. You can learn a lot through a step response.

[TiN]

I don't agree with the calibrated reference. You can learn a lot through a step response.

Didn't say we can't learn anything yet. I just mean we don't know real current values, only can estimate the stability of unknowns over the currents span.

I have completed some additional tests, using 4-terminal 1 ohm resistor as a shunt, with 3458A as detector on 100mV range.
Sweeps are 100mA, to keep thermals reasonable, as we don't have oil bath here .

I have used this resistor, build during livestream for first test. It is simple VPR221 1 Ohm 4-wire TO-220 BMF in a cast metal box, with Low Thermal posts.

Loop:
a) 3458B on 100mA range
b) F8508A, on 190mA range
c) K2002 on 200mA range

Plot using calibrator programmed value as ideal reference.


Now same data, but using reading from 1ohm/3458A shunt setup as ideal reference.



And repeated sweep but with IET SRL-1 standard resistors instead. Calibrator as ideal reference.
Please note, SRL-1 sweep is only -100 mA to 0 mA, so it's only half of the graph above.

 

Now same data, but using reading from SRL-1 1ohm/3458A shunt setup as ideal reference.



So far I can only see that 8508 and calibrator are tracking the best, while 3458B being the worst.
DIY 1 ohm also tracks well, better than SRL-1 in these coarse tests.

[maxwell3e10]

I ordered 0.5 Ohm Vishay Foil Resistors ‎Y14670R50000B9L‎  and Y16900R50000D9L‎ from Digikey. They claim 0.05ppm/K TC and 4ppm/W power coefficient, so in principle should be linear to 2 ppm. If the two resistors give similar results, perhaps one can trust it since they have slightly different construction.

[Dr. Frank]

I ordered 0.5 Ohm Vishay Foil Resistors ‎Y14670R50000B9L‎  and Y16900R50000D9L‎ from Digikey. They claim 0.05ppm/K TC and 4ppm/W power coefficient, so in principle should be linear to 2 ppm. If the two resistors give similar results, perhaps one can trust it since they have slightly different construction.

That's only correct, if you were able to stabilize / fix the surface temperature of the resistor.

So you have to mount it on a heat sink, and force-cool that to 25°C, see note (1) in the datasheet (of similar devices).
You might as well heat it to an appropriate temperature above R.T. and stabilize / ovenize that temperature.

That's exactly the remark, e61_phil has made, about the internal heat-resistance.

Worst case T.C is 2ppm/K, so at 4ppm/W, this internal thermal resistance might be about 2K/W.

If you just place the resistor on a heat sink, and let it warm up freely, then you'll get a much higher thermal drift, about 20 ppm, if the heat sink warms up from 25°C to 35°C (estimated) at 0.5 W P.D.

Frank

[maxwell3e10]

A big heat sink should have a thermal resistance of about 2 K/W. If one can believe that TC is typically less than 1 ppm near room temperature, than overall rise of the heat sink temperature should not have a big effect. Its mostly the internal temperature rise, which should be covered already by the power coefficient. But we'll see what happens in reality.

[e61_phil]

If costs aren't the biggest issue you could ask wekomm. I talked to a retired PTB guy some months ago and he told me wekomm has developed very stable 1R resistor. He said with this resistor they were able to measure 1A with an overall uncertainty of 5ppm.

[HighVoltage]

Today I tested this again.
This time with a different Burster 1240-1, SN: 610912

Fixed current range of 1A on all DCI instruments
Source: Keithley 2460
All Instruments had warmed up for 4 hours.
After changing the current, I waited 30 sec, before taking the readings.

Loop:
A) Burster 1240-1 Ohm on 34470 (1), 1V fixed range, 100 NPLC + Smoothing Filter ON
B) 34470A (2), 1A fixed range, 100 NPLC, + Smoothing Filter ON
C) DMM7510, 1A fixed range, 5 NPLC, + Smoothing Filter ON
D) 3458A, 100 NPLC

May be you guys can do your magic with the graphing again.

[tszaboo]

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.

Thermal EMF matters, because the sense leads need to be the same temperature to not create errors. Due to the physical layout it is usually not the case, also, often times sense resistors are not symmetrical in construction. It can be observed if you have a precision current source, and reverse the current on the DUT, the same current leads to different reading, while testing the same analog frontend, the offset is much smaller.

Keysight in their newer meter uses this as shunt:
https://www.isabellenhuette.de/fileadmin/user_upload/SMV.PDF
I tested some samples (10) from Isabellehütte, against a 3458A and other high precision shunts, and they showed excellent stability, much better than datasheet spec.
Measuring 1A is not easy, as I found (I had an error budget for a current source and sink of 300ppm over temperature, time humidity etc, few thousand was made from this). Most current shunts are low ohmic value and generate only a few mv or they have too much tempco for anything precision. If you have big budget, or it is a one off, there is always the RUG-Z, they make it up to 100Ohm, down to 1ppm, and with 250W rating self heating is not an issue. Just be careful with long term stability.
It is always easy to improve current measurement of a multimeter with an external shunt. In a 1000 EUR meter, they have a budget like 3 EUR for the shunt, if you need more precision, do it externally.

[HighVoltage]

I had the constant 100 mA running for over 1 hour and recorded the current.
Here are the result.

[tszaboo]

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.

Thermal EMF matters, because the sense leads need to be the same temperature to not create errors. Due to the physical layout it is usually not the case, also, often times sense resistors are not symmetrical in construction. It can be observed if you have a precision current source, and reverse the current on the DUT, the same current leads to different reading, while testing the same analog frontend, the offset is much smaller.

Keysight in their newer meter uses this as shunt:
https://www.isabellenhuette.de/fileadmin/user_upload/SMV.PDF
I tested some samples (10) from Isabellehütte, against a 3458A and other high precision shunts, and they showed excellent stability, much better than datasheet spec.
Measuring 1A is not easy, as I found (I had an error budget for a current source and sink of 300ppm over temperature, time humidity etc, few thousand was made from this). Most current shunts are low ohmic value and generate only a few mv or they have too much tempco for anything precision. If you have big budget, or it is a one off, there is always the RUG-Z, they make it up to 100Ohm, down to 1ppm, and with 250W rating self heating is not an issue. Just be careful with long term stability.
On the other hand, It is always easy to improve current measurement of a multimeter with an external shunt. In a 1000 EUR meter, they have a budget like 3 EUR for the shunt, if you need more precision, do it externally.

[maxwell3e10]

Here is the plot of the latest HighVoltage's data. It looks similar to before. DMM7510 has the smallest deviation, about 5-10 ppm. This time I dropped all points at -0.1 A because they were outliers.

For the time dependence, it would be interesting to measure at full 1 A for maximum thermal effect. Maybe measure all the meters at the same time to compare the transient.

[TiN]

Before going to next step, I ran also lowish current sweep, 19mA, in same setup with SRL-1 in loop.



As expected, 1 ohm + DVM give very bad result due to voltage noise.
a) 3458B show no problems on this span (using it's 100mA range), about +/-2ppm INL
b) 8508 stayed well under +/-0.5ppm vs source.
c) K2002 was within +/-3ppm vs source, using it's 20mA range.

[TiN]

If you guys tired of my colorful wiggly lines, let me know, I'll stop.



First try with DCCT (green line). 3 turns for primary, 10 ohm VHP-4 as burden resistor on secondary.
Setup photo, excuse the mess:

[Pipelie]

interesting， the curve shape of SRL-1 is kind of similar to the DCCT's.
I wonder which one is more close to the “truth”

the linearity SPEC of LEM IT-600 is less than 1ppm as good as I remember.

[HighVoltage]

If you guys tired of my colorful wiggly lines, let me know, I'll stop.

Total opposite, just keep going.

Amazing stability of the Fluke 8508A.

BTW, is that your 8508A, did I miss your post on it ?
 

[MiDi]

BTW, is that your 8508A, did I miss your post on it ?

Just my thought 

[Pipelie]

 
I don't think so.

He wishes but it's rental.

[maxwell3e10]

Here are the data for two 0.5 Ohm Vishay Foil Resistors ‎Y14670R50000B9L‎  and Y16900R50000D9L‎, plotted against each other.

They were mounted on two large heatsinks sitting in the lab and simply wired to BNC connectors for current and voltage sense. 
One is measured by DMM7510, the other by Keysight 34470A. We took four sweeps stepping the current up and down, with 1 minute dwell at each value. It looks pretty good, at a level of 1-2 ppm, at least up to 0.6 A.

[e61_phil]

Thanks!

BNC is not the best connector for DC precison.

I would also measure the step response. That will give good insights and one can rule out many external effects (source INL and so on)

[Kleinstein]

At the 1 ppm level one may also have to check if the INL of the DMM used to measure the voltage is good enough. In the low range, it is not just the ADC in the DMM, but also the amplifier that can contribute. INL in the 100 mV/200mV range can be worse than the prime 10 V range - though it's also more difficult to measure because of noise and thermal EMF.

For understanding the thermal contribution to INL, it might be worth to measure the thermal resistance (temperature rise on current) and the TC of the external shunt. For the temperature rise it is of cause only the case that can be measured, but that would at least take the heat sink out of the unknown.

As the thermal effect gets larger with more current (expect I³ for the TC related effect), one could extend the test to slightly larger current than one would normally us the shunt for a precision measurement. For the 0.5 Ohms resistors from Maxwell the 0.6 A test current is already such a case - this would be a 300 mV burden and thus more than normally used for high accuracy in this current range. At half the current it would be only 1/8 the error to be expected (for heating and TC).