HP 3458A fails calibration in 100mA range

[e61_phil]

Hi,

last week one of our HP 3458As fails the calibration in the 100mA range (limit is +/-49ppm). But I'm wondering why.

My simple understanding of the ACAL of the 3458A was this: Every time I run an ACAL all ranges will be adjusted against the two internal standards of the 3458A. Which are the 40k Vishay resistor and the LTZ1000. Therefore, I thought if the unit isn't broken and the 10V measurement is ok and the 10k measurement is also ok, all other ranges should be ok, after an ACAL.

My second assumption was/is the artifact calibration (which seems the only way of adjusting a 3458A) will allow the 3458A to adjust the stored values for the 40k and 7V, nothing more. Based on the new values all ranges are adjusted through ACAL.

I attached some values from the cal certs. The red values are measured before the artifact calibration and the green ones after.

I'm not sure which voltage range is used for 100mA, it seems to be the 100mV Range. Therefore, I noted the 100mV values also.

So, erverything looks fine except the 100mA range. How is that possible? And why is an artifact calibration able to cure that? The values for 10V and 10k where nearly perfect even before the artifact calibration. And why will a new ACAL run not bring the 100mA range out again?

Thanks
Philipp

[TiN]

It is bit unclear what is your exact question. You wondering why ACAL caused 100mA fail calibration before, or why it is not failing now after external calibration?

You assumption pretty was close, but only 50%, and ACAL operation described in good detail in calibration manual in appendix B. Calibration for current sources used in resistance/current also performed by using external 10 KOhm standard, not only measurement of internal 40KOhm reference resistor. As result external calibration provide:

Externally Derived Calibration Constants
Offset Constants: DC volts, 0.1 V to 10 V ranges, Front and rear input terminal paths; 6 offset constants
Two-wire resistance, 10 ? to 1 G? ranges, Front and rear input terminal paths; 18 offset constants
Four-wire resistance, 10 ? to 1 G? ranges, Front and rear input terminal paths; 18 offset constants
Internal Reference Constants:
Voltage - value of internal reference voltage; 1 constant
Resistance - value of internal reference resistor; 1 constant

And

Of the remaining 209 calibration constants in the instrument, 6 are determined through one-time external calibrations. These constants provide adjustments for frequency and period measurements, time
base interpolation, and the high frequency (beyond 2 MHz) AC response. This is SCAL calibration.

The remaining 197 constants are determined through internal ratio transfer measurements as previously described. These constants are also updated each time auto-calibration (ACAL ALL) is executed, reducing time, temperature, or environmentally-induced drift errors. This capability enhances measurement accuracies over extended time intervals and operating temperature ranges.

[e61_phil]

It is bit unclear what is your exact question. You wondering why ACAL caused 100mA fail calibration before, or why it is not failing now after external calibration?

My question was: If 7V and 40k are ok, why is the 100mA not moved away again by ACAL. But you explained it already. The artifact cal will do more than "adjusting" 40k and 7V.

Thanks! I will read the linked manual now

Conclusion for me is now: You can't trust an uncalibrated HP 3458A even if 10V and 10k are in perfect conditions (you shouldn't do it was clear to me, but I thought it can be done).

[TiN]

You can trust 3458A (and any other meter in this regard) only after you run full performance verifications and all functions for multiple times with desired period between the verifications, and ensured that all functions are meeting the specifications. Expecting otherwise, including judgement from single calibration does not provide you any confidence.

[Dr. Frank]

I don't understand, how this probable ACAL failure has been detected.
I assume, the 3458A passed its ACAL ALL procedure w/o error.
You're speaking about a cal certificate, so has your 3458A been calibrated AND verified in a cal lab, and they detected that problem during verification?
As the 3458A makes successive internal calibrations by 10:1 transfers, the 1A should also be out, as it might depend on the foregoing 100mA range auto calibration.

Maybe there's a problem with offsets, or the current relay is defect.

Frank

[e61_phil]

You can trust 3458A (and any other meter in this regard) only after you run full performance verifications and all functions for multiple times with desired period between the verifications, and ensured that all functions are meeting the specifications. Expecting otherwise, including judgement from single calibration does not provide you any confidence.

Yes, and even a full verification in a cal lab wouldn't show any possible errors

I don't understand, how this probable ACAL failure has been detected.
I assume, the 3458A passed its ACAL ALL procedure w/o error.
You're speaking about a cal certificate, so has your 3458A been calibrated AND verified in a cal lab, and they detected that problem during verification?
As the 3458A makes successive internal calibrations by 10:1 transfers, the 1A should also be out, as it might depend on the foregoing 100mA range auto calibration.

Maybe there's a problem with offsets, or the current relay is defect.

Frank

Yes, there wasn't any error display on the 3458A.

I think the calibration procedure we use here is the normal procedure:

The unit is verified on all ranges with an calibrator (or more depending on the instrument) and a calibration certificate (with values) will be created. This certificate says fail or pass. If everything is fine, the instrument will come back to our company.

If the instrument fails the verfication, the cal lab will try to adjust the failed range (or in case of the 3458A an artifact calibration will be done). After the adjustment, a new full verfication we be done with a second calibration certificate. So, if the instrument fails you get two certificates.


The 1A range was -23ppm compared to the FLuke 5700A before artifact calibration and -13 after.

[Cerebus]

Thar's ya problem:

The 1A range was -23ppm compared to the FLuke 5700A before artifact calibration and -13 after.

From Fluke 5700A specifications:
DC Current, 2.2A range, 99% confidence interval, 24 Hr absolute uncertainty +/-60 ppm + 15 nA, 90 days +/-70 ppm +15 nA

From 3458A specifications:
1A range 24 hour +/- 100ppm +10ppm, 90 day +/- 100ppm +10ppm

I think there's a little bit of an issue here with test uncertainty ratios.

[TiN]

5700A can be brought to 24hour spec, using ACAL as well. And for 100mA output 220mA range would be used, which at 95% (more typical for cal labs?) for 5700-II has 45ppm +/-0.8uA.
Also since 3458A shunts are all in series change/deviation in bottom shunts affect all the ranges (e.g. drift in 0.1 ohm resistance shunt for 1A would also cause extra error in 100mA range). I've learned this the hard way, which led to temporary suspension of my 3458TE experiments .

Something tell's me that observed error was due to PCR effects. However I'll let you fill be blanks, as we have no idea of calibration/verification protocol, and no measurement uncertainty was expanded. Without these time consuming steps rest is just speculation.

[e61_phil]

You are right, the uncertainty of the 5700A isn't low enough for a real 3458A verification. But we are aware of that fact. In the end the calibration certificate will count and not the 3458A datasheet. That isn't a big issue for us. And it is like Tin said: They run an artifact calibration of the 5700A before verifying a 3458A.
It was just my personal interest, how a single range can fail with ACAL without a defect.

@TIN: What dou you mean with PCR?

[Kleinstein]

I see 3 possible reasons why the 100 mA range might be bad despite ACAL being successful. One would be intermittent contact problems with one of the relays K203 / K204. It looks like there is no check for excessive drop at the relays and thus possibly current leakage through the protection - boots-trapping at the protection seems to by off the main amplifier, which is a little odd.

The second effect not fully compensated with ACAL is the power coefficient of the shunt resistors (PCR). This is mainly self heating if the shunts are tested with higher currents. ACAL for the 100 mA range is done with a rather small current. If PCR is indeed the problem, there is not much that can be done with a new calibration - it would be a problem of a relatively high TC of the 100 mA shunt.

A third (though unlikely) possible problem could be due to DC offsets. At least from TiNs description these are adjusted during artefact calibration. Due to the small current and thus low voltage, the ACAL process is especially sensitive to offsets for the 1 A and 100 mA ranges. However ACAL might (ideally should) work without the offsets - still it is question of how much time is spend on the ACAL. Ideally there would be the option of a separate slower more accurate ACAL for partial ranges - this could help especially with the 100 mA and 1 A.

[dacman]

The artifact calibration of the 3458A should include CAL 0 with a copper short (both front and rear terminals) which the ACAL does not do.

[e61_phil]

The artifact calibration of the 3458A should include CAL 0 with a copper short (both front and rear terminals) which the ACAL does not do.

Hmm, anyway it is imho a bit strange. An offset voltage in the input path should effect voltage measurements, but not current measurements.

[dacman]

I think the 3458A is highly reliable, but glitches do occur.  I've sent them in (to the manufacturer) for what turned out to be relay issues.  (I've also sent one in for what I though was a relay issue, but turned out to be the calibration RAMs.)  I run the ACAL daily (at my day job), whether or not I intend to use one, if for no other reason than to exercise the relays.

[dacman]

In the 3458A Calibration manual, both DCV and DCI Offsets are checked.  Without that information (which is not in the tables given) it cannot be determined whether the problem was an offset or a gain error.  (I would suspect it was an offset error.)  If these are not tested, I'd recommend adding them.