3458A drift with high voltage applied

[chilternview]

I've two 3458A's, and only recently have a stable high voltage source to test that range. I'm seeing one of them drift (up to a couple of hundred uV) after applying 1kV. The other, connected to the same source, has negligible drift. Both have been powered up > 8hrs. The effect is only really noticeable at applied voltages of around 500V and greater.

Has anyone else seen anything similar?

[strawberry]

something with 10Mohm divider or around that area?

[chilternview]

Where is that - I can't see it on the CLIP?

[PartialDischarge]

I'm more curious about the HV source, which one is it?

[chilternview]

I'm more curious about the HV source, which one is it?

Datron 4000A. Previously I'd thought the drift was in the Datron as one of my 3458's was at Keysight for repair/cal. But when that came back, it confirmed the observed drift was the 1st 3458.

My earlier statement of it being a few hundred uV was incorrect. Over a 10min period it drifted up about 5mV on the bad meter; the good one was stable within 5uV.

[Kleinstein]

The 10 MOhms divider can show significant heating and depending on the TC of the divider the can cause some error.  The accurcy specs allow for  12 ppm * (U/1000 V)²  as additional error. So 10 mV at 1000 V input would be still in specs.  The resitor TC  can vary between units. The high votlage effect can also have some effects beyond simple self heating times TC.

[chilternview]

That's interesting, Kleinstein. The different units must have very different tempcos for the resistors in that case. Attached is the measurements over about 10mins from applying 1kV, the blue curve is the 'good' meter and the yellow the 'bad' one. Red is the room temperature.

[Dr. Frank]

I've two 3458A's, and only recently have a stable high voltage source to test that range. I'm seeing one of them drift (up to a couple of hundred uV) after applying 1kV. The other, connected to the same source, has negligible drift. Both have been powered up > 8hrs. The effect is only really noticeable at applied voltages of around 500V and greater.

Has anyone else seen anything similar?

Hello,
yes, of course, but it's quite difficult, or requires special equipment (i.e. the FLUKE 752A) to properly extract this power coefficient effect, and to quantitatively measure this non-linearity. Latter is different for your both 3458A, please provide the vintage of each of your instruments, or their S/N, best photos of their individual 100:1 HV dividers, as it's of general interest.

We had a longer discussion here: https://www.eevblog.com/forum/metrology/influence-of-switch-resistance-in-hamon-dividers/msg4221751/#msg4221751

First off, the 3458A is specified for 12ppm non linearity @ 1kV, as Kleinstein already stated.
The effect arises from the self-heating of this 10MOhm divider, which increases from 100V to 1kV. In conjunction with the T.C. of its 100k and 9.9M divider resistors, but more with their thermal imbalance (1mW on 100k, but 99mW on 9.9M), its ratio varies over power dissipation.
Any HV measurement @ 1kV requires at least 2 minutes of thermal stabilization, before a proper measurement can be made.
This stabilization effect occurs inside all DMMs and inside the HV source, or calibrator, or HV dividers, like the 752A!

There are several techniques to compensate for this effect, but the 3458A has none of these implemented!

1. This non-linearity is mostly quadratic, see 3458A specification.
During calibration of the DMM (DUT) @ 100V, 500V and 1kV (ext. precision voltages) it calculates the parameters of this quadratic effect.
When measuring voltages between 100V and 1kV in the 1kV range, it can then mathematically compensate this effect.
This method is used e.g. in the 34401A

2. The 100:1 divider, which is a thin film resistor structure on a ceramic chip, has another heater element on the back side of this chip.
The DMM then heats the chip externally, so that the dissipation by the input voltage plus the heater dissipation is constant, i.e. their sum is always constant 250mW.
This keeps the divider resistors nearly on a constant temperature, so that the T.C. and thermal mismatch is mostly cancelled.
This method is used in the 34410/411, in all TrueVolt KS DMMs (e.g. 34465A)
In the Fluke calibrators 5440, 57x0A, the temperature is directly kept constant by means of a temperature sensor.

3. Reduce the T.C. of the divider resistors to near zero by pairwise +/- T.C. matching, and make a thermal match between the different resistor, by e.g. distributing the power dissipation to many individual resistors. This method is used in the 752A, and in the Datron 4902S Reference Divider. The topology of latter distributes the divider equally on about 200 low T.C. resistors, i.e. 0.5mW per resistor, so the heating effect is neglectable.

For my 3458A from 2000, I determined a non linearity of about -1.6ppm, instead of 12ppm, see link above.
The 752A was required, because only this instrument has sufficiently low specified uncertainty or non linearity of 0.5ppm @ 1kV.

We already discussed, that different vintages for the 3458A had worse or better dividers.

Frank

PS: You also can extract / distinguish this heating effect between your Datron 4000 and the 3458A.
Best and safest way is to use a HV reed relays for this purpose.

Switch on 1kV on your 4000, and let it settle for several minutes. The HV will then be stable for the rest of the experiment.
Now apply this 1kV to the 3458A in its manual 1kV range, by engaging the HV reed relays.
Let the 3458A run at a fast rate, so that you can observe the drift from its initial HV reading to the stabilized value a few minutes later.
Best would be to sample the readings into memory, so that you can afterwards analyze the voltage rise and drift on your PC.

This difference is then your power dissipation effect.
You can repeat this experiment at 100V and 500V, to see if this is really about quadratic.

[chilternview]

please provide the vintage of each of your instruments, or their S/N, best photos of their individual 100:1 HV dividers, as it's of general interest.

Dr Frank, the 'good' meter is of 1989 vintage with serial number 2823A0022x i.e. an early one. It's had its A2 board replaced recently by Keysight. The 'bad' meter is of 1995 vintage, serial number 2823A1508x and its cal was 6 months ago. It was within guardbands on the 1kV range at the time.

I understand the self-heating theory, however if one can have negligible effect, or at least maybe 1/10th the other, could this be indicative of a faulty divider?

[Dr. Frank]

please provide the vintage of each of your instruments, or their S/N, best photos of their individual 100:1 HV dividers, as it's of general interest.

Dr Frank, the 'good' meter is of 1989 vintage with serial number 2823A0022x i.e. an early one. It's had its A2 board replaced recently by Keysight. The 'bad' meter is of 1995 vintage, serial number 2823A1508x and its cal was 6 months ago. It was within guardbands on the 1kV range at the time.

I understand the self-heating theory, however if one can have negligible effect, or at least maybe 1/10th the other, could this be indicative of a faulty divider?

No, otherwise, you have to determine the value of this power non linearity.
If it's out of spec (12ppm), only then it might be defective.
Several 100µV in relation to 1kV are <1ppm , so it's still very good, I think.

Dr. Philipp told me, that the older DMMs (e.g. 1989) had even better dividers than younger ones (e.g.1995).
Around 1995, HP may have changed several components, evidently the 40k reference resistor to a more stable VHP101.
Mine from 2000 is very good, again(?).

Frank

[chilternview]

Thank you for the explanation Dr Frank. It's actually not uV, its about 5-6mV i.e. 5-6ppm warm up drift, but still within the 12ppm spec as you say.
The older, 'good' meter is -1ppm or better. I'd love to open both meters up and take a look at the dividers but I'm loathe to break the cal seals at the moment.

[strawberry]

RP7 applying heat to divider maybe

low possibility Q16

[chilternview]

PS: You also can extract / distinguish this heating effect between your Datron 4000 and the 3458A.
Best and safest way is to use a HV reed relays for this purpose.

Switch on 1kV on your 4000, and let it settle for several minutes. The HV will then be stable for the rest of the experiment.
Now apply this 1kV to the 3458A in its manual 1kV range, by engaging the HV reed relays.
Let the 3458A run at a fast rate, so that you can observe the drift from its initial HV reading to the stabilized value a few minutes later.
Best would be to sample the readings into memory, so that you can afterwards analyze the voltage rise and drift on your PC.

This difference is then your power dissipation effect.
You can repeat this experiment at 100V and 500V, to see if this is really about quadratic.

I will try this to quantify the effect more precisely in the near future.

Meanwhile I looked at the cal reports:

The meter with 1kV drift had for its +/- 1kV DC volt gain test:

Test        Minimum        Measured       Maximum       Uncert
+1000v    999.9775v    1000.0058v    1000.0259v    0.0057v
-1000v   -1000.0261v   -1000.0067v   -999.9777v    0.0057v

And the 'good' meter:

Test        Minimum        Measured       Maximum       Uncert
+1000v    999.9788v    1000.0000v    1000.0192v    0.0054v
-1000v   -1000.0194v   -999.9990v    -999.9790v     0.0054v

(in case anyone wonders why the limits are tighter for the 'good' meter, it's because it has option 002).

From these, the 'bad' meter reading 6mV high is to be expected. And it's well within spec.

Thank you Kleinstein and Dr Frank for awaking me to the self-heating effect of the input divider. A question: if the divider resistance values were increased, e.g to 99Mohm/1Mohm instead of 9.9Mohm/0.1Mohm, this would reduce the power dissipation by an order of magnitude too. Would this not improve the thermal effects, or is it harder to achieve a low tempco with high value resistors? Or would input bias currents lead to greater inaccuracy?

[strawberry]

I have external 1Gohm divider what limits AC BW to some 1kHz (due to some pF)
caddock probably make better tempco dividers

[Kleinstein]

Higher value resistors tend to be less stable. It would at least take more effort to get a similar stability.
The amplifier bias would have more effect, but this is mainly added offset that can be corrected. The possibly nonlinear and not that stable input resistance of the amplifier may effect the linearity and drift more with a higher impedance. Already with 100 K of source impedance the linearity could be a bit worse than with a low impedance source.
Higher resistance also means more noise: this would not be so bad for the 1000 V range, but the 100 V range would suffer. Much of the noise in the 100 V range is from the dividers noise. The noise also effects the time needed for ACAL: 10 times the resistance would need 10 times longer (or more if amplifier noise current is releavant) to get the same noise level.

The reduced BW is also a slight point, though there is a separate divider for the AC part and the AC part also has matching parallel capacitors to make it a dominant capacitive divder for the higher frequencies (e.g. similar to a scope probe). The effect is additional settling time when switching to the divider and also in the auto zero mode. A higher resistance divider might need extra waiting time.
An external higher impedance divider may also want an extra buffer, as the main amplifier may not be the best choice for a source with 1 M impedance.

The 10 M impedance is a reasonable overall compromise, though not ideal for the higher voltage end.

[Dr. Frank]

... A question: if the divider resistance values were increased, e.g to 99Mohm/1Mohm instead of 9.9Mohm/0.1Mohm, this would reduce the power dissipation by an order of magnitude too. Would this not improve the thermal effects, or is it harder to achieve a low tempco with high value resistors? Or would input bias currents lead to greater inaccuracy?

It's already very difficult to achieve low T.C. resistors for 10MOhm single resistors, which are mostly thin film technology.
At 100M it's even more difficult, so that's no good idea.
You might build a divider from many bulk metal foil resistors @ 1..2MOhm, so that's quite expensive.

In this link: https://www.eevblog.com/forum/metrology/influence-of-switch-resistance-in-hamon-dividers/msg4207087/#msg4207087
I calculated the error for a divider loaded by the 3458A input bias.

For 99MOhm, about 10pA and 1000V input this would be about -1ppm, but -10ppm for 100V input.. but you have to validate your specific DMM for input bias.
Because the bias current varies over temperature, it's very difficult to impossible to compensate this effect properly.

That also implies, that you would have too big errors in the 100V range!

The 100V range on any DMM using a 100:1 10MOhm divider is the most critical range, due to this bias effect, but also regarding offsets as well.
It's quite unfavorable to first divide 100V by 100, and then to measure internally on 1V level, as the 3458A has to amplify this 1V back to 10V.
An internal offset voltage of 1µV is already 1ppm error, whereas using a 10:1 divider, i.e. 100V => 10V would reduce offset errors to 1/10.

Frank

[chilternview]

I ran 3 tests fpr 10v, 100v, 1kv:
1) Run ACAL DC
2) Set Datron to required voltage with output off
3) Wait 10mins
4) Turn output on and start logging for 20mins (NPLC=100)
The results are in the graphs below and were logged to a file. Blue is the 'good' 3458, yellow is the 'bad' one, red is room temp.

For 1kv the total rise/fall was +7.188ppm ('yellow') and -1.846ppm ('blue').

[sahko123]

Yeah as mentioned throughout the thread I think it is a self heating issue in the resistor divider