adjustment methodology for reference value

[Rax]

This may be the dumbest question ever asked on this forum. But here it goes.

Assume I'd be adjusting a reference 7.00000V (the digits represent my ability to read it) in a bench DMM. This needs to be adjusted to +-.0002V of that value.

Do I adjust to the cusp of the flicker between 6.99999V and 7.00000V? Do I adjust to a solid 7.00000V? (assuming the latter is possible, which depends on a few factors, one being whether I have so many digits of reading that I can't have a perfectly solid, immutable reading, and the last count will likely have fluctuations).

I feel the first choice above is closer to the dead center of the range (evenly distanced between 6.99980V and 7.00020V). A truer 7.00000V. The second choice may be right at the center of the 7.00000V and 7.00001V range.

What's the correct adjustment methodology?

[guenthert]

   Are you telling me that your meter has a, even just 20m, short-term stability better than one least digit?  Not even speaking of the stability during the calibration period.
 
   And the current value of the source is known to better than 1ppm (given that you apparently intend to adjust a 6.5 digit meter)?

   There's no point to fret about fractions of the last digit, if the confidence interval of the actual value spans many counts.

[Kleinstein]

Higher resolution meters usually have noise of more 1 of the last digits.  A common situation is about a RMS value of about 1 of the last digits. So something like 6 steps peak to peak, maybe a little better.

Noise less than 1  digit step does happen with low resolution meters at the 2-3.5 digit level, but there it is usually advised to get a high resolution meter insted of trying to get better than 1 LSB.

The idea target is usually such that the average over a little longer time is at the right value. Still with limited stability and accurace of the ref. elemt to adjust too there is no need to overdo it. The good enough point is often more than 1 step off nominal.

[iMo]

In case the OP owned a 34401A (as a typical 6.5digits bench dmm) and the DUT reference was an LM399 the OP would see a "rock stable" 5th digit after the decimal point on the display most of the time (ie his/her 7.00000V).

[bdunham7]

What's the correct adjustment methodology?

Assuming your meter is somehow absolutely perfectly calibrated, linear and stable, what is the threshold for it to transition from 6.99999V to 7.00000V?

[Andreas]

Do I adjust to the cusp of the flicker between 6.99999V and 7.00000V?
What's the correct adjustment methodology?

I am not shure what you exactly want to express. (that flicker method would only work if the calibrator exacly outputs 6.999995V)

Many instruments (34401A / K2000) have internally a higher resolution and the display is rounded. (others may cut the output).
That is what you can see when you either record the values by interface to a computer or when you get out one more digit resolution by a "math" function.

with best regards

Andreas

[Rax]

Are you telling me that your meter has a, even just 20m, short-term stability better than one least digit?  Not even speaking of the stability during the calibration period.

What I am saying is that my 6.5 digit meter has absolutely no qualms locking perfectly on 7.00000V from this reference voltage, which it'd stay on for hours at a time, rock solid. For instance, just now, it drifted a bit to 6.99999V after 6 hours continuous run, but it could be the reference source, my meter's own internal reference, temperature variations and air movement, etc.

If I adjust the reference high (multiturn pot) it safely sails to 7.00001V and dwells there. I dial it low, it does the same to 6.99999V.

The question though was what's reasonable to do as methodology. What is the best workflow approach on this.

[Rax]

In case the OP owned a 34401A (as a typical 6.5digits bench dmm) and the DUT reference was an LM399 the OP would see a "rock stable" 5th digit after the decimal point on the display most of the time (ie his/her 7.00000V).

Spot on. SZA263 in this case, measured by an 8502A masterfully dialed in by one of the illustrious members here.

[Rax]

The idea target is usually such that the average over a little longer time is at the right value. Still with limited stability and accurace of the ref. elemt to adjust too there is no need to overdo it. The good enough point is often more than 1 step off nominal.

There's two kinds of measurements I am thinking of, both applying to my exact bench project right now.

• A reference voltage measurement with a 6.5 digit meter (LM199/299/399, SZA263, etc.). As others have pointed out, this can output a flawless 7.00000V. Having the possibility to adjust this reference voltage and the luxury of a perfectly stable output, and a very generously wide range of adjustment of the multiturn pot, the question here is what exact point should this be dialed to?
• A, say, 10.100000V output by a calibrator. I am seeing this much less stable, with variations between, maybe 10.099995V and 10.100005V. This is where I think Kleinstein's point applies - I've been trying to tweak my adjustment is such a way that I hit my target value with the middle of the range (essentially by averaging)

[Rax]

What's the correct adjustment methodology?

Assuming your meter is somehow absolutely perfectly calibrated, linear and stable, what is the threshold for it to transition from 6.99999V to 7.00000V?

Pertinent point and I'm not sure I have an understanding what that threshold is. My working assumption is that on such a minute variation I can infer a relatively linear function for this due to the "infinitesimal" differential (borrowing a bit from elsewhere).
[Added later after a bit of further ponder] Additionally, I am not aware of the 7V point in my meter to be a "switch," or "handover" point. Such as, maybe, say 19.99999V or 11.99999V and other such values would be, depending on meter and its specific design.

[alm]

If you want to adjust a value to as close to 7V as possible, then I'd argue you should turn it so it stops flickering between 6.999x and 7.000, but no further.

If you wanted to know the value it was adjusted to as closely as possible, but do not necessarily care which precise value, then there's a benefit to picking a value where the meter reading flickers, because in theory, assuming perfect linearity of the meter (aka fantasy land), you could extract an extra digit of resolution by tracking which fraction of the time it reads 7.000 and which fraction it reads 6.999. Although if the meter reads a higher resolution internally than the math average might achieve the same if it can show sufficient digits.

But again, in my view this discussion is purely academical because resolution is rarely the limiting factor. In almost all cases a DMM will have more resolution than accuracy or linearity.

[Rax]

in my view this discussion is purely academical because resolution is rarely the limiting factor. In almost all cases a DMM will have more resolution than accuracy or linearity.

I guess it sort of is - though I'd argue in the "geekly" extreme world of "hobby metrology" this is not totally out of the ballpark as a far as weird questions go  - and with a stable reference I think this is also a realistic question. The question being - the SM asks me to adjust to 7.00000V +-.00002V. Does this mean the point of flickering 6.99999/7.00000, or the 7.00000V "plateau?"

I gather you've answered that by saying the adjustment should pursue the point where the flickering stops, but not any further. Which I think makes sense.

[miro123]

Hello,
I've read the whole thread but It is still not clear to me what do you want to achieve.
I have few questions to fully understand your situation.
1. What type of DMM/calibrator do you use?
  - model?
  - does the DMM has an calibratoin history? If yes for how long?
  - Do you have any calibration or adjustment certificate on your DMM/calibrator? Are they NIST traceable?
 
2. Can you describe your voltage reference circuit?
3. Can you describe your test setup?

I need this information to roughly calculate your measurement errors an uncertainty.
I expect that your total error+ uncertainty is somewhere in  20...100 digits range. That is all in case you have NIST traceable certificate with defined error and uncertainty at that moment.

Can you reverse the measurement polarity? Does -7.00000V still stays spot on?

[David Hess]

Do I adjust to the cusp of the flicker between 6.99999V and 7.00000V? Do I adjust to a solid 7.00000V? (assuming the latter is possible, which depends on a few factors, one being whether I have so many digits of reading that I can't have a perfectly solid, immutable reading, and the last count will likely have fluctuations).

ADCs may or may not have the zero bit transition at zero or offset by 0.5 LSB, so the middle of the bits could be at a whole number of counts or half way in between whole number counts.

For a high resolution ADC, I would expect it to be irrelevant because the linearity and noise will always be worse than than 1/2 LSB, but it does often come up at 12 bits and lower resolution.

To further complicate the issue, high resolution ADCs may be monotonic while missing codes, so the exact value may not even exist.

[r6502]

Assume I'd be adjusting a reference 7.00000V (the digits represent my ability to read it) in a bench DMM. This needs to be adjusted to +-.0002V of that value.

I think, You did give the answer by yourself: request is to adjust the source to 7V ± 0.2mV, so anything between 6.9998V to 7.0002V is fine.

0.2mV is about 30ppm from 7V.  This is the typical uncertainty of a 6.5 digit bench multimeter in the 10V range. For the DMM65000 Keithley/Tec specifies here 25ppm for 1 year.

Guido

[tszaboo]

In case the OP owned a 34401A (as a typical 6.5digits bench dmm) and the DUT reference was an LM399 the OP would see a "rock stable" 5th digit after the decimal point on the display most of the time (ie his/her 7.00000V).

Spot on. SZA263 in this case, measured by an 8502A masterfully dialed in by one of the illustrious members here.

Your meter should be ~10 times more accurate than the DUT you want to calibrate or adjust (which is different). If it's less than 10 usually there is a lot of considerations to be made. So you cannot really just adjust something to 6.5 digits with a 6.5 digit meter.