Dr. Frank:
I see a basic problem here. We are discussing apples and oranges here, in terms of time periods of measurements. I am thinking of making one group of measurements between each reference pair of a bank of Voltage References (Vrefs) once a week over 15 years. See NBS Note 430. You are thinking of making measurements for a much shorter period of time, in your example 17 hours.
Hi Bill,
that's not the case. Exactly like you, I also make regular comparisons (every 2 or 4 weeks) of my bank of meanwhile 11 references, the 3458A itself is used as a transfer instrument only.
The 17h measurement has nothing to do with that, it only demonstrates the stability and noise of the 3458A relative to a DUT, but also that over a longer period of time there might occur some irregular bumps.
For 9 of these DUT I monitor their (more stable) raw zener output of 7.08 .. 7.16V. Then I have two 10V references.
Therefore the differential method is not suitable, so I have to make absolute measurements, using the extraordinary linearity of the 3458A and its good short term temperature stability, and afterwards calculating a gain correction.
I use 2 x 1min (16 samples) for each DUT, with reversing the polarity.
It's to be discussed, if 2x40 samples, like you do, really give a better uncertainty.
I think, that both methods are sufficient to achieve below 0.2ppm transfer accuracy, which is also sufficient for long term monitoring.
The 3458A simply cannot be stable enough, at 10 volts, over 15 years, or even a week, to give me the results I need to predict the drifts of the Voltage References. I agree that comparing two Vrefs in a short period of time, i.e. 5 minutes, you will get excellent results and I have done this. But try to do this again 7 days from the initial comparison and you will not be able to achieve stable enough results to get meaningful data, let alone over 15 years.
So the only way to remove the long term drifts of the 3458A is to use it as a Milli-volt meter on the 100 mV range. Then do 100 PLC, and take the mean of 40 measurements, reverse the leads and do another 40 measurements and then add the two readings, reversing the sign of the second reading, and divide by 2. I still get a lot of zener noise but over time the drifts can be determined. I didn't have a uV meter at the beginning of these measurements.
As explained above, I do not use the 3458A (any more) as a long term reference.
Anyhow, doing regular ACAL every day, and keeping the R.T. stable to within +/- 1°C, you will even get excellent stability over 1 week with the 3458A. My 3458A drifted only 1.7ppm over 7 years in DCV, because it is running only when used, and I have lowered its oven temperature to 65..70°C, i.e. adapted it to metrology requirements.
The LTZ 1000 references all drift between -0.3 .. -1.0ppm/year because they run 24/7 as usual.
The extrapolated mean value of this group meanwhile seems to be predictable to about 0.2ppm/year.. and that is now my actual volt representation.
The 3458A has only 100 nv resolution, on the 100 mV range, but that is .01 ppm of the 10 volt references and with averaging is good enough for my measurements.
Sorry no, it's got 10nV resolution in the 1V and in the 100mV ranges. The noise floor of nV meters is lower.
The K2182 has a 10 mV range and a resolution of 1 nV, the HP/A/K 34440A has a 1 mV range and a resolution of 0.1 nV. The K182 has a 3 mV range and a resolution of 1 nV.
I would imagine that with the SupraCon JVS they supply the 34420A because of the 0.01 ppm spec of the JVS or something like that. I have no idea how the JVS is recommended to be used. Again I would imagine that you would compare the JVS to the Vref under test, once a day, over a period of a week or more before reporting the absolute value of the output voltage.
Well, and here again I wonder if that really makes sense to have 10nV, even 1nV resolution with accordingly low noise instruments.
Any zener based reference shows an order of magnitude higher short term noise, at least these irregular jumps will ruin the wedding.
I would estimate from my latest measurements that you can determine the mean voltage of such references to no better than about 200nV, or 0.02ppm.
And that's easily accessible with a 3458A - in differential mode.
Changes by T.C. of < 0.03ppm/°C as specified for practically all voltage references will make such measurements even more questionable regarding their uncertainty.
Again the 3458A simply cannot get the required results using the 10 volt range if it was being used as a comparison meter between the Vref and the JVS. You can see from the JVS specs that you could expect to give a value to the Vref of around 0.01 ppm, depending upon the zener noise. Attached is a paper on the use of the SupraCon JVS with a F732A or F734A. This is discussing a one time calibration of the Vref, but to predict drift of the Vref this could be done over a week or more.
Bill
Well, the Supracon specification only shows some error bars, but no longterm differential measurement. There simply do not exist any LTFLU or LTZ based reference which would have this performance. Probably the averaged signal of 8 or 10 modules (734C, 7010N) might nearly reach this level.
Would be interesting to see one of your differential measurements, or such a long term noise measurement over 1h of your system.
I have done over night such a differential measurement, using the battery powered M7000 versus a mains powered LTZ 1000, 10V reference, 3458A as a null instrument.
Yes, that's really better compared to the absolute comparison method.
1h StD is about 40% less, and 'your' 40 point averaged peak to peak noise is 50% only, about 1µV
pp over 17h.
Take another factor of 1/ sqrt(2) down, to get the noise performance of a single zener, and you get the physical limit of about 800nV.
If you make measurements over 1 or 40 minutes only, the measured StD and pp noise might be even lower, but that'll not give you the real uncertainty for zeners.
Frank
Rem.:
absolute vs. differential method over 17h or 8h
pp noise, 40 point average is 1.9 µV vs. 1.0µV
StD over 1h is 220nV vs. 190nV
added picture of complete setup