HP 34401A Observations

[mvanocht]

Many years ago I bought a used HP 34401A meter.  From the serial number I estimate it was manufactured in February 1996.  Had it calibrated three times by Keysight (well, HP, Agilent, and Keysight) and this is a summary of the results.  The first time I sent it in was March 2005, and it was out of calibration  (that may have been the result of a sloppy calibration lab making adjustments between the time it was manufactured and the time I bought it).

Each time Keysight did the calibration they took “before” and “after calibration” data and they made adjustments to make it close to perfect. So in March 2005 they adjusted to be exactly right. The next time it was sent in was four years later (April 2009).  There is a column showing the ppm change over four years and the calculated “Drift per year”.  It was again adjusted to be exactly right.  Then eight years later (May 2017) it was sent in again and you can see the drift for the 100V and 1000V ranges has gone down substantially.  The 10V and 1V range show almost no drift.  The 100mV range shows some drift but still very little.

The way the meter has held its precision to me is astonishing!  It says a lot for the LM399 Reference.  I haven’t analyzed the other functions, but based on these numbers it would be a total waste of time and money to send it in every year for calibration.  Part of the reason for it’s low drift may be due to the stable 23° C (+/- 1° C) environment, but I think it is because the manufacturer is ultra-conservative in specifications.

[TiN]

Without knowing how small/big those adjustments were, you getting bit ahead of yourself on stability excitement.
But yea, in general manufacturer would have conservative specifications, otherwise they will need to scrap half of the production units due to natural variability in performance. .

[mvanocht]

The two columns marked "ppm change" ARE what the adjustments were. 

[TiN]

What are uncertainties of the measurement then? Either way, summary table in incomplete.

[splin]

Interesting numbers. Given that any additional calibration uncertainties, including noise, are amortized over each 4 year period and thus reduced by a factor of 4 compared to the drift, drift does look pretty good - comparable to an LTZ1000 with typical -1ppm/year? It's a pity the meter was adjusted each time as the calibration uncertainties would have been reduced by 8 relative to drift and by 12 when/if recalibrated in another 4 years.

As a 34401A owner it would be very interesting to see the drift on the current and resistance ranges - particularly the latter.

[mvanocht]

During "calibration" there are no adjustments in hardware, everything is done in software.  Corrections are "Y (Reading)= mX + b" where the "b" represents offset at zero and the "m" represents the slope taken from zero to full-scale. Therefore, having them "optimize" it for no error during calibration does not change how much drift happens until the next calibration.

Examining the schematic shows the same Input Divider (divide by 100) is used for the 100V and 1000V ranges.  Therefore it isn't too surprising that the errors on those two ranges somewhat follow each other.

As far as "uncertainties of the measurement" what I can say is that  the Loveland CO location which did the first calibration according to the "ANSI-ASQ National Accreditation board" is supposed to be able to generate voltages from 100mV - 1 kV with no more than +/- 0.43uV / V uncertainty.

The El Segundo, CA location where the last two calibrations were done is not as good...they are listed as +/- 0.63uV / V (they don't have a JJ like Loveland does).  Here is a link to the PDF showing their Scope of Accreditation for each laboratory to ISO/IEC 17025:2017:

https://www.keysight.com/main/editorial.jspx?cc=CN&lc=chi&ckey=2272436&id=2272436&cmpid=zzfindaccreditation

This particular meter (34401A) meter can only resolve 1 ppm so it seems to me the cal lab uncertainties are less than the resolution of the meter.  Please educate me if i am wrong.

[rhb]

Interesting numbers. Given that any additional calibration uncertainties, including noise, are amortized over each 4 year period and thus reduced by a factor of 4 compared to the drift, drift does look pretty good - comparable to an LTZ1000 with typical -1ppm/year? I

The reduction in error due to noise under Gaussian assumptions is sqrt(N) where N is the number of *measurements*.  In your case there are only single measurements so there is no reduction in error.

I have a pair of 34401As and a 3478A.  One of the 34401As was never calibrated after it left the factory. It's as accurate as my DMMCHeck Plus and Vref10   from Doug Malone which were calibrated on a 3478A before shipment to me.  They differ by 10-20 ppm but I have no  temperature control in my work area.  Both my 34401As are around 20 years old.  It's an amazing design.

[Xesfeder]

You saved my life. I can’t thank you enough.