It'd be interesting for me to see if it goes back down to an order of magnitude closer agreement (by which I mean being at 9.9999999V - 10.0000000V for any significant stretch of time.
Yes, we are. I'm not sure what all this "wild" "excursion" was about . I'll keep on watching it, and maybe all this is relatively normal in my "non-metrology-grade" environment (though tempco is not the cause), with many variables and little capacity to thoroughly diagnose on basis of a solid set of time-coordinated data with multiple streams (V, C, t, etc.). One keeps on learning! If so lucky.
Good evening Rax,
Maybe that's simply the usual hickups of very old instruments.
I copy the monthly comparisons of the 5442A and another LTZ reference, latter is continuously running.
They were measured and compared in relation to the mean of my whole reference group of 11.
The LTZ #1 tracking is relatively smooth, especially after I used shielded PTFE cables similar to Adrians, 10 samples with statistics with reversal (GPIB monitoring required), and always holding the breath and my hands still during the sampling time.
You also see, that the 5442A sometimes has hickups up to 1 ppm., which will vanish the next time, but can only be identified by using such regular comparisons.
Another INT CAL on that same measurement-day did never remove the hickup.
Recently, I changed my averaging method. I have found out, that during reversal of each measurement, I got a mostly constant -300nV offset between + and - values. This I simply had to assign to the constant non-linearity, or reversal error of my 3458A, which would be -0.03ppm, i.e. its nominal INL error, instead of any random e.m.f.
So I also doubt, that
any e.m.f. error will have big influence (~0.02ppm typically), like Dieter assumed recently:
https://www.eevblog.com/forum/metrology/fluke-5440a-calibration-procedure/msg5147310/#msg5147310When you're using proper PTFE with copper or gold plated lugs, those e.m.f. errors will stay nearly below the detection limit.
My new algorithm:
3458A @ 8 digits, NPLC 100
- screw lugs to the DUT, wait 1 min for thermal equilibrium
- take 16 measurements for ~ 1 min
- read statistics: average value = result (9 digits), StD will be 70 .. 300nV or 0.01 .. 0.04ppm
- if measurement shows more noise, repeat reading
Maybe it would be better in your case to use 7 digit, 2sec long measurements and use statistics of your 6048.
This will give you as well stable results to 8 digits , but also an idea / hint how stable your measurement, i.e. DUT vs. DMM is.
You might also identify external noise sources, like SMPS.
Frank
Btw.: Obviously, we have found the root cause of the deviation during the last external comparison.
When i get the corrected results, I hope that my baseline is confirmed within 0.3ppm.