I think it is done more simply than the 3458A, some clues are in the 24hr range error specification. From 10V>1V>0.1V the range error is 0.1>1>10 ppm. I don't know if the difference are from a worse ADC, or different Auto Cal scheme, or both.
If you have any free time with the 3458A, can you check the error from applying 100mVDC to the 10V range and then reading it from the 100mV range? I get a difference of about 2-5ppm of reading even after an Auto Cal.
Hello,
I won't do that test again .. just a look into the hp journal 4/89 and its specifications is necessary to explain the problem.
The 3458A can do accurate 10:1 transfers, i.e. from one range to the other, on the order of 0.3ppm.
That means, 10V and 100mV
gain should agree within 1ppm.
That is specified much more conservatively, i.e. the transfer 10V > 1V > 100mV is 1ppm each. The +0.3 and + 3 ppm / range can probably be assigned to offset errors.
Therefore, if one encounters bigger differences than 2 ppm between 100mV readings on the 10V and 100mV ranges in practise, it is also very probably due to such offset voltage effects.
If you have the possibility your measurement setup to cancel offsets, then you may get better results.
This is not a trivial aspect, as you often can not do precise zero measurements in your circuitry..
If I re-calibrate the 1V and 100mV ranges of my 5442A standard, I indeed have the possibility to output 0V first, null the offset, and then measure the 100mV output.
So I always achieve transfer accuracies < 1ppm even in the 100mV range..
Same scheme applies, when I measure 1V on the 10V range and 100mV on the 1V range, to check the foregoing AutoCal.
Each time I find, that the transfer are precise to << 1ppm, provided I had cancelled offsets before.
You may also experiment in that direction, if you own a stable, programmable DC voltage source.
Frank
Edit, because of an additional aspect!
A direct transfer 10V => 100mV range is about 10 times more unprecise than the path 10V => 1V => 100mV.
That's due to the linearity specification of typ. 0.02ppm of F.S. for the 3458A.
That gives about 0.3ppm for each 10:1 transfer, equivalently on a measurement at 1/10 of F.S., but for a 100:1 transfer, that's already 3ppm!
So, even if you properly cancel all offsets, 2-5 ppm error is to be expected, depending on the linearity of your R6581.