Agilent 34970A (34401A) linearity adjustment

[TizianoHV]

In a few weeks  I might have the opportunity to calibrate my meters at my university cal lab. My Agilent 34970A has really good temperature stability but linearity (negative polarity) is mediocre.
I don't have a 3458A so, to check linearity, I reversed the output voltage at 1V intervals and compared my 6.5D meters (HP 6114A sourcing voltage). On the positive Vs negative test my Keithley 2790 has 7uV error (offsets already compensated) but the 34970A has 16uV error. By comparing the two meter for positive voltages (K2790 as reference) they seem to agree quite well, but for negative voltages the 34970A shows some error. The errror seems "linear" and easily compensable by changing the negative voltage gain.

On the service manual they says: "The –10 Vdc calibration electronically enhances the Internal DMM’s
a-to-d converter linearity characteristic. This adjustment should
ONLY be performed after servicing the A-to-D converter or
replacement of the calibration RAM." with "ONLY" in capital.

Can I overwrite the older linearity calibration or this would be useless / worsening?

Thankyou!

[Kleinstein]

The error really looks like a slightly different gain for the positive and negative side. Even though the difference is seem mainly on the negative side one can not say the the linearty is bad there - the problem is that the 2 sides don't fit togehter, not one side bad.

It is not really clear how the 2 separate positive and negative 10 V readings are handled. Ideally this should be based on the expected INL contributions that could make a difference for the 2 polarities.  At least from my understanding of the ADC the main effect to expect would be something more U² like, though not very large. It is still possible to also get an error more like a separate gain factor - though I don't see a mechanism for this.  I don't know how the formare handles and corrects different readings - both versions would make some sense, but of cause only one of the 2 at a time.
A different gain for both sides would be the reasier to implement in the software.

There is a chance that a new calibration at -10 V could improve things. Unless there is also an offset problem I don't see why it should make things significant worse.

Another option would be to manually or afterwards with the data at the PC correct for the error. Some 1.6 ppm for the neg. side is not yet a big difference and in many cases one does not care.