. You do not want to add a regulator.
Hello,
Usually PSRR is limited and the references have a much larger noise than the batteries.
At the LM399 you have also to stabilize the heater voltage to get a stable output.
So you really want a regulator near the LM399.
With 90nV/sqrt(Hz) for the LM399 I would not worry to spend a low noise regulator.
Even the LT1763 with 20nV/sqrt(Hz) is below that figure.
And usually as calibrator the 0.1 ... 10 Hz noise is the more interesting since broadband noise is integrated away by the DMM.
Noise is also what limits resolution if you use of a 100 mV range instead of the 10V range of the instrument.
And of course every voltage range has its individual divider/amplification resistors in the DMM.
Usually the native (3 or 10V range) is the most stable. The stability in the 20mV or 100mV range may differ.
So practically you do not get a factor 100 improvement by this. Its more like a factor of 10.
And of course it does not help if all references drift in the same direction. (Which is very likely).
It was a big surprise for me when I did the first re-calibration of my LTZ#1 and LTZ#2.
I told our calibration department that I have a difference drift of 12uV between both.
And LTZ#2 has a zero drift against my best ADC at that time.
But the calibrator showed that both had drifted down. One by 8 the other by 19 uV.
So the ADC had drifted also nearly 3 ppm.
Edit: if you have a 34401A it might be better to use the ratio mode for relative measurements.
I really should do some stability measurements.
with best regard
Andreas