I'd like to ask for some patience..

practical stuff will be delivered definitely.

But first, here’s the summary of the requirements of the reference system, already with some instructional comments.

1) The Reference System should be capable of calibrating a HP34401A on each DCV range, over years. The HP34401As 24h specification is 15ppm for 10V, 20ppm on 1000V, 30ppm on 100mV.

For Good Metrological Practice, an uncertainty of <2ppm is required for each range of the Reference System.

That means, if the uncertainty (relative to SI) of the LTZ1000 output is known (i.e. calibrated once), all transfers to the different ranges must be done below that limit.

Also, all drift parameters should contribute in total below that limit.

2) The stability over time of a single LTZ1000 reference should reach realistically <= 1ppm/year:

LT specifies for the naked chip typ. 2µV/sqrt(khr.) @65°C, that’s 0.8ppm/yr.

This stability requirement (aging or deterioration) is most important, as it is not reversible, and cannot be mitigated or hardly be compensated (at most by elaborate trend analysis), like other instabilities.

3) The stability over temperature of the LTZ circuitry should reach realistically < 0.2ppm/K.

This instability is reversible, by returning to the nominal temperature, and can be cancelled statistically or even mitigated.

For that purpose, the reference system has to be operated in a stabilized / controlled environment. Temperature changes must stay within a few 1/10°C during 10 min, or over the measuring period (hours), and for long term stability measurements, the room temperature must be reproducible to e.g. +/- 2°C.

Otherwise, drift measurements on sub-ppm level are not possible.

Btw.: All other sources of similar instabilities, e.g. thermo electrical/mechanical force induced, humidity/leakage current, pressure, gravity, and so on, have to be analysed in value, relative to the ageing drift, i.e. if it's worthwhile to cancel them.

Like the temperature coefficient, those drift sources lead to spurious / reversible modulations of the reference output "only".

4) Simply for convenience, the raw 7,2V output of the LTZ1000 has to be attenuated and trimmed precisely to a plain value of 7,00000V.

This secondary output should be very stable over time and temperature.

This transfer might be done by a 34401A with 2ppm uncertainty (linearity error), or with a 720A (<0.2ppm), or a 3458A (<0.05ppm).

5) All reference outputs shall be amplified and buffered with low impedance by an exact factor of 10/7 to around 10V, so that the 34401A will accept this for calibration.

This amplification factor may be

**auto-calibrated** at any required time with an uncertainty of < 0.2ppm, so that the stability / uncertainty of the LTZ1000 is maintained in the 10V output.

6) A decade divider with precise 10% steps, 0.1ppm nonlinearity, should provide a means for calibrating the linearity error of a Fluke 332B to 1ppm.

7) The experimental verification of the stability of the LTZ reference and the complete system is required, due to Good Metrological Practice.

A theoretical model about the instabilities is not sufficient on its own.

Therefore, a measurement system is required which is capable to perform sub-ppm comparisons, of the DUT against a more stable standard, or against a group of equivalent stable references.

8.) A Reference Divider should provide precise ratios of 10:1 and 100:1 with uncertainties of < 0.2ppm and < 0.5ppm of output, latter one for a burden of 1kV also (like a Fluke 752A)

- To be continued –

Frank