My problem, conceptually, is the reference, a LTZ1000A. Now, lots of the advice I have been reading concerning references is to trim it in software.

The issue is how to trim out the LTZ1000A to exactly 10.48576V, divide that down to exactly 1.048576V, and maintain the excellent tempco.

.. With other references such as the ADR44x, fine trimming is achieved via a low-tempco pot.

On the LTZ1000A datasheet, they don't even suggest that method. Everything is done via low-tempco fixed resistors.

But how can you ever get a 1ppm precision out of the source with fixed resistors? You will never find the exact right value, not to that level of precision at least.

Anyway, that doesn't solve the problem of gross trimming and taking the output from 7.2V (nominal) to exactly 10.48576V. Is that possible or is that a fool's errand?

Both the LTZ1000A datasheet and the AN for the AD5791 gives a solution for adjusting the voltage to +7V or +10V/-10V, but I don't understand how that voltage can be guaranteed to more than a few digits without some sort of user adjustment.

It looks like it would be guaranteed to be precise, but not accurate.

But there is no adjustment here. I would have expected some low tempco pot in the op-amp feedback to trim that out. Does the pot ruin the tempco? Why is that not there? What am I missing here?

Thank you.

Hello JoeN,

you really have a complete misconception about such references like the LTZ1000 (also SZA263, LTFLU in the Fluke 732A/B).

These "reference amplifiers", as they are also called, deliver an extremely stable, fixed output, which has an

**artefact** character, meaning that they can not be trimmed to an otherwise intended value.

Their absolute output values are at first dependant on the topology of the circuit, i.e. a ~6.75V zener diode plus a ~ 0.45V BE junction, and on the oven temperature, which is also fixed.

A very small, but not intended, trimming is possible by variation of the driving currents of the zener and the transistors collector current.

So it's also completely impossible to "trim" the LTZ1000 to 10V or above, that idea is absolute nonsense, indeed.

You also confuse the terms "precision" and "accuracy", when you instead mean, that you need integral voltage output values to base 10 or base 2.

The artefact output of the LTZ1000, when once measured, may indeed be precise to < 1ppm absolutely, and indeed accurate to e.g. < 1ppm/yr or < 0.1ppm/K, as it will not change much over time and temperature.

These other references, you are referring to, also have a fixed core reference, like the 1.22.. 1.25V for the bandgap type.

These all are amplified by internal resistors and an OpAmp to the 10.00V or 10.24V.

Also here, the reference itself can

**not **be trimmed, but the amplification resistors only.

So the problem with instabilities caused by resistors apply to ALL such voltage references.

So you would omit all resistive dividers / amplifiers in first instance, and only use a precision DAC (high resolution and linearity, I.e. 24bit, <<1ppm linearity), to divide the LTZ1000 output precisely.

To get outputs of >= 10V, you would stack two LTZ1000s for a nominal 14,4V reference.

To get integral values (base 10 or base 2) out of that DAC, you only need one initial calibration for the DAC output, like 10V output equals 11650844 digital units, or 10.24V equals 11930465.

(example for 14.4000V , 24bit DAC).

Simply forget your idea of making one bit of the DAC an integral output value (like 1 bit equals 1.0000mV, or so).

This instead is done by software, using "fractional" numbers for integral outputs. (That REALLY works in practice!!)

Anyhow, to get other volt ranges (1V, 100V, 1kV), you always have to divide or amplify this DAC output with resistors, which again involves all of their instabilities.

For better understanding of these principles, I recommend to read the Fluke 5440B service manual, how to implement a really precise DAC (PWM type) and to overcome the resistor problem.

That's described in the addendum: "Recent Innovations in Direct Volts Calibrator Design".

Frank