Resistor stability in Voltage Reference amps - a way around?

[Cerebus]

I'm going to throw out for general consideration an idea I had while lying in bed last night.

We've discussed, several times in the various voltage reference threads, the issues surrounding long term and temperature stability in resistors in amplifiers used with voltage references to get other voltages than the 'natural' voltage produced by the reference. We've also looked at various chopper/zero-drift amplifiers as adjuncts to voltage references.

Well, what about combining a chopping amplifier type architecture with a technique that allows the gain of an amplifier to be set with zero tempco and zero drift - using a transformer to set the feedback ratio, and hence gain of an ac amplifier? Take the dc input to the amplifier and chop it, say with an LTC1043 or similar. Then an ac amplifier, with the feedback via a transformer - the turns ratio sets the gain. The turns ratio just is - it doesn't change, as far as I can see, with any of the usual things that give us problems. Finally a synchronous rectification stage, another LTC1043 and we're done. You can add the usual switched zero-drift offset correction in parallel with the other switching just to round things off.

Obviously, the gains available are only those that are permitted by rational numbers and any variable gain arrangement would have to be constructed from switching fixed taps.

Good idea? Is there some showstopper I've missed?

[uncle_bob]

I'm going to throw out for general consideration an idea I had while lying in bed last night.

We've discussed, several times in the various voltage reference threads, the issues surrounding long term and temperature stability in resistors in amplifiers used with voltage references to get other voltages than the 'natural' voltage produced by the reference. We've also looked at various chopper/zero-drift amplifiers as adjuncts to voltage references.

Well, what about combining a chopping amplifier type architecture with a technique that allows the gain of an amplifier to be set with zero tempco and zero drift - using a transformer to set the feedback ratio, and hence gain of an ac amplifier? Take the dc input to the amplifier and chop it, say with an LTC1043 or similar. Then an ac amplifier, with the feedback via a transformer - the turns ratio sets the gain. The turns ratio just is - it doesn't change, as far as I can see, with any of the usual things that give us problems. Finally a synchronous rectification stage, another LTC1043 and we're done. You can add the usual switched zero-drift offset correction in parallel with the other switching just to round things off.

Obviously, the gains available are only those that are permitted by rational numbers and any variable gain arrangement would have to be constructed from switching fixed taps.

Good idea? Is there some showstopper I've missed?

Hi

Basically the "convert it to AC" approach is the way it was done before about 1965 or so. You trade off AC parasitic issues against DC ones. You also add all of the DC -> AC conversion and AC -> DC conversion parasitics as well.

That said, the transformer part works ok. I have a variable ratio transformer sitting here that goes to 6 digits. It cost nothing compared to a similar DC (say Kelvin -Varley) device.

Bob

[Cerebus]

Hi

Basically the "convert it to AC" approach is the way it was done before about 1965 or so. You trade off AC parasitic issues against DC ones. You also add all of the DC -> AC conversion and AC -> DC conversion parasitics as well.

Yeah, nothing special there and I kinda assume that folks this will interest will (a) know that and (b) understand the trade-offs involved with a chopper type circuit.

That said, the transformer part works ok. I have a variable ratio transformer sitting here that goes to 6 digits. It cost nothing compared to a similar DC (say Kelvin -Varley) device.

The issue always seems to be with gain setting resistor stability and, while looking at a low-noise ac amplifier design that used transformer feedback to set the gain, I suddenly thought "Hold on, there's no gain drift there, none whatsoever". The attraction to me is that is seems simpler overall than something like PWM feedback and cheaper then ultra-stable resistors.

I'm sure that there's some fine detail that needs examining but I thought it looked promising enough to fly here and see if anyone can spot a significant flaw in the idea.

[Cerebus]

Anyway, if you have the money (and time) to try this transformer idea out, go for it!  The results (win, lose, or draw) would be quite interesting to everyone here I think.

I'll probably not actively do anything with this, or not for a good while at least, which I why I thought I'd punt it here in case someone else thought it was a good idea and wanted to have a play.

My personal reference needs only run to long term stable 7.xV type references, where I don't care what x is as long as I know it, or short term stable arbitrary (e.g. 10V) references, the latter of which you can get from ratio measurements and/or PWM.

[uncle_bob]

Hi

If you go with square wave, your conversion process is fairly straightforward, but your transformers are crazy. Bandwidth issues come in from all sides. Sine wave is *much* more transformer friendly. Conversion ... not so easy with sine wave. Digging through all that kind of thing (for a couple of years) will give you a reasonable list of the gotchas. At the most basic level, the answer is ... they tried very hard to make this all work with AC and then went back to DC that was a *lot* worse than what we have today.

Bob

[Marco]

If you take a hermetic film capacitor and ovenize it, is stability really going to be bad for a PWM scheme?

For instance : charge capacitor with constant current for a sawtooth voltage, reset to -Vref at each cycle, online calibrate the current source to give ground and reference voltage X time apart, stop current source Y time later for 10V ... as long as the non-linearity due to dielectric absorption is stable you don't even have to correct for it, can be handled during calibration against better references or meters.

[Dr. Frank]

Your idea is e.g. realized in the Datron 1271/1281 SelfCal circuitry.
They achieve about ppm uncertainty for decade ratios.

If the driving frequency is not too high, parasitic can be handled easily.
As the transformers ratio is absolutely stable, drifts may be neglectable.

So this might be a good idea and worth doing some experiments on.

Frank