Equipment or setup for ratio measurements

[slugrustle]

I often want to measure the ratio of two quantities, for example

• Input voltage and output voltage of an amplifier to measure gain
• Input current and output voltage of a current measurement chain to measure gain
• And so on

I usually hook up two DMMs, run trend charts to see when things have thermally stabilized, then set both meters to average and take a photo of both DMM screens once the average accumulates about 100 samples.  Or after the digits stop moving, either way.

But I think this is somewhat inadequate for more precision work.  I could improve the setup by triggering the measurements of both DMMs from the same source and writing an SCPI program to compute the ratio measurement for each simultaneous sample and average those computed values.  And maybe I'll do that; it would be an improvement.

Is there a better method than that?

Is there a device that's made to make ratio measurements?  I'm thinking like a 2-channel DMM where both channels can do voltage or current on different ranges, but it all hooks back to the same internal reference.

[bdunham7]

Do you mean DC ratio of a stable signal?  In that case, some DMMs have that as a built in feature, although AFAIK the samples aren't synchronized.  And they only do it for voltage over a limited range, so you'd have to prescale/convert any really large (or perhaps small) voltages and any currents.  Even with time-shifted sampling, if your signal is steady enough the fact that you are using one DMM with the same reference and input scaling gives you a pretty big advantage over 2 DMMs, at least for voltage ratios.

[slugrustle]

Yeah, DC ratio of a stable signal.  If the signal varies, it's usually due to thermally related shifts in a current sense shunt resistor or in my equipment.  Things are usually pretty settled after 30 minutes.

Although below say 0.25V, I wish I had a more stable voltage source than my ITECH IT6123B.  That thing is super nice in terms of being within 1mV across its range, typically, but at low voltages, even 100µV of drift matters.

Oh, perhaps I should have said: I thought to write this post while taking that measurement with 250mV and then 125mV input to hit a certain range on an amplifier.  I could see the signal drifting during the long term average, hence my thought to synchronize the meters.  Maybe I just need a more stable source for low voltages.

[bdunham7]

I could see the signal drifting during the long term average, hence my thought to synchronize the meters.  Maybe I just need a more stable source for low voltages.

If the drift is slow, the short time interval between alternate samples would be very small.  Your average and SD on the direct ratio measurement would give you a pretty good idea of the gain constant that you are looking for as well as the uncertainty of the measurement.  The only drawback is I think most meters with a ratio function require the two signals to have a common. 

[slugrustle]

Yeah, agreed. I think drift between samples, even at 10 NPLC integration time, would be negligible.

I'll look around at meters with ratio measurements.  But restrictions on ranges between the two measurements would probably be a show stopper, at least in some situations.  I'm sometimes looking at amps with high DC gain.

And maybe synchronizing my two DMMs, or even using bus triggering with SCPI and not even synchronizing via a trigger mechanism, would be good enough.

[Ian.M]

For  metrology grade DC voltage ratio measurements, see https://en.wikipedia.org/wiki/Kelvin%E2%80%93Varley_divider,
used with a galvo or other sensitive null meter.

[bdunham7]

I'll look around at meters with ratio measurements.  But restrictions on ranges between the two measurements would probably be a show stopper, at least in some situations.  I'm sometimes looking at amps with high DC gain.

And maybe synchronizing my two DMMs, or even using bus triggering with SCPI and not even synchronizing via a trigger mechanism, would be good enough.

I just tried this using a Fluke 8846A with a 1mV and a 10V input (completely separate and uncorrelated sources), nothing really warmed up and I got 100.0xxx µ (meaning .0001000xxx) and 10.00xxx k (meaning 10,00x.xx) where xxx was 300-400 counts.  So better than 0.1% ratio error with a 10,000X difference.  I can try a 34401A and see if it performs similarly.  I don't know what the limitations on inputs are, I'd have to read up on that.

Two DMMs may in fact be just fine depending on what accuracy you need and what they are.  But I wouldn't use the average function on the meters and then compare the results, I'd use logging software and do the math per sample as you proposed originally.  Of course the ratio function is sort of nice because it just gives you the answer directly.  What meters do you have to work with?

[Berni]

Yep some DMMs have built in ratio measurement, but this is normally only for voltage. Since the way that works is typically that you hook up the regular red and black inputs to one thing and then hook up the sense inputs to the other thing. The DMM then just toggles between measuring one or the other. So it is mostly a software feature that makes clever use of hardware that is already there in most DMMs.

If you want something that does ratios for anything, big and small values or ratios between currents and voltages and such, then two DMMs is the only infinitely flexible solution.

As long as what you are measuring is stable DC, you start the measurement at the same time on two DMMs of the same model you are going to get very good ratio results. Once DMMs warm up they are incredibly stable at short time scales. A larger source of error would likely be the two DMMs being a bit off in absolute reading. You could calibrate them against each other if you present them with a known very accurate ratio and do math backwards to remove the error.

[slugrustle]

For  metrology grade DC voltage ratio measurements, see https://en.wikipedia.org/wiki/Kelvin%E2%80%93Varley_divider,
used with a galvo or other sensitive null meter.

I think the input loading of the divider would be a bit heavy for some of my tests, and the input signal might drift a little faster than i could establish a good null by hand.

I just tried this using a Fluke 8846A with a 1mV and a 10V input (completely separate and uncorrelated sources), nothing really warmed up and I got 100.0xxx µ (meaning .0001000xxx) and 10.00xxx k (meaning 10,00x.xx) where xxx was 300-400 counts.  So better than 0.1% ratio error with a 10,000X difference.  I can try a 34401A and see if it performs similarly.  I don't know what the limitations on inputs are, I'd have to read up on that.

Two DMMs may in fact be just fine depending on what accuracy you need and what they are.  But I wouldn't use the average function on the meters and then compare the results, I'd use logging software and do the math per sample as you proposed originally.  Of course the ratio function is sort of nice because it just gives you the answer directly.  What meters do you have to work with?

Very cool result.  I've got two Rigol DM3068s.  I bought the first one because it was a low price for a 6.5 digit meter and the second to have a second meter with the same SCPI commands.  If I had a time machine, I'd go with Keysight.

If you want something that does ratios for anything, big and small values or ratios between currents and voltages and such, then two DMMs is the only infinitely flexible solution.

It is looking that way.

As long as what you are measuring is stable DC, you start the measurement at the same time on two DMMs of the same model you are going to get very good ratio results. Once DMMs warm up they are incredibly stable at short time scales. A larger source of error would likely be the two DMMs being a bit off in absolute reading. You could calibrate them against each other if you present them with a known very accurate ratio and do math backwards to remove the error.

The calibration thing is an interesting idea.  I'll start with the synchronizing strategy, which is already a step up from my current practice, and see how that goes.