ADI shows "How to Achieve 7.5-Digit Accuracy in Instrumentation Applications"

[laichh]

Not too long ago, 6½ digit multimeters were rare & expensive instrument. Today, you can built your own 7½ digit instrument like LEGO.



https://www.analog.com/en/resources/technical-articles/achieve-7-pt-5-digits-accuracy-instrumentation-apps-part1.html

https://www.analog.com/en/resources/technical-articles/achieve-7-pt-5-digits-accuracy-instrumentation-apps-part2.html

[Kleinstein]

There tests showed quite some difference in the INL when changing the reference. This is not that surprising as those capacitive ADCs react to changes in the reference and input drivers.  So while they have impressive INL specs for the chips, it still needs a test on the complete setup with drivers and the layout.

It is interesting that the AZ amplifiers seem to be good enough to drive the ADCs - it is not obvious from the data-sheets and good to know that there are relatively simple (1 OP-amp and not the 2 OP-amp solution in the CERN HPM7177) solutions that work.

The INL specs for the DMMs likely incluce quite some saftly factor for no doing a full test - so comparison to a test with 4 boards and a limited number of test voltages is not really fair - probably need to add a factor of 3 or so.

For the input current the usual DMMs are better than the ADA4522 - so for the input one may have to look for lower bias buffers. This is likely not an issue though as the noise at the very input is not that critical.

[BarrowBoy]

Not too long ago, 6½ digit multimeters were rare & expensive instrument. Today, you can built your own 7½ digit instrument like LEGO.

Very interesting. Once again full disclose, I'm involved in the distribution of 8.5 DMM's here in Australia. With your above statement, and reading through Part 2, can I conclude that the maximum input is 5V? I'm not a Engineer, and it's a serious question. Secondly, how much confidence would others have in your measurements using a device as this, until it was made  commercially available?

[KT88]

I would consider this more as a reference design. As is, it won't be a replacement for a bench DMM. The usecase would be more something like automated test equipment where cost is a driver. High channel count systemes would benefit a lot. Also, why having current and Ohms capability if one needs only accurate voltage measurements or other 'one-trick-pony' requirements...

[Kleinstein]

The ADC itself has a +-5 V or less input range with differential inputs.
The front end part dicussed in the article has an input divder to get a +-10 V (or similar range). It is more comparable to some data aquisition boards.

The circuit input is not fully compable to a good DMM. There is little input protection and the input bias current it higher than most 6 digit or better DMMs.
For the noise it can also matter how the noise depends on the frequency / time frame. A longer time frame may add low frequrency noise and most commercial DMMs use switching for some auto zero mode to reduce the low frequency noise.

The SAR type ADCs are quite good at high speed. So an interesting case can be the AC power measurement appliction. However here the ADC is only part of the circuit and the current part likely does not need the 10 V range.

The CERN HPM7177 (using AD7177 ADC) shows that these ready made ADC chips have impressive performance and compete with 7 and even 8 digit DMMs in some applicatoins.

[iMo]

The ADR1000 lobby here would do  ..
 

[laichh]

Meanwhile, TI has different opinion:

https://www.ti.com/tool/TIDA-010945

[ivo]

Very interesting. Once again full disclose, I'm involved in the distribution of 8.5 DMM's here in Australia. With your above statement, and reading through Part 2, can I conclude that the maximum input is 5V? I'm not a Engineer, and it's a serious question.

Designing something which can read a value reliably to a useful 7.5 digit precision is what's claimed here. That is laudable in itself. But commercial DMMs don't have to quake in their boots just yet, they have plenty of other features they offer that is not covered by the previous sentence. Can the <value> be anywhere from mV to hundreds or thousands of V? Can you stick mains AC up the signal path with no worries? Do you offer AC measurements, or resistance measurements, or current measurements? Can you read truly negative values? Is your measurement circuitry fully isolated? Do you have a nice display and interface and a rugged box? A commercial warranty and guarantee of accuracy offered? etc, I could go on. That said there's not a lot stopping someone from designing all these features extra around the platform to also make themselves a new commercial DMM; but that's just as much if not way way way more work than the "now simple" signal acquisition path.

Secondly, how much confidence would others have in your measurements using a device as this, until it was made  commercially available?

While it is a homebrew device, probably exactly as much confidence as they have respect and trust for that person on this forum or IRL.

[BarrowBoy]

Very interesting. Once again full disclose, I'm involved in the distribution of 8.5 DMM's here in Australia. With your above statement, and reading through Part 2, can I conclude that the maximum input is 5V? I'm not a Engineer, and it's a serious question.

Designing something which can read a value reliably to a useful 7.5 digit precision is what's claimed here. That is laudable in itself. But commercial DMMs don't have to quake in their boots just yet, they have plenty of other features they offer that is not covered by the previous sentence. Can the <value> be anywhere from mV to hundreds or thousands of V? Can you stick mains AC up the signal path with no worries? Do you offer AC measurements, or resistance measurements, or current measurements? Can you read truly negative values? Is your measurement circuitry fully isolated? Do you have a nice display and interface and a rugged box? A commercial warranty and guarantee of accuracy offered? etc, I could go on. That said there's not a lot stopping someone from designing all these features extra around the platform to also make themselves a new commercial DMM; but that's just as much if not way way way more work than the "now simple" signal acquisition path.

Secondly, how much confidence would others have in your measurements using a device as this, until it was made  commercially available?

While it is a homebrew device, probably exactly as much confidence as they have respect and trust for that person on this forum or IRL.

Agree with all above. As mentioned it was a question of interest. 

[CurtisSeizert]

I wonder what that source is. It seems to have linearity error on about the same level as the 3458A as the two are probably uncorrelated.

There tests showed quite some difference in the INL when changing the reference. This is not that surprising as those capacitive ADCs react to changes in the reference and input drivers.  So while they have impressive INL specs for the chips, it still needs a test on the complete setup with drivers and the layout.

It is interesting that the AZ amplifiers seem to be good enough to drive the ADCs - it is not obvious from the data-sheets and good to know that there are relatively simple (1 OP-amp and not the 2 OP-amp solution in the CERN HPM7177) solutions that work.

Notice that they did the INL tests at 62.5 kSPS rather than 1 MSPS. I would guess the AZ amps can't recover from the charge kickback quickly enough to get decent linearity at the higher sample rate. It would have been cool to see how they could do with a composite amp. Honestly, the thing that surprises me most in this article is that they were able to manage 200 ppb linearity error with 5k/1k25 attenuators. Presumably with some optimization and the use of a composite amp for the ADC driver, one could bring the INL down even more.

As others have said, being able to digitize from a low-impedance source at 24+ bits is not the same as having a proper voltmeter. Even without current and resistance measurements, no commercial op amp has everything people ask of a 7.5 digit meter input amp, and the meter needs noise immunity, isolation, input protection, etc. Then again, not everything needs to be a voltmeter.