ADC DC Measurement Error

I'm not taking the time to read the app note, but 3.4mV sounds about right for DC bias drift. Close and critical study of the data sheet is your friend, here.

In my experience most monolithic ADCs have crappy DC performance when compared to a cheap op-amp. I expect that your 12-bit example is NOT good to 2.44mV in the DC -- again, a close and critical study of the data sheet is called out.

The key consideration in all of this is DC bias (or offset), and its drift with temperature. In my experience that is generally the worst facet of an ADC's performance. When I need to get the most out of an ADC, I try to either choose a sensor that naturally has an AC output (i.e. an LVDT, resolver, or whatnot), or I put an analog switch in the front end of my electronics that's controlled by the microprocessor and I make the whole measurement system end-to-end chopper stabilized.

I'm pretty sure that there are 24-bit S-D parts that have much better DC performance.

Most d-s ADCs have phenomenal gain and offset specs. That particular ADI part is unusually bad in that respect, probably because it was designed to go into electric meters, where DC specs don't matter. Full-scale for the shunt input is only +-31 mV, so shunt thermocouple-voltage errors can dominate offset.

A little math processing will take out the offset errors when AC metering.

Thanks for the responses, I'm feeling like this part is no good for DC, it doesn't list a DNL or INL spec in the datasheet, is that parameter usually listed for ADC's?

I think I can correct the offset values of the ADC in software, but can you also correct the bad temperature coefficient?

Why are the offset temperature drift and DNL/INL parameters listed in the DC specifications of the ADC, do they not matter for AC performance?

Much thanks for helping me understand this stuff!

I should have been a bit stronger than "pretty sure". I suspect that a considerable amount of the chip expense goes into the DC performance on a "regular" S-D chip, which is why one that only needs AC performance might be attractive -- assuming that _you_ only need AC performance.

And yes -- it's pretty easy to null out offset errors in software.

AN-1304.pdf

It's usually listed for "general purpose" ADC's. For ADC's that are going into specific markets like audio or radio, the ADC will be rated for some performance specification that is related to DNL or INL, but comes out differently.

For instance, an "audio" ADC may list THD (which is a function of DNL and INL), while a "radio" ADC may list signal/noise ratio, harmonic generation, and 3rd-order intercept.

Realistically, only with something like chopper stabilization. You can _try_ to characterize each ADC for its temperature coefficient, and measure temperature, and correct for it. If you try it you'll either have lifetime employment on the manufacturing floor or you'll be fired.

In general AC performance specifications are ones where the signal speed matters. Below a certain rate of change, the impact of INL and DNL are the same no matter whether a signal takes a second to ramp up or an hour. Ditto with offset temperature drift -- it's a function of temperature, not speed.

In general DC performance specifications matter at any speed.

For AC metering, you only need to null out one, voltage or current. Voltage is easier, because it's a pretty steady sine wave. Auto-zero that, and multiply the result times the current waveform, and you have the power.

With a little noise dithering on the current input, you can do ANSI C12-grade metering with an 8-bit ADC.

AN-1304.pdf

Oh -- and a cynical observation from an instructor of mine from college: if it's not in the data sheet, it's because the manufacturer is really not proud of how well the part does in that regard. Never expect a part to be better than what's detailed in the data sheet.