Why aren't 16-bit resolution meters common?

[Circlotron]

16 bit A/D converters are everywhere and not particularly expensive. Even 24 bit A/Ds are readily available. Why then can't you find heaps of meters with a 65000 full scale?

[Zbig]

I'd say that's the indication that marketoids haven't yet taken the T&M field completely and, needless to say, I consider it a good thing. Resolution does not equal precision. There's much more to the DMM than the ADC alone, very far from that. Each part of the signal path greatly contributes to the precision of the measurement, each design decision and component selection. The last thing you want is a meter with an illusion of great accuracy - lots of digits jumping all over the place with no rhyme or reason, drifting with miniscule temperature changes, susceptible to slightest external interference, depending upon the phase of the moon and whether you look at it funny. That wouldn't be all that different from the megapixel wars we're seeing in the photo / smartphone industry - ever-increasing sensor resolutions with no corresponding progress in terms of optics. By sticking a 24-bit DAC in a cheap, poorly designed DMM, you'd end up with something akin to the bargain bin digital camera - tiny, plastic, single-element lens the size of a pinhead and a colorful sticker proudly boasting "bazilion megapixels" I, for one, don't have any need or desire for anything like that.

[Marco]

Meh ... a good tracking resistor divider is what, 5$ in singles? A temperature controlled reference can be build for next to nothing (still won't get you absolute accuracy, but getting short term differential accuracy is cheap, hell maybe you can even get rid of the precision divider this way). Auto-zero calibration can also be done for next to nothing.

Getting to 24+ bits is difficult, but the lack of budget intermediate precision meters is more a question of most people not giving a shit and artificial market separation.

[SeanB]

16 bit ADC that is both linear and monotonic to 16 bits is not something that is cheap. the common 16 bit converters are generally 12 bits before the noise is getting too much, and 10 bits when you start to get the missing code and monotonicity issues. Many onboard MCU 16 bit converters can only get 10 bits using averaging over 100 or more samples, and even then you have to sleep the processor during each conversion so that the noise is low enough to get more than 8 bit results.

You tend to find 16 bit ADC units that actually do a monotonic 16 bits, and which have no missing codes are made using high quality references, come in an expensive package and also have a price that is a lot higher than a 16 bit device for audio applications.

[Marco]

A 24 bit ADC with 10 ppm INL is a couple bucks in singles.

[krivx]

A 24 bit ADC with 10 ppm INL is a couple bucks in singles.

What's the part? You need your own reference right? FWIW I can't find any 24 bit ADC on digikey for $2

[free_electron]

dmm convertors are typically multislope charge balancers. totally different technology than those flash or sigma delta's used in cpu apps.

a multislope by design cancels out all errors. if its clock is stable then the conversion result is rock solid.

throwing 24 bit a/d at the problem solves nothing . you'll need oodles of maths to compensate for all the drift. and you'll need that for every range and feature.

in other words : classic converters are not suitable for such application

[Marco]

Oops, I meant 2 euro bucks Any way MAX11202 and MCP3911.

[dom0]

I think they have +- 1 V max differential input... not really useful IMHO.

[Marco]

Unless you want a single voltage range you'll need precision resistor dividers in a multimeter regardless of the ADC.

Needing amplification even on low voltage ranges makes everything a bit more noise sensitive, but at DMM update rates and with just 16 bits it's not a huge issue.

a multislope by design cancels out all errors.

As long as self heating doesn't affect any of the resistors, as long as the resistors for the different slopes track each other well enough with temperature and are stable enough for the duration of the measurement, as long as the resistance of the switches for selecting the different slopes is either negligible (which they won't be) or at least don't throw off the temperature tracking of the resistor network too much ... the requirements on the resistors in a multislope ADC are extreme (-ly expensive). If you want to get much more than 10 ppm linearity there might not be an alternative, but when you don't there are.

[Kleinstein]

The ADC's made for audio usually are rather bad at gain stability and thus not very useful for a meter.

There are some cheap sigma delta ADCs, that may be useful in a cheap meter: e.g. MCP3421 (18 Bit), LTC2440 (24 Bit), ...
However the resolution quoted for sigma delta ADCs is usually the limit where you get no missing codes. A reading on a LCD should be reasonable stable and linear. So you want something like less than about 0.2 digit DNL and noise. So you can't just convert LSB steps of these ADCs to display steps. So a 24 Bit ADC like the LTC2440 may be good for a 5 1/2 digit or at most 6 digit DMM. Still the ADC is just one part, and not the critical one any more.

The dual slope converters like ICL7106/7 found in old 3,5 digit DMMs have only about 12 Bit resolution, but something like 0.01 LSB DNL. So even if you have a normal 16 Bit ADC and suppress the lowest 4 Bits, leaving only 12 Bits, the DNL errors are likely larger. In these old days adjustments were done analog and thus there was a direct relation from ADC steps to display. Modern meters tend to use digital adjustment and thus need something like 3 Bits of extra resolution to keep rounding errors small.

In principle the multislope charge balancing converters found in high end DMMs (e.g. 6,5 Digits) are not that different than sigma delta converters. However the converters are optimized for good linearity using expensive parts (especially resistors arrays)  and separate analog and digital chips. Internally the ADC in a 6 digit DMM will also have something like 24-32 bits of resolution.

[Marco]

The MCP3911 looks unbeatable on paper, a bit of subtractive dither might be able to improve it too since it can supposedly do pretty high sample rates at 15 bit plus ENOB.