what happens to the signal/samples in a DMM?

[k8943]

I've watched a few multimeter teardowns and there don't seem to be a ton of components between, say, a voltage coming in and the ADC.

Anyone know how they make the most of a signal?

For example, Flukes seem to provide 4 numbers a second (display refreshes).

That leaves 250ms to capture a bunch of data and post-process it.

What electronic filtering?
How many samples? Like 100, 1000...?
How many bits the ADC? Couldn't find this anywhere.
What sorts of mathematical operation on the collected data?

Have messed about measuring voltages with a microcontroller and would like to know how the pros do it. 

[tszaboo]

DMMs dont have the typical SAR or Delta sigma ADC*. They use a so called slope ADC.
https://en.wikipedia.org/wiki/Integrating_ADC

It is basically charging a capacitor with some current and measuring time. There are no "bits", that you traditionally think of, when talking about an ADC.

*Of course some DMM will do it different way.

[k8943]

Cool! Thanks.

Looked at at datasheet of a random SAR from digikey and the notion of bits still somehow pertains: the number of bits in the result.

Do they take lots of readings - or does it really take 250ms for a SAR to do it's thing?

[TiN]

k8943, you might not be happy about it, but here is HP journal that covers ADC conversion and operation and other stuff happening inside HP 3458A bench/system meter. In advanced bench meters you can set own sampling time, desired resolution , post-processing filtering settings, addition math like log/mx a+b/relatives etc.

[k8943]

Happy? I'm delighted!

Thanks a bunch @TiN - will enjoying going through that in the evening.

Had also found what may be like the sort of thing found in multimeters and will browse the datasheet!

https://www.digikey.be/product-detail/en/texas-instruments/ADS8924BRGET/296-45080-1-ND/6562093

Thanks

[TiN]

Older manuals from HP, Fluke, Keithley often contained section "Theory of operation" and "Troubleshooting / service" which explain how device operate, perform measurement and provide it's performance. Too bad modern equipment does not go into these details anymore. 

[David Hess]

They integrate the input signal for a predetermined amount of time which is usually a multiple of 50/60 Hz.  This sets a lower limit of 20/16.6ms minimum per sample which might result in a sample rate of 25/30 samples per second (1) but often 100 milliseconds is chosen to cover both 50 and 60 Hz rejection.  This creates a sin(x)/x frequency response with nulls at the power line frequency and its harmonics removing them as a source of noise.

Longer integration times may be chosen to further reduce noise.  Datasheets may list "x PLC" or the number of power line cycles for the chosen resolution.

Dual slope converters, delta-sigma converters, and voltage-to-frequency converters all integrate their input in this way making them especially useful but a sampling converter could also be used and the integration done by adding (or boxcar averaging) all of the samples taken over the integration period.  Note that this does nothing to improve the error from integral non-linearity so do not count on getting better accuracy out of a microcontroller ADC by doing only this no matter how much the noise is reduced.

(1) Or a little faster depending on the implementation and how much time between samples is required.

[RobK_NL]

The datasheet for the venerable ICL7106, used in countless multimeters over the ages, contains a good description too:
http://www.farnell.com/datasheets/2041300.pdf

[David Hess]

Before the ICL7106, you had the Fairchild 3814 and the Siliconix LD120/LD121A and LD111(A)/LD110.  The Siliconix parts were probably used in some early handheld multimeters.

Usually you run out of integral non-linearity before running out of signal to noise so just averaging for more resolution is of limited benefit except insofar as it creates a sin(x)/x frequency response to remove power line noise.  Noisy sampling converters are an exception to this and may always benefit from averaging.

Discrete charge balancing voltage to frequency converters in one form or another used to be very common also and were probably found on many early handheld DMMs.  They can easily achieve 12 bit linearity and resolution so 3-3/4 digits.  Modern implementations can get to 20,000 counts or 4-1/2 digits of linearity with an operational amplifier and a charge pump IC which is 10 times better than most 4-1/2 digit multimeters are specified to.  Handheld DMM accuracy has not improved in decades.

If I was going to hack together a DMM from scratch, I would use a modern inexpensive sampling 16-bit successive approximation converter like a $10 LTC2376-16 and use the microcontroller to implement the integration.  This allows skipping the whole analog AC to DC and RMS conversion part because these are trivial to do digitally and may even be calculated in parallel and some DMMs made in the past couple of decades work something like this.  This will work up to at least 4-3/4 digits; above that, an external integrating converter will be required for better integral non-linearity.

[Kleinstein]

Quite a few modern DMMs use sigma delta conversion: the cheaper ones as part of a special DMM chip, many 5 digit ones separate ADC chips and higher resolution ones often use an integrating ADC made from separate OPs and lower resolution ADC, FPGA/µC, switches and so on - often also resembling a sigma delta ADC.

The integrating converters are often run with some 100 ms integration time and not much processing of the data is done. There can be some averaging / filtering and with high grade meters often there are alternating readings of 0 and the input signal and the result is the difference between those readings. With modern meters there usually is a numerical multiplication to convert from ADC steps to displayed voltage. So the displayed steps may not reflect the ADC resolution, but the ADC itself usually has a higher resolution.

The way of doing fast sampling and averaging is not limited to SAR converters. There are also a few fast SD converters (build for use in power meters) like the cheap MCP3911 that could be used this way.