8.5 digit multimeters and measuring performance of a 20bit DAC

[new299]

I've been working on a 20bit +/-10V DAC board, based around the AD5791 [1]. The board design is complete and I've had the layout looked over by an analogue design engineer. It works, but I'm now thinking about how to test its performance. It's been suggested that a 8.5 digit multimeter would be valuable in this regard (unfortunately I only have a 6.5 digit meter).

Looking around I see that there are many 24+ bit precision ADCs available. Multimeters like the 3458A appear to use a custom multislope ADC architecture. Why are commodity ADCs unsuitable? Is this due to poor linearity/"missing codes"? Poor DC performance? Is it possible that I might be able to use a commodity ADC to verify the performance of my DAC over a smaller voltage range?

I don't care much about the absolutely performance of the DAC (i.e. if there's some DC offset).

[1] http://41j.com/blog/2015/10/ad5791-board-rev4/

[exmadscientist]

Here be dragons... this is really not a beginner's project. I do hope you have read LTC AN86 cover-to-cover and back before attempting this project....

[Kleinstein]

The integrated 24 Bit ADCs are not very good when it comes to linearity - here the 6.5 Digit DMM is likely better.
You can do some tests with such an ADC, but not a complete characterization. It also needs more thought into it, to separate Errors of the DAC und ADC.

[lukier]

Can you say a bit more about your application?

It seems strange that you are buying one of the most precise and expensive single chip DACs on the market, but at the same time don't care about the DC performance.

I agree, AN86 from Linear is a must read. At such precision and linearity things get very very tricky. Thermocouple effects, noise, long term drift and so on. AD5791 eval board has space for LTZ1000 reference to really match the specs of this DAC.

3458A might be allright for characterization, it has superb ADC, but due to the sheer number of codes it will take a while to measure.

Common single chip ADCs are getting close to 6.5 digit meters (LTC2442 = +- 1ppm INL), but you will also need some signal conditioning front end with matching performance (e.g. 10V to 2.5V division). Characterizing them is also not trivial. If you have 140 dB THD sine generator you might be able to use the histogram method - it will require a lot of samples. More info here:

https://www.maximintegrated.com/en/app-notes/index.mvp/id/2085

[new299]

Can you say a bit more about your application?

It seems strange that you are buying one of the most precise and expensive single chip DACs on the market, but at the same time don't care about the DC performance.

It's for a scanning probe microscope. Basically to drive a piezo actuator. It's not critical if there's some DC offset (as that will just offset the position) but should ideally be linear and have have good resolution. Thanks for your other notes and suggestions.

[lukier]

It's for a scanning probe microscope. Basically to drive a piezo actuator. It's not critical if there's some DC offset (as that will just offset the position) but should ideally be linear and have have good resolution. Thanks for your other notes and suggestions.

Hmm, maybe PWM-DAC then. Not the fastest settling, but It should be linear and with dual MSB/LSB switches it can be precise. Check TiN's thread on HP 3245A.

[new299]

It's for a scanning probe microscope. Basically to drive a piezo actuator. It's not critical if there's some DC offset (as that will just offset the position) but should ideally be linear and have have good resolution. Thanks for your other notes and suggestions.

Hmm, maybe PWM-DAC then. Not the fastest settling, but It should be linear and with dual MSB/LSB switches it can be precise. Check TiN's thread on HP 3245A.

I guess at least initially I've been a little scared of anything PWM/with higher frequency components due to radiated/other noise. There are also pico/nanoamp current measurements happening in the system. But I will check out that thread thanks.

[Kleinstein]

A resonable integrated 24 Bit ADC should be better than the requirements for a scanning probe microscope. So for tests to this level it should be OK. So the AD5791 is pretty overkill. Likely even an average Audio DAC might be good enough. Piezos are usually not that linear and stable.

[Dr. Frank]

Hello new299,

according to the paper you linked about the AD5791 DAC, it should have an integral linearity of < 1ppm.
Noise vs. speed is another important parameter, which is not specified there, but may well influence your scanning microscope..

Several kHz of stepping is common for these, as far as I remember.
So, settling time comes on top.
High sampling rates usually reduce the effective resolution, that also goes for your DAC.


To characterize these 3 parameters, you'll need something better, i.e. linearity << 1ppm, 1ms A/D speed with high enough resolution, low noise.

6 1/2 digit DMMs from HP / Keysight are at par - at most - with your DAC.

Other 24bit ADC suffer either from non -linearity, or from noise or both, and therefore are not suitable.

The 3458A therefore is the best choice for your needs. It's got 0.02ppm linearity, < 0.1ppm / < 1uV /10V noise and resolves 6 1/2 digits down to 100usec aperture.

You may check TiNs thread about noise measurements on common DMMs, to search for a cheaper alternatives from Keithley or other manufacturers, if only noise has to be characterized.

Frank

[tszaboo]

Unless you have a pure resistive load, I dont think you can work without sense wires either. Also I dont think you can just ignore the reference voltage. The way I see it, unless your output is ratiometric, it has to be a lot better than an average joes reference voltage.
BTW those 24 bit ADCs have typically like 200 LSB INL.

[new299]

A resonable integrated 24 Bit ADC should be better than the requirements for a scanning probe microscope. So for tests to this level it should be OK. So the AD5791 is pretty overkill. Likely even an average Audio DAC might be good enough. Piezos are usually not that linear and stable.

Yes, it's likely overkill. However I'd like to get the whole system up and running before trying to reduce the DAC specs. I've heard bad things about audio DACs but I'm interested in trying them out. I guess part of this exercise is learning how to characterize the performance of the DAC better.

[new299]

8.5 digit corresponds to 28 bits, how can you verify 28 bits with 20 bits? Let along your test gear should be more accurate than DUT (unless you are using the error distribution and averaging tricks). At least a very ideal 32 bit ADC must be used, with insanely accurate reference, such as a well built LTZ1000ACH, and you need another uber accurate DVM to calibrate your LTZ1000ACH.

I'm trying to do the opposite I think. I.e. verify my 20bit DAC using an 8.5 digit DMM. Or alternatively find a cheaper solution. My understanding so far is that even with a LTZ1000ACH their is no 32bit ADC available which would match the specs of an 8.5 digit DMM. I'm interested in suggestions though.

[new299]

Unless you have a pure resistive load, I dont think you can work without sense wires either. Also I dont think you can just ignore the reference voltage. The way I see it, unless your output is ratiometric, it has to be a lot better than an average joes reference voltage.
BTW those 24 bit ADCs have typically like 200 LSB INL.

I don't think I can ignore the reference no. I guess I care more about it's short-term stability than absolute accuracy. By sense wires, I take it you mean "remote sense" type wires? I think if I was going in that direction it might be better to use closed loop Piezo actuators. But it's worth thinking about, thanks.