Estimating A/D noise in a Keithley 2002

[martinr33]

The K2002 has some detailed diagnostics. One of the tests, 20.1 "Test Cal Zero" reports the direct output from the A/D converter for 0V; the 7V reference; and the 1.75V reference. These voltages use switches that are very close to the reference, and thus independent of the input conditioning, protection and scaling system.
The number is a 9-digit positive integer, and zero is represented by a number like 153,656,331. By running the test multiple times (clumsy, but practicable), you can pull enough datapoints to see what the A/D is actually doing.

In the table below, I took 10 readings. The last column is the standard deviation of the 9 3-digit readings; I only recorded the last three digits for reading 2 - 10.
The number represents about 1.2 counts per 100 nanovolts.
Therefore, the SD of the meter is about 400nV on the 20V range. This is roughly equivalent to 200nV on a 10V range.

The SD of  the meter with a fixed voltage input can (rarely) get this low, or can go a bit higher. Depends on what the environment is up to.

The noise in the meter does not seem to see any incremental noise from the references.

At some point, I will need to implement TiN's op amp mod to the A/D board, and see if things change.


201.1   Reading          1          2      3      4      5       6       7      8      9     10     SD
Test Cal Zero   153656331   329   330   341   340   339   336   337   335   328   4.90    (means .5uV on the 20V range)
   7V ref          240021466   452   454   463   448   461   450   457   458   458   5.02
   1.75V ref     175247172   178   173   164   166   176   171   171   172   179   5.04

[Kleinstein]

When measuring zero or a voltage directly derived from the reference used for the ADC there will not be much noise visible from the reference. It is only if there is a second reference or significant filtering included that reference noise get visible. So not a surprise to see about the same noise level at 0 and 7 V - a little surprise to me that the 1.75 V level noise is similar low if this 1.75 V level is actually generated from the switched capacitors divider.

Just 10 readings is a rather short time series to calculated SD. It could be interesting to get a longer series so one could also see how much 1/f noise the ADC has. Comparing this to the noise seen in normal AZ mode, might give some insight in the AZ mode used by Keithley - it seem to be not very good with the 7510 and also the data one 2002 I have seen so far look a little suspicious.

[TiN]

op amp mod to the A/D board, and see if things change.

That did not render any visible difference,  in my case.

I did compare 2002 and 3458A before, and on base 10 VDC range their noise is about the same. Here's comparison graph with same LTZ 7V measured by both.
K2002 is purple line, 3458A - green samples. Here's another comparison, 10VDC from LTZ-powered 3245A, 2 x 2002 (olive+purple), 2 x 3458 (lime + blue) and 3 x 2001 (brown,red,green).

[martinr33]

Thanks for the comments. 

1) It is quite painful to get these numbers, which is why there are so few of them. They are only accessible through the diagnostics menu, and you have to keep exiting, re-entering and rerunning the diagnostic to get the number. However - they are also a direct read from the ADC, with no other software and minimal hardware in the way.
I wonder if there is a way to pull this result over GPIB.

2) More data won't tighten these numbers up - but it will expose long-term zero drift. 500nV seems a bit noisy to me, although that is on a 20V scale and likely with no line sync. 

3) The numbers are not derived from the reference - they are the absolute value from the ADC. That makes these numbers interesting because, unlike regular readings, there's no calculation involving the reference.  In this case, the reference noise is not visible. I'd attribute that to the 0.5uV standard deviation I observed in  the zero measurement, which represents baseline noise in the ADC.

4) These numbers have no filtering or averaging on them - they are just an immediate count. So I could be seeing power line noise. I've seen this particular meter report noise below this level on a 10V reading.

5) TiN's numbers appear to have a higher SD for both meters when measuring 10V - 800nV - which means that TiN's input might be a bit noisier, making a comparison hard.


 


 

[TiN]

However I don't quite understand what would you expect from meter, given it's own 7V reference noise in par of 1uVpk-pk anyway?
So even if ADC can deliver lower noise readings, it would be buried in reference noise anyway on actual signal measurement.

[Cerebus]

...
3) The numbers are not derived from the reference - they are the absolute value from the ADC. That makes these numbers interesting because, unlike regular readings, there's no calculation involving the reference.  In this case, the reference noise is not visible. I'd attribute that to the 0.5uV standard deviation I observed in  the zero measurement, which represents baseline noise in the ADC.
...

Yes, they are derived from the voltage reference because the reference is used to derive the charge balancing currents in the ADC. So when the ADC is measuring ground you'll be seeing uncorrelated reference noise from the ADC (obviously plus the ADC's self-noise); when the ADC is measuring the reference you'll be seeing correlated reference noise from the ADC which ought, to some extent, tend to balance itself out.

[Kleinstein]

The readings should should not include much reference noise (only a rather small part somewhere from higher frequencies used for modulation in the ADC).  The zero is likely just shifted by a kind or arbitrary number to make it always positive. The final zero reading with shorted input will include essentially no reference noise - so will the raw numbers. For the 7 V reference the input and the reference currents are derived from the same reference thus this will be a radiometric reading and thus also largely independent of the reference.

If the readings are taken by hand, they are also taken quite slow. This will cause them to include quite a lot of the low frequency noise, like the typical 10-0.1 Hz LF noise bands found in the datasheets. There are 6 OPs involved: 2 for each polarity, even with some gain in the reference current sources, the main integrator and likely the input buffer of the ADC. It is not a surprise to see RMS variations of this size. A real reading with AZ on would do two readings in short sequence and this way eliminate much of the low frequency noise (if AZ is done the right way). Despite of taking the difference of 2 readings this can be lower noise than a single reading. The test reading would be more like non AZ readings separated by several seconds - the Alan deviation is not that great for this too.

The readings with a shorted input and the 20 V range should also be mainly due to ADC noise. The input buffer is not expected to add really much noise compared to the ADC itself. I don't think the test mode will eliminate much more than the buffer amplifier and input protection.