Strange Hash measuring a Standard Cell

[Assafl]

So I am trying to learn how to stabilize the measurement at the low end of the DMM - and as a source I am using the temperature stabilized standard cell (in an unheated calorimeter).

Am getting decent readings, but can't figure out a certain "hash" of 1uV I am getting on the readings. In the chart below, the axis are as follows:
X axis is hours of the day (so roughly 31 hour span of readings)
Y axis Blue is in PPM units for the voltage (re a ref-level of 1.0176748 volts)
Y axis orange is Lab delta T in degrees Celsius (re a ref temp of 25C)

The HP34401A DMM is set to 100PLC, maximum resolution (@1V range), >10G and AutoZero on. Logging is on a RPi3.

The question I have is how do I deduce whether the has comes from the DMM or the cell (or some thermal EMF along the way)?




Notes:
1. I suspect the hash comes from the DMM so I tried shorting the DMM input. But I then didn't get the hash - so either it is an effect that happens at a full scale reading, or it is the standard cell's doing (but how could that be? does electrochemistry have effects that have 1uV "jumps"?).
2. Cables are copper and I put some lagging on the bananas (gold plated tellurium copper - assumedly...).
3. It should be possible to stabilize the reading further by heating the calorimeter so that the temperature regulation loop stabilizes completely.

[TiN]

Only true way to use multiple DMMs to determine if pattern matches. Preference number >3. So statistics will do the job for you
But in your specific case it looks like pink noise of 34401's reference. Similar I saw on flaky LTZ1000 chip on latest module build.

You don't see jumps on short due to zero input signal is incomparable to reference noise itself.

[Assafl]

Only true way to use multiple DMMs to determine if pattern matches. Preference number >3. So statistics will do the job for you
But in your specific case it looks like pink noise of 34401's reference. Similar I saw on flaky LTZ1000 chip on latest module build.

You don't see jumps on short due to zero input signal is incomparable to reference noise itself.

Thank you for the quick reply.

Silly me... You hit the nail on the head that my zeroing the input would (by virtue of the slope integration) be very much hidden in the noise at 0V. The measurement needs to be at (or nearly at) full range to show the hash. So another full 1V source should show the same response.

Trouble is non of the Datels are anywhere as stable to show the effect. I even tried using them with a KVD and that failed. Maybe I should try to put them all in the calorimeter... It seems that is similar to the Geller Labs identified hash (http://www.gellerlabs.com/DMMtempco.htm):

Agilent 34410A (from 2006) - was very near to zero ppm / c. Ironically during this testing, it became clear that the internal LM399 was from time to time causing a 1 ppm pk to pk "hash". The failing LM399 PDF (or simply intermittently noisy part) was replaced with a LM299AH-20 part. The "hash" noise is gone, however now the DMM has a tempco on the order of 0.25 ppm / c.

I was hoping the Muirhead would negate needing any more DMMs as it can show me whether the HP has wandered off. That will help me avoid becoming a voltnut.

I do wonder how does the mechanism of the intermittent noise work? Does it randomly start generating Avalanche noise? What cause it to start? I also guess it is possible to replace it - but it is well below the specs of the 34401A anyway...

 

[Kleinstein]

If it is about noise from the DMM internal reference, one could repeat a similar measurement in the 10 V range. This way the noise from the input amplifier and cables gets reduced. One would need something like stable 5-10 V source. Some batteries are supposed to be relatively stable.

This type of noise sometimes happens. Some devices are more prone than others, but there is no clear separation like LM399 has it and LM499 not. They are more like very similar. It is more like luck on how much of this noise a reference shows - so chances are it was this way from the beginning. One possible mechanism is a electronic state (usually due to an defect) can change the device current or voltage so much it gets visible - population (empty or populated) of such states has a certain probability and can thus change at a certain rate from populated or empty. Thus a more or less random on/off with a fixed step.

[Assafl]

Thank you - I am reading up on the popcorn/burst noise.

I am trying to put together a test using the 10V to see if I can create a voltage stable enough for this testing.

On the positive side, the 50$ standard cell seems to be very stable and very low noise (possibly more so than the LM399 - assuming I can stabilize the temp more).

[Assafl]

Indeed it seems like the HP34401A's LM399 that bursts with 1ppm noise. (unless both the standard cell and the Datel have the same popcorn noise).

So as per Kleinstein's suggestion I connected the Datel voltage calibrator set at maximum 19.999V through a KVD to get a ref level of 11.999V to the DMM.

Got the same hash as the Muirhead standard cell at 1.017..V - actually a bit higher that would reflect that the 11.999V is full range on the DMM.

The Y-axis is referenced to an 11.999v. The x-axis is in data points (about 4.5 hours worth of measurements).



BTW - by way of the S/N level in the measurements, I am not sure why I failed to make this out at 1.018v from the Datel (or 10.18v from the Datel and the KVD at 10:1). I guess I made a mistake in the connections. I'll try that again.

[Kleinstein]

If it is noise from the DMMs reference, the size should be always the same (around 1 ppm of the reading for this unit). So it should also be 1 ppm at 5 V - this would make sure it can not be a glitch for the ADC itself. Already 1.017 V and 11.99 V should be sufficiently different for the ADC.

With 1.018 V from the Datel and divider, there might be more noise, to make the burst noise less visible.

[Assafl]

If it is noise from the DMMs reference, the size should be always the same (around 1 ppm of the reading for this unit). So it should also be 1 ppm at 5 V - this would make sure it can not be a glitch for the ADC itself. Already 1.017 V and 11.99 V should be sufficiently different for the ADC.

With 1.018 V from the Datel and divider, there might be more noise, to make the burst noise less visible.

Yup - I see what you mean. Since PPM is already relative to the reference (whether it is 10V or 5V). If it were uV (as ~PPM happen to be in the ~1.018V reading) - then for a 5V or a 0.5V reading the voltage would be 5uV or 500nV (but would be hard to discern at the 7th digit resolution ~300nV resolution).

As it stands - the 1PPM hash is rather hard to discern as it resides in the last 3 counts of the 7.x digit display, and 1 count (at most) for the 6.5 digit display. It also resides completely below the specifications and therefore passed calibration with aplomb...

So I don't think it is worth trying to fix this (even though the LM399 is socketed) for two reasons:
1. It is fairly easy to figure this out from a log.
3. I will rationalize this - I can use this has as a marker: for example, by seeing the hash I know noise is pretty low...

Assaf

[David Hess]

It looks like popcorn noise to me.  It does not necessarily have to come from the reference; any bipolar transistor in the signal path can produce it.  It is difficult to test for because of the long test time required and it can still manifest later anyway.

[Kleinstein]

A test at 5 V or similar could tell if the noise is from the reference (or an amplifier closely related to it), or an amplifier in the ADC. As the relative size is the same in the 1 V and 10 V range it can not come from the input amplifier.

Even if you change the LM399, things might not get better. The noise is still well within the noise specs of the LM399.