USA Cal Club Round 3

[RandallMcRee]

Round 3 members,

The person scheduled for the kit is not able to take it, for personal reasons.

Does anyone currently waiting want to go next? Please PM me.

Thanks,
Randall

[alm]

Mathematically, there clearly is a correlation, though?  On four different DMMs measuring at the same time, in parallel?  I'm not sure what you mean...  are you saying the correlation that is visible is just a mathematical artifact and should be discounted?

I think temperature is a confounding factor here. The measurement depends on temperature, and relative humidity depends on temperature (even if the amount of moisture in the air remains the same), so the measurement could also be modeled as a function of humidity, using humidity as proxy for temperature. I'd think that if you were doing a regression analysis, then using temperature with multiple lag values would explain the variance just as well as humidity. It is likely to me that there are different time constants at play for temperature: connectors, DUT and different parts of the meters won't all react to temperature changes equally fast.

[SilverSolder]

Mathematically, there clearly is a correlation, though?  On four different DMMs measuring at the same time, in parallel?  I'm not sure what you mean...  are you saying the correlation that is visible is just a mathematical artifact and should be discounted?

I think temperature is a confounding factor here. The measurement depends on temperature, and relative humidity depends on temperature (even if the amount of moisture in the air remains the same), so the measurement could also be modeled as a function of humidity, using humidity as proxy for temperature. I'd think that if you were doing a regression analysis, then using temperature with multiple lag values would explain the variance just as well as humidity. It is likely to me that there are different time constants at play for temperature: connectors, DUT and different parts of the meters won't all react to temperature changes equally fast.

Interesting thought, that could well be true.  It doesn't seem intuitively right that things should be this sensitive to RH...


For the next cal club round,  I will run the same experiment for a much longer time period and hopefully shed some light on the matter!

[CatalinaWOW]

Mathematically, there clearly is a correlation, though?  On four different DMMs measuring at the same time, in parallel?  I'm not sure what you mean...  are you saying the correlation that is visible is just a mathematical artifact and should be discounted?

I think temperature is a confounding factor here. The measurement depends on temperature, and relative humidity depends on temperature (even if the amount of moisture in the air remains the same), so the measurement could also be modeled as a function of humidity, using humidity as proxy for temperature. I'd think that if you were doing a regression analysis, then using temperature with multiple lag values would explain the variance just as well as humidity. It is likely to me that there are different time constants at play for temperature: connectors, DUT and different parts of the meters won't all react to temperature changes equally fast.

Interesting thought, that could well be true.  It doesn't seem intuitively right that things should be this sensitive to RH...


For the next cal club round,  I will run the same experiment for a much longer time period and hopefully shed some light on the matter!

If you are interested in humidity effects you might want to create your own reference (there are several threads on the subject) and conduct your experiments using that.  Even without tracing your reference to a standard it can be used to evaluate variation.  And the answer will come much more quickly.  The cycle time of the club reference is measured in years.

[dmmguy]

Hello everyone,

Does this club also include Canada?

Can you send the kit to Canada?

Thanks

[RandallMcRee]

Hello everyone,

Does this club also include Canada?

Can you send the kit to Canada?

Thanks

Unfortunately, going outside the United States entails extra costs and custom forms. We have decided not to do so, at least on this round.

[RandallMcRee]

Hello all,

Unfortunately, the unthinkable has happened!
The FX reference sustained some damage.

Although minor, the fix requires de-certification.  Therefore, it will be a month or so whilst Illya graciously performs surgery and re-certifies.

Thanks Illya (bow)

Randall
Your bearer of bad news!

[rsjsouza]

Well, I am the idiot that broke one of the knobs of the voltage reference. For that, I am very sorry for Randall, Illya and the rest of the cal club for the delay.

Randall asked me to provide a summary of "not what to do". I can tell that one should never work with such delicate instruments when one is sleep deprived and with the judgment impaired. I was fixing a bad connection and exerted pressure beyond reasonable on the knob. What was supposed to be a relaxing activity after a long day turned out to be a nightmare.

Again, please apologize for the delays and the mess.

[SilverSolder]

Don't feel bad,  you have a great future in the QA department! 

[dmmguy]

Hi, thank you for you answer.

[RandallMcRee]

Hello all,

Unfortunately, the unthinkable has happened!
The FX reference sustained some damage.

Although minor, the fix requires de-certification.  Therefore, it will be a month or so whilst Illya graciously performs surgery and re-certifies.

Thanks Illya (bow)

Randall
Your bearer of bad news!

Some updates on the FX reference. Also Ilya has characterized the 1k/10k and PX reference for our future reference.

[rhb]

FWIW the transfer of heat from the device to the atmosphere depends upon the specific heat of the air which is a function of water vapor, barometric pressure and temperature.  Backing out correlated noise is easy.

Reg

[alm]

FWIW the transfer of heat from the device to the atmosphere depends upon the specific heat of the air which is a function of water vapor, barometric pressure and temperature.  Backing out correlated noise is easy.

What role do you think heat transfer plays in this data? I'd argue if it plays any role at all, the temperature sweep is too fast, and I trust the sweep to have been done sufficiently slowly so the system is very close to steady state at any point.

[RandallMcRee]

FX reference is back in play...we apologize for the delay.

Sending on to bdunham7 soonish.

Thanks,
Randall

[bdunham7]

FX received in good shape.  First step is to make sure everything sort of works.  Here's the first hour of logging on a Fluke 8846A.  I guess it's at least nominally functional.

[RandallMcRee]

How long was your 8846A on before starting the run? Was the temperature constant?

It kinda, sorta, looks like things started to stabilize in the last quarter of that run. The FX itself should not be making "popcorn" jumps.

Thanks,
Randall

[bdunham7]

How long was your 8846A on before starting the run? Was the temperature constant?

It kinda, sorta, looks like things started to stabilize in the last quarter of that run. The FX itself should not be making "popcorn" jumps.

The 8846A was warmed up and stable, the FX was fresh out of the box.  I didn't have a temperature sensor on it yet, so it may have been quite a few degrees cooler than ambient.  Ambient was 23 +/-2C.  I probably didn't have the best test lead setup, I'm not obsessing over that yet because I don't have any 8.5-digit meters and 0.2ppm of deviation is just the noise level.  I'm running a (hopefully) 24 hour log of two meters plus the temperature of the FX case.  So far so good--here's four hours.  The FX and 8846A are warmed up and stable, the 34401A had only been on for a half hour beforehand.

[bdunham7]

OK, the first step is getting some idea of how well I'm going to be able to measure and compare things on my home bench using the FX.  I ran the FX against the two meters for 24 hours and you can see the HVAC cycling at night, the house cooling off and warming up, some doors getting opened up.  Our housecleaner came this morning and I had to transfer a sick cat to this room as well. So nothing was ideal, but it never is, right?  So I'm going to conclude that this is about as good as it is going to get with what I have.

[bdunham7]

One of the ways to use the FX would be to use the DC Ratio function of the meter with the FX as the reference, then some math as needed to correct for the difference between the FX voltage and the nominal 10V.  This can be a first-order approximation of just adding in 3.6ppm since the difference is so small here.  However, these meters are both close enough that I'm not entirely sure that this method wil be an improvement. 

I replotted this using the specified tolerances of the meters as the bounds of the chart.  It is 29ppm 24ppm + 5 counts for the 8846A @ k=2.58, 35ppm + 5 counts for the 34401A @ k=2.  The yellow line is the nominal voltage of the FX.  It seems clear that both meters are performing much better than expected.  The 34401A is about 25 years old and still has a calibration count of 35 (no changes).  The 8846A is esssentially new.

[bdunham7]

Here is an example using the FX reference and the ratio function of the 34401A to track the tempco of a Fluke 731B 10V reference.  I knew the tempco wasn't great on this unit, I had roughly estimated it to be 1.5 ppm/F (ambient) from previous usage and testing.  The line, which I just drew manually, represents -2.5ppm/C.  Note that here I'm measuring the temp of the internal case around the ref amp and precision resistors rather then ambient.  I'm using a K-type bead thermocouple with Kapton tape to hold it to the case, reading with a Fluke 289.  This leads to some interesting issues that anyone mixing temperature into their logging needs to account for--look at the the graph and see if you can spot the anomaly.

Anyway, given the outstanding performance of the two meters in question there isn't too much improvement to be had, but there is a little.  You can see the 34401A very closely tracking the 8846A with a delta of between 4 and 5ppm, which would be approximately the sum of the nominal difference (from 10V) of the FX reference plus the previously measured error of the 8846A (about 0.7ppm).  If I had a worse meter with a DC ratio function I'd use that for an example, but I haven't characterized my two monsters, the 8505A and 8506A, quite yet.  And I'm not even sure they're "worse".

[bdunham7]

To continue some hopefully helpful experiments on what uses the FX can be put to for those of us of normal means, I've done a plot of a VREF10 reference (from Doug @ voltagestandard.com) using the 34401A with the FX as the ratio reference.  I present the chart and the scatter plot without any commentary or conclusions and I invite any readers to comment on what they see.  This is the latest version of the VREF10 which is enclosed and is specified to 10ppm for 8 months.  The temperature readings were taken by the internal sensor of the F289, not an external thermocouple.

[CatalinaWOW]

The nice thing about digital filtering is that it is possible to implement zero lag filters as long as you don't need the first group of samples (the group of the same length as your filter).  I would suggest a simple boxcar average of the half dozen or dozen samples prior to and after your reference time point to reduce the high frequency noise from the ratio data.  That should make the correlation with temperature much clearer.  Then you can assume that TINs measurements of the reference dependence on temperature are more accurate than any other contributors in the system and factor that out. 

You will be left guessing whether it is your instrument or your local reference that is varying with temperature, but there are a couple of ways to tease that out.

All of this is made more complicated because there are so many questions in the thermometry.  Does the value at the temperature sensor represent the actual temperature at the references or the instrument?  What are the thermal time constants of the instrument, references and sensor? 

[bdunham7]

One of the more general things I'd like to determine and show is what the practical limits of accuracy are for someone like me in an average home lab.  My room is pretty good but not perfect if I take a few precautions.  My cables and setup are admittedly mediocre in this example, at least by metrology standards.  I think my primary sources of deviation here--in descending order--will be the accuracy of the DC ratio measurement itself, the DUT reference, wiring and lastly the small 0.05ppm/C tempco of the FX.  But here are the charts, redone with a 40-sample moving boxcar average, for you to peruse.  One interesting issue regarding the 34401A itself disappears when you apply averaging.  I'm not sure what else we're looking at.  I've added a .zip of the CSV file original data in case someone else would like to torture the data.

[CatalinaWOW]

I am assuming the interesting issue is the quantization of the data and the missing codes apparent in the correlation scatter plot.  I have similar issues on my much older 3456 and 3457.   Actually only apparent on the 3457 when using the extra digit register.  The loop behavior shown in the correlation plot is what I have assumed is caused by thermal lags in "something" that matters, and might still apply even though you have instrumented what should be the critical area, but I don't really know what it is.  I'm just not enough of a volt nut to chase it down further.  Also problem is smaller in my lab as the temp swings are smaller. 

[Andreas]

I am assuming the interesting issue is the quantization of the data and the missing codes apparent in the correlation scatter plot.

Hello,

The 34401A has a problem with rounding (limited resolution on data output). I would try to interchange the inputs to get below 1.0 for the ratio. (so you might gain one digit).

with best regards

Andreas