Comparing calibration (DC 10V) on Fluke/Keysight/Keithley bench DMM's

[HKJ]

I got the new Keysight and Keithley bench DMM and wanted to see how my old Fluke DMM's has handled the passage of time.
I do not have any precision referenc, only one of the 4 voltage ebay ones, but running all DMM's in parallel will show how well they compare.
I changed to 10 NLPC and increased resolution on Keithley, the other DMM's are running default configuration.



Looks like the two Fluke and the Keithley agree and the Keysight is a bit out, but lets clean it up a bit:



With 1 minute averaging (in the graphing software) the curves are much easier to see.



5 minutes averaging makes them even smoother.



And without the Keysight, it looks like the Flukes are still good.

All meters basically agree on the variation, it looks like it is from the reference.

But there is some variation between the DMM's:




Note: My setup is not a high precision setup, just ordinary test leads between the DMM's and the reference and no shielding. I am reading a value each second.

[Andreas]

Hello,

try to keep air currents away from the reference + the leads.
(cover them with cotton or clothes).

With best regards

Andreas

[HKJ]

try to keep air currents away from the reference + the leads.
(cover them with cotton or clothes).

Some of it may be air currents, but that does not really explain why the differences mostly tracks.

[Andreas]

Hello,

thats why I suggested to cover the leads and the reference (photo ?)

with best regards

Andreas

[HKJ]

thats why I suggested to cover the leads and the reference (photo ?)

The reference is in a box:

[Andreas]

Hello,

as I expected a TO-99 (metal can) package.
The reference leads (material=kovar) have around 39uV/K thermal voltage.
The self heating of the reference is at least 2-3 K over room temperature.

So I would try to thermally isolate the kovar leads against environment.
(on both sides of the PCB).

Of course the reference itself (a bandgap reference) has also some noise
and the DC/DC converter on the PCB will also contribute some effects.
Perhaps you can increase filtering of the 1/f noise by setting instruments to 100 NPLC.
(one measurement all 2 seconds).

with best regards

Andreas

[lars]

If it is an AD584, as I guess, the results seems very reasonable. My experience is about 3ppm p-p noise, the datasheet I just found at AD says 50uV p-p 0.1-10Hz noise typ. As it is flicker noise you can filter it a little but I should say it is difficult to get below 1ppm for some hours with the AD584. For the difference charts the LM399 in the Flukes starts to set the limit I guess.

The TO99 is good for humidity but not necessarily long-term drift. 15 years ago I worked with the AD581 and 584 my experience was that occasional short (hours) start-ups was ok but longer tests made it difficult to predict drift. Just some months could give 20ppm drift on some units. I also have one AD581and AD584 in boxes that I just startup every second month for three hours. Over 10 years I couldn´t see the drift due to my uncertainty (10ppm). But compared to REF102CM and AD587LQ they shows up to 3ppm p-p variations over say a year. Regression line to the other references shows less than 1ppm/year difference.
The problem with short start-ups at months between is that you cannot use it for other tests so I prefer to have some references continious powered. After a year many non-heated references drift very little. I should also say that so far I have no reference in plastic case that is not humidity sensitive.

Lars

[HKJ]

The reference is AD584.
It looks like I have about 10uV variation in the output when the noise is filtered out. With 39uV/K it might be the explanation.

I recorded some more hours (The night):


The bump is probably the charge circuit in the reference. I did run it with power connected (I do not know how long time it will run without power).

I might try some more runs, but with less DMM's connected.

[Andreas]

The reference is AD584.
It looks like I have about 10uV variation in the output when the noise is filtered out. With 39uV/K it might be the explanation.

Hello,

the 39uV/K is only for the solder junction (temperature difference) of the package.

The AD584 itself also has a temperature drift.
My AD584 has around 4 ppm/K (measured in 5V configuration near 25 deg C).

With best regards

Andreas

[HKJ]

The AD584 itself also has a temperature drift.
My AD584 has around 4 ppm/K (measured in 5V configuration near 25 deg C).

I would guess that it is about the same at 10V and that means 40uV variation for 1 degree change and that the charging heated the reference a bit more than 1 degree.

[saturation]

Thanks HKJ,  assuming the Keysight and Keithley have current calibration certificates from the factory, the Keysight initially appears off, but the big question is what is the true value of the DUT?  If its assumed to be 10.000 00 VDC, the Keysight is the least off.

If the Keithley is right, then the Keysight is off by ~ 7ppm.  If the Keysight is right, then the Keithley is off by that amount.  The difference is likely Keysight's calibration methods versus Keithley, which likely is the same used by Fluke, given how stable the measurements apear are over time.

Given the shorter test period, all the DMMs are very stable, ~ within  +/- 1ppm of a central value.  In the 12+h graph the Keysight appears more stable by about 1ppm.

[HKJ]

They both have factory standard calibration, nothing more. The reference is not really calibrated.

I would not conclude anything about stability for the 12h period, that might very likely be the reference.

It is interesting that the two Flukes matches the Keithley, they are a couple of years out of calibration.

[saturation]

Since Fluke and Keithley are sister companies within Danaher, they could share similar cal methods.  While I agree not to conclude on anything, you had all meters connected simultaneously in the same room, isn't it?  Then the drift of the DMMs are under similar constraints, but for small ppm changes many minor setup things can cause the difference.

Compared to these tests, posted on another thread on the forum, referenced against calibrated equipment and a 3458a:

http://imgur.com/gallery/Wbx7q

The noise floor and the burden voltage, if true, would limit the accuracy of the Keysight.

They both have factory standard calibration, nothing more. The reference is not really calibrated.

I would not conclude anything about stability for the 12h period, that might very likely be the reference.

It is interesting that the two Flukes matches the Keithley, they are a couple of years out of calibration.

[HKJ]

While I agree not to conclude on anything, you had all meters connected simultaneously in the same room, isn't it?  Then the drift of the DMMs are under similar constraints, but for small ppm changes many minor setup things can cause the difference.

Yes the meters are stacked on my table, one stack with Keysight and another stack with the 3 others, not more than 30cm between the two farthest apart. I do not have any AC or forced ventilation in the room, i.e. air currents are fairly minor.

[eas]

The bump is probably the charge circuit in the reference. I did run it with power connected (I do not know how long time it will run without power).

I have one of those references. The "cal" sheet is dated from 2/15. Mine has a bare pouch cell, while I've seen others with reused camera or cellphone batteries. I can't remember how many mAh I estimated it holds, but I found that it will power my reference for 24-36 hours with a single Keithley 2000 DMM ( > 1 GOhm input impedance). Can't remember which voltage I was using.

I'm curious is the can marked KH or LH?  The Chinese maker used to sell both varieties, but their supply of LH parts must have dried up because their TaoBao store has only listed the KH since I found it in February. Most of the eBay sellers haven't bothered updating their listings. LH has better initial accuracy and lower tempco than the KH. Noise and long term stability are supposed to be the same.

[HKJ]

Mine is the LH type and thanks for the time estimate.