DMM tempco comparison and measurement methods.

[TiN]

This thread is to start and discuss methods and various measurement/validation procedures that apply to commercial instruments temperature coefficient specification.

This thread is draft work in progress
We do invite any participants/company representatives and metrology experts to chime in  .

Based on discussion with Dr.Frank started first here about 3458A's temperature stability of main 10V DC range.

---

I have verified TC of 10V range on two HP 3458A meters, not three (yet).

Third meter awaiting new A3 (3458 "C") and forth and latest one has stable A3, but not tested in any details yet (awaiting front panel/new terminal block parts, that is meter "D" in my terms). I do NOT say that all 3458A should be this low tempco, just merely stating data I get.

Background info:

All details about 3458A STD meter.. This unit have Agilent A3 from 2014 with Elantec comparator patch board due to EOL of original DIP comparators.
A9 here is STD, but with LTZ1000CH chip, since original ACH was a jumper. A9 is modified with 100K BMF for lower oven temp.

All details about 3458 "B" meter. This unit have brand new A3 ADC from 2016 with modern ALTERA CPLD instead of Fujitsu gate array.
A9 here is also STD. A9 is modified with 100K BMF VAR for lower oven temp.

I also did minor fix-ups last year covered in short vid:
.

Please note, i use simple box method for tempco calculation, not the more complete discrete math.
As example - measure constant 10V by meter at ambient (or TEMP? reading) at Tlow as 0.0 ppm reference. Then ambient change by significant amount (e.g. +10C), wait till reading settle again - record result as Thigh = 1.0 ppm. Calculate meter 10V range TC as Tlow - Thigh / DeltaT , in this theoretical example 0.0 - 1.0 / 10 = 0.1 ppm/K. But i do always log all thousands of samples during temperature variation to see correlation (or lack of) between readings and temperature, as example data below shows too.

Same method I use to calculate all my tempco (for resistors, meters, references, etc), only difference if it's DUT that subject to temperature change (we keep ambient stable for this), or the meter (we keep measured standard isothermal).

Dr. Frank, I found data plot with 3458A and 3458B meter data.

10V DC range

* First test run, using two KX LTZREF at constant +24.00 temperature. Test with 3458A and 3458B. Each meter measure different KX REF.
* Another test run, using FX #3 LTZ1000A reference at constant +37.00 temperature. This time with 3458A, 3458B and one of K2002's. All three meters measure same LTZ.

OHMF Resistance 4W ranges

* Test with 3458A,3458B,2002-4,2002-6 meters. Each meter measure own 4-wire resistor (hermetic VHP BMF) in isothermal box

You can download CSV raw data for any analysis or processing.

Data format header:

Code: [Select]

date;hp3458a;dmm2;temp;temp2;ext_temp;rh;pressure;rtd_temp;tec_curr;

HP3458A is hp3458a column,
dmm2 is 3458B unit data.
temp is HP3458A's TEMP? output value
temp2 for second meter
ext_temp, rh, pressure columns are readback from BME280
rtd_temp is feedback temperature of Honeywell HEL-705 PT100 sensor, mounted to small enclosed die cast metal chamber with two different LTZ KX references. This reading is PID input of Keithley 2510, used to control 40W TEC to maintain reading +/-0.005c.
tec_curr is drive current to the TEC element.

[TiN]

Reserved for various methods post related to characterization of instrument's temperature stability.

[TiN]

Reserved for test results collection and example data sets and python scripts used in this project.

[TiN]

Reserved for used reference links and research papers.

[TiN]

Reserved for summary

[Andreas]

Hello,

Here a T.C. measurement on a K2000 with 4 NTCs distributed within the K2000
Finally I have added a internal NTC at the RS232 connector (so accessible from outside via connector).

https://www.eevblog.com/forum/metrology/project-pimp-a-keithley-2000/msg1106833/#msg1106833

Result: 0.55 ppm/K change for the readings in 10V range (-0.55 ppm/K for correction).

with best regards

Andreas

[Andreas]

Hello,

Another (early measurement) of a 4.8 Digit (80000 counts) commercial hand multimeter (Digitek DT80000) in a heated temperature controlled box attached to my very first LM399#1 reference.

Result +83.9 ppm/K for linear coefficient.

guess this will be the "winning highest" value 

with best regards

Andreas

[TiN]

I was doing some more testing on this subject, and here's initial data I have to share:



All four 3458 configured same way, NPLC100, DCV10, AZER ON, no ACAL.
Temperature cycled by turning room aircon on/off, while all four meters logging same 10VDC LTZ1000 source output (Wavetek 7000).

Trend plot generated by Excel (poly 2 function) and formula shown on each meter data. Vertical scale is 1 uV/division (0.1 ppm).
My reference meters are green "A" and "B", so you can see they earn their keep for a reason.

I will need to work on 3458 "D", as while +0.28 ppm/K TC is in spec, it's horrible in my opinion, comparing to what other 3458A doing.

Same data but in time domain looks like this, with light-green line being Ambient temperature (BME280 sensor):

[e61_phil]

I did a similar test with 2x 3458A and a Fluke 8508A. Voltage source was Fluke 5730A (10V).

A linear regression through the whole data gave the following TCs:
Fluke 8508A: -0.01 ppm/K
HP 3458A#1: -0.24 ppm/K
HP 3458A#2: -0.18 ppm/K

[TiN]

Nice. Now you making me in need to expand test with more meters too. 
Your data however leaves question, about drift of source itself. Could be 5730 output changed -0.2ppm/K , so 3458s actually showing closer to truth value, than 8508A.
I verified my 3458A to be near zero TC before by separate isothermal reference output, hence I'm confident in source in test above.

[e61_phil]

Your data however leaves question, about drift of source itself. Could be 5730 output changed -0.2ppm/K , so 3458s actually showing closer to truth value, than 8508A.

I can't gurantee it by now, but it is very unlikely. Fluke specifies the 5730 with 0.15ppm/K + 0.2µV outside the +/-5°C from calibration.

My 3458A at home shows even worse TC than these two and I know from others similiar TC for 3458A. So that is more or less what I expected.

Edit: Attached a comparision between my LTZ1000-Box against my Fluke 5440B. I would expect the same stability from the 5730.

What do you mean with "showing closer to truth value"? TC or absolute voltage?

[TiN]

TC only of course. Proper way would be to anchor either reference point (source or the DMM) to fixed stable temperature by thermal chamber, and variate temperature for the rest of the system to determine the deviation. But I know having calibrator or large DMM isn't very convenient to put into thermal chamber That's reason why I used LTZ board in little TEC box instead of large calibrator for original tests.

My next step is to find out who is the contributor to big TC on 3458D unit, so will be interesting rabbit to chase.

Another problem on this level of testing is that these low tempco data captures are very easy to confuse with hysteresis or other effects, like noise, humidity changes or even power-induced shifts. So one has to be careful and do lot of lengthy testing with large enough temperature deltas to be confident. 

[e61_phil]

It was just a quick test, because the room was heated up and cooled down for another reason. If I have time anytime, I will put the 8508A in a climate chamber.

But I'm pretty sure the 5730A is very stable over temperatue. I did some tests at 1kV against another thermal stabilized system and both stayed absolutely stable over temperature.

Would be very interesting if you find the TC source of your 3458A. Mine is performing worse than my Agilent 34401A TC wise.

[Kleinstein]

There is not much searching for the reason of the 3458 TC without ACAL: there are resistor ratios (mainly 2 ratios) inside the infamous U180 chip that set the ADC gain. So the main part is very likely due to not that perfect resistor TC matching.  Matching to well better than 1 ppm/K is still good.

Small contribution could be from OP bias currents and offset drift in the reference amplification ( 7 V to some +-12 V), but this should be not that much: 0.6 µV/K (specs for LT1001AM) relative to a 7 V reference is in the 0.1 ppm/K range.

From the gain drift the 34401 is not that much different: a large part could there also be due to the matching in a TaN based resistor array - maybe lower grade, but essentially a similar technology and source.

[Dr. Frank]

Currently, I've got no time yet, to set up this test, sorry.

I will do it as follows:

3458A, one LTZ1000 zero T.C. reference (or 5442A), 2 DMMs for 2 temperature sensors (precision NTC or Pt100).

One T-sensor attached near the 3458As LTZ reference, the other near U180.

Monitor program taking samples from both T-sensors, the 3458A readings of the external 7V or 10V reference, and frequently taking CAL? 175, CAL? 72 readings after each ACAL DCV.


The program will take all these readings, when the 3458A is just switched on and cold (probably 18°C in Taipeh, 21°C in my lab).
Every 1°C of warming-up, until about 35°C in my case, an ACVAL DCV is done (about 15 times in total), and directly after finishing, one reading, and the other parameters as mentioned above. One might crank up the end temperature by blocking the rear fan, so that the 3458A itself heats up to ~ 45°C.
That warming-up takes about 2h, so these en-passant measurements should be fast enough to be precise / stable enough for this use-case.

The immediate reading of the external reference, over the temperature of the internal LTZ board will directly deliver the T.C. of this internal reference, i.e. the specified T.C. WITH ACAL.

The CAL? 72 gain factor over the temperature of the U180 will directly deliver the effective T.C. of the resistor network, as I have already published somewhere else.

The combination of both T.C.s give the specified T.C. WITHOUT ACAL.

Frank

Here's that diagram, using TEMP? only, giving a big offset from the real U180 temperature, but basically a good estimate for its T.C.
 

[TiN]

I'm not sure going that deep is too viable for comparison purposes. After all most user's would be interested in TC stability of instrument as a whole (black box), given the unknown signal digitized by meter at stable temperature or into reading versus known stable signal into meter at changing temperature. And even if meter's ADC have 1ppm/K TC but is cancelled out by REF -1ppm/K TC, such user would not care much if output in data is 0ppm/K (rough example, but it's not impossible outcome).

Also another reason of external only "black-box type" testing is ability to repeat close to same conditions in tests by different people here. We cannot expect everyone to open up meters to hook probes near ADC/REFs or read gain constants (which are often unavailable to readback other secretive meters like 8508A). As much I'd love to support this deeper testing, my "golden" meters are not going to be opened up to preserve calibration/reduce damage risks, so that already eliminates half of DUTs from competition  .

Another point, when meter is warming up from power off state there are other effects can be in play (humidity, thermal gradients variations, etc), which are lesser problem when meter already running for extended periods, and only overall temperature change in graceful steps.

Btw, I never seen 1.xxx gain factor before. Mostly it's 0.98x or 0.99x .

Kleinstein
Maybe, if this theory holds then swapping ADC A3's between meter C and D should reflect equal swap of TC result in readings.
This test to be done end of the week. If data does not confirm to theory, we could have temperature EMFs generated/affecting the signal path somewhere on A1 or connection joints. It does not take much to generate 0.2ppm/K error, just 2uV imbalance for 10V input signal. Hint for that visible on purple 3458C data, when fast temperature change from +23.6C caused +0.5ppm blip, which hours later recovered to original 0.0ppm artifact value once temperature settled around +20C.

[Dr. Frank]

Hi Illya,

you're correct, that only few people will jeopardize their instruments in such a way.
In case of an 8508A, it's not necessary, anyhow, as it does not have an ACAL feature, and the external temperature will be good enough.
But in any case, there will be no way to distinguish reference T.C. and ADC / signal conditioning T.C.

The core question for me is, whether the 3458A LTZ reference is in any way T.C. - compensated, or - selected by hp, which I can hardly believe by many pictures I've seen.. there's always this 200k resistor visible, which is additionally not intended for the A version, and for sure can not compensate for all specimen of the LTZ1000A.

And 2nd core question, I think it was yours, if an ACAL DCV is really necessary in general, as your unbelievably low T.C. instruments (3 of that kind!) all do not need that, obviously 

Anyhow, you also could do that test, by omitting both external T-sensors, and only taking TEMP? or CAL? 72 as a rough estimate for both internal temperatures. 

Worst thing happening would be a slightly distorted temperature axis, as these T-offsets may differ in value and over set-up.

These other parameters like humidity and so on I would regard as being of 2nd order.

Frank

[e61_phil]

In case of a known stable source over temperature one could change the room temperature again. Let everything stabilize at lowest point. Run ACAL and after that repeated the ACAL after stabilizing on each other point in temperature. If ACAL is able to bring back the reading to the correct value, it is not a problem of the LTZ1000. With these informations one can calculate the contribution of LTZ1000 and the rest.
This could be done without opening the 3458A.

[TiN]

A9 tempco can be estimated separately w/o meter, but that involve teardown too. I'll do it for A9's in 3458C and D units. Have another A9 which I measured to have better than 0.02ppm/K stability.

ACAL DCV still does more than just correction for 10V base range, so we opening large pandora box, as same testing would need to be performed for rest of the ranges too.

For that feat i'll need to validate tempco of extended setups with 752A and K262. Or go alternative route, as there is a small balcony and current cal "lab" room have window access to it. So I could make bit longer cable and place 3458A's outside, open to ambient, while signal sources stay in airconditioned room, relatively stable temp. 

You right, after seeing so small TC on 10V I stopped using temperature/dependent ACAL DCV in all my python scripts that log 7V/10V refs only. Only do ACAL DCV once before data log start and that's it, as it just wears relay without any benefit in typical common use when logging DC REF.

8508 was just an example. K2002 does not have ACAL for DCV either, however it can give back range gain and LTZ's value over GPIB.

[Kleinstein]

Some of the older meters (E.g Keithley 196 and likely others of that age)  do an gain adjustment for the ADC concurrent to the normal conversions. Due to the extra ADC reference the Keithley 2001 also kind of may have to do it this way - though I have no direct info on this.
So chances are good the old K196 meter would have a pretty good TC - mainly the reference - likely an LM399.   

This would be similar to the ACAL72 of the 3458 for nearly every reading.

The special point about the ACAL in the 3458 is using the ADC to correct the gain stages / divider of the other ranges and between shunts. Just ADC gain adjustment is more common - it's also less demanding on the ADC.

With the 8508 spending so much time on a single conversion, I am not sure they might even also do that - it helps with stability and also linearity (e.g. thermal effects), but at the cost of more noise / slower reading.

[e61_phil]

My 3458A do every night at 4:30 an ACAL DCV. This night I forgot to close my window and therefore the inner temperature (TEMP?) went down from 38.6°C last night to 34.7°C. That gives a shift in the CAL?72 value of 1.24ppm -> 0.32ppm/K

CAL?72 seems to be really the best and easiest way to figure out which contribution the A3 board has to the TC.

[Dr. Frank]

Hi,
at the 14th day of the WAVE 2019 event, there was in parallel a very interesting podcast about ASIC packaging, where I posed my question:

'Btw.: Are you still manufacturing the infamous U180 , that is the A/D ceramic hybrid of the 3458A, or are you living from stock of such 30 year old chips?'

'Keysight Podcasts...
 The Keysight component that you referenced, which was used in our multimeter products, has largely been replaced by new technology in an ASIC we released around 8 years ago. However, it’s not unusual for us to maintain very long lifespans for some of our ASICs – sometimes 10-20 years! -Jesse'

I wonder if Jesse really refers to that hp3458A ASIC U180, as this used ONLY in this DMM, not in other multimeter products, and obviously the state machine / Gate Array U210 has been updated in the recent years.

Maybe TiN can tell, if there is a new package for U180 on A3 replacement boards?

Frank

[TiN]

Both of my "new" A3's (bought in 2017 and 2018) that manufactured in 2016 per datecodes have visually identical U180 package. Digital gate array/state machine is replaced with patch board and ALTERA MAX V CPLD (I did read the bitstream with USB Blaster ) So I think you got generic answer instead.

[aronake]

Lowering the temperature setpoint is a common tweak to the 3458a A9 board, but has anybody done tweaks to adjust the TC?

I guess the easiest way to do this would be to do some temperature sweeps and change the 200K resistor to whatever value cancel out as much TC as possible.

Anyone upgraded resistors with success? R415 (R2 in LTZ1000 reference design) doesn't look too impressive on the A9. But it is a very aged resistor by now, that ought to make it time drift stable but not great from TC perspective.

Better to build a new A9 or tweak the original? Tweak the original have the downside that an original component get tweaked. Making a new that it would be a new LTZ1000 that would need quite some time to stabilize. And the amount of tweaking matters too in relation to how much the original will be changed and not original anymore.

I have a set of Vishay hermetically sealed 120R 70K 1K/13K laying around, so a bit tempted to use these on the original board. But as boring as it sounds it seems best would be to keep original resistors on A9, add 100K to adjust temperature and then change 200K to cancel out TC.

[dietert1]

Most people prefer making a separate reference using those gold plated parts. If you put that reference including its circuit into a (small !) external oven, then you can determine the DVM TC based on the usual day/night temperature cycle, i mean after some days it will be pretty clear.
This is an honest setup very similar to the usual application of the DVM. Should the TC of the DVM be to small to determine, so be it..

Regards, Dieter