optimal configuration of HP 3458A to measure 10k

[e61_phil]

Hi,

I know, we discussed that already severeal times, but it is hidden in various threads.

So the question is: "How to configure the 3458A to get the most accurate 10k reading"

My current configuration is the following:

OHMF
RANGE 10000
OCOMP ON
DELAY 2
APER 1
NDIG 8

Is there anything to improve?

I take several measurements with this configuration and calculate the average.

Best
Philipp

[TiN]

DELAY depends on cable, some cables might need as high as 5, but most high quality cables are fine with 1-3 sec.
you may want to NULL meter with short prior to DUT if possible. Or not

Also beware tempco, 3458A (at least mine) are not that great in this respect, so you either run ACAL every 0.5c or characterize tempco of the meter itself by using thermally anchored DUT resistor first.
Keithley 2002 is much better for tempco in this respect on 10K range.

[Dr. Frank]

Singular measurement samples will have too big an error, as the noise / standard deviation in Ohm mode is at least 0.2ppm of range.
Therefore it's mandatory to invoke the statistics function for measurements.

By measuring the StD also, you will have control over possible erroneous measurements, as can happen easily by external disturbance. If you observe an excessive StD value, just skip the measurement.

With these above settings, take 16 readings and use mean value for the Ohm value, and the StD value as its uncertainty value (plus calibration uncertainty, of course).

The setting for my 10k resistors and cabling, and the 'workflow' for taking measurements is therefore:

OHMF
RANGE 10000
NDIG 8
OCOMP ON
DELAY 1
APER 1

NRDGS 16
TRIG HOLD
MATH STAT
TRIG SGL or TRIG

=>

RMATH NSAMP
RMATH MEAN
RMATH SDEV

I have stored the settings in the first block to a State, to be accessed by e.g. RSTATE 2, and the MATH functions are assigned to several Keys, copied that configuration from the FLUKE  3458A/HFL.

A single measurement then lasts about one minute.

Once configured that way, you can easily repeat the measurement, or measure the next resistor by pressing

MATH STAT
TRIG


PS:
MATH NULL is not recommended, because OCOMP and 4W function already compensate NULL errors. Additionally, for using NULL, you would have to change the cabling configuration and would probably subtract a residual Offset which has nothing to do with your 10k DUT.

Also, MATH NULL will delete the STAT function, and it's difficult to invoke both functions in a logically correct manner.

Frank

[PartialDischarge]

Add AZERO ON

[Dr. Frank]

Add AZERO ON

AZERO is on by default, but is not necessary due to OCOMP ON, as this automatically subtracts  voltage offsets from the DUTs wires, but also internal voltage offsets.

If you select OCOMP ON, the 3458A does not use a dedicated AZERO cycle any more, I strongly suppose, because the measurement time is not increased over about 4 seconds / sample.

[PartialDischarge]

AZERO is on by default,

True

but is not necessary due to OCOMP ON, as this automatically subtracts  Offsets, from DUTs wires, but also internal voltage offsets.
If you select OCOMP ON, AZERO is not used by the 3458A any more, I suppose, because the measurement time is not increased over about 4 seconds / sample.

Are you sure?? I'm not that sure about that, and the manual does not suggest that exclusive behaviour anywhere. Someone even posted this question regarding his suspicion on the inner workings of OCOMP and AZERO
https://community.keysight.com/thread/20408

[Dr. Frank]

AZERO is on by default,

True

but is not necessary due to OCOMP ON, as this automatically subtracts  Offsets, from DUTs wires, but also internal voltage offsets.
If you select OCOMP ON, AZERO is not used by the 3458A any more, I suppose, because the measurement time is not increased over about 4 seconds / sample.

Are you sure?? I'm not that sure about that, and the manual does not suggest that exclusive behaviour anywhere. Someone even posted this question regarding his suspicion on the inner workings of OCOMP and AZERO
https://community.keysight.com/thread/20408

I'm pretty sure, because I wondered, why the sampling time did not double by using OCOMP ON.

The manual is lacking some information at several instances

[e61_phil]

Thanks for your comments!
It seems to be already the right setup, that I used.

Singular measurement samples will have too big an error, as the noise / standard deviation in Ohm mode is at least 0.2ppm of range.
Therefore it's mandatory to invoke the statistics function for measurements.

I do all statistics on the computer.

As an example I attached a measurement of my SR1010-1k Transfer Standard (R1-R10).
Std. Deviation is 0.38ppm over 1000 Measurements
Temperature was well within 22°C +/- 0.1°C

Best
Philipp

[Alex Nikitin]

Also beware tempco, 3458A (at least mine) are not that great in this respect, so you either run ACAL every 0.5c or characterize tempco of the meter itself by using thermally anchored DUT resistor first.
Keithley 2002 is much better for tempco in this respect on 10K range.

Yes, the tempco of our 3458A on 10K is somewhere between 1 and 2 ppm/C  .  It is about the same on my 3456A meter at 10K.

Cheers

Alex

[Dr. Frank]

Also beware tempco, 3458A (at least mine) are not that great in this respect, so you either run ACAL every 0.5c or characterize tempco of the meter itself by using thermally anchored DUT resistor first.
Keithley 2002 is much better for tempco in this respect on 10K range.

Yes, the tempco of our 3458A on 10K is somewhere between 1 and 2 ppm/C  .  It is about the same on my 3456A meter at 10K.

Cheers

Alex

I assume, that's the T.C. w/o ACAL?
It's specified 3.1ppm/K,  but 1ppm/K with (or right after) ACAL

Frank

[TiN]

Yes, of course. ACAL helps, but not very practical if ambient is unstable due to 860 seconds wait time.

[e61_phil]

Why is the K2002 that much better? Is the VHP101-40k only used for the ACAL and not for 10k in the 3458A? Or is it the circuit around?

And how is it done in the K2002?

[TiN]

It's kinda obvious if one look at 2002 current source and 3458A current source schematics and resistors used.
And no, 40K not used in resistance measurement directly.

[Alex Nikitin]

Also beware tempco, 3458A (at least mine) are not that great in this respect, so you either run ACAL every 0.5c or characterize tempco of the meter itself by using thermally anchored DUT resistor first.
Keithley 2002 is much better for tempco in this respect on 10K range.

Yes, the tempco of our 3458A on 10K is somewhere between 1 and 2 ppm/C  .  It is about the same on my 3456A meter at 10K.

Cheers

Alex

I assume, that's the T.C. w/o ACAL?
It's specified 3.1ppm/K,  but 1ppm/K with (or right after) ACAL

Frank

Yes, without ACAL. ACAL helps but doesn't remove the temperature dependency completely (so even after ACAL the measured value somewhat depends on the ambient temperature).

Cheers

Alex

[dl1640]

I'd rather wonder how to make a correct null on 3458 before measuring a voltage source, 1) null by copper short on 3458 binding posts, 2) null by shorting measurement leads, 3) null by measuring voltage source set at 0V and output on, 4) dont null..

[TiN]

Depends on what you trying to do. Usually it's 3.

[CalMachine]

I'd rather wonder how to make a correct null on 3458 before measuring a voltage source, 1) null by copper short on 3458 binding posts, 2) null by shorting measurement leads, 3) null by measuring voltage source set at 0V and output on, 4) dont null..

Depends on where you want your plane of reference to be.  At the end of your leads?  At the front of your meter?  At the front of your source?

[cellularmitosis]

I'd rather wonder how to make a correct null on 3458 before measuring a voltage source, 1) null by copper short on 3458 binding posts, 2) null by shorting measurement leads, 3) null by measuring voltage source set at 0V and output on, 4) dont null..

Depends on where you want your plane of reference to be.  At the end of your leads?  At the front of your meter?  At the front of your source?

What might determine that?  Perhaps, "Am I shipping the cables with this reference, or not?"

[CalMachine]

I'd rather wonder how to make a correct null on 3458 before measuring a voltage source, 1) null by copper short on 3458 binding posts, 2) null by shorting measurement leads, 3) null by measuring voltage source set at 0V and output on, 4) dont null..

Depends on where you want your plane of reference to be.  At the end of your leads?  At the front of your meter?  At the front of your source?

What might determine that?  Perhaps, "Am I shipping the cables with this reference, or not?"

Generally personal preference.

[dl1640]

IMO it is good to use method 3 if check "gain", but to check "zero" output from the source, one would consider to null out any cabling effect and/or DMM "residual" reading beforehand. Some would use method 2, to short two leads and let them hanging in the air, or short them at either of the source terminal, then take a null, then change the cable config to gain measurement, this means the cable config at null measurement and gain measurement are diffierent, I am not sure if such null is correct for precise application. Most of time I do not have to measure down to a few or sub ppm, so I do not null, method 4.

Please correct me if I'm wrong.

Also, I think method 1 can be used to check the "true" null of DMM, but need a good copper short.

[CalMachine]

IMO it is good to use method 3 if check "gain", but to check "zero" output from the source, one would consider to null out any cabling effect and/or DMM "residual" reading beforehand. Some would use method 2, to short two leads and let them hanging in the air, or short them at either of the source terminal, then take a null, then change the cable config to gain measurement, this means the cable config at null measurement and gain measurement are diffierent, I am not sure if such null is correct for precise application. Most of time I do not have to measure down to a few or sub ppm, so I do not null, method 4.

Please correct me if I'm wrong.

Also, I think method 1 can be used to check the "true" null of DMM, but need a good copper short.

Yes, as you pointed out, each scenario is applicable to certain situations!

[dl1640]

For precise DCV measurement, I would consider a remote controlled switch to form and release a short at DUT end. This is method 2.


For Ohm measurement, I use 4-wire, OCOMP ON, and Delay 1, NPLC 20~100.

[e61_phil]

For precise DCV measurement, I would consider a remote controlled switch to form and release a short at DUT end. This is method 2.

This controlled switch should be very good. Most relay contacts I know are specified with around 2µV.

[dl1640]

Or, if the source has binding posts, then just leave the cable config as gain measurement (Hi-Hi, Lo-Lo), and place a good copper short in between the binding posts, I think this is a good null, then remove copper short and measure gain.

[voltampere]

I hope the answer to my question is not just a trivial thing that i could have found in the manual myself!

I am scratching my head regarding the use of the APER vs the NPLC command.
According the manual it should have the same rersult! ( as far as i understood it ).

Meaning if i set APER 1 then the integration time should be 1 second.
And if i say NPLC 50 then it should also be 1 sec, right ( on a 50 Hz model ).

I tried the APER command like this

DELAY 1
APER 1
NRDGS 15

And it will take roughly 89 sec to complete the 15 readings. ( hand measured, +/- 1 sec ).

If i use
Delay 1
NPLC 50
NRDGS 15

then it will take roughly 204 sec to complete 15 readings.
OCOMP ON for both sessions.

It is the first time i used the APER command because Dr. Frank mentioned it in his reply!

Furthermore:

If i command APER 1
then i get the following results when querying:
NPLC ? -> 50
APER ? -> 1.0 E+0

If i command NPLC 50
i get the following results:
NPLC ? -> 50.0 E+0
APER ? -> 200.0 E-03

I hope somebody can shine a light for me!

Regards

Joerg

EDIT:
Typo ; it looked like i used the same commands; instead it should read NPLC 50 on the second run ( with 204 sec )

[Echo88]

Please take a look a Dr Franks nice summary in a related thread: https://www.eevblog.com/forum/metrology/hp3458a-ocomp-delay-related-problems/msg1464930/#msg1464930
If you use a setting in which NPLC is greater than 10 NPLC, the 3458A just averages the samples of multiple 10 NPLC measurements. Also delay and OCOMP are used on every 10NPLC, so this leads to such a large measurement time. The APER-command integrates over the whole time instead of doing the mathematical 10NPLC/DELAY/OCOMP-averaging, therefore saving sample time.

Here are a few measurements i did which show the timing on a 3458A:

#OCOMP OFF, Delay default, NPLC 10,AZ OFF -> 0.2s
#OCOMP OFF, Delay default, NPLC 10,AZ ON -> 0.4s
#OCOMP OFF, Delay 1, NPLC 10,AZ ON -> 1.4s
#OCOMP OFF, Delay 2, NPLC 10,AZ ON -> 2.4s

#OCOMP ON, Delay default, NPLC 10, AZ OFF -> 0.4s
#OCOMP ON, Delay default, NPLC 10,AZ ON -> 0.8s
#OCOMP ON, Delay 1, NPLC 10,AZ ON -> 2.77s
#OCOMP ON, Delay 1, NPLC 10,AZ OFF -> 2.38s
#OCOMP ON, Delay 2, NPLC 10,AZ OFF -> 4.33s

#OCOMP OFF, Delay default, NPLC 100,AZ ON -> 4.05s
#OCOMP ON, Delay default, NPLC 100,AZ ON -> 8.25s
#OCOMP ON, Delay 1, NPLC 100,AZ ON -> 27.26s

#OCOMP ON, Delay 1, NPLC 50,AZ ON -> 13.65s
#OCOMP ON, Delay 1, APER 1,AZ ON -> 5.97s

[voltampere]

Thank you for pointing!
Should have found that by myself 

That makes sense and your numbers are equal to mine.

But still it doesn´t explain why a query APER? gives an aperture of 0,2 sec after i set 50 NPLC.

Any explanation for it?

[dl1640]

I didnt see difference between your 2 sets of cmds, APER is straightforward but it is only settable between 500ns to 1s.
NPLC 50 is identical to APER 1 in 50Hz system, unless you specify by LFREQ LINE which will set integration time to a actually measured line frequency, dont if it really helps to improve any measurement.

Yes, APER 1 saves time of measurenent than NPLC 50.

If you use a setting in which NPLC is greater than 10 NPLC, the 3458A just averages the samples of multiple 10 NPLC measurements. Also delay and OCOMP are used on every 10NPLC

I set NPLC=50, DELAY=1, measure dc current
the data update rate is less than 3s, that is the DELAY is not inserted into each 10PLC
so, 50PLC = average of 10PLC is true and the user manual does say that
but DELAY is used on each 10PLC, is not true, for all conditions at least

also, if I set 50PLC, the APER? is returned as 200ms, yes it is

[Dr. Frank]

NPLC 50 overrides the APER 1 setting, and splits the continuous 1sec aperture.

As NPLC 50 consists instead of 5 measurements, APER? delivers the 200ms aperture of each NPLC 10 measurement.

I recommend to test a longer DELAY if using NPLC 50 instead of APER 1.

Frank

[aronake]

When making 10K ohm measurements on 3458a, can somehow usage of the guard terminal and guard switch be of any help?