having standards for a certain readout range - effect on readout

[acts238willy]

The 'laying on its side' trapezoid, with the 'ppm of range' rectangle in the middle
and the two triangles above and below it assigned 'ppm of readout'....
Question-
If I use a standard to note the offset of the range and correct for it,
the remaining error is just the two 'ppm of readout' triangles...
Right?

I'm trying to improve the accuracy of resistance readouts on a DMM.

[chuckb]

Willy

Which DVM are you using?

[alm]

I would say no. The error is the uncertainty for any single measurement, and will consist of contributions from various sources, for example the noise of the voltage reference and ADC. Now you might expect two measurements taken minutes apart to be closer than two measurements taken months apart is. commonly described by specifications like 1 year accuracy, 24 hour accuracy and transfer accuracy. If the manufacturer did not specify this, or you need tighter specifications, you could characterize this yourself if you had the resources. But uncertainty will never be zero.

[acts238willy]

Hi, Chuck!  I'm trying to improve the accuracy of the
ohms measurements on a 3458.

I guess what this idea attempts is the elimination of the range
selection error, say going from 10 ohms to 10K ohms.  In the
3458, all ohms ranges are autocal'd to the 40K internal reference.
From there, all ranges depend on the linearity of the 3458.

My thought was that using a known standard at the selected
range, the linearity and drift of the 3458 would be cancelled...
in effect, making the measurement a 'transfer' instead.

[ap]

The 3458A does NOT have a transfer specification for Ohms. The 3458A HFL does. You could use this spec and claim that the standard 3458A uses the same stable resistor (at least the newer 3458As with the markings on the resistor case; see elsewhere here on eevblog) and the voltage reference (and its potential impact, at least for short term measurements) and the other parts are equivalent. However, you have no formal proof for this claim on paper. If you do a test, you will find that the standard 3458A meets that spec. In a metrology application, this may not be of much help though. The question is what proof you finally need. Often it is a spec you can formally refer to. If you need a meter with transfer specs for Ohms, the Fluke 8508A and the Datron/Wavetek 1281 support this, besides the HFL. The Datron Ohms transfer spec is lower uncertainty than the HFL spec.

[acts238willy]

Thanks, AP!

I have an SR 104 to consult for ohms at 10K and a good
732B to check for 10V. My old beater lab mules have a few
year's worth of data and are almost HFL good... but I always
get a Labview Home (R) starting point for any ppm work.

Looks like the only way to get really good Ohms is to do the
old ESI decade-stepping thing with the shorting bars on the
SR 1010 boxes?  This bypasses the 3458 internal linearity
on the ohms scales to give a few ppm at each ohms decade.

[acts238willy]

The 40K reference in the 3458 is at the front of the A1 board,
surrounded by the current measuring components...

Not very impressive next to an SR-104........

[ap]

So the topic boils down to the question on how to derive low uncertainty decade values starting from 10VDC and 10KOhm (up and down).
I do this using a Flue 752A for Volts, respectively a 3458A for the 10V to 1V transfer (based on the very good A/D linearity).
For Ohms I have in fact started with the ESI SR1010 but I have now moved to the Guildline 9975, at least for the lower values up to 100k.
The issue is how the uncertainties of all steps in a transfer pile up and result in lowest uncertainties achievable without having to use someting calibrated besides the two standards (e.g., the 752A is self-calibrating, needs to be done before use). So yes, SR1010 for 1/100 or 1/10 transfer, or 9975 for 1/10 transfer, or 1281 for 1:1 transfer. Keep in mind, for the SR1010s, you still need a 1:1 transfer meter.

[amspire]

For resistance, you can make up some Hamon resistance transfer boxes.

This is a box with 10 or 12 identical resistors. With 0.1% resistors, you can get 1ppm accuracy. 0.01% resistors theoretically gives you 0.01ppm accuracy. Typical quality 0.1% resistors will probably actually match to 0.03% and that would give 0.1ppm accuracy.

Say you have 10x  100K 0.1% resistors. The way it works normally is you first have all the resistors in parallel. This will give 10K. You measure it on the 10K and compare it to your standard and work out that the resistance is 10.00123K ohms. Now put all the resistors in series. You now have 1.000123 M +/- 1ppm.

Now you cannot get 100K with this box but what you can do is put 9 resistors in parallel that gives you 11.111111K - lets say it is 11.11112 K. Now make 3 groups of 3 parallel resistors with the same 9 resistors. Put the three groups in parallel. You now have 90K x 11.11112K/11.11111K = 90.00072K +/- 1ppm.

These boxes often have 12 resistors rather then 10 to give more patterns. Say 12 in parallel to give 8.33333333K. Then 4 series groups of 3 parallel resistors to give 133.33333K or 3 series groups of 4 parallel resistors to give 75K.

It is unfortunately not possible with this 100K transfer box to transfer from 100K to 10K or 1M directly. You don't get the transfer accuracy unless the two measurements use the same set of resistors and if the resistors are connected in symmetrical groups. With 10 resistors, the only symetrical pattern are all in series, all in parallel, 5 groups of 2 in series (250K) or 2 groups of 5 in series (40K).

Good choices for boxes are:

10 x 100M resistors for 10M/100M/1G
10 x 1M resistors for 100K/1M/10M
10 x 10K for 1K/10K/100K
10 x 100 ohms for 10 ohms/100 ohms/1K

Now that last box would need a thought on how to get very low resistance interconnects and the accuracy will be lower.

The Fluke 752A is accurate to about 0.2ppm, but you do need stable voltage sources and you have to watch for the effect of the meters input impedance. It should be fine on the voltage ranges that have the very high (>1G) impedance.

[ap]

What you describe is what is meant by the SR1010s above. In addition you can take 3x3 in series and put these three sets in parallel, gives you the original resistance. In the SR1010 manual you can also see details on how to do that transfer and the error generated this way.

[acts238willy]

I want to get rid of the 'ppm of range' portion of the measurement error.

That would make the readout just subject to the 'ppm of readout',  making
it more of a 'transfer', right?  I know the 3458 doesn't have that in writing.

The ppm of range makes up almost all of the error at small values.
Having a standard for each decade would get rid of the dependence
on the accuracy of the internal meter decades?

[ap]

You may want that, but it is not vaild. Even with a meter specified for ohms transfer measurements (Datron 1281 calls it 10 minute stability), there is a range-related contribution. E.g. in the 10k range, it is 0.2+0.1 ppm (reading+range). You need to multiply by sqrt 2 as you do 2 measurements, comparing two resistors. So a transfer of 10k has about 0.4ppm of mu. At 2k it is about 1ppm.

[acts238willy]

Thanks, ap.
Looks like an ESI resistor magic combo tool and a
good standard for the range selected is the way to
get ppm. I can be happy with single-digit ppms in
the low ohms ranges on the 3458.

I try not to be overly optimistic about any kind of
readout in the 'couple of ppms' for accuracy in my
3458s.  There is one A9 reference board that looks
like it's as good as my 10V reference herd., though.

Any Keysight A/D converter from the factory wanders
a couple of tenths, while keeping well within the
 bulletin 18 daily drift.... so folks are really stretching it
to chart voltages to 8 figures...

[chuckb]

You may want to review all the posts in the topic "how are 8.5 digit meters calibrated?".

Well, the 3458A already has a fan.
As Dave mentioned on EEVBlog #426, 3457A teardown, a fan would be no good for ultra precise instruments, and I agree, as the airflow might cause thermal gradients and thermo voltages on the internal circuitry.

A better thermal concept would have made the instrument more stable, especially the Ohm function.
Latter one is 7 1/2 digits only, and sadly for such an instrument, no transfer accuracy is defined.

But it's possible to make it more stable:

Use the 3458A in a stable, metrology grade environment, i.e. at 20..25°C only.
Environmental temperature should not vary more than 2K over the year, and not more than 0.2K during the measurement.

Don't put the 3458A in a rack, use it always as a desktop device, and always keep the fan filter clean, so that the interior is max. 13K above environmental temperature (i.e. 36°C max.).

Now it's possible to reduce the LTZ1000A heater temperature to 65°C, which will improve the 1 year stability by a theoretical factor of eight.

Calibration temperature should equal measurement temperature.
 
Under those conditions, it's also possible to get Ohm transfer uncertainties of about 0.2ppm.
I 'm able to demonstrate that on the comparison measurements of  my DIY 10k standards, designed with Vishay Z201 resistors.

The absolute uncertainty of the Ohm function stays mediocre, but that's not very different to the other instruments.

Anyhow, the 3458A for me is the best instrument, due to it ultra linear ADC (typ. 0.02ppm of input), its ultra precise transfer accuracies, and its unique AutoCal feature.

Frank

[acts238willy]

Hi, Chuck!!!!

  The scary thing I read here somewhere on the blog was that
after dropping the running temp down to 65C from 90-some C, that
the reference itself would enter some kind of instability and
never be as accurate as it was at the 90C temp.  That would
take at least 6 months to evaluate, using 732A/B standards.

I'm running some A9 3458 LTZ1000 references now in my lab
mules to see if some of these old A9s are 002 stable. I have
sold some 3458s with these good references already and I'm
cooking some more right now.... (smile)

After 6 months, I may try to sell them with a trade-in. Then, put
the traded A9s into the 6 month drift program to see if any gems
emerge....