What is the next best ratio measurement device after the 3458A?

[e61_phil]

Hi,

I would like to measure ratios. The voltage ratios between two decade resistors connected in series for example (like Fluke 5450A calibration). To avoid trouble with my wife, I have to postpone the purchase of a 3458A

According to https://www.eevblog.com/forum/testgear/8-5-digit-dmm/msg270530/#msg270530 and some manuals there is no DMM which specifiy a transfer accuracy better than 0.1ppm (0.05ppm+0.05ppm HP 3458A). Even the Datron 1281 is specified with 0.2+0.05ppm in the manual. (The Fluke 8508A seems to be linear to 0.035ppm FS, but a 8508A isn't an option)

I own a (noisy) Solartron 7071 (7081) and a Datron 1071 (7.5 digit but no transfer specifications found) as the highest resolution meters. These meters should be also far away from 0.1ppm linearity.


In https://www.eevblog.com/forum/testgear/hp34401-measurement-of-linearity/msg356304/#msg356304 Dr. Frank showed a very good linearity of his Fluke 5442A (below 0.1ppm of input).

Therefore, my question: If I could verify the linearity of my Fluke 5440B also to below 0.1ppm (I have access to some 3458A, but I have to carry the 5440B to the 3458A ) isn't it a good ratio measurement machine in combination with my beloved HP 3456A acting as nullmeter? (I also own a Fluke 845 AR Nullmeter, but the setup with the 3456A could be fully automated).

Or is there any cheaper  multimeter out there with very good linearity?

What do you think?

[Dr. Frank]

Hello,
The referenced link was a one pass linearity test only, and one quadrant.
Somewhere else, I have published a multiple pass test on both quadrants.
The median linearity of my 5442 was about < 0.3 ppm, additionally both quadrants behaved differently in gain and offset.
As you see, this 0.1 ppm linearity is not achievable each time, that's a general phenomenom of A/D and D/A characterizations.
The 3458A linearity test vs JJA in hpj 4/98 also shows fluctuations during different passes. The 0.02ppm linearity is therefore a typical, median value only.
If you characterize your 5440, you may also achieve a median linearity of 0.2 ... 0.3ppm for one quadrant, but that's no guarantee.

So, the Fluke 720A with its 0.1ppm guaranteed linearity of input is the next best solution.
Frank

Edit: Added complete 8 pass linearity measurement

[e61_phil]

I would like to have an automated setup to make an offset compensated ratio measurement with short time between the measurements. With such a setup it should also possible to make all the ratio transfers in the 5450A automatically. My Knick JS 3010 can act als voltage and current calibrator. I would use this to supply voltage/current to the series resistors.

Rewiring is only needed to transfer the 10k reference to the 5450A 10k.

Therefore,  I don't want a KVD

Another option/upgrade could be a computer controlled switchable 1:10/1:1 hamon divider which divides the output of my 5440B. Very similar to the described technique in the 5440A handbook, but automated.


@Dr. Frank: How do you calibrate the 1R resistor in the 5450A? The manual says one should use a 1R reference resistor (same as stated in the 5700A calibration process). Do you transfer the 10R to 1R?

I would like to know the 1R also as precise as possible (within my capabilities) to transfer this further to my 0,1R shunts.

Thanks.
Philipp

[Dr. Frank]

I use my 3458A only, and live with the mediocre calculated uncertainty for the 10 Ohm range at 1 Ohm
Frank

[acbern]

I am using SR1010s for transfer of 10k to other resistance values. But these are not cheap either, depending on where you get them.

[zlymex]

Fluke 8508A was designed and build based on 1281 at UK Wavetek/Fluke after Fluke acquired Wavetek. The ADC(and Vref etc) were almost the copy of the 1281, so you cannot expect the linearity anything much better than that of 1281. This also evidenced by the same slow speed and panel arrangement of both meters(binding posts on the left, dual displays, similar number of buttons)


The 8508A manual does specify the linearity, however, there is this Transfer Uncertainty of 0.12ppm+0.1ppm for reference, which specified by HP/Agilent as the same figure in 3458A manual. That 0.1ppm is of range of 20V and is 2uV, compared with 0.5uV(0.05 ppm of 10V for 3458A), four times larger than 3458A. Therefore, 8508A is not superior in this regard, it may not even be the next best thing after 3458A.

In order to check the linearity, probably 11 points is enough: 0%, 10%,,, 100%. If that is the case, there is an old but effective way of doing it:
build you own divider string consist of 10 identical resistors(like a Hamon), and buy or build a battery operated, isolated(thus with very low leakage, very high CMRR) null meter to check with the 5440B outputs with the taps of the sting. I sometimes use my Metra Hit 30M(0.1uV last digit, battery operated) for this purpose. The divider ratios of the string can be calibrated by another 1/10th of the voltage (floated, need battery powered again) by 'moving check' of voltages of every resistor-voltage with the help of the same null meter. Have a look at this article for similar configuration: 'A Sub-ppm Automated 1-10 Volt DC Measuring System'.

BTW, 8508A has much better uncertainty than 3458A when measure 1 Ohm, probably because it is the only hi-digits DMM with 100mA test current on low Ohm range.

[e61_phil]

Hi zlymex,

thanks for your great additional info.

I found the 0,035ppm FS here: http://www.infinet.com.tw/001/member/download/fpm/thereferencemultimeter&ratiomeasurements.pdf In the end, they say the same as you.


A divider is a nice idea to check linearity without a 3458A. I think andreas uses a divider to calibrate the linearity of his LTC2400 ADCs. I will have a look at his technique. If I remember correctly he hasn't tuned the steps to be exactly 1/10th. He is measuring from one point of the divider (x/10) to 0/10 and to 10/10 or something like that.

[zlymex]

I found the 0,035ppm FS here: http://www.infinet.com.tw/001/member/download/fpm/thereferencemultimeter&ratiomeasurements.pdf In the end, they say the same as you.

I recall I had that info 6 years ago which looks like a market campaign seminar material to me that I tends to ignore. If 8508A has that kind of remarkable linearity, why not specified in the datasheet in the first place? How can one get a linearity better than repeatability/noise?

BTW, Fluke does't specify the noise either, another critical piece of data that Fluke is ashamed to mention IMO.

However, there is this paragraph in the specification tells the right thing:
http://assets.fluke.com/images/products/fpm/bench_instruments/8508A_Extended_Specs_Rev_B.pdf
--------I quote
 When making measurements on the
same range, these errors will affect both measurements
and effectively cancel, leaving short
term noise and linearity as the dominant errors.
The 20 minute Transfer Uncertainty specifications
are provided to describe the performance
obtained when making ratio measurements on
the same range.
--------End of my quote.

Therefore, Fluke does know these two dominant errors are very important, which make up the Transfer Uncertainty.

[e61_phil]

Yes noise and drift is a great problem for ratio, I think.

It isn't easy to meet the 0.1ppm in 10min specs with the 3458A, but possible (my experience with some LTZ1000 tests). Therefore, it must even much more difficult to measure the 0.02ppm linearity. Are there some addinational tricks I don't know?

[zlymex]

Yes noise and drift is a great problem for ratio, I think.

It isn't easy to meet the 0.1ppm in 10min specs with the 3458A, but possible (my experience with some LTZ1000 tests). Therefore, it must even much more difficult to measure the 0.02ppm linearity. Are there some addinational tricks I don't know?

Tricks I can think of
- repeat measurement  + recording. Later the data may be unpicked and averaged to decrease the type A uncertainty.
- automatic switch of different voltage to ease the thermal EMF variation and speed up the test duration
- reverse switch (better to be auto and repeat) to eliminate large portion of thermal EMF
- stable and nice environment,  good power supply and filter, wind shield on binding post, low thermal EMF test leads, enough warm up time.

[tszaboo]

Probably some analog circuits + a universal counter. It would not be that hard to make an analog integrator, null detector, make it as slow as you want, and then use the counter to make a ratio. You need a few good quality film capacitor analog switches few opamps and off you go?

[wiss]

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

[e61_phil]

Probably some analog circuits + a universal counter. It would not be that hard to make an analog integrator, null detector, make it as slow as you want, and then use the counter to make a ratio. You need a few good quality film capacitor analog switches few opamps and off you go?

Such an easy approach won't work for more than 4.5 digits. The HP Journal which describes the ADC from the 3458A has described the problems a bit.

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is very interesting!

[tszaboo]

Probably some analog circuits + a universal counter. It would not be that hard to make an analog integrator, null detector, make it as slow as you want, and then use the counter to make a ratio. You need a few good quality film capacitor analog switches few opamps and off you go?

Such an easy approach won't work for more than 4.5 digits. The HP Journal which describes the ADC from the 3458A has described the problems a bit.

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is very interesting!

Well, I did it for some 1-2ppm linearity/accuracy, in noisy environment. It was not needed to be better. It is not rocket science. Take a look at the LT1043 datasheet 0.005% V/F Converter. With simple components, like a LM358 and a LT1009, they created a 50ppm accurate converter.
It is not that hard to match the specs of high end meters for 1 range. To start with, you dont need an accurate voltage reference. You need a clock source, which has small drift in a 10 second timeframe. Any timer counter can do that. The oldschool 1989 active parts are easily outclassed today's laser trimmed opamps and chopper amplifiers. There are analog switches which behave nicely, almost no leakage, almost like the Jfets they had to use 30 years ago.
It is actually a good experiance to build these kind of circuits, PWM dividers and voltage to time/frequency converters are really nice. Especially, if you want to have a 3458a.

[Kleinstein]

One difficulty with an analog integrator is linearity of capacitors and even resistors. So if you really want, it can get pretty expensive and large with vacuum capacitors.

However it is possible to test DNL this way. The difficult part is INL.

[gamalot]

Application Note 14 ?

http://cds.linear.com/docs/en/application-note/an14f.pdf

[Kleinstein]

In the AN14, there is a version with 1:10^8 range, that is 1 Hz to 100 MHz. But this is not linearity. It even looks like using adjustment pot for linearity improvements. Another Version called high linearity claims 7 ppm (INL I suppose), which I would consider very good for a U/f converter.

Usually U/f converters have quite some problems with linearity due to non constant frequency combined with dielectric absorbtion and charge injetion.  This is more like the old style before multi cycle charge balancing converters with CPU / µC control became the standard. Another trouble is that you can't synchronize with the line frequency very well.

DNL tests are relatively easy, even to high levels from histograms.
INL at some point could be checked with stacking of references - its not a full test of cause but only a few points to see slow varying components like square or cubic contributions.

[zlymex]

........., then get the 3-year repair agreement (failures on this beast can be REALLY EXPENSIVE), ........

Good advice. A very common failure of 3458A is the 'big bad chip'(U180 on A3 board). Here is another photo of collection of the replaced U180, most of which result in halt of the meter. There are many other meters that also suffer from less critical problems such as out of daily drift spec due to the same chip.
There is only one way to solve the chip problem, that is, to replace the whole A3 board, unless you can find the source for good U180s.

[zlymex]

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is a very intersting article, the technique is similar to the DIY divider of NIM, where they also use multi-test and equation solving.

However, I don't think it works here because
- the measurements are all direct, floated, with large noise, large CMRR problems and bias current difference of Hi and Lo input of the meter.
- OP requires a ratio device to test the linearity of a source, not requiring a device to test the linearity of a meter.

[wiss]

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is a very intersting article, the technique is similar to the DIY divider of NIM, where they also use multi-test and equation solving.

However, I don't think it works here because
- the measurements are all direct, floated, with large noise, large CMRR problems and bias current difference of Hi and Lo input of the meter.

Yes, the measurements are floated and this puts significant requirements on the grounding scheme. And the bias currents has to be low enough to not effect the measurement, it will only work for a high-Z meter. And the resistors has to be of low enough resistance to not generate high thermal noise.

- OP requires a ratio device to test the linearity of a source, not requiring a device to test the linearity of a meter.

But a "stack of references", which this essentially is, will be linear, the voltages will add as they should, no?
Nowhere I require the resistors to actually be linear, only to have a stable voltage over them.

[tszaboo]

In the AN14, there is a version with 1:10^8 range, that is 1 Hz to 100 MHz. But this is not linearity. It even looks like using adjustment pot for linearity improvements. Another Version called high linearity claims 7 ppm (INL I suppose), which I would consider very good for a U/f converter.

Usually U/f converters have quite some problems with linearity due to non constant frequency combined with dielectric absorbtion and charge injetion.  This is more like the old style before multi cycle charge balancing converters with CPU / µC control became the standard. Another trouble is that you can't synchronize with the line frequency very well.

DNL tests are relatively easy, even to high levels from histograms.
INL at some point could be checked with stacking of references - its not a full test of cause but only a few points to see slow varying components like square or cubic contributions.

I paid $4500 for my 3458A from a test equipment dealer [with calibration].  I think I'll stick with that.  By the time I got anything near 8.5-digits and 0.1ppm INL, I would have spent 100X of that $4500 in my own time and parts.  No thanks...

So, to answer the OP's question: "The next best thing after a 3458A...?"; well, the answer is a USED 3458A in good condition.  Get it calibrated at Keysight (the "cheap calibration") which will verify all functions work, then get the 3-year repair agreement (failures on this beast can be REALLY EXPENSIVE), and calibrate it just before the 3-year contract is up with the expensive "Standards Lab" calibration at Keysight (then buy another 3-year contract)... and so on.

Well, if the goal is to have an accurate meter, sure, a used 3458A is a good buy, though a 34470A is probably better, if you sacrifice some accuracy, and get it "half the price", more features. On the other hand, the gained knowledge about these circuits is really valuable.
I did design circuits, which are more accurate than a 6.5 digit multimeter in one or two ranges. I tested recently released opamps, passive parts, topologies. Redesigned boards, because mechanical stress related drifts, replaced capacitors because micro-phonic effects. The few ppm region is really fun.

[Alex Nikitin]

One more option to get a sub-ppm linearity check for not much money is to use the AD5791 evaluation board. My sample is below 1ppm INL over 0-10V scale (measured by HP3458A) and there is a space on the board to put LTZ1000 reference on (which I did). As the stability of the DAC output ratio is better than 0.2ppm, it should be possible to calibrate at least some points using HP3458A to about that level of non-linearity (or even compensate for it with a secondary correction DAC). Also this DAC can be used in a bridge configuration to measure the ratio directly using a null-meter output to control the DAC for balancing the bridge.

Cheers

Alex

[zlymex]

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is a very intersting article, the technique is similar to the DIY divider of NIM, where they also use multi-test and equation solving.

However, I don't think it works here because
- the measurements are all direct, floated, with large noise, large CMRR problems and bias current difference of Hi and Lo input of the meter.

Yes, the measurements are floated and this puts significant requirements on the grounding scheme. And the bias currents has to be low enough to not effect the measurement, it will only work for a high-Z meter. And the resistors has to be of low enough resistance to not generate high thermal noise.

- OP requires a ratio device to test the linearity of a source, not requiring a device to test the linearity of a meter.

But a "stack of references", which this essentially is, will be linear, the voltages will add as they should, no?
Nowhere I require the resistors to actually be linear, only to have a stable voltage over them.

Grounding scheme may solve the problem of external noise and interference, but no help for common mode error of the meter if the meter is "floated".
Low bias current reduce the error due to Hi(Red color pen) input of the meter but not for Lo(Black color pen) input, because the Lo input leakage to ground is not specified, may be much larger than bias current even for high-Z meters. Also, there may well exist noise and interference at Lo input to the case(internal noise and interference). So if the Lo input is connected to the middle of a resistor string, it is problematic for sub-ppm level measurement. In the Cutkosky Divider system such as MI's 6000B (http://www.mintl.com/media/pdfs/6000b), where the DMM must be floated, but that DMM has a guard input terminal, and there is a dedicated DAC to generate the appropriate guarding voltage to solve the floating problems.

Linearity refers to the differences of a range of supposed-to-be proportional values from the specified. A stable "stack of references" on it's own has no linearity unless their individual voltage values are precisely specified, but that will void the beauty of the article, isn't it?

[wiss]

Maybe try the method in this document?

https://www.eevblog.com/forum/testgear/dmm-linearity/msg698554/#msg698554

That is a very intersting article, the technique is similar to the DIY divider of NIM, where they also use multi-test and equation solving.

However, I don't think it works here because
- the measurements are all direct, floated, with large noise, large CMRR problems and bias current difference of Hi and Lo input of the meter.

Yes, the measurements are floated and this puts significant requirements on the grounding scheme. And the bias currents has to be low enough to not effect the measurement, it will only work for a high-Z meter. And the resistors has to be of low enough resistance to not generate high thermal noise.

- OP requires a ratio device to test the linearity of a source, not requiring a device to test the linearity of a meter.

But a "stack of references", which this essentially is, will be linear, the voltages will add as they should, no?
Nowhere I require the resistors to actually be linear, only to have a stable voltage over them.

Grounding scheme may solve the problem of external noise and interference, but no help for common mode error of the meter if the meter is "floated".
Low bias current reduce the error due to Hi(Red color pen) input of the meter but not for Lo(Black color pen) input, because the Lo input leakage to ground is not specified, may be much larger than bias current even for high-Z meters. Also, there may well exist noise and interference at Lo input to the case(internal noise and interference). So if the Lo input is connected to the middle of a resistor string, it is problematic for sub-ppm level measurement. In the Cutkosky Divider system such as MI's 6000B (http://www.mintl.com/media/pdfs/6000b), where the DMM must be floated, but that DMM has a guard input terminal, and there is a dedicated DAC to generate the appropriate guarding voltage to solve the floating problems.

If there is a high enough V-low bias should show as an error in the measurement, it would not show as less error than there is (unless the other errors are "exactly" the opposite of that introduced by the bias), I certainly agree.
And I do not think I would like to try this down to 0.1 ppm...

Linearity refers to the differences of a range of supposed-to-be proportional values from the specified. A stable "stack of references" on it's own has no linearity unless their individual voltage values are precisely specified, but that will void the beauty of the article, isn't it?

What I mean is that if you stack V_1 on top of V_2 the sum should measure (V_1 + V_2), no matter what the actual values of V_1 and V_2 are.
Problems could be introduced by thermal EMF in switches and the like...

[zlymex]

What I mean is that if you stack V_1 on top of V_2 the sum should measure (V_1 + V_2), no matter what the actual values of V_1 and V_2 are.
Problems could be introduced by thermal EMF in switches and the like...

That is correct.
However, the stack of voltages of not precisely known(although they are stable and roughly equal in value) cannot be used to check the linearity of Fluke 5440B.