Sub-ppm voltage measure.

[MegaVolt]

I'm trying to understand how it is possible to measure voltages below 1ppm and so far I see only one method: compare two very close voltages using a null meter.

I have compiled a table for comparing different devices.

From it we can draw conclusions:
1. The best null meter is a nanovoltmeter.
2. The best nanovoltmeter is the HP 34420a (for good reason it is used in the set of Johnson standard )
3. Ordinary null meters approach nanovoltmeters only if the difference between the compared voltages is less than 1ppm.

Tell me what other methods are there to get the best accuracy?

In the table, you can change the values ​​of the compared voltages and the difference between them. Lines that do not fit the range are automatically crossed out.

The table does not include temperature.

[ArthurDent]

I personally don't like any digital instrument for reading any null. As I have mentioned elsewhere in the forums, I did compare two 10.00000VDC sources using my HP419A analog null meter and it worked perfectly for the resolution I was looking for. The equipment did indicate the two supplies were stable within about 1uV which is certainly good enough for my non commercial use. True, my HP419A is over 50 years old and there may be some better newer meters now but I would not use a meter with a digital display for this application.  YMMV.

[TiN]

Using 845A or EM nanovolt meters/amplifiers on 100uV range to compare standards is unlikely (assuming you want to compare two 10V standards). You didn't say anything about what is the measurement, is it 10V-10V or maybe 1000V-10V ? In latter case, you'd need much more expensive gear, not only nanovoltmeters, such as F720A KVD or reference divider like 752A or a precision calibrator.

[martinr33]

You might add the Keithley 182 to your table.  Its minimum range of 3mV gives it some nice headroom over 1mV devices, even though it has 1 fewer digit.

[iMo]

A $10 DIY null meter with any DMM would do <100nV resolution..
An example (I did it with an LTC1050+TL061+9V battery and it worked):

+/- 10uV input --> +/- 10.01mV output

PS: in case you would use 1N4148 protection diodes at the input - paint them black

[MegaVolt]

I personally don't like any digital instrument for reading any null. As I have mentioned elsewhere in the forums, I did compare two 10.00000VDC sources using my HP419A analog null meter and it worked perfectly for the resolution I was looking for. The equipment did indicate the two supplies were stable within about 1uV which is certainly good enough for my non commercial use. True, my HP419A is over 50 years old and there may be some better newer meters now but I would not use a meter with a digital display for this application.  YMMV.

HP419A has an error of 2% i.e. over the selected range of 100 μV, we will have an error of 2 μV. A digital instrument gives 100 times greater accuracy with 10 times the range margin.

Guys working with the Johnson standard use a digital device.

Using 845A or EM nanovolt meters/amplifiers on 100uV range to compare standards is unlikely (assuming you want to compare two 10V standards). You didn't say anything about what is the measurement, is it 10V-10V or maybe 1000V-10V ? In latter case, you'd need much more expensive gear, not only nanovoltmeters, such as F720A KVD or reference divider like 752A or a precision calibrator.

I am considering a light case of 10V-10V.
The range is selected from the calculation of 10 ppm of the total error between the two standards. Even if we accept 1ppm of the total error (which, in my opinion, is the ideal case), the nanovoltmeter still beats the ordinary null meter.

You might add the Keithley 182 to your table.  Its minimum range of 3mV gives it some nice headroom over 1mV devices, even though it has 1 fewer digit.

Thanks for this add-on. 181 and 182 are close to leaders.

A $10 DIY null meter with any DMM would do <100nV resolution..

Resolution is not equal to accuracy

+/- 10uV input --> +/- 10.01mV output

I see an accuracy of 0.1%, and as I understand it, without a guarantee. The best industrial multimeter gives only 0.5%. But I added this device to the table

[iMo]

The accuracy of the above DIY Null meter is given by the connected DMM's parameters, and partially by the accuracy of the 100/100k (99.9k) feedback divider (inclusive its ratio TC).
The resolution of the above DIY Null meter is mostly given by the noise parameters of the chopper opamp and by the DMM's parameters..

[MegaVolt]

The accuracy of the above DIY Null meter is given by the connected DMM's parameters, and partially by the accuracy of the 100/100k (99.9k) feedback divider (inclusive its ratio TC).
The resolution of the above DIY Null meter is mostly given by the noise parameters of the chopper opamp and by the DMM's parameters..

I think this is not all errors. The input resistor together with the input current of the operational amplifier gives microvolts of error. Which should be temperature dependent. Partially, they can be reset during calibration ...

But then again, if we are talking about measurements, it would be nice to have a comparison of the parameters of a homemade device with some kind of metrological device.

[Conrad Hoffman]

The beauty of null measurements is that when you zero a null meter, the error is close to zero. The percentage errors apply as you go up the scale, so the problem only becomes difficult when the two voltages aren't sufficiently close together. A $10 DIY meter can easily outperform 50 year old designs, especially when those designs use neon choppers. I have both the HP419 and Fluke 845 and both can be very good, but both required quite a bit of service and TLC to get back to spec. See the long thread here on the 845. IMO, the biggest problems with any null meter design are noise, input bias currents/leakage and thermals. Accuracy is rarely a concern. It's quite amazing how well they did back before "modern" null meters, using galvanometers and optical magnification.

[BravoV]

Is there any "recommended" null meter circuit that uses analog meter, but utilizes modern semiconductor component, as I have a NOS null analog gathering dust.

[beanflying]

I personally don't like any digital instrument for reading any null. As I have mentioned elsewhere in the forums, I did compare two 10.00000VDC sources using my HP419A analog null meter and it worked perfectly for the resolution I was looking for. The equipment did indicate the two supplies were stable within about 1uV which is certainly good enough for my non commercial use. True, my HP419A is over 50 years old and there may be some better newer meters now but I would not use a meter with a digital display for this application.  YMMV.

HP419A has an error of 2% i.e. over the selected range of 100 μV, we will have an error of 2 μV. A digital instrument gives 100 times greater accuracy with 10 times the range margin.

Guys working with the Johnson standard use a digital device.

snippy...

You need to read the entire specification for the meter it is 2% at FSD on the meter. We chatted about it yesterday briefly in the TEA thread. Near Null as Conrad states it is nothing like 2%

This thread might be worth a read too for you https://www.eevblog.com/forum/metrology/null-voltmeter/

[SilverSolder]

I recently used the David Jones uCurrent as a null detector (on an AC bridge measurement, in conjunction with a scope) -  worked fantastically well!

Might work on DC too, but I didn't try that.  Using AC gets rid of any offset problems (in exchange for a bunch of new issues, but they seem manageable).

[MegaVolt]

The beauty of null measurements is that when you zero a null meter, the error is close to zero. The percentage errors apply as you go up the scale, so the problem only becomes difficult when the two voltages aren't sufficiently close together.

Difference between the two voltages set 10ppm. This is 100 μV and the gain error of the null meter is already noticeable.

And then we need some other device that will help equalize the initial voltages. I plan to make a second table where I will calculate the errors of the KVD or the calibrator.

If we have a Johnson standard that can produce any voltage, then in this situation, the usual analog null meter wins without a doubt. But if we try to measure the discrepancies of the two old 732a, then we can easily have 5ppm on each. And null meter will not help us in any way

[MegaVolt]

You need to read the entire specification for the meter it is 2% at FSD on the meter. We chatted about it yesterday briefly in the TEA thread. Near Null as Conrad states it is nothing like 2%

This thread might be worth a read too for you https://www.eevblog.com/forum/metrology/null-voltmeter/

I discuss measuring two signals spaced at 10ppm (100 uV). For the 100 μV range, this is precisely the full range and it is necessary to take into account exactly 2%. If you see my mistake, point to it directly.

I already read that thread about null meters.

[MegaVolt]

I recently used the David Jones uCurrent as a null detector (on an AC bridge measurement, in conjunction with a scope) -  worked fantastically well!

Oh, that's a great idea. I had thought of this before. Use some kind of pico-meter to compare voltages. It is necessary to calculate the numbers for this. But I don’t understand what test resistance to stop at.

[beanflying]

You need to read the entire specification for the meter it is 2% at FSD on the meter. We chatted about it yesterday briefly in the TEA thread. Near Null as Conrad states it is nothing like 2%

This thread might be worth a read too for you https://www.eevblog.com/forum/metrology/null-voltmeter/

I discuss measuring two signals spaced at 10ppm (100 uV). For the 100 μV range, this is precisely the full range and it is necessary to take into account exactly 2%. If you see my mistake, point to it directly.

I already read that thread about null meters.

In your opening post you made no mention of any offset and that is not where analogue null meters are at their best is the point that has been mentioned already. Not a mistake but if you don't offer a full list of requirements how do you get the answers you need?

Your blanket assumption that the 419A is a 2% class instrument is not an accurate description as it takes into only the worst case possibility. In an ideal world the closer to a true null you can get the better these meters all work you do what you can to avoid using them at the edges. Without looking at the Fluke and Kiethley and others spec or going over the linked thread I don't think any of them were much better or worse than the HP.

[MegaVolt]

If we want to improve the accuracy of a zero meter, we can use the KVD 720a to match the voltages.

But its accuracy is worse than that of a zero meter in the 100μV range.

Those. to compare voltages closer than 10ppm, the use of the KVD 720a will only worsen accuracy

Maybe we can use a calibrator like the Fluke 5720? But alas, it also gives an error comparable to zero meter in the range of 100 μV.

Using Keithley 260 will improve the accuracy of zero meters in the 100 μV range twice. In this case, the entire error will lie with Keithley 260 and the null meter error will be negligible.

If you know what else you can form 100 μV with sufficient accuracy, write.

[MegaVolt]

In your opening post you made no mention of any offset and that is not where analogue null meters are at their best is the point that has been mentioned already. Not a mistake but if you don't offer a full list of requirements how do you get the answers you need?

The table attached to the first post contains a mention of 10 ppm in the second line. Also on the topic page, I counted 3 references about 10 ppm. And I'm glad that we are now talking about the same thing

Your blanket assumption that the 419A is a 2% class instrument is not an accurate description as it takes into only the worst case possibility. In an ideal world the closer to a true null you can get the better these meters all work you do what you can to avoid using them at the edges.

What are some ways to make voltage  from two standards close to each other? If we can’t touch the standards themselves?

[ArthurDent]

MegaVolt – “HP419A has an error of 2% i.e. over the selected range of 100 μV, we will have an error of 2 μV. A digital instrument gives 100 times greater accuracy with 10 times the range margin.”

As Conrad Hoffman and others have pointed out, your assumption on how the accuracy of the null meter works is incorrect and the accuracy of the null meter itself is almost meaningless. Thinking of a null meter as a balance type scale might make it easier to understand. The scale with both sides unloaded is adjusted to read zero, then the unknown is placed on one side and calibrated weights are placed on the other side to rebalance the indicator pointer to zero and the unknown weight is equal to the calibrated weights used to obtain the balance. The accuracy of the measurement depends on the accuracy of the calibrated weights. You can increase the resolution of the reading by using a magnifying glass on the zero pointer.

What I did was similar when I was comparing two very stable 10.00000VDC sources using my HP419A null meter that was set to read 3 microvolts full scale and the readings were staying within 1uV.  Check the link to see the HP419A in action.  https://youtu.be/wTGxBJNNhWc . I was resolving a difference of 1uV out of 10 volts. The 2%+ accuracy of the 1uV meter reading I was seeing would be +/-.02uV but that is out of the entire 10.00000VDC reference, not just out of the 3uv range of the meter. The accuracy of the 10 volt reading depends almost entirely on the accuracy of the 10.00000VDC reference, the same as in a digital meter and the same laws apply to the errors you will see using either method.   

Resolution and accuracy are two different things.   

[MegaVolt]

The scale with both sides unloaded is adjusted to read zero, then the unknown is placed on one side and calibrated weights are placed on the other side to rebalance the indicator pointer to zero and the unknown weight is equal to the calibrated weights used to obtain the balance.

I understand that well. But the question of how exactly to compare the two voltages (10V) and (10V + 10ppm) so that they become equal with good accuracy has not yet been solved.

A little higher, I tried to count errors from the KVD or the calibrator. Only Keithley 260 can offer good accuracy.

What are other ways to make these two tensions look similar?

Those. 34420a can measure the difference with an accuracy of 22.5 nV. How can we form an additional 100 μV (10 ppm) with an accuracy better than 22.5 nV (better than 0.0225%) in order to reveal all the beauty of a null meter?

[iMo]

Perhaps "differential" meter would be the right name here. OP wants to measure the difference between V1 and V2 with as high as possible accuracy.
The difference will be < 100uV.
For example:

[MegaVolt]

Perhaps "differential" meter would be the right name here. OP wants to measure the difference between V1 and V2 with as high as possible accuracy.
The difference will be < 100uV.
For example:

The new scheme has the same 0.1% error and remains on the honorable 7th line )

But I venture to suggest that the circuit does not take into account the error of the feedback resistors, the input op-amp bias (5 μV) and the voltage drop across the input resistor (0.5 μV)

[iMo]

Then use 100/99900 divider, add a trimmer to reset the offset of the first opamp, and the 0.5uV drop on the input resistor most probably does not exist
Be creative..

[MegaVolt]

Then use 100/99900 divider, add a trimmer to reset the offset of the first opamp, and the 0.5uV drop on the input resistor most probably does not exist
Be creative..

  I am sure that if you calibrate this amplifier according to the Johnson standard, it will be able to hold this calibration for some time with perfect accuracy )))

[Kleinstein]

Having a gain stable to some 0.02% is not very easy, but also not that difficult. With suitable resistor networks the gain can be stable enough - still the availability at low resistance can be limited. So it may need relatively expensive wire wound of foil resistors.
The more critical part could be getting the offset stable to really better than 100 nV over a longer time.

The old classic analog Nullmeters have a limited accuracy - alone from the analog movement. AFIAk there are not many modern, digital Nullmeters available. AFAIK the HP34420 is not that suitable for battery powered operation.  One may get around this if the DUT can be powered from battery.