9½ Digit Multimeter, feasible?

[Kleinstein]

What are your propositions to improve on existing so called 8.5 digit DMMs like HPAK3458A/F8508A in regards to schematics/construction moerm?

The HP3458 is kind of the benchmark meter, but it is still not made for just good INL and low noise. It still has some compromises to also get high speed.
Today one can split the jobs for low speed at highest resolution very low INL and high speed at still good resolution over 2 ADCs, a bit like done in the Fluke8858 or DMM6500. There are now good SAR type ADCs that are an easy and good solution for the 1 kSPS-1MSPS range.

The 3458 still uses a relatively low 12 V reference - a higher voltage there and equal reistors at the integrator input would improve on noise and make matching with the switches easier. Not a big deal, but an easy 10%. Another point is the rather fast modulation to get good high speed performance. Slower modulation could improve on the noise and to some degree also the INL, though there is the point dielectric absorbtion. To still allow for slower modulation there should be the option to used a kind of 2nd order integration for the FB (at bit like MS4 and NI meters) or as a larger step a more contineous PWM like FB as in the F8508 / Prema / Solartron meters. Another missed oportunity in the 3458 is filtering the reference: the ADC also react the higher frequency (e.g. 100  kHz range) noise from the reference, that is relatively easy to filter.

An odd point in the F8508 is that for the 20 V range the signal is divided by 2 before the ADC. In my DMM I have used a different way to get a 20 V range. The input signal is split in a U/2 and -U/2 part. The auto zero cycle now measures these 2 parts instead of what HP does with 0 and the input signal, kind of loosing on the range by fixing one side to zero instead of also allowing for a vairable input signal and thus loosing half the theoretical possible input range. As positive side effect the input is essentially sampled all the time and thus less noise bandwith for the input amplifier and also for the signal source. By combining a positive and negative reading the even order contributions to the INL are canceled out.

There is a principle limation from the resistors at the ADC input - for more than about a factor of 2 improvement in the noise one has to go low resistance and use really good, well matched resistors to fight thermal effects from self heating.

To improve on the INL there is the option to do some dithering, e.g. with an added offset or using moderately different versions of the MS conversions. This can reduce the effect of possible local linearity errors, similar to idele tones.

For the short time stabilty there would be the option to have thermal stabilization - nothing new, but it adds costs and power consumption. With modern parts the power consumption may not be that bad to start with - modern µCs and FPGAs consume quite a bit less than some 35 years ago when the 3458 was designed. Even OP-amps can be less power hungy (e.g. the OPA205 takes something like 1/10 the power of an OP07/LT1001 and is still lower noise  higher BW).

It may not be enough to get a full factor of 10 of improvement, but still quiet some possibilities.

[NNNI]

Sorry to butt in to the conversation, but Kleinstein, could you elaborate upon the second order feedback in Multislope IV? I would also appreciate it if you could point me to resources regarding the same.

Also interesting would be a somewhat more detailed explanation about the U/2 and -U/2 method, maybe you have already posted about it in one of the multislope threads?

Thanks,
NNNI

[coromonadalix]

was there some compare between the old 3458 and the new 3458   "AKA  Black model" ??

an 9 .5 digit dmm if it could exist, would not be useful for many DIY techs and some in the TEA,  it would be surely used in a very stable environment or long term checks / drifts / references voltages  etc .... 

not as an all day long dmm ??  at already at 7.5, 8.5 digits,  you should be satisfied ?   no ?

and calibrating such ... what would be the $$ since we know a 3458 is in the 1 to 1.2k$ range ...

[Kleinstein]

The simple feedback in the multi-slope ADC is directly from the integrator and thus a bit like a 1st order SD ADC. 1st order SD ADCs are known to have strong idle tones, that is relatively static patterns in the feedback, that can cause local INL errors e.g. from dielectric absorbtion.  2nd oder SD ADCs show less severe idel tones. So the idea is to have a 2nd integrator (or an 1+ integrator) that is than used for the feedback during the run-up. This way the average voltage gets close to zero and the simple patterns near simple H/L ratios are more mixed up.  The idea is not new and used in the patent US6970118  (NI meters). The run-down part would still be from the 1st order integrator part. The HP MS4 is some-what similar to the NI-meters, but looking at the patent it does not look like it uses 2nd order integration in the FB. Anway they use super fast modulation and a more PWM like FB and thus no issues with DA for this reason.

I use the U/2 and -U/2 way in my DMM and it is described in the MS-thread. The idea is that the low side of the signal is no longer at the ADC ground but also moving, e.g. to minus the positive side to create a differential signal with little common mode voltage. By moving the low side, there is no actual gain or attenuator in the signal path. The inverter to move the low side only effects the CM voltage and thus no offset, gain or linearity error from there.  The AZ cycle than alternates between the 2 sides instead of zero and signal and this way read the differential signal with the larger range.
For the 4 wire resistance mode it is common practive to use the switching between 2 signals for a differential reading (sense Lo to sense Hi). The +- U/2 scheme is moveing "sense Lo" to maximize the usable range.

The new black KS3458 is still very close to the original - mainly replacing old obsolete parts (e.g. the comparators and gate array) with newer ones. Still essentially the same specs and performance.

[moerm]

But to talk in examples that can be followed by a larger part of the readers: If you have a 1MHz 8-bit ADC available and need a 100Hz 12-bit then you don't need to go shopping. Normal distributed noise is reduced by sqrt(n) if n is the number of averaged measurements. That means if the ADC delivers 6 effective bits then you gain sqrt(10 000)=100 approx= 2^6.5 so your averaged result should theoretically have 12.5 effective bits. If your target are 12 data bits of which 10 should be effective then this method can work without any problem.

Of course I agree you can't have any magic algorithm that creates a usable 12-bit signal out of a 8-bit ADC without drastically reducing its sample rate. "Garbage in garbage out" stays always valid.

I know but frankly I consider that a fairy tale too. Because NO, you can *not* get 12 bits out of an 8-bit ADC, period.
But you can get something like 12 bits out of a system/instrument if the sample rate is sufficiently high. What you actually get though, and that should be understood, is 8 (well, actually more like 6) *actually measured* bits out of the ADC plus another 4 (actually more like 6) bits out of statistics i.e. basically out of thin air).

But then, that statistical "information" is not completely useless because it's at least based on what statistics need: a *large dataset*. With that one at least has some basis for "funny" statistics games.

But an oscilloscope with (nowadays) billions of samples per second is one situation. With a DMM it's a very different situation because you do not have a sufficiently large data set.

And *that*, DMMs were the topic here.

[moerm]

Contributor moerm is trolling this thread.
There is nothing like a 1 ppm physical limit. Zener references with a stability of 0.02 ppm over several months have been demonstrated before. One can find the reports in this forum. And integrating ADCs stable to that level are commercially available.
The 1 ppm limit is something created by the calibration lab market. If Fluke or Keysight labs with quantum standards don't certify to better than 1 ppm it's their choice. Quantum voltage standards are known to be good for about 0.0001 ppm, ten times better than necessary for a 9½ digit meter.

Regards, Dieter

First, why an ad hominem ("troll")?

"no 1 ppm limit, zener refs ... 0.02 ppm ..."
(a) Context! My statement was in the context of DMM.
(b) I was talking about accuracy and precision, not about stability.

"The 1 ppm limit is something created by the calibration lab market. If Fluke or Keysight labs with quantum standards don't certify to better than 1 ppm it's their choice."

Yeah, because those corporation suddenly decide to not turn something that would be very, very attractive for many customers into money! Well noted, after selling "8.5 digits" DMM, which actually are 6 digits DMM in terms of accuracy and precision - and that what measuring is about - for decades. Totally credible. Do you happen to have evidence that shows those corporations having crossed the 1 ppm barrier with DMMs?

I call BS!

The reality is that even the best standards the have are basically about 1 ppm standards, but 4 of them and only based on a bit of averaging those 4 they achieve a bit better than 1 ppm.

"Quantum voltage standards are known to be good for about 0.0001 ppm, ten times better than necessary for a 9½ digit meter."

Cool, really, I'd be enchanted if only there wasn't a "small" problem: "Known to be good for ..." very, very highly likely means "calculated" but *not measured*! And I understand that because *with what* should they measure that?

Feel free to tell me more about "Quantum voltage standards" once you can provide true, consistent, and repeatable measurements. Until then I take "quantum"-[anything] as what I think it is, a mix of fairy tale with a garment of some math.

Don't get me wrong, I'd be very, very pleased if one day we had (even just) 0.1 ppm DMMs and/or standards, preferably over 90 days - but until then I'll verify and check wild claims.

[moerm]

@Kleinstein

Bravo! That's why I respect you, although IMO you are moving in a territory that is cumbersome, tricky and has a somewhat low reward-for-efforts ratio.

Not only do you obviously have lots and lots of relevant knowledge but also are one of the few who actually work and "fight" for small improvements. And that, besides some potential (but unlikely) breakthrough in physics and/or electronics seems to be (and is in my mind) the only way towards more accuracy and precision.

One hint though (again, I studied physics), albeit a bit loosely worded: heat means entropy and should hence be considered an enemy (here, in this context). Yes, I know the most stable references like LTZ1000 are heated, but actually IMO that is a tradeoff for a specific situation, namely the expected (in normal usage) rather wide (+-5°C) ambient temperature frame in which they are used. Or put differently there is a good reason for JJAs are being held close to absolute zero degrees: low entropy.

If "yeah, yeah only about +-1ppm but over a (relatively) large temperature range and hence real world practical" is the situation, then heating, in order to at least achieve a *stable* temperature is a "good deal" - but when hunting to get an "8.5 digits" DMM a few tenths of a ppm better, heating is not a good companion.

I wish you good progress in your work

[dietert1]

National metrology institutes compare their JJA standards more or less regularly and know them to agree within about 0.0001 ppm. Anybody interested in voltage metrology has seen those reports, i mean they are easy to find if somebody is interested.
That's what we call trolling: Members not interested at all in a subject writing long essays full of confusion. That 1 ppm limit proposal is telling it all.

[moerm]

National metrology institutes compare their JJA standards more or less regularly and know them to agree within about 0.0001 ppm. (repeated personal attack deleted)

That may well be but *how* do they measure that? I don't care about "agreeing", that's just words, I want to see *facts* and data based on *measurements*.

So?

[Phil1977]

It seems moerm trusts authorities like metrology institutes but he seems to not trust statistics.

But do you know who trusts in statistics? The whole science and consequently also the metrology institutes.

It´s just very very short-sighted to disrespect statistics if you want any precision and accuracy.

[dietert1]

For example: https://en.wikipedia.org/wiki/Volt and its references.
JJA standards work so well that the method has been adopted as the metrological definition of Volt in 2019. Since then the Volt can be derived from a known frequency using a JJA setup. It delivers voltages up to about 10 V. A precision comparison of two 10 V standards requires a nanovoltmeter for a direct comparison. Accuracy of the comparison can be about 1 or 0.1 nV.

https://en.wikipedia.org/wiki/Hertz
Frequencies can nowadays be measured to about 15 digits and another quantum standard serves as metrological definition. As frequencies can be obtained to about 12 digits from GPS, this means anybody with the JJA setup can instantiate the Volt to about 10 digits. National metrology institutes can do better than GPS using their own frequency standards.

Wikipedia has enough on statistics, too. I mean if somebody wants to learn.

Regards, Dieter

[Echo88]

JJAs can be intercompared via nullmeters like a HPAK34420A: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=32759
https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=32308

Abstract: The majority of Josephson Voltage Standard(JVS) intercomparisons have been performed by using Zener voltage transfer standards and a protocol based on the Measurement Assurance Program (MAP) with uncertainties in the range of a few parts in 10^8 at 10 V that are limited by the Zener characteristics.
In order to improve the uncertainty of the comparison, protocols using a compact Josephson voltage standard (CJVS) as the transfer standard have been
developed. The uncertainty using the CJVS in the comparison can be in the range of a few parts in 10^9 at 10 V.
The array-to-array direct comparison using the conventional JVS or programmable JVS (PJVS) can further improve the uncertainty of the comparison to a few parts in 10^10.

[moerm]

It seems moerm trusts authorities like metrology institutes but he seems to not trust statistics.

But do you know who trusts in statistics? The whole science and consequently also the metrology institutes.

It´s just very very short-sighted to disrespect statistics if you want any precision and accuracy.

No, I do trust metrology institutes to (quite) some degree but not fully. In fact you yourself laid out the reason: "The whole science and consequently also the metrology institutes" - which btw is largely true but not fully; there are some fields in science where statistics plays a rather minor role, if any.

The main two reasons: while statistics can be useful it by its very nature always has some distance from reality. So, averaging on a scope can help to get a clearer (literally) image, on a multimeter it can get rid of (random) noise, but it's only meaningfull iff based on real measurement data. As soon as statistics (or more precisely statistical methods) go beyond actual physical data that may be useful only for interpretation.

That said, metrology institutes and I (and all of us) have a common problem: we have to do with the best that's available, which in my case (and highly likely yours too) is national standards and in their case is not really sufficiently accurate and precise (and frankly IMO questionable) instruments. That's just how the world is.

Second reason: science, or more precisely "science" that is, to provide two examples, idiots telling us that sex is determinded *sociologically* (or whatever BS) but definitely *not* biologically. And idiots telling us that the laws of physics are valid in the *whole universe*, which is obviously not a tenable statement, although quite likely true, because it's based on a ridiculously tiny empirical space (not even one complete solar system). Now, if some secondary teacher tells that nonsense he can be forgiven, but if a scientist tells it he obviously missed some basic and important lectures.

In case you're interested look up "quantors". Math (and Math is my yardstick) tells us that if a proposition is not true for even a single element of a set it can't hold for the set.

Btw. "The whole science and consequently also the metrology institutes" is an argumentum ad verecundiam and would not stand in (real) science. "The whole science" also told us for centuries that the earth is flat and the sun circles around it ...

Volt (half-cooked wikipedia link plus some quotes from it)

Hertz (another wikipedia link)

Thank you for the links which you seem to consider a reference.

Unfortunately though you only provided links supporting your belief system but *ignored my question*, so I'll repeat it for your convenience: *how* do they measure that?

(maybe also see my response to @echo88)

[moerm]

JJAs can be intercompared via nullmeters like a HPAK34420A: https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=32759
https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=32308

Abstract: The majority of Josephson Voltage Standard(JVS) intercomparisons have been performed by using Zener voltage transfer standards and a protocol based on the Measurement Assurance Program (MAP) with uncertainties in the range of a few parts in 10^8 at 10 V that are limited by the Zener characteristics.
In order to improve the uncertainty of the comparison, protocols using a compact Josephson voltage standard (CJVS) as the transfer standard have been
developed. The uncertainty using the CJVS in the comparison can be in the range of a few parts in 10^9 at 10 V.
The array-to-array direct comparison using the conventional JVS or programmable JVS (PJVS) can further improve the uncertainty of the comparison to a few parts in 10^10.

First, thank you for responding to my question to @dietert1, I appreciate that.

But while I very much welcome NIST's and other's striving for ever so slightly improving what they have I do not really trust it, sorry. For multiple reasons, a major one being the HPAK34420A being the weak link in more than one regard. For one it's as far as I know *not* used (nor I guess even usable) at cryo-temp (say below 70° K) as is a JJA/JVA and hence has way more noise, temp-drift, etc. Moreover it's datasheet talks about 2ppm basic 24hr (presumably +-1°C) accuracy (and only in 1 range albeit gladly in the one that seems to be used) which is orders of magnitude higher than the required level for ppb accuracy and precision. Also note that an additional range "uncertainty" of 1ppm (1000 ppb!) applies. And there's also DC noise and DC noise vs source resistance, the former being close to half a ppm in the 10V range. Furthermore the numbers provided seem to be for the 6.5 digits mode (vs the nominal 7.5 digits capability). Unfortunately the counts of the meter/ADC seem to not be mentioned so I presume it's 12 mio. which in itself indicates (common for (not only) them cheating) hint: even 20 mio would only be 7.3 digits, but may explain their talking of 6.5 digits measurements.

One laudable point though: at least they don't simply use banana jacks (as most 7.5 and even 8.5 digits DMM regrettably do) but a shielded jack/connection.

All in all what I see is this: They try hard, they really do, but I see no solid basis for assertions like "only a few ppbs ("uncertainty")" or even "fractions of 1 ppb". And I certainly see no solid basis (at all) for "10^-10" precision, let alone in the context of checking accordance between multiple standards, even less so, when hours of flight and oceans are between the "homes" of the two (or more) standards.

Summary: I appreciate their work, and no doubt they *do* try hard, but (a) they have to work with what's available (e.g. questionable "super-meters") plus while the logic "let's use our capability to measure (and produce) frequency with high precision and accuracy to also create a very precise and accurate voltage" seems to make sense and highly likely works, one can only be really sure of what one can reliably measure.
Example: certainly quite unlikely, but can one really say with certainty that the known mechanisms, laws, and techniques that work so well in the voltage ranges we *can* reliably measure, still hold true in significantly smaller ranges? Maybe some weird factor enters the game that is absent or insignificant/irrelevant in the ranges we can measure?
Again, quite unlikely, but we can't be certain unless we can measure and verify (and *fully understand*).

Epilog

So what? I for one am quite happy with a 1 ppm over 1year +-5°C reference and knowing that say, 10 times better would be available if needed.

Also: Hey, I'm just me, just some guy and not of importance. Obviously I *can* learn and change my mind if someone convince me, and quite some people did in my life. I am grateful for the effort they made.
You (not specifically you, but those involved in this discussion) did not convince me so far. But so what? Again, I'm just me and not important. Just let me think in my way as I'm totally ready to not force-feed to you what and how I think. But maybe you think a bit about what I said, just as I thought about what you said (and even glanced over what you linked (except wikipedia).

[The Soulman]

Obviously the HPAK34420A  or similar would be used as a null-meter:
https://en.wikipedia.org/wiki/Null_detector

When I was a little kid (long time ago) someone thought me: "you can't talk and listen at the same time".
Learning requires listening/reading and not much talking/writing.

[Phil1977]

He´s free to think and to some extent to write what he wants, but I stop reading if someone gets esoteric.

And one of the biggest differences between esoterics and science is the usage of maths and statistics 

[Kleinstein]

The reference accuracy for a null meter is not relevant. That part of the error is proportional to the reading and at the intendet zero reading also zero. So the HP34420 is OK as a nullmeter.
For the really sensitive tests with the JJ references they sometimes also use superconducting null meters and this way avoid also the thermal EMF from going from low temperature to room temperature. These can be way more sensitive, down to the sub pV level to allow precise comparsisons even with single Josephson junctions. However they are by design null-meters with a low input impedance (e.g. µH range inductance, no resistance).

@moerm: I totally agree that there is limited need for even higher resolution meters. The quoted resolution is for the noise limit and naturally the accuracy is quite a bit lower (e.g. 1 digit). Even just the transfer accuracy for a small range is not the same as the resolution. To allow testing a meter to certain level one needs noise levels that are something like a factor of 10 lower. The noise part is one of the easier parts to handle. There are even parts where one has to compromise between noise and accuracy (e.g. low resistors for less noise, but more INL from self heating).
The question is more if a meter with accuracy of lets say better than 0.1 ppm for some time frame is feasible.

[Echo88]

Did i read this correctly "a JJA seems to work"? Its the basis for SI-units Mr physics guy.

Youre completely right Dieter, he is a troll.

I wonder if user rhb has already done his aging drift elimination of zeners by statistics as claimed, he seems like an antithesis to moern who hates statistics. Anyway, time for hobbystuff and the quest for lower noise refs, instead of wasting it on trolls.

[dietert1]

There's no forest in antarctica, so he's coming here. There may be other, hidden reasons as well.

[mendip_discovery]

Sooooo

The International types have been voltage measuring for a while.

Zenner Voltages,
https://www.bipm.org/kcdb/comparison?id=1769
1 V,
https://www.bipm.org/documents/d/guest/bipm-em-k11_graph-1-v-may-2024
10 V
https://www.bipm.org/documents/d/guest/bipm-em-k11_graph-10-v-may-2024

JJA,
https://www.bipm.org/kcdb/comparison?id=1779
1 V,
https://www.bipm.org/documents/20126/45253381/RESULTS+BIPM.EM-K10.a-Sept-2021+GRAPH.pdf/4c9b4b53-82b7-f517-e6f6-ea8913a01ccb
10 V,
https://www.bipm.org/documents/20126/45253381/RESULTS+BIPM.EM-K10.b-Sept-2021+GRAPH.pdf/64bdc9c1-872a-c484-3084-d35d30ea864f

Zener gets to uV/V but the JJA gets down to the nV/V range.

https://www.bipm.org/documents/20126/45254323/APMP.EM.BIPM-K11.1.pdf/cdb6332a-c7b8-0334-eec6-0220d6256901
Fluke 732B,
Temperature,
17°C to 27°C
10 V output: -8.5 nV/V (°C)^-1, standard uncertainty 4.7 nV/V (°C)^-1.
1.018 V output: -13.5 nV/V (°C)^-1, standard uncertainty 6.1 nV/V (°C)^-1.

Humidity
Each of the two voltage outputs was measured at 35%, 70% and again at 35% relative humidity. The temperature was set at 20°C and the duration of measurements at each humidity setting was between 6 and 8 days, with the voltage measured and recorded automatically every hour over this period. For these automatic measurements, a second Zener (owned by NMIA) was used as a reference at constant relative humidity.
The results were:
10 V output: +0.56 nV/V (1%)^-1, standard uncertainty 0.07 nV/V (1%)^-1, with a time-constant of approximately 1.5 days. Thus for a change in relative humidity of (for example) 45% to 46%, the voltage rises by 0.56 nV/V.
1.018 V output: any response to humidity change was masked by the noise of this output, expressed by a relative standard deviation of about 87 nV/V. By contrast, the relative standard deviation of the 10 V output was about 38 nV/V. These two measures of noise include the noise of the NMIA reference Zener mentioned above.

Pressure,
The response of the Fluke to a change in pressure was immediate, with no observable settling time.
10 V output: +1.6 nV/V (hPa)^-1, standard uncertainty 0.2 nV/V (hPa)^-1.
1.018 V output: +1.7 nV/V (hPa)^-1, standard uncertainty 0.2 nV/V (hPa)^-1.


From this, I can see we can measure deeper. I have yet to see a multimeter with correction abilities for ambient pressure and humidity. But if I was building a 9.5-digit meter I guess you would have to start taking that into account.

81.96 hpa variance over the past year where I live so I have at least 131 nV/V of error already.

[dietert1]

Above there has been some discussion about a complete voltmeter built inside a (near) hermetic enclosure. I want an IP66 kind of enclosure with a rubber gasket that is easy to work with. Those are available in many different sizes.
Making it an oven for constant temperature is easy, using some PID controller with heaters or Peltier elements. For humidity i have seen good results, too, as we can buffer humidity with some silica gel inside the enclosure. We found seasonal variation of less than +/- 1 % and +/- 0.1 % should be possible.
Recently i started to look at constant pressure solutions and it seems a bit more difficult. Research of leakage modes isn't complete yet. In the end we have those nice integrated digital sensors, like the BME680. So one might use their results for numerical compensation.

Regards, Dieter

[uer166]

But to talk in examples that can be followed by a larger part of the readers: If you have a 1MHz 8-bit ADC available and need a 100Hz 12-bit then you don't need to go shopping. Normal distributed noise is reduced by sqrt(n) if n is the number of averaged measurements. That means if the ADC delivers 6 effective bits then you gain sqrt(10 000)=100 approx= 2^6.5 so your averaged result should theoretically have 12.5 effective bits. If your target are 12 data bits of which 10 should be effective then this method can work without any problem.

Of course I agree you can't have any magic algorithm that creates a usable 12-bit signal out of a 8-bit ADC without drastically reducing its sample rate. "Garbage in garbage out" stays always valid.

I know but frankly I consider that a fairy tale too. Because NO, you can *not* get 12 bits out of an 8-bit ADC, period.
But you can get something like 12 bits out of a system/instrument if the sample rate is sufficiently high. What you actually get though, and that should be understood, is 8 (well, actually more like 6) *actually measured* bits out of the ADC plus another 4 (actually more like 6) bits out of statistics i.e. basically out of thin air).

But then, that statistical "information" is not completely useless because it's at least based on what statistics need: a *large dataset*. With that one at least has some basis for "funny" statistics games.

But an oscilloscope with (nowadays) billions of samples per second is one situation. With a DMM it's a very different situation because you do not have a sufficiently large data set.

And *that*, DMMs were the topic here.

And what about Sigma-Delta ADCs? Are those not 1-bit ADCs that average a reading however many times you need to get to 24+ bits?

With gaussian noise or intentional noise shaping, you definitely CAN make a 12-bit ADC out of a 8-bit one, it ain't magic..

[moerm]

The reference accuracy for a null meter is not relevant. That part of the error is proportional to the reading and at the intendet zero reading also zero. So the HP34420 is OK as a nullmeter.
For the really sensitive tests with the JJ references they sometimes also use superconducting null meters and this way avoid also the thermal EMF from going from low temperature to room temperature. These can be way more sensitive, down to the sub pV level to allow precise comparsisons even with single Josephson junctions. However they are by design null-meters with a low input impedance (e.g. µH range inductance, no resistance).

@moerm: I totally agree that there is limited need for even higher resolution meters. The quoted resolution is for the noise limit and naturally the accuracy is quite a bit lower (e.g. 1 digit). Even just the transfer accuracy for a small range is not the same as the resolution. To allow testing a meter to certain level one needs noise levels that are something like a factor of 10 lower. The noise part is one of the easier parts to handle. There are even parts where one has to compromise between noise and accuracy (e.g. low resistors for less noise, but more INL from self heating).
The question is more if a meter with accuracy of lets say better than 0.1 ppm for some time frame is feasible.

As usual a fine post showing deep knowledge and lots of experience that is, a post one actually can learn from or at least think about and ponder. Thank you!

Probably even more importantly a post that doesn't boil down to "meh, you attack my belief system so I'll fight not only what you say but even you as a person!" like quite a few here seem to approach my posts.

A man who can't calmly listen and react to a view he dislikes is both, someone with a weak belief and a weak man. You obviously are not such a man, and such, a man I'm taking seriously and whose arguments I ponder, think about, and search for knowledge that extends mine.

Btw, I appreciate all the work they do at NIST and other similar institutions. I don't really trust the let's say far out edges (like "< 1 ppb") and think that they stand on somewhat floating grounds there, but I nevertheless appreciate their experiments, attempts, and work.
My main concern is the widely spreading diluted understanding - and practice - of science which I increasingly often have to put into '"'.
Example: the "holy" gravity constant is *not* constant, but hey, as long as the people herds eat the BS it's OK (it seems). And soon it's preached as "everybody knows" and "wikipedia" blabla. And of bloody course the corporations are more than happy to build on that and to sell BS like e.g. a "6.5 digits DMM" with actually less than 6.1 Digits (based on counts) and an "uncertainty" of not ~0.3, not 1 but >20 ppm in its best range (10V) that is, that DMM is in fact not even a 5.5 digits one but closer to a 4.5 digits one (Note: I calculated the average over 2 measurements, one measuring a value of 1/1000 (10 mV) and one of 95% of the range (9.5V) that is at the periphery of typical measurements).

But some fervent believers accuse me of disliking math, being esoteric, unscientific, etc....

[moerm]

He´s free to think and to some extent to write what he wants, but I stop reading if someone gets esoteric.

And one of the biggest differences between esoterics and science is the usage of maths and statistics 

Uhm, I'm quite deep into math almost every (work) day and I'm not against statistics generally but I'm against statistical "magic". In fact I even clearly said that math for me is the yard stick.
So you dislike esoteric but magic is OK? - OK, noted.

Did i read this correctly "a JJA seems to work"? Its the basis for SI-units Mr physics guy.

Well, I guess a few centuries ago I would have been burned as a heretic by the vatican ...

Btw. SI units didn't exist then but they had "standards" too. Like e.g. the width  of "an average men's thumb".

Btw I can sell you the original Eiffel tower and if you buy two I'll even give you 30% rebate!

Kindly do yourself and me a favour and stop making attempts to paint me as stupid in order to shut me up.

There's no forest in antarctica, so he's coming here. There may be other, hidden reasons as well.

Oh no, how terrible, you got me red-handed! OK, I'll confess: I indeed have a hidden agenda. I'm sent by a fervent esoteric (secret, of course!) group to, once settled in, sell space travel tickets to a "nearby, possibly earth-like planet only seven hundred thousand light years away"!

And what about Sigma-Delta ADCs? Are those not 1-bit ADCs that average a reading however many times you need to get to 24+ bits?

With gaussian noise or intentional noise shaping, you definitely CAN make a 12-bit ADC out of a 8-bit one, it ain't magic..

As I said, I'm not here to force feed my view, not even to convince anyone. So, if you feel that what you wrote really is true and correct I simply wish you fun and success building a 3.5 digits DMM around a 8-bit SD ADC (if you find one you like).

[Phil1977]

It´s really time to stop this discussion because it´s that ridiculous that it must be trolling...

Anyhow, though statistically seen its useless: Enhancing ADC-resolution by noise shaping and averaging or Sigma-Delta-ADCs is neither esoteric nor magic. It´s pure maths, and for someone with a scientific or engineering background it´s not even hard to understand why it works.

It´s nothing like noise-reduction algorithms for images or music that are using psychological or AI models to interpret the signal. I could understand if you are against these methods in "pure metrology" - but resolution enhancement is absolutely inside the entropy laws.

Statistics, entropy and especially the second law of thermodynamics are really hard to argue with. Empirical knowledge and experience of most smart people of the world give more than enough trust that they are valid.