Is the Fluke 5440B really an artifact CAL instrument?

[e61_phil]

Hi,

again Fluke 5440B questions, sorry..

I wonder if the Fluke 5440B is an artifact calibration instrument which can calibrate all ranges from only one 10V source.

The manual will guide you to all ranges within the external calibration procedure and you can make adjustments in all of the ranges. Therefore, I think this calibration will tell the 5440B the real factors. After this the 5440B can compare the actual measured factors with the factors measured at calibration and compensate for it. It seems to be more like the AutoCAL function of the 34470A for example. Which can correct for temperature and some drift, but you have to calibrate the instrument in all ranges.

If I'm correct the 5440B can't be calibrated only with 10V if the unit has lost all of the calibration data. And if anyone has entered bad calibration values for example in the 1000V range this will not be corrected, if you calibrate only to a an external standard (732x) with 10V.

Is my understanding correct?

Thanks
Philipp

[zlymex]

The internal calibration removes offsets and gain errors for 10V, 20V, 250V and 1000V ranges.
But, internal calibration cannot remove Vref error just like any other DMM(34470A included).
Also, the internal calibration has a small flaw(at least for 250V and 1000V range) that the calibration voltage is much smaller than working voltage for the sample string(of ten 200k).
That ten 200k string is in a way a Hamon that can be paralleled to 20k during internal calibration.

There are several steps in the external calibration and the procedure is semi-automatic(semi-manual).
Only one voltage standard(10V) is required for external calibration for all ranges.
However, a voltage divider(752A) and a null meter are also required.

[e61_phil]

Thus, just like the AutoCal of the 34470A (or any other AutoCal).

If these gain factors were factors against any artificial internal voltage which is created out of the 13V reference and the 10V calibration, then the calibration for all other ranges are only needed once in a lifetime. However, I'm afraid that the gain factors are related to the 13V reference. That will mean, if the 5440B has drifted to like say +10ppm it isn't enough to do a calibration within the 10V range. You need to calibrate all ranges with a voltage divider, to get the new gain factors matching the actual 13V:(

10V is much easier to get than a 752A.

Unfortunately, I misunderstood Dr. Frank here:
"The internal 13V reference is calibrated via an external 10V standard.  The scale factors of the 4 main ranges are determined by an auto cal process, which can be repeated at any time. An internal resistor divider delivers 220mV and 2.2V ranges, which have to be calibrated against external transfer standards (i.e. 752A or 3458A)." (amplifier.cd)

I hoped I get away with only 10V and without a Fluke 752A if I use the ranges 22V and above only.

[e61_phil]

I'm with you DiligentMinds.

My (wrong) understanding from the text on amplifier.cd was: You need do compare the 10V and the rest is done by the instrument. An external nullmeter is imho okay to name it "artifact calibration". But if you need more instruments (like Fluke 752A) it is of course not an artifact calibrating unit.

[RobK_NL]

10V is much easier to get than a 752A.

I want to get rid of mine, so if you're interested, drop me a PM.

[Dr. Frank]

Well, the 5440/5442 is in fact an artifact calibration (AUTOCAL) instrument, with some limitations:
- 0.1ppm resolution (of gain shift)
- small gain shifts only
- ranges 11V, 22V, 250V and 1100V only (cannot autocal 2.2V and 220mV ranges)

Fluke did not yet (at about 1986) call it clearly an artifact calibration instrument, maybe because they were not self-confident about this, and as the conservative metrology community would have grilled them.
When they introduced the artifact calibration with the 5700A, (early 1990-ties) they initially had a lot of trouble to be accepted by the NBS, and others, and a lot of additional verification had to be performed, before this 57xxA series became the artifact instrument standards, as of to-day.

The 3458A ACAL was accepted much easier, as somebody reported here, due to the close cooperation of NBS and HP on the JJ array research.

Anyhow, if you read the service manual very carefully, especially paragraph 3-27, 3-28 about internal / external calibration, and the appendix "Recent innovations in DCV calibrator design", page A-11, it's quite evident, that the 544x acts exactly as an artifact instrument. (In fact the manuals are quite odd to read, it took me a while to discover this fact, between the lines, or out of what is NOT said)

One decisive hint is, that during calibration, you can always omit the complete external calibration, of course except the 10V calibration, that is the 10V (ext) to 13V (int) transfer.
After this initial step, you are allowed to take the short cut, and only do the internal calibration, for the above mentioned 4 higher ranges, w/o any specified penalty.
So up to this point, you only need an external 10V reference, but no 752A.

In the whole manual there is no restriction for that abbreviated procedure, so it should bring all ranges to their 24h limits, provided the internal 13V reference is still in cal.

Maybe, if the internal resistors had drifted too much, these small correction gain factors would get too big, and a complete external calibration would be necessary.
But as old instruments probably do not drift that much any more, this does also not occur. (Never got an internal cal error the last 7 years..)

Maybe also, that an extensive external calibration would provide greater resolution than 0.1ppm (see A-11), but this is not relevant for an ~0.5ppm grade instrument.
The internal/external cal uncertainty is also not clearly specified for neither case, and the Fluke 752A also provides 0.2 and 0.5ppm transfer uncertainties only.

The 5440B and the 57x0A both have a similar internal transfer / ACAL system inside, especially both have ovenized divider resistors.
This allows precise gain calibration of all DCV ranges, at low bridge excitation voltages, even for the High Voltage ranges.

Hint @ zlymex: the 5440B has no design flaw in this sense at all, because the temperature of the divider resistors is always constant, either with the 1000V burden (30sec stabilization required), or with the -13V bridge configuration. Other HV effects can be neglected, I think. Also, if that would be a design flaw, the 57xx instruments would have the very same, which is accepted to not be the case.
Also, the 752A is calibrated at 20V level, even for the 100:1 / 1kV range, generating exceptional uncertainty, by T.C. matching, instead of ovenizing.

The 544x is even superior over the 57xx, as it uses the Hamon configuration, which is more precise (like the 752A) than the A/D or D/A dependent 10:1 transfers inside the 3458A (~0.3ppm) and the 57xx (< 1ppm ).
The 544x is superior over the 3458A at least in the 1000V range, as the 3458A lacks all kind of thermal compensation methods.


I also have compared the internal calibration function of my 5442A, against the 3458A and my Hamon type 100:1, 10:1 divider, and never found a contradiction to the assumed artifact calibration functionality of the 5442B.
All three agree always within about typ. 0.2ppm, whereas the 3458A is off by about -3ppm at 1000V, compared to the other instruments.

As a conclusion, the 544xB is definitely an artifact predecessor of the accepted 57xx instruments, but FLUKE maneuvered verbally around that fact, these days.

Frank

PS for e61_phil: Like all other ACAL instruments, this internal calibration is necessary once a day, at least, to compensate for short termed drifts.
Also, you interpreted my statements on amplifier.cd correctly, I think, maybe the German text is even more precise.

[e61_phil]

I hopped you will answer here, thanks Dr. Frank


If I understand you correct, I can stop the calibration after the 10V range and the other ranges will follow the "new" 13V?


What about a little experiment?
I got a "new" 5440B from singer at work. I don't have any calibration history and it seems to be about 6ppm down. The unit will go to the calibration lab in about two weeks. So, why not do a small test? I could start a 10V calibration and use 9,99V as "standard". After this step I will stop the calibration procedure. If the unit is able to do an artifact calibration, the 100V Range should give 99,9V.

What do you think?

[Dr. Frank]

If I understand you correct, I can stop the calibration after the 10V range and the other ranges will follow the "new" 13V?

See page 3-21: No, you do not stop the calibration after step 12 or 13, in this sense, as you initially have already initiated the internal calibration, in step 1.

What about a little experiment?
I got a "new" 5440B from singer at work. I don't have any calibration history and it seems to be about 6ppm down. The unit will go to the calibration lab in about two weeks. So, why not do a small test? I could start a 10V calibration and use 9,99V as "standard". After this step I will stop the calibration procedure. If the unit is able to do an artifact calibration, the 100V Range should give 99,9V.

What do you think?

Not good an idea, as this deviation will be too big, and maybe spoil completely the calibration of your own instrument.. because the internal calibration can handle "small" deviations only (whatever 'small' is, it's not specified). And what do you want to check, anyhow?

Principally, you need to compare each of the 4 ranges/ratios against an independent ratio instrument.
If you use the calibrated Singer instrument as such, you might detect or confirm this ratio scaling effect to some degree of precision.
The Singer instrument should be calibrated completely with a 752A (extended calibration), otherwise you compare identical methods. (That's not independent from each other)

In the end, as soon as you engage the internal cal procedure, this will virtually overwrite the preceding calibration of the lab, won't it?


I would suggest instead to use a 3458A, or a 752A, to cross check these 4 ratios, up to 300V (which does not yet yield thermal non-linearities on the 3458A).

Frank

[e61_phil]

The singer 5440B will be calibrated with a 732A and a 752A in two weeks. Therefore, I thought it doesn't matter to play around with the singer unit before calibration. Everything I will mess up should be fixed after the calibration.

My intention was to verify, if the unit does really an artifact calibration if I will only calibrate the 10V range against a 732A. Ok, 9,99V is very much. I can also try to use a "standard" which is only about 10-20ppm shifted against the actual value. I need a deviation which I can measure for sure in another range (with a 34470A or 3458A).

1. measure 100V for example with a 3458A
2. "calibrate" the 10V of the 5440B to +20ppm
2. measure 100V again

If the unit will perform an artifact calibration, the 100V should also shifted by +20ppm.

[Dr. Frank]

The singer 5440B will be calibrated with a 732A and a 752A in two weeks. Therefore, I thought it doesn't matter to play around with the singer unit before calibration. Everything I will mess up should be fixed after the calibration.

My intention was to verify, if the unit does really an artifact calibration if I will only calibrate the 10V range against a 732A. Ok, 9,99V is very much. I can also try to use a "standard" which is only about 10-20ppm shifted against the actual value. I need a deviation which I can measure for sure in another range.

Both instruments might have the very same systematic deviation in their internal cal procedure, so you won't detect this by comparing identical techniques..

But you might do this test, without any re-calibration of your own unit; the freshly calibrated Singer box will very probably deviate sufficiently (a few ppm maybe) in the 10V range, and so will all other ranges do, ratio-wise.
Therefore, you only will have to compare each of these four ranges and calculate the ratios, should be exactly be 1:2:25:100, within < 1ppm.

Frank

[e61_phil]

What do you mean with "both instruments"? My test setup will only include a 3458A (or 34470A) and the singer 5440B. I will only change the calibration of the 10V (singer 5440B) and measure the 100V (out of the singer 5440B) before and after this "calibration" of the 10V range.

Sorry, for the confusions due to my bad english...

[Dr. Frank]

What do you mean with "both instruments"? My test setup will only include a 3458A (or 34470A) and the singer 5440B. I will only change the calibration of the 10V (singer 5440B) and measure the 100V (out of the singer 5440B) before and after this "calibration" of the 10V range.

Sorry, for the confusions due to my bad english...

Your English is fine, I think..

In first instance, you did not mention 3458A, or 34470A. So I thought, you would compare two 5440Bs against each other..

Anyhow, if you have a 3458A, you can rely on its de facto ratio / ACAL functionality (precise to about 0.02ppm / 0.3ppm), and simply determine the ratio values between the 11V, 22V, 250V and 1100V ranges at first w/o de-calibrating anything. Choose clever values, to check range-ratios of the 5440B on the same 3458A range, and avoid self-heating effect of 3458A HV divider..
The precise ratio values, expected to be within < 1ppm (non-linearity of the D/A included) should be a first proof.

Then you might simulate a full external calibration of the 5440B, with slightly wrong ratios in the upper 3 ranges (leaving the 11V as it is), maybe 20ppm off, again using the 3458A.
After an internal cal of the 5440B, the ratios should be precise again.

I think, the 34470A may only be suitable to detect that sufficiently well, if it had been adjusted (not only calibrated!) recently, to <1ppm range / ratio uncertainty, like in the initial test report.
Although I assume, and also have measured, that its ACAL function now works very precisely with the latest FW 2.14 (*), this is not specified at all. Specification says something of +/-20ppm between ranges, within 24h..

Frank

(*) The 34465A/34470A had a FW flaw, so that the 100V and 1kV calibration were off by up to 20ppm, from the initial test report, and therefore also the ratio relative to the 10V range was off by  that amount. So if your 34470A had been calibrated with FW < 2.14, you can forget that experiment.

[zlymex]

10V is much easier to get than a 752A.

That's true in a way now days, but in the old days 732A was the only decent 10V available from Fluke and had similar physical size as 752A, 732A probably costed not much less than 752A. Any one got the original list price for those two?

Also, 752A is self-calibratable at 0.2ppm and 0.5ppm while 732B must be externally calibrated and rated at 3ppm per year(old: 6 ppm per year), that will make the 10V more troublesome than 752A divider.

[zlymex]

Hint @ zlymex: the 5440B has no design flaw in this sense at all, because the temperature of the divider resistors is always constant, either with the 1000V burden (30sec stabilization required), or with the -13V bridge configuration. Other HV effects can be neglected, I think. Also, if that would be a design flaw, the 57xx instruments would have the very same, which is accepted to not be the case.

'Calibration at working voltage' has been a golden rule(at least here in China) for critical voltage dividers to overcome non-linearity.
It's true that wire-wound resistors show negligible voltage non-linearity especially thermally stabled, but there are other factors such as leakage on bare PCB tracks and relays. Leakage current has very severe non-linearity especially at high voltage. For sub-ppm grade 1000V voltage dividers without internal guarding, it IS a design flaw.

[e61_phil]

I think, the 34470A may only be suitable to detect that sufficiently well, if it had been adjusted (not only calibrated!) recently, to <1ppm range / ratio uncertainty, like in the initial test report.

My first intention was to measure 100V before and after this 10V decalibration only. To detect a 10ppm shift even my 3456A at home (3ppm transfer spec@100V) should be suitable.
However, your ratio measurement suggestions are much nicer to check the ability of the 5440.

10V is much easier to get than a 752A.

That's true in a way now days, but in the old days 732A was the only decent 10V available from Fluke and had similar physical size as 752A, 732A probably costed not much less than 752A.

As a hobbyist volt-nut it is even hard to get an 732A/B. Nevertheless, many people (like me) has access to calibrated instruments (for example a few 3458A with calibration data) and use these instruments immediately after calibration to transfer the 10V from the calibrated instruments to the DIY 10V standard. (Many people are not allowed to take the instruments with them and don't want to carry the 35kg monster to the workplace).

Therefore, I think the 5440B is great because of the autocal ability and the only need for 10V to calibrate the whole instrument except the DIV ranges. I had a Fluke 343A before, but such a calibrator is nearly useless without the whole calibration equipment (except as a stable and tuneable voltage source).

'Calibration at working voltage' has been a golden rule(at least here in China) for critical voltage dividers to overcome non-linearity.

I read a lot of the voltage dividers for the KATRIN experiment (18kV). The PTB uses their 100kV divider to calibrate the KATRIN divider at working voltage. The design of the KATRIN divider is well described in some german PHD theses. But how can one make such measurements at working voltage without a second divider to compare?

[Dr. Frank]

'Calibration at working voltage' has been a golden rule(at least here in China) for critical voltage dividers to overcome non-linearity.
It's true that wire-wound resistors show negligible voltage non-linearity especially thermally stabled, but there are other factors such as leakage on bare PCB tracks and relays. Leakage current has very severe non-linearity especially at high voltage. For sub-ppm grade 1000V voltage dividers without internal guarding, it IS a design flaw.

Hello zlymex,
do you want to tell, that you, in China, calibrate the 752A, or the 720A, or whatever < 1ppm ratio instruments at 'working voltage', i.e. at 1000V?
I really doubt that..

The 544x and the 57xx also are not specified and intended as a sub-ppm grade instrument @ 1000V, (that's the 752A only, which achieves 0.5ppm) but the HV ranges of the 544x and 57xx in relation to the basic 11V range are accurate by this internal cal routine to 1ppm, obviously.

I can't reproduce, whether or not a dedicated guarding mechanism is foreseen in these instruments, to mitigate leakage effects, but obviously the design is sufficient to meet the spec.

Strange, that you still call instruments with these very good characteristics a design flaw.

Frank

[zlymex]

Hello Dr. Frank,

Firstly, there is no guarantee for Fluke specs. For instance, there are many 742A and 732B are out of spec.
Secondly, rule is a rule, people observe it or not, that is another matter. The head of Division of Electricity and Magnetism at NIM(Mr. Shao) had written and spoken that rule(calibrate at working voltage) many times.
Thirdly, at NIM(National Institute of Metrology, China NIST equivelant), they don't use 752A and 720A as serious dividers, they use 4902S, and better still, they build their own true guarded divider, and all calibrated at working voltage.
Fourthly, the input resistance of 720A is 100k(at 1.0 input) which is not very large and thus the leakage probably less significant. However, the divider string of 5440B is 2 Meg(for 1000V range), making it more vulnerable for leakage on the not very short bare PCB tracks and relay contacts that easily get dirty because of the circulation.
Lastly, if something is good, doesn't necessarily mean it must be flawless.

To get to my point, calibrating at non-working voltage is a small flaw, and no guarding on ppm-level divider for 1000V at mega-ohm string is another flaw.

BTW, there is a guarding circuit inside 752A, but not true guarding.

[zlymex]

Therefore, I think the 5440B is great because of the autocal ability and the only need for 10V to calibrate the whole instrument except the DIV ranges. I had a Fluke 343A before, but such a calibrator is nearly useless without the whole calibration equipment (except as a stable and tuneable voltage source).

True, I think 5440B is great too, otherwise I wouldn't brought the second one.

'Calibration at working voltage' has been a golden rule(at least here in China) for critical voltage dividers to overcome non-linearity.

I read a lot of the voltage dividers for the KATRIN experiment (18kV). The PTB uses their 100kV divider to calibrate the KATRIN divider at working voltage. The design of the KATRIN divider is well described in some german PHD theses. But how can one make such measurements at working voltage without a second divider to compare?

Yes, calibration at working voltage requires a second(and better) divider. Picture below is the Chinese regulation(JJG 1007-2005) for calibration high voltage dividers where B is the high voltage generator.

[e61_phil]

Yes, calibration at working voltage requires a second(and better) divider. Picture below is the Chinese regulation(JJG 1007-2005) for calibration high voltage dividers where B is the high voltage generator.

But how do you get such a better divider in the first place?

[zlymex]

Yes, calibration at working voltage requires a second(and better) divider. Picture below is the Chinese regulation(JJG 1007-2005) for calibration high voltage dividers where B is the high voltage generator.

But how do you get such a better divider in the first place?

There is this Datron 4902S available that specified as 0.2ppm for 1000V:10V divider. They use tons of hermetic foil resistors(from Vishay), matched, guarded, compensated.

More still, NIM said there are some limitations for 4902S so they built one of their own and claimed to be better than 4902S.

[Dr. Frank]

Hello Dr. Frank,

Firstly, there is no guarantee for Fluke specs. For instance, there are many 742A and 732B are out of spec.

This is a trivia, I think.
You always have a statistical distribution of the specified parameters. At a typical confidence level of 95%, about 5% of the instruments fall out of the limits.
Fluke published a study about 732A and 732B references, showing exactly this, but this included also the drift of the SZA263 in typically positive, and the LTFLU in typically negative direction.
It's also trivial, that calibration instruments may drift out of specification over time, and even earlier than the specified 1 year.
If the design and verification is done correctly, this should also cover the 95% confidence level.

Secondly, rule is a rule, people observe it or not, that is another matter. The head of Division of Electricity and Magnetism at NIM(Mr. Shao) had written and spoken that rule(calibrate at working voltage) many times.

Sorry, I don't know Mr. Shao, but he for sure also cannot overrule the laws of electrical engineering, or physics.
It's not possible to calibrate a 752A or other Hamon type dividers at 'working level', for obvious reasons, e.g. that would destroy the divider resistors.

A different story would be, to use a very different technique of calibrating reference / decade dividers, so that each of the resistors also sees the same voltage burden during calibration, as during usage.

Thirdly, at NIM(National Institute of Metrology, China NIST equivelant), they don't use 752A and 720A as serious dividers, they use 4902S, and better still, they build their own true guarded divider, and all calibrated at working voltage.

Well, I don't know the adjustment technique of the 4902S, as I did not yet find a manual, or a description of the circuitry.
If you have that available, please share.

If it's a Hamon type calibration technique again, it's also not possible to do the adjustment on working voltage.
If it's more a technique like in the 720A, where each of the about 100 resistors may be adjusted at 10V, by comparing it in a bridge configuration, you would only get the mediocre ratio specifications for 10:1, 100:1, like the 720A, for sure.

Please, share the NIM technique, how to achieve Hamon type uncertainty (or better) at  working voltage adjustment.. I really can't figure out, how this works.. 

Fourthly, the input resistance of 720A is 100k(at 1.0 input) which is not very large and thus the leakage probably less significant. However, the divider string of 5440B is 2 Meg(for 1000V range), making it more vulnerable for leakage on the not very short bare PCB tracks and relay contacts that easily get dirty because of the circulation.
Lastly, if something is good, doesn't necessarily mean it must be flawless.

Yes, that's also already known about the 720A; but in practical terms, the 720A is anyhow not suitable for very precise 10:1 or 100:1 ratios.
Leakage currents are taken into account for the 720As specification also, like in the 752A ones.

The 5440B may suffer from leakage currents, but obviously it meets or exceeds its specifications anyhow.
Would be interesting, instead, if you had some experience or research results about older 5440B instruments, how and where to possibly clean the instrument, to mitigate such leakage effects and restore its initial performance.
In the end, the 5440B is not intended as a sub-ppm ratio instrument, like the 752A, or like any other of your superior NIM dividers, so your criterion of 'design flaws' in the 5440B is a bit academic, in this case.

To get to my point, calibrating at non-working voltage is a small flaw, and no guarding on ppm-level divider for 1000V at mega-ohm string is another flaw.

BTW, there is a guarding circuit inside 752A, but not true guarding.

Do you have any reference, or documents, how such ratio dividers standards are built, so that they achieve the Hamon type uncertainty, but were really also adjusted at working level voltages?

That would be really helpful.

Btw.: In the 752A manual, they summarized and evaluated all possible sources of errors, which leads to the theoretically calculated specification, which again is good metrological practice.
I also assume, that all these institutions also make an evaluation of their self-built dividers, at first.

I also know about several ratio divider comparison papers from different National Standards Institutes, where they really do a ring-comparison of 752As vs. 4902S, vs. self-built versions under working voltage conditions, i.e. at 1000V:10V mode.
But that comparison again is a different story than the different adjustment methods, I think.


I really would appreciate your practical guidance about designing more precise ratio divider standards!

Frank

[zlymex]

Hello Dr. Frank,

This time I won't explain correspondingly, but I'll say what ever I know about precision dividers, limited only by the verbal agreement to one of my friend who shared me with something divider related that I promised not to disclose.

1. I have 4902/S manual(of 31 pages) in electronic(scanned) form, but I cannot share it because the one who gave it to me obtained it not in a legitimate way.
I'll attach the 2nd content page anyway, this will explain a little about how 4902/S is calibrated: by using the dedicated 4901 thru 4904, allowing 4902/S to be calibrated at full working voltage(1000V max).
There is an Fluke article 'Migrating from dc voltage dividers to modern reference multimeters'  http://support.fluke.com/calibration-sales/Download/Asset/2114953_6200_ENG_A_W.pdf
Bottom-right photo of the first page is the Datron 4900 system, top is 4901, middle is 4903, bottom is 4902/S, left is 4904.


2. There are three basic principle types for precision dividers, namely KVD(as in 720A), Hamon(as in 752A) and the type that 4902/S uses, which I think is called cascade(by translation from Chinese, the manual didn't say it). There are many dividing ratios, not 2 as for 752A, not near infinite as for 720A, but all fixed voltages and marked on the panel. For instance, 600V:10V can be used, but this range cannot be used for 60V:1V. This will make sure that the divider is working on exact the same voltage as calibrated for every possible ranges.

3.There are some photos of 4902/S here: http://bbs.38hot.net/thread-7032-1-1.html
Yes, I once get hold of that unit for a while, teardown and tested. I could keep it if I like, but gave up in the end. The same unit is now on sale here: https://world.taobao.com/item/42045355947.htm

4. I don't have the schematics, it is not included in the manual, it is a separate file called Circuit Diagram 430582.
I haven't draw the diagram myself either. It actually very simple, just nine 100V 4-W strings, plus ten 10V, all fully guarded. The bottom part of the PCB photo is attached, anyone can figure it out together with the photo of the panel and the top PCB.

5. The DIY divider by NIM
There was an early article 'Calibration of High Correctness DC Resistance Voltage Divider', where they compared 4902S with their own divider called TH12 at working voltage(1000V max) and obtained a compatible results: http://gfjl.org/thread-8073-1-1.html
The article is in Chinese but the abstract is in English. It is this article that saying there are three types and 4902S belongs to cascade type.
Later, NIM build another divider, the photo is still on their website: http://www.nim.ac.cn/divisions/4

Top part is of course 4902S, bottom is 4000A, middle part is their DIY.
I have the report of the DIY at hand, again I cannot share it which in Chinese anyway. It is also a cascade type, but with an auxiliary divider and 9 high CMRR wireless-data-link null meters inside, to facilitates the calibration by 'two step shift methed'. The divider string is made up of some matched 400 Vishay resistors of untold type. Looking at the scattered diagram of the tested TCR, most of the TCR fall in -1.5ppm/K to 2.0ppm/K band. Also, unlike 4902/S that use common film resistors, they use very high precision resistor for guarding network thus keep any voltage differences to within 10mV. The result, thru uncertainty analysis, they claimed to be 5 time better than the previous one, 20 to 50 ppb type A uncertainty and 80ppb combined uncertainty for 1000V:10V.
In the last part of the report, they compared 752A with it and said there were two weaknesses for 752A: only two ratios and calibrate at low voltage.

Edit, there is probably a fourth type of precision divider, Binary Voltage Divider, like MI's 8000B, with an 20 Channel Scanner, Accuracy <0.05 ppm, Range to 1200 Volts, able to Calibration of Fluke 5700A/5720A, and Self Calibration.
http://www.mintl.com/media/pdfs/8000B.pdf

[Vgkid]

The method employed by the Datron4902s, and the NIM divider. Remind me of the Voot Boxes produced by L&N/ Guildline/(possibly others). Which used strings of resistors in an ovenized oil bath, to act as a voltage divider.

[e61_phil]

Today I played with the 5440B and a 3458A:

EXT CAL 10V + 10ppm no other calibrations
After this calibration the 100V output was also shifted by +10ppm

-> artifact calibration possible q.e.d.


For further tests I "calibrated" the Fluke 5440B to match the 10V from the 3458A.

Ratio error in 100V Range between 10V and 100V seems to be about 1ppm

Edit:

99,999677 against 9,999979 results in 1,13ppm error

But why is the reading in the 100V range about 3,2ppm low? Is this the sum of errors? 2ppm from the Autocal of the 3458A and 1,13ppm from the 5440B = 3,2ppm?

I wouldn't expect 2ppm error from the Autocal (this is the difference between the ranges in the specs). And the last calibration certificate (last month) showed 10V about 0,2ppm to low and 100V about 0,9ppm low.

[e61_phil]

If anybody is interested, I've done some linearity measurements in the 10V range of the 3458A

Fluke 5440B   HP 3458A
0   -0,00000008
1   1,0000048
2   2,0000065
3   3,0000083
4   4,000009
5   5,0000096
6   6,0000098
7   7,0000105
8   8,0000102
9   9,0000106
10   10,0000106