KVD for ACV?...

[Rax]

Hi all,
As I just acquired an AC voltage standard (Fluke 510A/1KHz) and having a KVD (ESI RV722), can I reliably divide the output (10VAC) with the divider, and output precise fractional AC voltages?

I think this is supposed to work very well with low DC voltages, which would allow the meter to look at this from its stealth 10G input impedance, but I'm not sure what the accuracy of the output would be with AC voltage.

Thank you!

[Vgkid]

You want an inductive decade divider.
Gertsch is a common maker , Eaton,  Palico,  esi

[srb1954]

You want an inductive decade divider.
Gertsch is a common maker , Eaton,  Palico,  esi

Sometimes also known as a ratio transformer. Gertsch is the best known brand but they are not cheap.

For AC signals ratio transformers have the advantage that their division ratios are inherently fixed and stable so they don't need periodic re-calibration like a KVD. However, they are somewhat limited in their usable frequency range and the maximum voltage that can be applied.

[Vgkid]

You want an inductive decade divider.
Gertsch is a common maker , Eaton,  Palico,  esi

Sometimes also known as a ratio transformer. Gertsch is the best known brand but they are not cheap.

For AC signals ratio transformers have the advantage that their division ratios are inherently fixed and stable so they don't need periodic re-calibration like a KVD. However, they are somewhat limited in their usable frequency range and the maximum voltage that can be applied.

Gertsch units are not that expensive, tegam/ eaton/esi units are.

[Rax]

So, a regular KVD is not appropriate for this - that's what I'm hearing.

I guess the Vac rating on my KVD gave me unfounded hopes... 

[Vgkid]

I believe that there used to be an ac response note in the esi rv-xxx manual.

[Martian Tech]

RTFM.

"The RV 722 is suited for use at low frequency ac as well as dc." [page 1-2]
"For ac measurements the detector is nearly always tuned and exhibits a fairly high Q..." [App Note 29, page 12]

https://www.ietlabs.com/pdf/application_notes/RV722_app_note.pdf

[Conrad Hoffman]

A KVD should work at low frequencies, but I don't know what the limit is. The price for ratio transformers varies widely as they aren't used as often as they once were. General Radio also made a couple that are quite excellent, but rarer than hen's teeth.

https://www.ietlabs.com/pdf/Manuals/GR/1493%20Precision%20Deacde%20Transformer.pdf

They also made a version with one or two less decades. I have one and a friend has the 1493, but those are the only two I've ever seen in the wild.

Henry Hall did a paper on precision AC ratios where he got down into the ppb range. Doesn't seem possible, but he did it.

[Rax]

RTFM.

"The RV 722 is suited for use at low frequency ac as well as dc." [page 1-2]
https://www.ietlabs.com/pdf/application_notes/RV722_app_note.pdf

Well, fine - and thank you - but it's interestingly tucked in the "Description" section (the very last sentence), and there's not a word about AC in the specs. Therefore, this tells me almost nothing. For instance, it'd be interesting to... disclose what exactly is "low frequency." Is it up to 1kHz, 10kHz, 20kHz (audio), is it 100kHz (typical "LF" in some quarters), or are we getting near the MHz'? Which I'm sure we're not.

"For ac measurements the detector is nearly always tuned and exhibits a fairly high Q..." [App Note 29, page 12]
https://www.ietlabs.com/pdf/application_notes/RV722_app_note.pdf

I don't think this is quite directly related, my question was specifically regarding a standard ACV source as divided by a KVD. I'd want to learn if I can output such divided ACVs into meters during calibration adjustment.

[bdunham7]

If the AC reference has a low source impedance and you ignore any parastic capacitance or inductance in the KVD, then the main remaining issue is the loading of the meter and the output impedance of the KVD at whatever setting you are using..  Assume you have 1M + 100pF for loading and you can then calculate what errors might be introduced.  For a 0.1 ratio on your KVD, I think the loading errors willl be more than you would like, probably more than 1%. 

[Rax]

If the AC reference has a low source impedance and you ignore any parastic capacitance or inductance in the KVD, then the main remaining issue is the loading of the meter and the output impedance of the KVD at whatever setting you are using..  Assume you have 1M + 100pF for loading and you can then calculate what errors might be introduced.  For a 0.1 ratio on your KVD, I think the loading errors willl be more than you would like, probably more than 1%.

Thank you, b, that's helpful. I assume the parasitic capacitance and inductance to be relatively sizable (relative to the frequency of the source, hence the "low frequency" requirement). But maybe this approach at least gives me a starting point to estimate under what conditions and which ranges I can actually trust the KVD to output ACVs that can be trusted.

[Vgkid]

While not helpful model wise, genrad, and esi do use mica card resistors, but of different winding techniques.
This should give you a guesstimate for frequency response.
Looking at the math, some accuracy can be had at low frequencies. Though I wonder how the inductance of the wiring plays into it.

[srb1954]

If the AC reference has a low source impedance and you ignore any parastic capacitance or inductance in the KVD, then the main remaining issue is the loading of the meter and the output impedance of the KVD at whatever setting you are using..  Assume you have 1M + 100pF for loading and you can then calculate what errors might be introduced.  For a 0.1 ratio on your KVD, I think the loading errors willl be more than you would like, probably more than 1%.

Thank you, b, that's helpful. I assume the parasitic capacitance and inductance to be relatively sizable (relative to the frequency of the source, hence the "low frequency" requirement). But maybe this approach at least gives me a starting point to estimate under what conditions and which ranges I can actually trust the KVD to output ACVs that can be trusted.

The calculation of the AC effect of loading is complicated by the considerable variation in the source resistance presented by the KVD as the settings are changed. There is a graph in the RV722 manual that shows the source resistance varying by a factor of about 100 as the KVD settings are adjusted over its range. You would have to take the worst case figure in that graph to determine whether your external capacitive/resistive loading causes excessive error at your frequency of interest.

Normally a KVD is used in a DC bridge circuit with a null detector so the load resistance presented at the null is practically infinite and the KVD source resistance doesn't have any effect on the setting accuracy.

 

[Kleinstein]

There are 2 problems with a KVD with AC. One is that AC voltmeters are usually not very high impedance, but more like 1 M or maybe 10 M and at least have some input capacitance. The 2nd is parasitic capacitance also inside the KVD. Because of the usually relatively large resitors parasitic inductivity is less of an issue.

With resistors in the 100 K range and parasitic capacitance in the 10s of pF the cross over from resistive to capacitve is in the 100 kHz range. For good accuracy one should be well (e.g. a fator 1000 for 0.1% range) below that. So in this case low frequency should mean about mains frequency with already a reduced accuracy.

[alm]

I'd look if you can find research articles from the time these dividers were common where they use them for AC or compare them to IVDs. What I could quickly find is this paper which in the introduction very briefly mentions that calibrating resistive dividers for AC voltage at high audio frequencies is problematic because of parasitic inductance and capacitance. This is even after calibration (correcting for frequency-dependence). So without any corrections I think Kleinstein's estimate would be reasonable.

There may well be more articles out there that may not use the words Kelvin-Varley or that are not very well indexed. For example, try skimming NIST special publication 300 volume 3, which is a collection of NIST papers from the sixties.

If you can't find any reference to people using them for AC, then that's probably not a sign that it works well

[Conrad Hoffman]

Not sure when/if I can get to this, but since I have multiple ratio transformers and KVDs, I should be able to do a comparison and see exactly what the KVDs are capable of. I need to clean up the bench, so this might be a good excuse. That GR manual I linked shows using a null meter and some parts for phase compensation, which I need to understand.

[Rax]

Awesome input, thank you all.

What about in joint venture with a null meter? I am thinking that possibly the parasitic effects may get canceled in a "zero reading" setup, with no (or infinitesimally small) currents flowing.

But along these lines, can anyone describe a practical application for ACV? I am not sure I fully grasp it in application (involving a null detector, that is). I'd essentially have an ACV standard (510A), a KVD (RV722), a calibrated meter for characterization, and a whole bunch of other things that may come in handy but can't imagine they do.

[Kleinstein]

An AC nullmeter it a bit tricky. Usually the null detectors for AC have to also look at the phase (e.g. as a lock in amplifier), so one knows which way to adjust. Otherwise one gets a more or less broad minimum.
Reading null helps with the input impedance of the meter, but not with capacitance to ground and internal capacitance, that is in parallel to parts of the divider.

A problem with the KVD is that it has a variable impedance of maybe some 50 - 100 K at the center and getting smaller to the ends, but not necessary in a simple way.
The big advantag of the transformer based dividers it that the outputs are low impedance, like 50 ohm or mybe even 10 ohm range. So parasitic capacitances are much less important.
Still they idelly use auxiliary dividers and driven shields / triax connectors if really high accuracy is aimed for.

[Rax]

An AC nullmeter it a bit tricky. Usually the null detectors for AC have to also look at the phase (e.g. as a lock in amplifier), so one knows which way to adjust. Otherwise one gets a more or less broad minimum.

The article mentioned by Martian Tech may offer some clues:

"For ac measurements the detector is nearly always tuned and exhibits a fairly high Q..." [App Note 29, page 12]
https://www.ietlabs.com/pdf/application_notes/RV722_app_note.pdf

[alm]

Not sure when/if I can get to this, but since I have multiple ratio transformers and KVDs, I should be able to do a comparison and see exactly what the KVDs are capable of. I need to clean up the bench, so this might be a good excuse. That GR manual I linked shows using a null meter and some parts for phase compensation, which I need to understand.

The ESI DT72A manual also describes using a R with a variable C circuit to make the phase shift of two dividers the same. I'd think a generic AC millivoltmeter might be good enough after you adjust the phase shift for minimum reading between the two dividers, although you may have to readjust every time you adjust a divider substantially.

I was also considering comparing a resistive KVD (Fluke 720A) with an IVD (ESI DT72A) , but I was planning to use absolute readings of a DMM to compare, which won't be 1 ppm accurate, but then I don't expect great performance from the KVD. Meter input impedance, which would be 1 MOhm in parallel with some dozens of pF best case, and even lower for some AC measurement standards (Wavetek 4920) will certainly be a concern. But the closest I have to an AC null meter is the Fluke 8920, which I'm not sure is up to the task.

But along these lines, can anyone describe a practical application for ACV? I am not sure I fully grasp it in application (involving a null detector, that is). I'd essentially have an ACV standard (510A), a KVD (RV722), a calibrated meter for characterization, and a whole bunch of other things that may come in handy but can't imagine they do.

See for example the application section of ESI DT72A (and probably manuals for other ratio transformers). The main use would be using the AC null detector to compare the divided output to that of another divider.

[guenthert]

You want an AC null detector.  A lock-in amplifier can act as a (high impedance) sensitive AC null detector.  As a bonus, they typically have a reference signal output, which can be used as an AC voltage supply for the bridge (you probably will need an AC amplifier though).

[alm]

I looked at the Fluke 720A manual, and the only way I can see the load impedance of a DMM in ACV range not destroy any accuracy, even at very low frequencies, is to use the divider at ratios like \$2\cdot10^{-4}\$ and lower. That way the output impedance is low enough that variations in the DMMs input impedance will not swamp the uncertainty budget, but uncertainty due to the KVD linearity of 0.1 ppm of input will be 0.05%, and you will still need to correct for the loading impedance. The lower the ratio, the higher the linearity error and the lower the loading error. So I'd argue KVDs like this are not useful for AC unless you can increase the load impedance using either potentiometric technique (null meter against another divider) or some sort of buffer amplifier.

If the only usable way to use a KVD for ACV is using a null meter because the loading due to the input impedance of a DMM is too high, then I'd argue that fact alone is enough to severely limit their usability for ACV. Since you'd probably need to determine correction coefficients for the KVD for ACV using another divider, what would be the point of subsequently using that KVD to compare it to yet a third divider?

I don't see how an AC null meter would work with another non phase locked source like adjusting an ACV calibrator or stable function generator to a Fluke 510A.

[Alex Nikitin]

The ancient KVD I have in my home lab is specified for AC work up to 4kHz, with 0.001% error (same as at DC) up to 400Hz.

Cheers

Alex

[Rax]

Not sure when/if I can get to this, but since I have multiple ratio transformers and KVDs, I should be able to do a comparison and see exactly what the KVDs are capable of. I need to clean up the bench, so this might be a good excuse. That GR manual I linked shows using a null meter and some parts for phase compensation, which I need to understand.

Looking forward, Conrad! And thanks!!

[Conrad Hoffman]

GR appears to use the common 1232 as a null meter. Maybe OK with the ratio trans, but very low input impedance for a resistive divider. They sold the 1232-P2 preamp, which seems to be 100 meg input and a gain of 0.7, which would improve things, but not as much as desired. Also has high shunt capacitance. It's also single ended, which makes me a bit wary. (I don't have one) Seems like an opamp buffer would be easy enough.

[Vgkid]

The 1232a preamp is just a fet input 2n3457

[Rax]

The ancient KVD I have in my home lab is specified for AC work up to 4kHz, with 0.001% error (same as at DC) up to 400Hz.

Cheers

Alex

That's a useful data point, and it kind of makes sense to me. I should probably not expect the ppm performance (in DC) my unit has (.1ppm) in AC under any scenario.

[alm]

GR appears to use the common 1232 as a null meter. Maybe OK with the ratio trans, but very low input impedance for a resistive divider.

Does the input impedance matter if you're measuring between two dividers that are matched in amplitude and phase? The DT72A manual shows a recipe where one of the divider outputs is grounded to work with grounded detectors.

A buffer amplifier could make it actually useful for real loads instead of purely potentiometric measurements, though.

[Kleinstein]

If measuring zero the input impedance of the meter would not matter, but the capacitance to ground can matter.  Chances are a meter will have some capacitance to ground too, especially if mains powered. So it would at least load down one side.
Similar just the cables will have capacitance (e.g. some 1 pF per meter) and this may need active driven shields to avoid loading. E.g. an auxiliary divider or directly drive the source via a transformer to get the output about at ground.
AC bridge circuits can be quite a bit more complicated than a simple DC circuit.

[alm]

If you ground one side of the meter (the output of one of the dividers), and adjust the other divider so amplitude and phase are matched, would capacitive loading of the meter still be a problem?

[Kleinstein]

With briging the divider output to a virtual ground one can avoid much of the capacitive loading. However this needs additional effort, usually some sort of transformer drive to than drive both ends of the dividers and maybe extra adjustment (may need capacitive and resistive trim) to get really close to zero.
This is nothing new but old standard, especially in times when semsitive AC meters often needed mains power or at least large batteries.
https://nvlpubs.nist.gov/nistpubs/jres/40/jresv40n3p245_a1b.pdf

[alm]

See figure 2.1c (page 13) of the ESI DT72A manual for an example of this.

[alm]

I just came across the "DC and Low Frequency AC Ratio Measurements" application note in the already linked https://www.ietlabs.com/pdf/application_notes/RV722_app_note.pdf (starts on page 34). It mentions using resistive dividers for DC and low frequency AC, and inductive dividers for higher frequencies. But unfortunately its not very specific about what low frequency means.

Since this app note is from the sixties, it only talks about bridge configurations and not about lower impedance DMMs as loads. Rax, do you have the 10kOhm or 100 kOhm input impedance version of the RV722? According to the manual the RV722 existed in both versions. The 10 kOhm would probably be more usable with a 1 MOhm ACV DMM as load.

[Kleinstein]

The 1 M DMM input impedance could at least be compensated to a certain degree. If the error is small, one could do a test with additional series resistance/capacitance so that the output impedance about doubles and use the 2 readings than to correct. Without any corretion 1 M load may be a problem already.

[alm]

I'm worried about the stability of this 1 MOhm impedance, so that's why in my suggestion about what setting to use on the KVD (2e-4 or lower assuming a 100 kOhm KVD), I picked a setting where the uncertainty due to load impedance was attenuated about a factor of 100 compared to the uncertainty due to KVD linearity. So the that if the load was 1 MOhm +/- 10 kOhm (1%), its contribution to uncertainty would be roughly the same as the KVDs linearity's contribution (which is attenuated by a factor 1/x where x is the setting of the KVD, like 2e-4). Then you could calibrate the setup by building a bridge from a KVD and IVD with a 1 MOhm load across the KVD (wiper to low).

But I admit that I never measured the stability of input impedance in ACV mode, and that obviously over frequency range this will become more complicated. Is the DMM more or less resistive than a 'resistive' voltage divider?

[Rax]

Rax, do you have the 10kOhm or 100 kOhm input impedance version of the RV722?

I have the 100k version.

[1audio]

FWIW I have an RV722 a 7 decade Gertsh ratiotran (and an HF ratio tran) and a Fluke 931B AC differential meter. If I can find time and bench space I'll try comparing them. (I also have the Fluke 8060A and 8922a) It will be interesting.

However chasing PPM accuracy for AC is a fools errand. Its really hard to do and the number of error sources are enormous. Even though the Gertsch is 7 digits I'm not sure what instument can reliably detect that last digits change. Just determining flatness of your source is really difficult. I have several thermal voltage converters but their accuracy vs frequency degrades a lot as frequency goes up. References to .0001% are more like wishful thinking in our world. Maybe more real at NIST with the JJ based DAC (I should offer one to the audiophile market- $500K per channel and you need a staff physicist to keep it running).
What I suggest would be to compare the KVD to the IVD using the differential meter (1 meg/8 pF ) by just comparing the output. (Its meaning of differential is different from a differential input.) Nulling AC strikes me as really questionable if you do not have a differential detector. The stray cap to everything is the leakage issues at DC amplified. The nice part of the IVD is that its output impedance is lower than the input.

[Conrad Hoffman]

So, I cleaned up the bench and did a test. Turns out I have a RV622A (100 kohm input), not a 722. Used that against a North Atlantic Industries RB-504 ratio box (ratio transformer). Various "null" detectors were tried: HP3478A DMM, Tektronix 1A7A differential plug-in and a floating GR 1232A AC bridge null meter. Signal was from a Wavetek 185 at 50-400 Hz.

The short answer is this was a dismal failure. The phase shift problem is so severe that one can only null about 3 digits, not exactly metrology lab precision. Just being read with a DMM, either box divides as expected. The easiest thing to do would be to simply calculate the error on on the RV unit due to loading and factor that in. The impedance of the transformer box is low enough that it can be ignored in many/most cases.

I'm going to try the phase compensator shown in the manuals and see how well I can do with that. There's still the matter of capacitive loading of the RV box, so one has to factor that in, depending on frequency. Next test will just use the KVD and see how bad the loading problem really is, and if it's easy to compensate.

There are some GR articles, one in the May 1968 Experimenter, talking about how they could achieve 200 ppb with the ratio transformers. I think I've got a ways to go yet.

[Alex Nikitin]

I suspect that one of the problems might be a (relatively) high harmonics content of the Wavetek 185 output.

Cheers

Alex

[Conrad Hoffman]

No question it's a rounded off triangle wave, but the THD is pretty low and the problem is so bad I don't think I'm anywhere near that limit. I'm going to do something more practical and see how big the errors are just using the KVD to calibrate a meter, since that's all the OPs got right now. It does raise the question of what to use as a generator for AC calibrations. I have a few, but all have some drawback. The 185 is extremely amplitude stable over time and frequency, but not flawlessly clean. My various GR generators will do much higher voltages, but aren't stable or clean. I have an HP with similar issues, plus a broken one! I think Thaler made an AC chip with much better performance, but I'm guessing that's been gone for a long time. I suppose I could run the sine wave from my THD analyzer through an amp, but that's more effort than I'm up to. FWIW, when I do my own meters I use an old Fluke 540B to compare with a DC reference. I don't have the calibration constant for it, but assume it's close to unity.

[Rax]

using the KVD to calibrate a meter, since that's all the OPs got right now. It does raise the question of what to use as a generator for AC calibrations.

Just to help clarify, what I'd use with this is a 10V AC standard sine from a 510A. My Audio Precision box need not apply due to dB-realm accuracy.

I've also started to look at Gertsch units. Would an RT5 serve the purpose? Which would be to scale standard AC voltages for different calibration points.

Arguably, I guess I may currently essentially be piecing together a multifunction calibrator (just doing it with 25 different components).

[Conrad Hoffman]

This is crude at best, but shows what you're up against. The signal generator was an ancient GR BFO, since that's the only thing on the bench that goes to 10 Vrms. The input voltage was monitored with an HP3455A. The target meter was a HP3478a because it was handy. The 3478A was just plugged into the ratio transformer, and then the KVD, at the same settings, noting the reading for each. Nothing was very stable, but the results are still clear.

[1audio]

The Gertsch ratio transformers are all OK if you understand their limitations. Basically 50 Hz to 10 KHz and don't overdrive them. There are versions with 4 5 6 7 and even 8 decades. They are intrinsically accurate. Calibration would be to check for failure (bad switch).

For AC calibration your 510 is a good start. It needs to be calibrated as a baseline.

Here is some info on the 510 mods I did to get lower distortion. https://www.diyaudio.com/archive/blogs/comments/comment3920.html 

Unfortunately for AC there are several gotchas- distortion affects accuracy. 1% distortion can degrade the accuracy significantly.  Frequency response is not an easy one to check. Very few oscillators are really flat. Many will have errors at higher frequencies and even cabling can be an issue. It will keep you very busy.

[Conrad Hoffman]

The 510 is interesting. Does it really manage to hold 10 ppm (typical) over a 0-10 mA load? That seems very good.

[Rax]

The 510 is interesting. Does it really manage to hold 10 ppm (typical) over a 0-10 mA load? That seems very good.

If you mean the 24hr accuracy, I think it's more like 100ppm (.01%).

[Rax]

Here is some info on the 510 mods I did to get lower distortion. https://www.diyaudio.com/archive/blogs/comments/comment3920.html 

Thanks a lot for this. One quick question - I can't find an R51 on the schematic if my life depended on it. But are you sure that's not a typo, and instead you refer to R61? I think a CCS would make sense there.