Precision binary (R2R) divider

[VintageNut]

Reading the datasheet for the Measurements International model 8000B binary divider has me wondering if a DIY implementation could be devised.
The MI8000 datasheet states 0.01 ppm resolution and 0.05 ppm accuracy.

The attraction is that the system is self-calibrating and has better resolution than a Fluke 720A KVD.

My assumption is that MI characterizes every stage of the R2R ladder individually and saves each stage's error to use in the calculation of total deviation.

A 30-stage R2R ladder would give a resolution of 10^-9.

The DIY implementation that I am thinking of is to use a DMM with switch cards to perform the programming of each stage and to perform bypassing for measuring each stage.

The downside of this architecture is that the entire input voltage will be across one stage limiting the usefulness of the architecture to low voltages. It would seem impractical to make this exact architecture able to accept 1000V input. The MI system has a high voltage extender that changes the sub-ppm goodness of the base divider to 2ppm for the high voltage enabled two-box system.

Has anyone found a discussion of a practical implementation of a DIY version of a sub-ppm accurate R2R divider ?

[zlymex]

There was an IEEE article by TSAO giving more detailed and more accurate block diagram than 8000B DS.

[VintageNut]

Thank you Zlymex. That answers some questions.

Do you have the article?

[zlymex]

Thank you Zlymex. That answers some questions.

Do you have the article?

Here it is.

Add circuit: In case anyone is interested in the Cutkosky divider on which 8000B is based.
All resistors are equal in value, R1b and R1c can be omitted.

[Kleinstein]

The  Cutkosky divider is more similar to a binary version of the KV divider than to an R2R DAC. A slight advantage might be the use of polarity changing switches and the possibility to use resistors with R and 2R value instead of separate resistors like shown. Though at the MSB one might still use 3 identical resistors instead.

The advantage is that the total resistance is constant. The advantage of a R2R chain would be having half the number of switches (contacts), but at the price of a non constant and possibly low load of the total divider.

The interesting part is still how to do automatic adjustments / measurement of the correction values. 

[VintageNut]

Hello Zlymex

The article is very helpful. Do you have the other articles referenced? Some more pieces of the puzzle will help.

6. Self-checking resistive ratios, A.M. Thompson

5. On the output resistance of self-checking voltage dividers S. H. Tsao

2. A sub-ppm automated 1-10 volt dc measuring system B.F. Field

1. A reference 25-bit resistive voltage divider S.H. Tsao

[JS]

That unit is huge, must be a lot of goodness inside!

Dealing with the few MSBs is quite a thing to archive sub ppm performance, I don't know how the trimming is done there, you do need some serious way of doing so. It states to be self checking, as long as you can measure the deviation you have enough precision to correct the INL. For checking it I don't know, swapping the polarity and comparing with it self (some sample and hold or ratio reference) would be a first degree of self checking.

JS

[zlymex]

Hello Zlymex

The article is very helpful. Do you have the other articles referenced? Some more pieces of the puzzle will help.

6. Self-checking resistive ratios, A.M. Thompson

5. On the output resistance of self-checking voltage dividers S. H. Tsao

2. A sub-ppm automated 1-10 volt dc measuring system B.F. Field

1. A reference 25-bit resistive voltage divider S.H. Tsao

I have them all, and I'll attach them here.
It may takes a little while to upload because some file is large.

[zlymex]

Have to start new posts to get around that 2M per post limit

[zlymex]

More.....

[zlymex]

And the last.
You may have to rename these five .zip files to .rar before decompress. I use WinRar but the forum does not accept .rar extension so I have to rename them before upload. I can decompress those .zip files direct by WinRar though. I tried to compress to zip file, but have to split as well due to the forum limit, but .zip file split to z01, z02,,, which not allowed.

[VintageNut]

Thank you very much, Zlymex. This will keep me busy trying to figure out what the original inventors did.

[zlymex]

.......... and the possibility to use resistors with R and 2R value instead of separate resistors like shown. Though at the MSB one might still use 3 identical resistors instead.
.........

Another meaning to use identical resistor is the self-calibration where voltages is referred at the middle point of 2R.

I guess 8000B must use mechanical relays for the switches to minimize  the contact resistance at least for MSBs.

[VintageNut]

.......... and the possibility to use resistors with R and 2R value instead of separate resistors like shown. Though at the MSB one might still use 3 identical resistors instead.
.........

Another meaning to use identical resistor is the self-calibration where voltages is referred at the middle point of 2R.

I guess 8000B must use mechanical relays for the switches to minimize  the contact resistance at least for MSBs.

Is there an article for using the middle of the 2R element as the reference point?

[zlymex]

.......... and the possibility to use resistors with R and 2R value instead of separate resistors like shown. Though at the MSB one might still use 3 identical resistors instead.
.........

Another meaning to use identical resistor is the self-calibration where voltages is referred at the middle point of 2R.

I guess 8000B must use mechanical relays for the switches to minimize  the contact resistance at least for MSBs.

Is there an article for using the middle of the 2R element as the reference point?

See attachment.

[VintageNut]

Once again thank you very much. That is quite a lot of info and will keep me busy studying and deciding what to try first.

It is interesting that the 1984 article by Bruce Field uses a common 10:1 divider. The method of calibrating each step of the divider is very clever.

 I think that using an ordinary DMM with a switch card could replicate this calibration of the divider.

You could automatically calibrate the 10:1 string, then take the measurement(s) that you want to compare to the divider and then re-run the calibration to see if the divider drifted.

[TiN]

Thanks zLymex!

Rehosted PDFs without stupid size limits.

A_25-Bit_Reference_Resistive_Voltage_Divider.pdf

Microcomputer-Based_Calibration_Systems_for_Variable_Voltage_Dividers.pdf

[Cerebus]

And the last.
You may have to rename these five .zip files to .rar before decompress. I use WinRar but the forum does not accept .rar extension so I have to rename them before upload. I can decompress those .zip files direct by WinRar though. I tried to compress to zip file, but have to split as well due to the forum limit, but .zip file split to z01, z02,,, which not allowed.

Thanks mate. That little pile has sorted out my bedtime reading for tonight.

[Echo88]

Has anyone ever actually built a KVD based on Conrad Hoffmans suggestion to use selected resistors out of an normal batch? Im interested in the achievable performance: linearity/short term stability and so on, of course no one can predict what the long term drift of those resistors will be.

Also: The "A 25-Bit Reference Resistive Voltage Divider" document is very interesting, since they use selected +-20ppm and normal relays, while achieving excellent results.
Are there more documents available on cutkosky dividers? Especially error comparisons between a normal automated KVD and cutkosky-divider considering switch resistance would be nice.

[VintageNut]

Has anyone ever actually built a KVD based on Conrad Hoffmans suggestion to use selected resistors out of an normal batch? Im interested in the achievable performance: linearity/short term stability and so on, of course no one can predict what the long term drift of those resistors will be.

Also: The "A 25-Bit Reference Resistive Voltage Divider" document is very interesting, since they use selected +-20ppm and normal relays, while achieving excellent results.
Are there more documents available on cutkosky dividers? Especially error comparisons between a normal automated KVD and cutkosky-divider considering switch resistance would be nice.

...normal automated KVD...   

Do you have an example of such an animal?

[Echo88]

@VintageNut: Instead of a rotating switch a lot of bistable relays are used, while maintaing the original KVD-concept with 11 resistors in each decade.

[VintageNut]

@VintageNut: Instead of a rotating switch a lot of bistable relays are used, while maintaing the original KVD-concept with 11 resistors in each decade.

What I am asking for is a link or documentation of a working example.

[Echo88]

Ive never seen an automated KVD, it was just an idea of mine and it certainly is possible to build it this way.

[VintageNut]

Ive never seen an automated KVD, it was just an idea of mine and it certainly is possible to build it this way.

Your use of the word "normal" seemed to imply that it was something that existed.

The problems with contact resistance of relays will make automation of a KVD a very difficult engineering task.

I suspect that Measurements International goes to heroic lengths to make characterize the contact resistance of the relays.

There is a picture of a method that Fluke uses to rotate the knobs of a KVD using stepper motors. This tells me that it is more cost effective to build a robot to turn the knobs of a KVD than it is to invent a way to switch relays inside a KVD.

[Echo88]

Yeah, but were talking about self built dividers, not modifying existing KVDs. Of course it is easier and more suitable to use steppers to adjust the rotating switches on existing KVDs to automate them. I would love to have an Fluke 720. 

So either one gets those super low emf/resistance rotating switches (where and for how much money?) or designs something new based on COTO-Relays (guaranteed low emf) or those proprietary Guildline-relays: latching relays which press a goldplated copperfoil on gold plated copper traces on the PCB, which results in <50nV emf error. Either way, one can parallel many less costly relays to achieve lower switching resistance, like it is done in the pdf of the cutkosky-divider.

I suspect MI uses the same relay-concept like Guildline uses. Either way, im still very sceptical about selecting standard 1% 50ppm resistors to use for a divider, since no one can guarantee what the long time drift will be.   

Edit: Source for the Guildline-relay-description http://www.dataproof.com/Guarded%20Scanner.PDF Page 2

Edit2: https://xdevs.com/review/dp160_scanner/ Pictures of the mentioned relays.

[VintageNut]

Yeah, but were talking about self built dividers, not modifying existing KVDs. Of course it is easier and more suitable to use steppers to adjust the rotating switches on existing KVDs to automate them. I would love to have an Fluke 720. 

So either one gets those super low emf/resistance rotating switches (where and for how much money?) or designs something new based on COTO-Relays (guaranteed low emf) or those proprietary Guildline-relays: latching relays which press a goldplated copperfoil on gold plated copper traces on the PCB, which results in <50nV emf error. Either way, one can parallel many less costly relays to achieve lower switching resistance, like it is done in the pdf of the cutkosky-divider.

I suspect MI uses the same relay-concept like Guildline uses. Either way, im still very sceptical about selecting standard 1% 50ppm resistors to use for a divider, since no one can guarantee what the long time drift will be.   

Edit: Source for the Guildline-relay-description http://www.dataproof.com/Guarded%20Scanner.PDF Page 2

If you really want a KVD, buy one used. I purchased two of them at different times from FleaBay. Both operate well enough to do some volt nutting.

You have to be patient and wait for one that you can afford.

There was a 722 for $300-ish recently. That is a fair price.

[timb]

... or those proprietary Guildline-relays: latching relays which press a goldplated copperfoil on gold plated copper traces on the PCB, which results in <50nV emf error ...

This sounds a lot like the way Tek made switches in their old gear. It was used extensively in their plugins, both on scopes (like the 7000 series) and general purpose T&M gear (TM 500 series).

Basically, they had these little metal leaf springs they would rivet into the PCB. Connected to the other end of the leaf (by a piece of insulating Teflon) was a gold plated copper "fork" contact. It was U-shaped, so when the leaf spring was pressed down, the U would bridge two adjacent pads on the PCB, completing the circuit.

So they would have as many of these contacts as needed, side by side in an array. Over top of this array they affixed a plastic cylinder that had ridges placed wherever a particular contact was supposed to be closed in a particular position. (The ridges would press down on the leaf spring.) So, as the cylinder rotated, the entire array of contacts would open or close as needed for that particular switch position; essentially, it worked just like a music box!

The U contact kept firm pressure with the PCB, so it was very low resistance. It was also small, so it had very low capacitance and inductance (hence they could run high frequency signal over them as well).

I know in several instances it was used for selecting divider taps in some of their calibrators. They eventually developed an electronically controlled version of the technology (they used some sort of micro miniature solenoid) for their GPIB controlled TM 5000 "all in one" scope calibration plugin.

Here's a view of a board with the cylinders in place:



Here's another board showing the contacts:

[C]

Guess I am missing something here.

A cal lab visit is comparing your's to their's.
If you build two or have two then what is problem using one to check or calibrate the two.
Old timers just used a good meter and put a mark on dial for a step.
Between the two you only have three readings, Zero, minus step & plus step. Just need repeatable readings for a short time from meter.

If you want to use switches instead of relays, you might think of using an RC servo to move switch position. Check total system for what is cheapest option with needed control.

[Echo88]

@ VintageNut: 300€ for a KVD is acceptable by me, but what really interests me is to check which specifications one could achieve with available components (no selected super expensive hermetic resistors) and a careful setup (regarding current leakage, emf, thermostatic....). At the moment the whole concept is just an idea of mine.

@timb: Very interesting, thanks! I think your post is missing the second picture or my browser doesnt show it. Do you have more information on those microminiatur solenoids or similar relays?

@C: The point of my first post in this thread, was to ask if anyone has ever built a KVD like the one Conrad Hoffman suggested with simple components and what errors/drift were observed. Calibrating the KVD only makes sense, if the achieved drift is acceptable. Since i have no 3458A, i dont have the capability to regularly check the linearity of a self built KVD and would have to rely on expensive calibrations at calibration labs.

[C]

Calibrating the KVD only makes sense, if the achieved drift is acceptable. Since i have no 3458A, i dont have the capability to regularly check the linearity of a self built KVD and would have to rely on expensive calibrations at calibration labs.

No, just the second self built KVD.
The quick test is just run both KVD's  through all steps with Zero read meter. Meter does not need to be calibrated, just repeatable.
Might be good to read up on how old timers did some things.

[babysitter]

While its going off-topic to discuss switches for KVD here...

... as someone who spent a reasonable part of his childhood in a phone exchange office playing between racks full of them, I wonder about using one of these stepped or motorized rotary switches, sucessors of strowgers nice idea.

I was unable to get the real stuff yet but found a cheap eastern european one which looks like a clone.

I should really go and measure it.

[zlymex]

I don't think COTO relays have very good low emf property if they are read based, because Kovar must used which has very large emf. Also, latching reeds are rare and the heat generated by the coil make the emf worse.

In order to automate a good KVD, the contact resistance must be very low especially for the first decade thus requiring a lot of relays in parallel.

[Echo88]

COTO and Pickering (dont really know where to buy pickering-relays, so i dont consider them at all) are the only real relay-manufacturers i know of, which specify such low emf-error for their relays and are used in high end equipment, while your statement that reed-relays arent that suitable to low-emf/resistance in general is true. Guess the COTO-relays achieve their excellent emf-error by the special construction: http://patents.justia.com/patent/4084142

The fact that one needs to use many parallel relays, to reduce the switching resistance to acceptable levels in KVDs is the main problem, since it gets costly so fast.

Maybe the relays can be selfbuilt like in the DataProof 160 based on bistable latching solenoids (havent found any cheaply) or like timb mentioned stepper motors with little excenters, that press down a u-shaped copper contact spring. But of course thats only suitable once they cost way less than other relay-solutions.

[Cerebus]

...and Pickering (dont really know where to buy pickering-relays, so i dont consider them at all) ...

Pickering seem to be another one of those companies that seem to make it difficult to acquire their products. A shame as their kit looks really good - as far as I can tell - but a datasheet is no substitute for a few samples in the hand.

[guenthert]

[..]
The fact that one needs to use many parallel relays, to reduce the switching resistance to acceptable levels in KVDs is the main problem, since it gets costly so fast.
[..]

Isn't that the reason, not to use the KVD in the first place, but the topic's binary divider?  If every resistor in that schematic is of the same (nominal) value R, it can be large, can't it?  If it is, say, 1MOhm, then the contact resistance of the relays (always in series with at least one R) is insignificant (as long as one doesn't strive for accuracy levels better than 0.1ppm), no?

[VintageNut]

[..]
The fact that one needs to use many parallel relays, to reduce the switching resistance to acceptable levels in KVDs is the main problem, since it gets costly so fast.
[..]

Isn't that the reason, not to use the KVD in the first place, but the topic's binary divider?  If every resistor in that schematic is of the same (nominal) value R, it can be large, can't it?  If it is, say, 1MOhm, then the contact resistance of the relays (always in series with at least one R) is insignificant (as long as one doesn't strive for accuracy levels better than 0.1ppm), no?

That is a very interesting idea. Relay contact resistance would be no big deal.

On the other hand, what about leakage and guarding? The best old-school high resistance bridges required guarding.

[guenthert]

On the other hand, what about leakage and guarding? The best old-school high resistance bridges required guarding.

Well, the problem is mentioned in http://home.mit.bme.hu/~kollar/IMEKO-procfiles-for-web/congresses/WC-16th-Wien-2000/Papers/Topic%2004/Vargha.PDF but not further investigated. 

[Kleinstein]

One downside of the binary version of the KVD is that the output resistance is quite high - so one might use 1 M resistors, but this could lead to an output impedance in the 10 M range, which is not all practical. It is more like problem with the binary version that one has not that much choice in resistance. The advantage of the binary version is more like all the same resistors, which can simplify adjustment and control circuit. Also the number of relays would be a little smaller - but more active contacts !

The normal decimal KVD can use the 1:5 stepped resistors for the first two decades and than extra shunt resistors and thus the same resistor values for all lower decades. The shunting resistors also reduce the influence of the contact resistance in the lower decades.