Poor Man's KVD - What is the accuracy / resolution achievable?

[zhtoor]

Hello,

here is my (naive) approach to a poor man's kvd:-

instead of making 6 decades let us make 3 "centades".

first centade is YR-NEOHM 10K 15ppm/degC, 0.1% resistors (unselected), 101 in series, (positions marked 00 thru 99)
second centade is YR-NEOHM 1K 15ppm/degC 0.1% resistors (unselected), 101 in series, (positions marked 00 thru 99)
third centade is YR-NEOHM 100R 15ppm/degC 0.1% resistors (unselected), 101 in series, (positions marked 00 thru 99)

second centade is paded down to 20K (from 101K) and the third centade is padded down to 2K (from 10.1K)
the kvd input impedance is 1.01M (can be padded down to 1M).

the division ratio will be selected by selecting 2-digits at a time. (using jumpers like Conrad Hoffman's design)

what would be the performance of this contraption?
regards and comments required from the learned members of this forum.

[IconicPCB]

An interesting thought experiment...otherwise not of much use.

Consider the need for 101 well matched tracking resistors versus only 11 of similarly performing resistors for the first decade second and may be third decade.

Consider a binary solution for a minimum resistor count ( say 24 bit R -2R ladder )

[MosherIV]

instead of making 6 decades let us make 3 "centades".

Errrrrmmmmmm. You do not have a Kelvin Varley Divider!

The cleverness of the KVD is in
A) the dual connection to the next decade (which can be done using a multilayer rotary switch)
B) the connection to the next decade perfectly creates the correct resistance ratios to divide down the voltage.

If you deviate away from the decade arrangement - you no longer have a KVD!
Not sure how you would make the switch with the dual connection.
The kvd needs 11 switch positons, you would need 111 per centade.

You can change the values of the resistors but the ratios must stay the same.
Then there is the thermal matching/stability of the resistors (so they stay in ratio)

[zhtoor]

An interesting thought experiment...otherwise not of much use.

Consider the need for 101 well matched tracking resistors versus only 11 of similarly performing resistors for the first decade second and may be third decade.

Consider a binary solution for a minimum resistor count ( say 24 bit R -2R ladder )

thanks. but the question was: "what resolution / accuracy is achievable" with this?

regards.

[MosherIV]

but the question was: "what resolution / accuracy is achievable" with this?

Resolution - assuming you can get it to work, it is not giving you any more resolution in the lowest units which is what you are trying to achieve

Accuracy - as IconicPCB has said - it will be worse because it will be harder to thermally match so many resistors.

Conrad Hoffman has told you that you can make a KVD with ordinary resistors, just do not expect good performance from it. There is on old thread in this forum with links to Conrad's magazine articles about building a home metrology lab. There is a pretty good write up on how to make a KVD from ordinary resistors.

[zhtoor]

but the question was: "what resolution / accuracy is achievable" with this?

Resolution - assuming you can get it to work, it is not giving you any more resolution in the lowest units which is what you are trying to achieve

Accuracy - as IconicPCB has said - it will be worse because it will be harder to thermally match so many resistors.

Conrad Hoffman has told you that you can make a KVD with ordinary resistors, just do not expect good performance from it. There is on old thread in this forum with links to Conrad's magazine articles about building a home metrology lab. There is a pretty good write up on how to make a KVD from ordinary resistors.

thanks for the input.

now let me ask this question in another way:-

assume a Rich Man's (KV)D built with a string of 1000001 1K 0.1% resistors (unselected),
15ppm/degC again, positions marked 000000 thru 999999.

now what accuracy and resolution is achievable with this divider?

regards.

[Conrad Hoffman]

Makes my brain hurt- do a spreadsheet. On the centade arrangement, I think it's no KVD, but it might have some merit. Download the manual for the General Radio 1608 CGRL (LCR) bridge and study the schematic. They got something like 4 digits using a centade arrangement with good accuracy, and they never selected the resistors like we might do. It saved them a lot of work, but they also had the ability to do the mechanical switch based on a heavy patterned PCB. Building that same switch today would probably cost a fortune.

[zhtoor]

Download the manual for the General Radio 1608 CGRL (LCR) bridge and study the schematic. They got something like 4 digits using a centade arrangement with good accuracy, and they never selected the resistors like we might do.

thank you sir!

what a piece of work GR1608 was!, brilliant and i wonder what other jewels history has in store!.

regards and stay well.

[Squantor]

Sorry for hijacking the thread but what about a KVD using the vishay NOMCA 8 resistor arrays and using analog switches to switch those. You get a digitally controlled KVD but not with convienient decimal steps but would that a big problem? Or instead of analog switches use relays? Maybe I should make a different topic on this idea?

[Gyro]

Just curious zhtoor, what is your intended use for the KVD? Instrument calibration? Would you be better with a multi-stage Hamon divider? You would only get a fixed divide by 10 on each stage but it is inherently self calibrating. Again Conrad is your man.

[IconicPCB]

NOMCA8 is an interesting package. While the resistance value is specified at 0.1% the actual resistor match in the package is specified at 0.015%
Quite usable so long as it is understood the scaling would be in one sevenths of a decade and each stage would need to be resistance adjusted to suit the requirement.The spread of values in the product is reasonable too.

[Robert763]

Just curious zhtoor, what is your intended use for the KVD? Instrument calibration? Would you be better with a multi-stage Hamon divider? You would only get a fixed divide by 10 on each stage but it is inherently self calibrating. Again Conrad is your man.

If going solidstate on the switching a PWM set-up is the best way to go. This can be followed by fixed (Hamon) dividers if you need very low levels.

[The Soulman]

PWM, in theory, can be very precise but pitfalls could be rise and fall times and output filtering.
A higher frequency makes rise and fall time errors more significant.
A lower frequency makes filtering more problematic.
A low duty cycle as ,1  percent (thousand to one ratio) can worsen things all together and I imagine it will
be hard to do that with some good accuracy.

larger ratios are not a huge problem with Kelvin-Varley or Hamon dividers.

[Alex Nikitin]

PWM, in theory, can be very precise but pitfalls could be rise and fall times and output filtering.
A higher frequency makes rise and fall time errors more significant.
A lower frequency makes filtering more problematic.
A low duty cycle as ,1  percent (thousand to one ratio) can worsen things all together and I imagine it will
be hard to do that with some good accuracy.

larger ratios are not a huge problem with Kelvin-Varley or Hamon dividers.

Hmm, have a look at the PWM circuit from the Keithley 263. Very simple, built with 74HC and 4000 series chips, all components are available still from Farnell/Digikey, etc. The stability is very good and the tempco around room temperatures is only 1-2ppm/C (from full scale) including the reference and scaling amp. The resolution is limited to 100ppm (the LSB in theory is 50ppm, however the scaling of smaller steps is uneven by design) but the accuracy is very good.

Cheers

Alex

[Kleinstein]

In many respects PWM can be easier than a KVD to get a high resolution DAC. There are a few difficulties, but parts can be relatively simple. Creating the digital PWM signal is rather difficult today. The more difficult parts are filtering (may need good capacitors) and highly linear switching.

The circuit of the Fluke 5700 could give some inspiration, especially if simplified (e.g. leave out the time 2 gain option, maybe use CMOS switching chips).

[Echo88]

Or just use a AD5791 + LTZ1000. But i dont know if it still fits the poor man budget.

[zhtoor]

i would prefer an old-school setup with a high-stability source (LTZ1K, battery etc.), a KVD and a battery operated null-detector
in a room maybe with a strip chart recorder and no other electronics to mess up my readings.

regards.

[jeroen79]

If you deviate away from the decade arrangement - you no longer have a KVD!

You can use base 10, binary or any other arrangement for a Kelvin Varley divider.
base 10 decades are just easy to use for humans as we use a base 10 system for most things.
All that is needed for an n step stage is n+1 resistors R and the next stage being 2R in total.

[MosherIV]

You can use base 10, binary or any other arrangement for a Kelvin Varley divider.
base 10 decades are just easy to use for humans as we use a base 10 system for most things.
All that is needed for an n step stage is n+1 resistors R and the next stage being 2R in total.

Agreed. I understand that, you understand that but from what the OP was describing I do not think they understand it.

They were under the missconception that creating a 100 division  based divider and reducing the number of stages would some how maintain the precission (no of decimal places).

[zhtoor]

You can use base 10, binary or any other arrangement for a Kelvin Varley divider.
base 10 decades are just easy to use for humans as we use a base 10 system for most things.
All that is needed for an n step stage is n+1 resistors R and the next stage being 2R in total.

Agreed. I understand that, you understand that but from what the OP was describing I do not think they understand it.

They were under the missconception that creating a 100 division  based divider and reducing the number of stages would some how maintain the precission (no of decimal places).

the real question was: "what is the accuracy and resolution *ACHIEVABLE* ?" using the mentioned tolerances of the resistors.

kelvin varley is just a technique which can be applied from base 2 (3 resistors) to upwards say base 100 (101 resistors).

regards.

[Conrad Hoffman]

Another useful thing to look at is the schematic for the Analogic/Data Precision 8200 current/voltage calibrator. It uses an LM399 for the reference and an interesting hex or octal summed divider scheme.