Is this binary Kelvin-Varley divider?

[technix]

I'll admit I am becoming a bit volt nut. The same seller sold me the used LM399 also have lots of used Vishay 20K 0.1% 2ppm precision resistors in grab bags of 20. Since someone mentioned it to me I am thinking a binary Kelvin-Varley divider. A binary Kelvin-Varley divider use three resistors and one DPDT relay per bit, and my calculation tells me that all the resistors is of the same value (and hence this 20K precision resistor grab bag can become very useful)

Is the following schematic a 5-bit binary Kelvin-Varley divider? Does this scale well up to 24 bits?

If this scales well, slap in 3 TPIC6B595 this 24-relay 71-resistor mess will become the 24-bit SPI precision DAC module for my voltage standard. (LM399 only have 6.5 digit precision, all those extra bits is for compensating and calibrating.)

Also, since the entire voltage standard draws power from an ATX12V power supply, is it a good idea to run this relay array on its own 12V rail (using the 4-pin 12V socket originally intended for CPUs,) with every TPIC6B595 chip heatsinked, in case I have to output 0xFFFFFF for a certain value?

[Kleinstein]

With 0.1% resistors, there is not much sense in going much beyond 10 Bits. It will still be stable but may not be monotonic. One might get a little better with selected resistors, to get well matching ones for the first few stages.  The divider does not scale especially well to higher resolution, as there are quite a lot of relay contacts involved. It's not that bad, but also not especially good. So I don't think it is worth going much beyond something like 16 Bits. As a rough view, I don't see much advantage over a more classical R2R DAC, except that loading to the Ref. source is lower and independent of code.

At high resolution the relays should be bistable types. So they only consume power when switched. This avoids heating of the relays and thus reduces thermal EMF to cause large errors. Normally the relay drivers don't need special cooling. 12 V or similar coil voltage is Ok. A separated supply (e.g. from rectifier on) may be helpful.

Using a ATX or similar computer supply is a bad idea, as this can produce quite some EMI problems. The ref. should not consume a lot a power, so something like a 2x12 V , 5 VA conventional transformer could be all it takes.

[technix]

With 0.1% resistors, there is not much sense in going much beyond 10 Bits. It will still be stable but may not be monotonic. One might get a little better with selected resistors, to get well matching ones for the first few stages.  The divider does not scale especially well to higher resolution, as there are quite a lot of relay contacts involved. It's not that bad, but also not especially good. So I don't think it is worth going much beyond something like 16 Bits. As a rough view, I don't see much advantage over a more classical R2R DAC, except that loading to the Ref. source is lower and independent of code.

At high resolution the relays should be bistable types. So they only consume power when switched. This avoids heating of the relays and thus reduces thermal EMF to cause large errors. Normally the relay drivers don't need special cooling. 12 V or similar coil voltage is Ok. A separated supply (e.g. from rectifier on) may be helpful.

Using a ATX or similar computer supply is a bad idea, as this can produce quite some EMI problems. The ref. should not consume a lot a power, so something like a 2x12 V , 5 VA conventional transformer could be all it takes.

So if this binary KVD is not a good idea, please suggest a good DAC with good temperature stability. Initial precision is not exactly important as it can be calibrated away if the resolution is high enough.

[Dr. Frank]

I suggest also an R2R ladder, with DPDT relays, mixed binary and  decimal scheme.
The first two MSB digits then can be made very linear by adding  trim pots for each resistor.

So you can achieve 6 very linear decimal places, i.e. 1ppm linearity, with a limited number of resistors.

Please download the manual of the FLUKE 3330B voltage / current calibrator, where this technique is used.
As far as I remember, this comes closest to your idea, using 24 identical resistors , but that has maybe to be adapted.

P.S.: Welcome in the club!

Frank

[moffy]

I was just reading the datasheet for the dual 12bit DAC, MCP4922 and they have an application that turns it into an almost 24 bit voltage reference, worth a look. Also the AD5791 is a 20 bit DAC which has an application note written: http://www.analog.com/library/analogdialogue/archives/44-04/AD5791.pdf about how to use it as a 1ppm voltage reference.

[technix]

I was just reading the datasheet for the dual 12bit DAC, MCP4922 and they have an application that turns it into an almost 24 bit voltage reference, worth a look. Also the AD5791 is a 20 bit DAC which has an application note written: http://www.analog.com/library/analogdialogue/archives/44-04/AD5791.pdf about how to use it as a 1ppm voltage reference.

I am going to investigate MCP4922. AD5791 is ridiculously expensive here from Shenzhen and I cannot afford it at all (even if I can, it is three times as expensive as LTZ1000, and I would went for that)

[Kleinstein]

The alternative of using a high resolution ADC and two lower quality DACs is possible.
However the MCP4922 is more on the very cheep side, and better 12 to 16 Bits DACs  (e.g. LTC1590, but many alternatives) should be used.
There is another thread (DIY SMU) here in the forum using two LTC2756 using this route of two DACs and a ADC.

The MCP4922 may also have to much noise - so the MCP4922 and an ADC is more a replacement for a good (but slow) 16 bit DAC, for something like a power supply.

[technix]

The alternative of using a high resolution ADC and two lower quality DACs is possible.
However the MCP4922 is more on the very cheep side, and better 12 to 16 Bits DACs  (e.g. LTC1590, but many alternatives) should be used.
There is another thread (DIY SMU) here in the forum using two LTC2756 using this route of two DACs and a ADC.

The MCP4922 may also have to much noise - so the MCP4922 and an ADC is more a replacement for a good (but slow) 16 bit DAC, for something like a power supply.

Prices of those parts get ridiculously expensive real fast. I am half way there deciding between forking $50 on an DAC and using a whole bunch of matched resistors and a few 74HC595s to roll my own R-2R DAC. Wire-wound resistors using the same material will have very small relative tempco and the principle of operation of R-2R only requires a stable relative resistance.

[Marco]

Prices of those parts get ridiculously expensive real fast. I am half way there deciding between forking $50 on an DAC and using a whole bunch of matched resistors and a few 74HC595s to roll my own R-2R DAC. Wire-wound resistors using the same material will have very small relative tempco and the principle of operation of R-2R only requires a stable relative resistance.

Unless you can software calibrate it stability doesn't get you much ... out of the box you're still stuck with very poor INL and non-monotonicity.

[technix]

Prices of those parts get ridiculously expensive real fast. I am half way there deciding between forking $50 on an DAC and using a whole bunch of matched resistors and a few 74HC595s to roll my own R-2R DAC. Wire-wound resistors using the same material will have very small relative tempco and the principle of operation of R-2R only requires a stable relative resistance.

Unless you can software calibrate it stability doesn't get you much ... out of the box you're still stuck with very poor INL and non-monotonicity.

There is no calibration data that can fill up one 24C512 or two (and I happen to have stocked a few) so digital calibrating is non-issue.