Kelvin Varley divider base 2

[IconicPCB]

A Kelvin Varley divider scaled down to a base 2 format is similar to R-2R ladder network in terems of number of switches and identical resistor components.

Have there been any papers /publications presenting analytic sensitivity comparisons between the two topologies?

How would the division ratios of the two topologies compare with respect to a deltaR in identical ( bit )  positions in the two topologies?

Which then brings me to the question of practicabilities of either of the two topologies?

OK I know google is my friend but... please humor me.

EDIT: statistical arrays anyone?

[Kleinstein]

I have a vague memory of an article about a base 2 KV type divider, but can't find a reference.

The number of switches is the same as the R2R chain, but one need DPDT instead of SPDT switches. So the R2R chain is simpler in this respect. I am also afraid the switch resistance is more important (alone from using 2 as many contacts).
 

[IconicPCB]

Yes, The point about DPDT SPDT switch is a valid one.

Although the question of sensitivity of division ratio versus resistance increment is still there

[kleiner Rainer]

Cutkosky:

https://xdevs.com/doc/_Metrology/A_25-Bit_Reference_Resistive_Voltage_Divider.pdf

Greetings,

Rainer

[EmmanuelFaure]

And :
https://www.eevblog.com/forum/metrology/precision-binary-(r2r)-divider/

[dietert1]

A modern solution should include one of those high precision DACs, like the MAX5719.

Add some more bits on the MSB end, for example by making a quad isolated DAC similar to the IOTech DAC488/Keithley 213 and chain those DACs to one. That DAC for example has a basic resolution of 13 Bits. They used 12 Bit DAC parts and added the sign bit using a switched inverter. It generates +/- 10 V, so one step is 2.5 mV. I found that those steps are accurate to +/- 100 uV (INL) and stable to +/- 50 uV, although the DACs use very simple voltage references and outdated LF422 OpAmps. The 8 bit calibration DACs also included in that instrument can be used to interpolate. So in the end you can make it a +/- 40 V DAC with +/- 50 uV precision, all this based on 12 Bit DAC parts.

Regards, Dieter

[razvan784]

The MAX5719 has a maximum INL of 20 ppm. You may be thinking about the AD5719, which has a 1 ppm INL.
Regarding your 40V / 50uV example, to achieve that with 12-bit DACs is certainly impressive, and I'm sure that better performance can be achieved. However, such performance is quite hard to verify at the level discussed here. For example, the 25-bit Reference Divider article discusses a divider that is self-calibrating and accurate to 0.1 ppm. It has dedicated switches tied into the resistor network that are needed for self-calibration - a feature that is absent in modern integrated DACs. But maybe a different self-cal scheme can be devised - any ideas?

[dietert1]

I would focus onto the five most leading bits, the rest of it could be an integrated circuit. First one would try to make 1 bit that is good for 0.5 ppm, including some self-checking scheme,
Think about a sign bit like in the IOTECH DAC488 example. One could use a LT5400 voltage divider to make the inverter. Those have a tracking error of < 0.2 ppm typical. Below that the air gets thin and one needs to find "golden" parts from larger lots. Recently i found a DALE RS-2B that has it's TC curve maximum (TC = 0) at 25 °C. But that's a rare find. Their specification is +/- 20 ppm/K.

Regards, Dieter