thought experiment - self-controlled voltage reference

[RandallMcRee]

Found that I have been working on this idea independently and have started another thread on simulation of voltage references. I simulated the case of four references and measuring the differences.

Using least squares equation solver and adding one or two sum constraints to the system of equations allows the system to be solved on each iteration and actual voltages to be precisely tracked (e.g within noise and measurement uncertainties).

Java simulation here:
https://github.com/RMcRee/ModelVoltageReference

Nor is my least squares idea as original as I had imagined. Here is a 1974 NIST paper talking about the same thing with gage blocks:
http://emtoolbox.nist.gov/Publications/NBSIR74-587.asp

[rs20]

What do you mean by "is viable", "can be precisely tracked", etc? Where are your numbers, comparing the performance of your algorithm to, e.g.:

a) A single uncorrected reference
b) Simply averaging the four references (i.e., a sqrt(4) = 2 attenuation)

[RandallMcRee]

Good questions....
I am working on updating the simulation with answers. It may be a few days as life and work intervene.

[RandallMcRee]

I updated github with my latest simulations.
It would seem that only if the range of drift and temperature is restricted can the method be useful.

Long-term drift (I believe in the range of real voltage references) cause the LSQ solution to wander away from the actual values.

So it is of limited utility.

I certainly learned a lot. If for example, you have four accurate initial measurements and one reference (only) drifts it is possible to use differential measurements and LSQ to quantify that. But it is not realistic to have only one reference at a time drift.

 It is also easy to embed least squares solutions in a microcontroller see, for example, file QRDecomp44VRefs.