I'm thinking there are basically two ways to obtain good accuracy in something like a multimeter: use accurate components and such, or use less accurate components and then calibrate the meter. I intend to use a hybrid approach: a couple of accurate components, to allow me to use the meter to calibrate itself – not by comparing to itself (which obviously doesn't work), but by measuring the less-accurate components.

Obviously the values of the less-accurate components won't change. But the scale factor they produce, whether as part of the voltage divider for voltage measurement, or the current shunts for current measurement, will – and that can be accounted for in software. No, I'm not handing a hardware problem over to software; I'm saying the software would have to handle it anyway, so the compiler may as well compute the constants from actual component values instead of me calculating the constants from the selected component values.

First, an accurate voltage divider will be very important, and is easy to make. Just measure the voltage across each resistor, and you can easily calculate the exact division ratio – the voltage measurement doesn't need to be accurate either (the ADC reading is good enough), but resolution helps. Make sure the resistors are of similar values to minimise ADC gain error.

Most of all we need accurate voltage measurement. This requires an accurate voltage reference, which I will buy. The complication is that the ADC will have gain error, which must be calibrated out. I should be able to do that by using an accurate voltage divider to measure an accurate fraction of the reference, which gives me both what the ADC actually reports, and what it should report. I can't just measure the reference directly, because the particular ADC I'm using usually reads high.

I also need to be able to measure resistances accurately, primarily for the current shunts and also for actual use – and it is probably the easiest way of getting accurate voltage dividers too. The best way to do this will be to buy one accurate resistor and use it in an accurate voltage divider – we can then easily calculate the accurate value of other resisters.

I will still need low temperature coefficient resistors, but I won't need tight tolerances for most of them.

So one accurate voltage reference and one accurate resistor (I'm looking at parts with 0.01% tolerances and temperature coefficients of about 10 ppm/K – after all, I'm going to end out with an accuracy lower than whatever theirs is, hence my aim of 0.1% or better overall), and I should have good accuracy on every range of every measurement type.