I am not sure how an FIR filter will increase your measurement accuracy. I thought that the standard way to increase measurement precision for scales is to average (like you are doing). The error decreases with the square root of averaged samples. So to add 4 bits of accuracy to your measurement, you average 256 samples (which takes 10 sec!). Your integrating ADC naturally averages as well. So there is a whole lot of averaging going on.

What microcontroller/FPGA/whatever is measuring time on the A/D? Lets say you've got a high-speed Cygnal that runs at 50MHz. One sampling period lasts 50MHz*40ms=2,000,000 cycles. That is 21-bits right there. And that's without de-integration. Are you using a higher resolution time measurement?

Sounds interesting.

- Lee