AVR-Chat

(apparently this didn't make it, the first two times I sent it, so 
I'm plugging it in again from the web interface.)


I'm not sure if this is an on-chip problem, or off chip..

I have a T26, measuring current by a differential measurement, using 
the G=20 amplifier.  

Explaining the whole thing in detail would be tedious, but I'm doing 
high side current measurement with resistors scaling down the inputs 
so that I don't rail out the ADC.  I am using an external 2.5V 
reference, and I've verified that it's in tolerance. (2.458V)

In the existing setup, I am seeing about 180 counts on the ADC, but 
running the math on the external parts says that I should be seeing 
closer to 90 counts.  IOW, the system thinks the current is about 
twice what it really is.  

This is a mature system, not a prototype, and has been in production 
for some time.. Date codes on the chips don't seem to matter (I've 
tried various ones all the way back to "ES"), and the existing 
firmware works fine in older "known good" boards. This does seem to 
be a problem on some number of boards.  Others seem ok, but I am 
waiting on confirmation of that. The ADC is not overclocked.


What could cause the ADC readings to be almost (but not exactly) 
twice what they should be, assuming that it's not a firmware change, 
and that the components pass visual inspection, plus the voltage 
measurements confirm the component values and PCB connectivity?