the input channel select control A0 can not work properly, that is when A0=0,
it will sample INA2 and INB2, not INA1 and INB1. Is it possible? Another
problem is that using the A channel and B channel to sample the same voltage
(2.5V reference), and the results are not the same, for example: Ax channel
---> Result: 7FD6, Bx channel: --->Result: 807C.
The A0 will sample INx1 when LOW and INx2 when HIGH. They must have a mistake
on their logic controlling this as we have looked into this in the past.
perhaps they have an older datasheet? This was an error on the last revision.
There is a difference in the offset and gain errors from CHA and CHB. This is
probably what they are observing. The offset error for CHA is in opposite
direction of offset error for CHB (one is positive and one is negative).
I do notice the 10uF reference decoupling capacitor which is probably not be
enough. This should be increased to 22uF or even better 47uF and placed very
close to the ADC, preferably right at the REF pin.
On the reference itself, the AD1582 is not able to drive the dynamics of the
reference. It will require a suitable buffer such as the AD8031 or AD8606
otherwise there will be problems. Or they can just use a good reference such as
AD780, ADR421, ADR431 or ADR441.