Hi Miguel,
After a bit of back-and-forth between Sigma Studio and Matlab, I've confirmed the coefficient mapping that I hypothesized above. (Though one needs to invert the signs of the 'An' coefficients when moving between the two environments).
Knowing that, I was able to compare my Matlab-generated filter to the Sigma Studio generated filter, and confirm that YES, they produced different coefficients, and therefore, different responses.
It came down to coefficient quantization. Sigma Studio's fixed point representation isn't precise enough for values in the range of 1E-7, so some were set to 0. I should have know better...
In addition to facilitating coefficient export from the 'General Filter' cell in a future release, perhaps it would be good to issue a warning when compiling if any filter coefficients are on the order of a single LSB, in order to save silly engineers like me from themselves...
Thanks,
P.