Hey, I wanted to post a correction to this for anyone who might come across this. I had previously read a discussion from Tom Rondeau and some other GNU gods discussing the sign of the feedback taps needing to be negated; when working with the test bench data, that is still the case.
Normally the gain reported from the feedback filters is subtracted, as in the equation that Mark posted above. However, when looking at the actual function in iir_filter.cc it turns out that the feedback coefficients are added.
acc = d_fftaps * static_cast<gr_complexd>(input);
for(i = 1; i < n; i ++)
acc += (d_fftaps[i] * static_cast<gr_complexd>(d_prev_input[latest_n + i]));
for(i = 1; i < m; i ++)
acc += (d_fbtaps[i] * static_cast<gr_complexd>(d_prev_output[latest_m + i]));
This means you can either do one of two things when dealing with the feedback. You can either negate the value of the reported coefficient when using it in the equation above, or change the operation of the equation. Get your taps from fm_emph.py as above, and utilize them in the following equation:
y(n) = b0*x(n) + b1*x(n-1) + a1*y(n-1)
When doing this, the calculated values match what we printed to a file while under operation.