In looking at the coefficients, it appears that, not only will some values cast to zero, many of the others will be severely limited in resolution. Do you know if there was a reason why the code not written to scale the coefficients with a larger number than 32768 and then cast into Longs?