This patch changes code to allocate filter poles and zeros to biquads
to allow seeing poor SNR due to strong scaling vs. available fixed point
filter word length. The issue was severe especially with Matlab version
of tf2sos() function. The replacement is straighforward z, p frequency
sort. The scaling of biquads was changed to max -6 dB instead of previous
-3 dB to be more safe against internal signal clipping.
The IIR design was changed entirely to be based on zeros and poles
(z, p, k) in intermediate storate instead of polynomial coefficients.
This improves the accuracy especially of low frequencies high-pass
designs.
The default max. IIR order is increased from 12 to 16.
Finally since the grpdelay() function in octave-signal package calculates
incorrect result the group delay computation and plotting is disabled
for now in Octave. It works in Matlab.
Signed-off-by: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
206deaf457