mirror of https://github.com/thesofproject/sof.git
95 lines
1.9 KiB
Matlab
95 lines
1.9 KiB
Matlab
function fbr = eq_fir_blob_quant(b, bits, strip_trailing_zeros)
|
|
|
|
%% Quantize FIR coefficients and return vector with length,
|
|
% out shift, and coefficients to be used in the setup blob.
|
|
%
|
|
% fbr = eq_fir_blob_resp(b, bits)
|
|
% b - FIR coefficients
|
|
% bits - optional number of bits, defaults to 16
|
|
%
|
|
% fbr - vector with length, in shift, out shift, and quantized coefficients
|
|
%
|
|
|
|
% SPDX-License-Identifier: BSD-3-Clause
|
|
%
|
|
% Copyright (c) 2016, Intel Corporation. All rights reserved.
|
|
%
|
|
% Author: Seppo Ingalsuo <seppo.ingalsuo@linux.intel.com>
|
|
|
|
b = (b(:))';
|
|
|
|
if nargin < 3
|
|
strip_trailing_zeros = 1;
|
|
end
|
|
|
|
if nargin < 2
|
|
bits = 16;
|
|
end
|
|
|
|
%% Quantize
|
|
[bq, shift] = eq_fir_quantize(b, bits);
|
|
|
|
%% Check trailing zeros
|
|
nf = length(bq);
|
|
nz = nf;
|
|
while bq(nz) == 0
|
|
nz = nz - 1;
|
|
end
|
|
if nz < nf && strip_trailing_zeros
|
|
nb = nz + 1;
|
|
fprintf(1, 'Note: Filter length was reduced ');
|
|
fprintf(1, 'to %d -> %d due to trailing zeros.\n', nf, nb);
|
|
bq = bq(1:nb);
|
|
else
|
|
nb = nf;
|
|
end
|
|
|
|
%% Make length multiple of four (optimized FIR core)
|
|
mod4 = mod(nb, 4);
|
|
if mod4 > 0
|
|
pad = zeros(1,4-mod4);
|
|
bqp = [bq pad];
|
|
nnew = length(bqp);
|
|
fprintf(1,'Note: Filter length was %d, padded length into %d.\n', ...
|
|
nb, nnew);
|
|
else
|
|
fprintf(1,'Note: Filter length is %d\n', nb);
|
|
nnew = nb;
|
|
bqp = bq;
|
|
end
|
|
|
|
%% Pack data into FIR coefficient format
|
|
% int16_t length
|
|
% int16_t out_shift
|
|
% uint32_t reserved[4]
|
|
% int16_t coef[]
|
|
fbr = [nnew shift 0 0 0 0 0 0 0 0 bqp];
|
|
|
|
end
|
|
|
|
function [bq, shift] = eq_fir_quantize(b, bits)
|
|
|
|
% [bq, shift] = eq_fir_quantize(b, bits)
|
|
%
|
|
% Inputs
|
|
% b - FIR coefficients
|
|
% bits - number bits for 2s complement coefficient
|
|
%
|
|
% Outputs
|
|
% bq - quantized coefficients
|
|
% shift - shift right parameter to apply after FIR computation to
|
|
% compensate for coefficients scaling
|
|
%
|
|
|
|
scale = 2^(bits-1);
|
|
|
|
%% Output shift for coefficients
|
|
m = max(abs(b));
|
|
shift = -ceil(log(m+1/scale)/log(2));
|
|
bsr = b*2^shift;
|
|
|
|
%% Quantize to Q1.bits-1 format, e.g. Q1.15 for 16 bits
|
|
bq = eq_coef_quant(bsr, bits, bits-1);
|
|
|
|
end
|