ref: 92376697c6ad73d352a6fecae30e3dfcc86b4c49
dir: /libcelt/bands.c/
/* (C) 2007-2008 Jean-Marc Valin, CSIRO
*/
/*
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions
are met:
- Redistributions of source code must retain the above copyright
notice, this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright
notice, this list of conditions and the following disclaimer in the
documentation and/or other materials provided with the distribution.
- Neither the name of the Xiph.org Foundation nor the names of its
contributors may be used to endorse or promote products derived from
this software without specific prior written permission.
THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*/
#ifdef HAVE_CONFIG_H
#include "config.h"
#endif
#include <math.h>
#include "bands.h"
#include "modes.h"
#include "vq.h"
#include "cwrs.h"
#include "os_support.h"
void exp_rotation(celt_norm_t *X, int len, celt_word16_t theta, int dir, int stride, int iter)
{
int i, k;
celt_word16_t c, s;
/* c = cos(theta); s = dir*sin(theta); but we're approximating here for small theta */
c = Q15ONE-MULT16_16_Q15(QCONST16(.5f,15),MULT16_16_Q15(theta,theta));
s = dir*theta;
for (k=0;k<iter;k++)
{
/* We could use MULT16_16_P15 instead of MULT16_16_Q15 for more accuracy,
but at this point, I really don't think it's necessary */
for (i=0;i<len-stride;i++)
{
celt_norm_t x1, x2;
x1 = X[i];
x2 = X[i+stride];
X[i] = MULT16_16_Q15(c,x1) - MULT16_16_Q15(s,x2);
X[i+stride] = MULT16_16_Q15(c,x2) + MULT16_16_Q15(s,x1);
}
for (i=len-2*stride-1;i>=0;i--)
{
celt_norm_t x1, x2;
x1 = X[i];
x2 = X[i+stride];
X[i] = MULT16_16_Q15(c,x1) - MULT16_16_Q15(s,x2);
X[i+stride] = MULT16_16_Q15(c,x2) + MULT16_16_Q15(s,x1);
}
}
}
/* Normalise each band such that the energy is one. */
void normalise_bands(const CELTMode *m, const celt_sig_t *freq, celt_norm_t *X, celt_ener_t *bank)
{
int i, c, B, C;
const int *eBands = m->eBands;
B = m->nbMdctBlocks;
C = m->nbChannels;
for (c=0;c<C;c++)
{
for (i=0;i<m->nbEBands;i++)
{
int j;
float g;
float sum = 1e-10;
for (j=B*eBands[i];j<B*eBands[i+1];j++)
sum += SIG_SCALING_1*SIG_SCALING_1*freq[j*C+c]*freq[j*C+c];
bank[i*C+c] = ENER_SCALING*sqrt(sum);
g = 1.f/(1e-10+ENER_SCALING_1*bank[i*C+c]*sqrt(C));
for (j=B*eBands[i];j<B*eBands[i+1];j++)
X[j*C+c] = NORM_SCALING*SIG_SCALING_1*freq[j*C+c]*g;
}
}
for (i=B*C*eBands[m->nbEBands];i<B*C*eBands[m->nbEBands+1];i++)
X[i] = 0;
}
#ifdef FIXED_POINT
void renormalise_bands(const CELTMode *m, celt_norm_t *X)
{
int i;
VARDECL(celt_ener_t *tmpE);
VARDECL(celt_sig_t *freq);
SAVE_STACK;
ALLOC(tmpE, m->nbEBands*m->nbChannels, celt_ener_t);
ALLOC(freq, m->nbMdctBlocks*m->nbChannels*m->eBands[m->nbEBands+1], celt_sig_t);
for (i=0;i<m->nbMdctBlocks*m->nbChannels*m->eBands[m->nbEBands+1];i++)
freq[i] = SHL32(EXTEND32(X[i]), 10);
normalise_bands(m, freq, X, tmpE);
RESTORE_STACK;
}
#else
void renormalise_bands(const CELTMode *m, celt_norm_t *X)
{
VARDECL(celt_ener_t *tmpE);
SAVE_STACK;
ALLOC(tmpE, m->nbEBands*m->nbChannels, celt_ener_t);
normalise_bands(m, X, X, tmpE);
RESTORE_STACK;
}
#endif
/* De-normalise the energy to produce the synthesis from the unit-energy bands */
void denormalise_bands(const CELTMode *m, const celt_norm_t *X, celt_sig_t *freq, const celt_ener_t *bank)
{
int i, c, B, C;
const celt_word16_t sqrtC_1[2] = {QCONST16(1.f, 14), QCONST16(1.414214f, 14)};
const int *eBands = m->eBands;
B = m->nbMdctBlocks;
C = m->nbChannels;
if (C>2)
celt_fatal("denormalise_bands() not implemented for >2 channels");
for (c=0;c<C;c++)
{
for (i=0;i<m->nbEBands;i++)
{
int j;
celt_word32_t g = MULT16_32_Q14(sqrtC_1[C-1],bank[i*C+c]);
for (j=B*eBands[i];j<B*eBands[i+1];j++)
freq[j*C+c] = MULT16_32_Q14(X[j*C+c], g);
}
}
for (i=B*C*eBands[m->nbEBands];i<B*C*eBands[m->nbEBands+1];i++)
freq[i] = 0;
}
/* Compute the best gain for each "pitch band" */
void compute_pitch_gain(const CELTMode *m, const celt_norm_t *X, const celt_norm_t *P, celt_pgain_t *gains)
{
int i, B;
const int *pBands = m->pBands;
B = m->nbMdctBlocks*m->nbChannels;
for (i=0;i<m->nbPBands;i++)
{
celt_word32_t Sxy=0, Sxx=0;
int j;
/* We know we're not going to overflow because Sxx can't be more than 1 (Q28) */
for (j=B*pBands[i];j<B*pBands[i+1];j++)
{
Sxy = MAC16_16(Sxy, X[j], P[j]);
Sxx = MAC16_16(Sxx, X[j], X[j]);
}
/* No negative gain allowed */
if (Sxy < 0)
Sxy = 0;
/* Not sure how that would happen, just making sure */
if (Sxy > Sxx)
Sxy = Sxx;
/* We need to be a bit conservative (multiply gain by 0.9), otherwise the
residual doesn't quantise well */
Sxy = MULT16_32_Q15(QCONST16(.9f, 15), Sxy);
/* gain = Sxy/Sxx */
gains[i] = DIV32_16(Sxy,ADD32(SHR32(Sxx, PGAIN_SHIFT),EPSILON));
/*printf ("%f ", 1-sqrt(1-gain*gain));*/
}
/*if(rand()%10==0)
{
for (i=0;i<m->nbPBands;i++)
printf ("%f ", 1-sqrt(1-gains[i]*gains[i]));
printf ("\n");
}*/
}
/* Apply the (quantised) gain to each "pitch band" */
void pitch_quant_bands(const CELTMode *m, celt_norm_t *P, const celt_pgain_t *gains)
{
int i, B;
const int *pBands = m->pBands;
B = m->nbMdctBlocks*m->nbChannels;
for (i=0;i<m->nbPBands;i++)
{
int j;
for (j=B*pBands[i];j<B*pBands[i+1];j++)
P[j] = MULT16_16_Q15(gains[i], P[j]);
/*printf ("%f ", gain);*/
}
for (i=B*pBands[m->nbPBands];i<B*pBands[m->nbPBands+1];i++)
P[i] = 0;
}
/* Quantisation of the residual */
void quant_bands(const CELTMode *m, celt_norm_t *X, celt_norm_t *P, celt_mask_t *W, int total_bits, ec_enc *enc)
{
int i, j, B, bits;
const int *eBands = m->eBands;
celt_word16_t alpha;
VARDECL(celt_norm_t *norm);
VARDECL(int *pulses);
VARDECL(int *offsets);
SAVE_STACK;
B = m->nbMdctBlocks*m->nbChannels;
ALLOC(norm, B*eBands[m->nbEBands+1], celt_norm_t);
ALLOC(pulses, m->nbEBands, int);
ALLOC(offsets, m->nbEBands, int);
for (i=0;i<m->nbEBands;i++)
offsets[i] = 0;
/* Use a single-bit margin to guard against overrunning (make sure it's enough) */
bits = total_bits - ec_enc_tell(enc, 0) - 1;
compute_allocation(m, offsets, bits, pulses);
/*printf("bits left: %d\n", bits);
for (i=0;i<m->nbEBands;i++)
printf ("%d ", pulses[i]);
printf ("\n");*/
/*printf ("%d %d\n", ec_enc_tell(enc, 0), compute_allocation(m, m->nbPulses));*/
for (i=0;i<m->nbEBands;i++)
{
int q;
celt_word16_t theta;
float n;
q = pulses[i];
/*Scale factor of .0625f is just there to prevent overflows in fixed-point
(has no effect on float)*/
n = .0625f*sqrt(B*(eBands[i+1]-eBands[i]));
theta = Q15ONE*.007*(B*(eBands[i+1]-eBands[i]))/(.1f+q);
/* If pitch isn't available, use intra-frame prediction */
if (eBands[i] >= m->pitchEnd || q<=0)
{
q -= 1;
alpha = 0;
if (q<0)
intra_fold(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), norm, P+B*eBands[i], B, eBands[i], eBands[m->nbEBands+1]);
else
intra_prediction(X+B*eBands[i], W+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, norm, P+B*eBands[i], B, eBands[i], enc);
} else {
alpha = QCONST16(.7f,15);
}
if (q > 0)
{
exp_rotation(P+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, -1, B, 8);
exp_rotation(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, -1, B, 8);
alg_quant(X+B*eBands[i], W+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], alpha, enc);
exp_rotation(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, 1, B, 8);
}
for (j=B*eBands[i];j<B*eBands[i+1];j++)
norm[j] = X[j] * n;
}
for (i=B*eBands[m->nbEBands];i<B*eBands[m->nbEBands+1];i++)
X[i] = 0;
RESTORE_STACK;
}
/* Decoding of the residual */
void unquant_bands(const CELTMode *m, celt_norm_t *X, celt_norm_t *P, int total_bits, ec_dec *dec)
{
int i, j, B, bits;
const int *eBands = m->eBands;
celt_word16_t alpha;
VARDECL(celt_norm_t *norm);
VARDECL(int *pulses);
VARDECL(int *offsets);
SAVE_STACK;
B = m->nbMdctBlocks*m->nbChannels;
ALLOC(norm, B*eBands[m->nbEBands+1], celt_norm_t);
ALLOC(pulses, m->nbEBands, int);
ALLOC(offsets, m->nbEBands, int);
for (i=0;i<m->nbEBands;i++)
offsets[i] = 0;
/* Use a single-bit margin to guard against overrunning (make sure it's enough) */
bits = total_bits - ec_dec_tell(dec, 0) - 1;
compute_allocation(m, offsets, bits, pulses);
for (i=0;i<m->nbEBands;i++)
{
int q;
celt_word16_t theta;
float n;
q = pulses[i];
/*Scale factor of .0625f is just there to prevent overflows in fixed-point
(has no effect on float)*/
n = .0625f*sqrt(B*(eBands[i+1]-eBands[i]));
theta = Q15ONE*.007*(B*(eBands[i+1]-eBands[i]))/(.1f+q);
/* If pitch isn't available, use intra-frame prediction */
if (eBands[i] >= m->pitchEnd || q<=0)
{
q -= 1;
alpha = 0;
if (q<0)
intra_fold(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), norm, P+B*eBands[i], B, eBands[i], eBands[m->nbEBands+1]);
else
intra_unquant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, norm, P+B*eBands[i], B, eBands[i], dec);
} else {
alpha = QCONST16(.7f,15);
}
if (q > 0)
{
exp_rotation(P+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, -1, B, 8);
alg_unquant(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), q, P+B*eBands[i], alpha, dec);
exp_rotation(X+B*eBands[i], B*(eBands[i+1]-eBands[i]), theta, 1, B, 8);
}
for (j=B*eBands[i];j<B*eBands[i+1];j++)
norm[j] = X[j] * n;
}
for (i=B*eBands[m->nbEBands];i<B*eBands[m->nbEBands+1];i++)
X[i] = 0;
RESTORE_STACK;
}
void stereo_mix(const CELTMode *m, celt_norm_t *X, const celt_ener_t *bank, int dir)
{
int i, B, C;
const int *eBands = m->eBands;
B = m->nbMdctBlocks;
C = m->nbChannels;
for (i=0;i<m->nbEBands;i++)
{
int j;
float left, right;
float a1, a2;
left = bank[i*C];
right = bank[i*C+1];
a1 = left/sqrt(.01+left*left+right*right);
a2 = dir*right/sqrt(.01+left*left+right*right);
for (j=B*eBands[i];j<B*eBands[i+1];j++)
{
float r, l;
l = X[j*C];
r = X[j*C+1];
X[j*C] = a1*l + a2*r;
X[j*C+1] = a1*r - a2*l;
}
}
for (i=B*C*eBands[m->nbEBands];i<B*C*eBands[m->nbEBands+1];i++)
X[i] = 0;
}