FreeCalypso > hg > gsm-codec-lib
view libtwamr/cor_h.c @ 581:e2d5cad04cbf
libgsmhr1 RxFE: store CN R0+LPC separately from speech
In the original GSM 06.06 code the ECU for speech mode is entirely
separate from the CN generator, maintaining separate state. (The
main intertie between them is the speech vs CN state variable,
distinguishing between speech and CN BFIs, in addition to the
CN-specific function of distinguishing between initial and update
SIDs.)
In the present RxFE implementation I initially thought that we could
use the same saved_frame buffer for both ECU and CN, overwriting
just the first 4 params (R0 and LPC) when a valid SID comes in.
However, I now realize it was a bad idea: the original code has a
corner case (long sequence of speech-mode BFIs to put the ECU in
state 6, then SID and CN-mode BFIs, then a good speech frame) that
would be broken by that buffer reuse approach. We could eliminate
this corner case by resetting the ECU state when passing through
a CN insertion period, but doing so would needlessly increase
the behavioral diffs between GSM 06.06 and our version.
Solution: use a separate CN-specific buffer for CN R0+LPC parameters,
and match the behavior of GSM 06.06 code in this regard.
author | Mychaela Falconia <falcon@freecalypso.org> |
---|---|
date | Thu, 13 Feb 2025 10:02:45 +0000 |
parents | 5401aaf7acb0 |
children |
line wrap: on
line source
/* ***************************************************************************** * * GSM AMR-NB speech codec R98 Version 7.6.0 December 12, 2001 * R99 Version 3.3.0 * REL-4 Version 4.1.0 * ****************************************************************************** * * File : cor_h.c * Purpose : correlation functions for codebook search * ***************************************************************************** */ /* ***************************************************************************** * MODULE INCLUDE FILE AND VERSION ID ***************************************************************************** */ #include "namespace.h" #include "cor_h.h" /* ***************************************************************************** * INCLUDE FILES ***************************************************************************** */ #include "typedef.h" #include "basic_op.h" #include "no_count.h" #include "inv_sqrt.h" #include "cnst.h" /* ***************************************************************************** * PUBLIC PROGRAM CODE ***************************************************************************** */ /************************************************************************* * * FUNCTION: cor_h_x() * * PURPOSE: Computes correlation between target signal "x[]" and * impulse response"h[]". * * DESCRIPTION: * The correlation is given by: * d[n] = sum_{i=n}^{L-1} x[i] h[i-n] n=0,...,L-1 * * d[n] is normalized such that the sum of 5 maxima of d[n] corresponding * to each position track does not saturate. * *************************************************************************/ void cor_h_x ( Word16 h[], /* (i): impulse response of weighted synthesis filter */ Word16 x[], /* (i): target */ Word16 dn[], /* (o): correlation between target and h[] */ Word16 sf /* (i): scaling factor: 2 for 12.2, 1 for others */ ) { cor_h_x2(h, x, dn, sf, NB_TRACK, STEP); } /************************************************************************* * * FUNCTION: cor_h_x2() * * PURPOSE: Computes correlation between target signal "x[]" and * impulse response"h[]". * * DESCRIPTION: * See cor_h_x, d[n] can be normalized regards to sum of the * five MR122 maxima or the four MR102 maxima. * *************************************************************************/ void cor_h_x2 ( Word16 h[], /* (i): impulse response of weighted synthesis filter */ Word16 x[], /* (i): target */ Word16 dn[], /* (o): correlation between target and h[] */ Word16 sf, /* (i): scaling factor: 2 for 12.2, 1 for others */ Word16 nb_track,/* (i): the number of ACB tracks */ Word16 step /* (i): step size from one pulse position to the next in one track */ ) { Word16 i, j, k; Word32 s, y32[L_CODE], max, tot; /* first keep the result on 32 bits and find absolute maximum */ tot = 5; move32 (); for (k = 0; k < nb_track; k++) { max = 0; move32 (); for (i = k; i < L_CODE; i += step) { s = 0; move32 (); for (j = i; j < L_CODE; j++) s = L_mac (s, x[j], h[j - i]); y32[i] = s; move32 (); s = L_abs (s); test (); if (L_sub (s, max) > (Word32) 0L) max = s; move32 (); } tot = L_add (tot, L_shr (max, 1)); } j = sub (norm_l (tot), sf); for (i = 0; i < L_CODE; i++) { dn[i] = round (L_shl (y32[i], j)); move16 (); } } /************************************************************************* * * FUNCTION: cor_h() * * PURPOSE: Computes correlations of h[] needed for the codebook search; * and includes the sign information into the correlations. * * DESCRIPTION: The correlations are given by * rr[i][j] = sum_{n=i}^{L-1} h[n-i] h[n-j]; i>=j; i,j=0,...,L-1 * * and the sign information is included by * rr[i][j] = rr[i][j]*sign[i]*sign[j] * *************************************************************************/ void cor_h ( Word16 h[], /* (i) : impulse response of weighted synthesis filter */ Word16 sign[], /* (i) : sign of d[n] */ Word16 rr[][L_CODE] /* (o) : matrix of autocorrelation */ ) { Word16 i, j, k, dec, h2[L_CODE]; Word32 s; /* Scaling for maximum precision */ s = 2; move32 (); for (i = 0; i < L_CODE; i++) s = L_mac (s, h[i], h[i]); j = sub (extract_h (s), 32767); test (); if (j == 0) { for (i = 0; i < L_CODE; i++) { h2[i] = shr (h[i], 1); move16 (); } } else { s = L_shr (s, 1); k = extract_h (L_shl (Inv_sqrt (s), 7)); k = mult (k, 32440); /* k = 0.99*k */ for (i = 0; i < L_CODE; i++) { h2[i] = round (L_shl (L_mult (h[i], k), 9)); move16 (); } } /* build matrix rr[] */ s = 0; move32 (); i = L_CODE - 1; for (k = 0; k < L_CODE; k++, i--) { s = L_mac (s, h2[k], h2[k]); rr[i][i] = round (s); move16 (); } for (dec = 1; dec < L_CODE; dec++) { s = 0; move32 (); j = L_CODE - 1; i = sub (j, dec); for (k = 0; k < (L_CODE - dec); k++, i--, j--) { s = L_mac (s, h2[k], h2[k + dec]); rr[j][i] = mult (round (s), mult (sign[i], sign[j])); move16 (); rr[i][j] = rr[j][i]; move16 (); } } }