os/mm/mmlibs/mmfw/Codecs/Src/Gsm610CodecCommon/rpeltp.cpp
author sl@SLION-WIN7.fritz.box
Fri, 15 Jun 2012 03:10:57 +0200
changeset 0 bde4ae8d615e
permissions -rw-r--r--
First public contribution.
     1 // Copyright (c) 2000-2009 Nokia Corporation and/or its subsidiary(-ies).
     2 // All rights reserved.
     3 // This component and the accompanying materials are made available
     4 // under the terms of "Eclipse Public License v1.0"
     5 // which accompanies this distribution, and is available
     6 // at the URL "http://www.eclipse.org/legal/epl-v10.html".
     7 //
     8 // Initial Contributors:
     9 // Nokia Corporation - initial contribution.
    10 //
    11 // Contributors:
    12 //
    13 // Description:
    14 //
    15 
    16 #include "types.h"
    17 #include "rpeltp.h"
    18 #include "basicop.h"
    19 #include "tables.h"
    20 #include "gsm610fr.h"
    21 
    22 /*
    23 ** Static variables are allocated as globals in order to make it
    24 ** possible to clear them in run time (reset codec). This might be
    25 ** useful e.g. in possible EC code
    26 */
    27 
    28 
    29 /*
    30 ** void reset_encoder(CGSM610FR_Encoder* aEncoder)
    31 **
    32 ** Function clears encoder variables.
    33 ** Input:
    34 **   None
    35 ** Output:
    36 **   Clear z1, L_z2, mp, LARpp_Prev[0..7], u[0..7], dp[0..119]
    37 ** Return value:
    38 **   None
    39 */
    40 void reset_encoder(CGSM610FR_Encoder* aEncoder)
    41 {
    42   int i;
    43 
    44   aEncoder->z1 = 0;
    45   aEncoder->L_z2 = 0;
    46   aEncoder->mp = 0;
    47 
    48   for ( i = 0; i <= 7; i++ )
    49     aEncoder->LARpp_prev[i] = 0;
    50   for ( i = 0; i <= 7; i++ )
    51     aEncoder->u[i] = 0;
    52   for ( i = 0; i <= 119; i++ )
    53     aEncoder->dp[i] = 0;
    54 }
    55 
    56 
    57 /*
    58 ** void reset_decoder(CGSM610FR_Encoder* aDecoder)
    59 **
    60 ** Function clears decoder variables.
    61 ** Input:
    62 **   None
    63 ** Output:
    64 **   Clear LARpp_Prev[0..7], v[0..8], drp[0..119], nrp
    65 ** Return value:
    66 **   None
    67 */
    68 void reset_decoder(CGSM610FR_Decoder* aDecoder)
    69 {
    70   int i;
    71 
    72   for ( i = 0; i <= 7; i++ )
    73     aDecoder->LARrpp_prev[i] = 0;
    74   for ( i = 0; i <= 8; i++ )
    75     aDecoder->v[i] = 0;
    76   aDecoder->msr = 0;
    77   for ( i = 0; i <= 119; i++ )
    78     aDecoder->drp[i] = 0;
    79   aDecoder->nrp = 40;
    80 }
    81 
    82 /*
    83 #  4.2.1. Downscaling of the input signal
    84 #
    85 #  4.2.2. Offset compensation
    86 #
    87 #  This part implements a high-pass filter and requires extended
    88 #  arithmetic precision for the recursive part of this filter.
    89 #
    90 #  The input of this procedure is the array so[0..159] and the output
    91 #  array sof[0..159].
    92 #
    93 #  Keep z1 and L_z2 in memory for the next frame.
    94 #  Initial value: z1=0; L_z2=0;
    95 
    96 @  Downscaling and offset compensation are combined in order to spare
    97 @  unnecessary data moves.
    98 */
    99 
   100 void prepr( CGSM610FR_Encoder* aEncoder, int2 sof[], int2 so[] )
   101 {
   102   int k;
   103 
   104   int2 msp;
   105   int2 temp;
   106   int4 L_s2;
   107 /*
   108 # 4.2.1. Downscaling of the input signal
   109 # |== FOR k=0 to 159:
   110 # | so[k] = sop[k] >> 3;
   111 # | so[k] = so[k] << 2;
   112 # |== NEXT k:
   113 */
   114 
   115 /*
   116 # |== FOR k = 0 to 159:
   117 # |Compute the non-recursive part.
   118 # | s1 = sub( so[k], z1 );
   119 # | z1 = so[k];
   120 # |Compute the recursive part.
   121 # | L_s2 = s1;
   122 # | L_s2 = L_s2 << 15;
   123 # |Execution of a 31 by 16 bits multiplication.
   124 # | msp = L_z2 >> 15;
   125 # | lsp = L_sub( L_z2, ( msp << 15 ) );
   126 # | temp = mult_r( lsp, 32735 );
   127 # | L_s2 = L_add( L_s2, temp );
   128 # | L_z2 = L_add( L_mult( msp, 32735 ) >> 1, L_s2 );
   129 # |Compute sof[k] with rounding.
   130 # | sof[k] = L_add( L_z2, 16384 ) >> 15;
   131 # |== NEXT k:
   132 */
   133   for (k=0; k <= 159; k++) {
   134 
   135     /* Downscaling */
   136     temp = shl( shr( so[k], 3 ), 2 );
   137 
   138     /* Compute the non-recursive part. */
   139     /* Compute the recursive part. */
   140 
   141     L_s2 = L_deposit_l( sub( temp, aEncoder->z1 ) );
   142     aEncoder->z1 = temp;
   143 
   144 
   145     L_s2 = L_shl( L_s2, 15 );
   146     /* Execution of a 31 by 16 bits multiplication. */
   147     msp = extract_l( L_shr( aEncoder->L_z2, 15 ) );
   148     temp = extract_l( L_sub( aEncoder->L_z2, L_shl( L_deposit_l( msp ), 15 ) ) );
   149     temp = mult_r( temp, 32735 );
   150     L_s2 = L_add( L_s2, L_deposit_l( temp ) );
   151     aEncoder->L_z2 = L_add( L_shr( L_mult( msp, 32735 ), 1 ), L_s2 );
   152     /* Compute sof[k] with rounding. */
   153     sof[k] = extract_l( L_shr( L_add( aEncoder->L_z2, (int4) 16384 ), 15 ) );
   154   }
   155 }
   156 
   157 /*
   158 #  4.2.3. Preemphasis
   159 #
   160 #  Keep mp in memory for the next frame.
   161 #  Initial value: mp=0;
   162 */
   163 void preemp( CGSM610FR_Encoder* aEncoder, int2 s[], int2 sof[] )
   164 {
   165   int k;
   166   int2 temp;
   167 /*
   168 # |== FOR k=0 to 159:
   169 # | s[k] = add( sof[k], mult_r( mp, -28180 ) );
   170 # | mp = sof[k];
   171 # |== NEXT k:
   172 */
   173 
   174 /*
   175 @ Reverse looping in order to make it possible to
   176 @ update filter delay mp only at the end of the loop
   177 */
   178   temp = sof[159]; /* make overwrite possible */
   179 
   180   for ( k = 159; k >= 1; k-- )
   181     s[k] = add( sof[k], mult_r( sof[k-1], -28180 ) );
   182 
   183   s[0] = add( sof[0], mult_r( aEncoder->mp, -28180 ) );
   184 
   185   aEncoder->mp = temp;
   186 }
   187 
   188 /*
   189 #  4.2.4. Autocorrelation
   190 #
   191 #  The goal is to compute the array L_ACF[k]. The signal s[i] must be
   192 #  scaled in order to avoid an overflow situation.
   193 *
   194 * output:
   195 *       scalauto (return value)
   196 *
   197 */
   198 int2 autoc( int4 L_ACF[], int2 s[] )
   199 {
   200   int k, i;
   201 
   202   int2 smax;
   203   int2 temp;
   204   int4 L_temp2;
   205   int2 scalauto;
   206 /*
   207 # Dynamic scaling of the array s[0..159].
   208 #
   209 # Search for the maximum.
   210 #
   211 # smax=0;
   212 # |== FOR k = 0 to 159:
   213 # | temp = abs( s[k] );
   214 # | IF ( temp > smax ) THEN smax = temp;
   215 # |== NEXT k;
   216 */
   217   smax = 0;
   218   for ( k = 0; k <= 159; k++ ) {
   219     temp = abs_s( s[k] );
   220     if ( sub( temp, smax ) > 0 )
   221       smax = temp;
   222   }
   223 /*
   224 # Computation of the scaling factor.
   225 #
   226 # IF ( smax == 0 ) THEN scalauto = 0;
   227 #    ELSE scalauto = sub( 4, norm( smax << 16 ) );
   228 */
   229   if ( smax == 0 )
   230     scalauto = 0;
   231   else
   232     scalauto = sub( 4, norm_l( L_deposit_h( smax ) ) );
   233 /* 
   234 # Scaling of the array s[0..159].
   235 # IF ( scalauto > 0 ) THEN
   236 #    | temp = 16384 >> sub( scalauto, 1 );
   237 #    |== FOR k=0 to 159:
   238 #    |    s[k] = mult_r( s[k], temp );
   239 #    |== NEXT k:
   240 */
   241   if ( scalauto > 0 ) {
   242     temp = shr( 16384, sub( scalauto, 1 ) );
   243     for ( k = 0; k <= 159; k++ ) 
   244       s[k] = mult_r( s[k], temp );
   245   }
   246 /*
   247 # Compute the L_ACF[..].
   248 # |== FOR k=0 to 8:
   249 # |     L_ACF[k] = 0;
   250 # |==== FOR i=k to 159:
   251 # |         L_temp = L_mult( s[i], s[i-k] );
   252 # |         L_ACF[k] = L_add( L_ACF[k], L_temp );
   253 # |==== NEXT i:
   254 # |== NEXT k:
   255 */
   256   for ( k = 0; k <= 8; k++ ) {
   257     L_temp2 = 0;
   258     for ( i = k; i <= 159; i++ )
   259 	L_temp2 = L_mac( L_temp2, s[i], s[i-k] );
   260 
   261     L_ACF[k] = L_temp2;
   262   }
   263 /*
   264 # Rescaling of the array s[0..159].
   265 #
   266 # IF ( scalauto > 0 ) THEN
   267 #   |== FOR k = 0 to 159:
   268 #   |    s[k] = s[k] << scalauto;
   269 #   |== NEXT k:
   270 */
   271   if ( scalauto > 0 ) {
   272     for ( k = 0; k <= 159; k++ )
   273 	 s[k] = shl( s[k], scalauto );
   274   }
   275   return(scalauto); /* scalauto is retuned to be used also in vad */
   276   
   277 }
   278 
   279 
   280 /*
   281 #  4.2.5. Computation of the reflection coefficients
   282 */
   283 void schur( int2 r[], int4 L_ACF[] )
   284 {
   285   int k, i, n, m;
   286 
   287   int2 P[9];
   288   int2 K[7];
   289   int2 ACF[9];
   290   int2 normacf;
   291 
   292 /*
   293 # Schur recursion with 16 bits arithmetic
   294 #
   295 #    IF ( L_ACF[0] == 0 ) THEN
   296 #                   |== FOR i=1 to 8:
   297 #                   |    r[i] = 0;
   298 #                   |== NEXT i:
   299 #                   |    EXIT; / continue with section 4.2.6/
   300 #    normacf = norm( L_ACF[0] ); / temp is spec replaced with normacf /
   301 #    |== FOR k=0 to 8:
   302 #    |    ACF[k] = ( L_ACF[k] << normacf ) >> 16;
   303 #    |== NEXT k:
   304 */
   305   if ( L_ACF[0] == 0 ) {
   306     for ( i = 0; i <= 7; i++)
   307 	 r[i] = 0;
   308     return; /* continue with section 4.2.6 */
   309   }
   310   normacf = norm_l( L_ACF[0] );
   311 
   312   for ( k = 0; k <= 8; k++ )
   313     ACF[k] = extract_h( L_shl( L_ACF[k], normacf ) );
   314 /*
   315 # Initialize array P[..] and K[..] for the recursion.
   316 #
   317 #    |== FOR i=1 to 7:
   318 #    |    K[9-i] = ACF[i];
   319 #    |== NEXT i:
   320 #
   321 #    |== FOR i=0 to 8:
   322 #    |    P[i] = ACF[i];
   323 #    |== NEXT i:
   324 */
   325   for ( i = 1; i <= 7; i++ )
   326     K[7-i] = ACF[i];
   327 
   328   for ( i = 0; i <= 8; i++ )
   329     P[i] = ACF[i];
   330 /*
   331 # Compute reflection coefficients
   332 #    |== FOR n=1 to 8:
   333 #    |    IF ( P[0] < abs( P[1] ) ) THEN
   334 #    |                        |== FOR i=n to 8:
   335 #    |                        |    r[i] = 0;
   336 #    |                        |== NEXT i:
   337 #    |                        |    EXIT; /continue with
   338 #    |                        |    section 4.2.6./
   339 #    |    r[n] = div( abs( P[1] ), P[0] );
   340 #    |    IF ( P[1] > 0 ) THEN r[n] = sub( 0, r[n] );
   341 #    |
   342 #    |    IF ( n == 8 ) THEN EXIT; /continue with section 4.2.6/
   343 #    |    Schur recursion
   344 #    |    P[0] = add( P[0], mult_r( P[1], r[n] ) );
   345 #    |==== FOR m=1 to 8-n:
   346 #    |         P[m] = add( P[m+1], mult_r( K[9-m], r[n] ) );
   347 #    |         K[9-m] = add( K[9-m], mult_r( P[m+1], r[n] ) );
   348 #    |==== NEXT m:
   349 #    |
   350 #    |== NEXT n:
   351 */
   352 
   353   for ( n = 0; n <= 7; n++ )  {
   354     if ( sub( P[0], abs_s( P[1] ) ) < 0 ) {
   355 	 for ( i = n; i <= 7; i++ )
   356 	   r[i] = 0;
   357 	 return; /* continue with section 4.2.6. */
   358     }
   359 
   360     if ( P[1] > 0 )
   361 	 r[n] = negate( div_s( P[1], P[0] ) );
   362     else
   363 	 r[n] = div_s( negate( P[1] ), P[0] );
   364 
   365     if ( sub(int2 (n), 7) == 0 )
   366 	 return; /* continue with section 4.2.6 */
   367 
   368     /* Schur recursion */
   369     P[0] = add( P[0], mult_r( P[1], r[n] ) );
   370     for ( m = 1; m <= 7-n; m++ ) {
   371 /*
   372  *    mac_r cannot be used because it rounds the result after
   373  *    addition when add( xx, mult_r ) rounds first the result
   374  *    of the product. That is why the following two lines cannot
   375  *    be used instead of the curently used lines.
   376  *
   377  *    P[m] = mac_r( L_deposit_l( P[m+1] ), K[7-m], r[n] );
   378  *    K[7-m] = mac_r( L_deposit_l( K[7-m] ), P[m+1], r[n] );
   379 */
   380       P[m] = add( P[m+1], mult_r( K[7-m], r[n] ) );
   381       K[7-m] = add( K[7-m], mult_r( P[m+1], r[n] ) );
   382     }
   383   }
   384 }
   385 
   386 /*
   387 #  4.2.6. Transformation of reflection coefficients to Log.-Area Ratios -----
   388 #
   389 #  The following scaling for r[..] and LAR[..] has been used:
   390 #
   391 #  r[..] = integer( real_r[..]*32768. ); -1. <= real_r < 1.
   392 #  LAR[..] = integer( real_LAR[..]*16384. );
   393 #  with -1.625 <= real_LAR <= 1.625
   394 */
   395 
   396 void larcomp( int2 LAR[], int2 r[] )
   397 {
   398   int i;
   399 
   400   int2 temp;
   401 /*
   402 # Computation of the LAR[1..8] from the r[1..8]
   403 #    |== FOR i=1 to 8:
   404 #    |    temp = abs( r[i] );
   405 #    |    IF ( temp < 22118 ) THEN temp = temp >> 1;
   406 #    |         else if ( temp < 31130 ) THEN temp = sub( temp, 11059 );
   407 #    |              else temp = sub( temp, 26112 ) << 2;
   408 #    |    LAR[i] = temp;
   409 #    |    IF ( r[i] < 0 ) THEN LAR[i] = sub( 0, LAR[i] );
   410 #    |== NEXT i:
   411 */
   412   for ( i = 1; i <= 8; i++ ) {
   413 	int j = i-1;
   414     temp = abs_s( r[j] );
   415 
   416     if ( sub( temp, 22118 ) < 0 )
   417       temp = shr( temp, 1 );
   418     else if ( sub( temp, 31130 ) < 0 )
   419       temp = sub( temp, 11059 );
   420     else
   421       temp = shl( sub( temp, 26112 ), 2 );
   422 
   423     if ( r[j] < 0 )
   424       LAR[j] = negate( temp );
   425     else
   426       LAR[j] = temp;
   427   }
   428 }
   429 
   430 
   431 /*
   432 #  4.2.7. Quantization and coding of the Log.-Area Ratios
   433 #
   434 #  This procedure needs fpur tables; following equations give the
   435 #  optimum scaling for the constants:
   436 #
   437 #  A[1..8]=integer( real_A[1..8]*1024 ); 8 values (see table 4.1)
   438 #  B[1..8]=integer( real_B[1..8]*512 );  8 values (see table 4.1)
   439 #  MAC[1..8]=maximum of the LARc[1..8];  8 values (see table 4.1)
   440 #  MAC[1..8]=minimum of the LARc[1..8];  8 values (see table 4.1)
   441 */
   442 
   443 void codlar( int2 LARc[], int2 LAR[] )
   444 {
   445 
   446   int i;
   447 
   448   int2 temp;
   449 /*
   450 # Computation for quantizing and coding the LAR[1..8]
   451 #
   452 #    |== FOR i=1 to 8:
   453 #    |    temp = mult( A[i], LAR[i] );
   454 #    |    temp = add( temp, B[i] );
   455 #    |    temp = add( temp, 256 );      for rounding
   456 #    |    LARc[i] = temp >> 9;
   457 #    |
   458 #    | Check if LARc[i] lies between MIN and MAX
   459 #    |    IF ( LARc[i] > MAC[i] ) THEN LARc[i] = MAC[i];
   460 #    |    IF ( LARc[i] < MIC[i] ) THEN LARc[i] = MIC[i];
   461 #    |    LARc[i] = sub( LARc[i], MIC[i] ); / See note below /
   462 #    |== NEXT i:
   463 #
   464 # NOTE: The equation is used to make all the LARc[i] positive.
   465 */
   466   for ( i = 1; i <= 8; i++ ) {
   467 	int j = i-1;
   468     temp = mult( A[j], LAR[j] );
   469     temp = add( temp, B[j] );
   470     temp = add( temp, 256 ); /* for rounding */
   471     temp = shr( temp, 9 );
   472     /* Check if LARc[i] lies between MIN and MAX */
   473     if ( sub( temp, MAC[j] ) > 0 )
   474       LARc[j] = sub( MAC[j], MIC[j] );
   475     else if ( sub( temp, MIC[j] ) < 0 )
   476       LARc[j] = 0;
   477     else
   478       LARc[j] = sub( temp, MIC[j] );
   479   }
   480 }
   481 
   482 
   483 /*
   484 #  4.2.8 Decoding of the coded Log.-Area Ratios
   485 #
   486 #  This procedure requires for efficient implementation two variables.
   487 #
   488 #  INVA[1..8]=integer((32768*8)/(real_A[1..8]);    8 values (table 4.2)
   489 #  MIC[1..8]=minimum value of the LARc[1..8];      8 values (table 4.2)
   490 */
   491 
   492 void declar( int2 LARpp[], int2 LARc[] )
   493 {
   494   int i;
   495 
   496   int2 temp1;
   497   int2 temp2;
   498 /*
   499 # Compute the LARpp[1..8].
   500 #
   501 #    |== FOR i=1 to 8:
   502 #    |    temp1 = add( LARc[i], MIC[i] ) << 10; /See note below/
   503 #    |    temp2 = B[i] << 1;
   504 #    |    temp1 = sub( temp1, temp2 );
   505 #    |    temp1 = mult_r( INVA[i], temp1 );
   506 #    |    LARpp[i] = add( temp1, temp1 );
   507 #    |== NEXT i:
   508 #
   509 # NOTE: The addition of MIC[i] is used to restore the sign of LARc[i].
   510 */
   511   for ( i = 1; i <= 8; i++ ) {
   512 	int j = i-1;
   513     temp1 = shl( add( LARc[j], MIC[j] ), 10 );
   514     temp2 = shl( B[j], 1 );
   515     temp1 = sub( temp1, temp2 );
   516     temp1 = mult_r( INVA[j], temp1 );
   517     LARpp[j] = add( temp1, temp1 );
   518   }
   519 }
   520 
   521 
   522 /*
   523 #  4.2.9. Computation of the quantized reflection coefficients
   524 #
   525 #  Within each frame of 160 anallyzed speech samples the short term
   526 #  analysissss and synthesis filters operate with four different sets of
   527 #  coefficients, derived from the previous set of decoded 
   528 #  LARs(LARpp(j-1)) and the actual set of decoded LARs (LARpp(j)).
   529 #
   530 # 4.2.9.1 Interpolation of the LARpp[1..8] to get LARp[1..8]
   531 */
   532 
   533 void cparc1( int2 LARp[], int2 LARpp_prev[], int2 LARpp[] )
   534 {
   535   int i;
   536 
   537   int2 temp;
   538 /*
   539 #    FOR k_start=0 to k_end = 12.
   540 #         
   541 #    |==== FOR i=1 to 8:
   542 #    |         LARp[i] = add( ( LARpp(j-1)[i] >> 2 ) ,( LARpp[i] >> 2 ) );
   543 #    |         LARp[i] = add( LARp[i], ( LARpp(j-1)[i] >> 1 ) );
   544 #    |==== NEXT i:
   545 */
   546   /* k_start=0 to k_end=12 */
   547 
   548   for ( i = 1; i <= 8; i++ ) {
   549 	int j = i-1;
   550     temp = add( shr( LARpp_prev[j], 2 ), shr( LARpp[j], 2 ) );
   551     LARp[j] = add( temp, shr( LARpp_prev[j], 1 ) );
   552   }
   553 }
   554  
   555 
   556 void cparc2( int2 LARp[], int2 LARpp_prev[], int2 LARpp[] )
   557 {
   558   int i;
   559   
   560 /*
   561 #    FOR k_start=13 to k_end = 26.
   562 #    |==== FOR i=1 to 8:
   563 #    |         LARp[i] = add( ( LARpp(j-1)[i] >> 1 ), ( LARpp[i] >> 1 ) );
   564 #    |==== NEXT i:
   565 */
   566   /* k_start=13 to k_end=26 */
   567 	
   568   for (i=1; i <= 8; i++) {
   569 	int j = i-1;
   570     LARp[j] = add( shr( LARpp_prev[j], 1 ), shr( LARpp[j], 1 ) );
   571   }
   572 }
   573 
   574 
   575 void cparc3( int2 LARp[], int2 LARpp_prev[], int2 LARpp[] )
   576 {
   577   int i;
   578 
   579   int2 temp;
   580   
   581 /*
   582 #    FOR k_start=27 to k_end = 39.
   583 #    |==== FOR i=1 to 8:
   584 #    |         LARp[i] = add( ( LARpp(j-1)[i] >> 2 ), ( LARpp[i] >> 2 ) );
   585 #    |         LARp[i] = add( LARp[i], ( LARpp[i] >> 1 ) );
   586 #    |==== NEXT i:
   587 */
   588   /* k_start=27 to k_end=39 */
   589 	
   590   for ( i = 1; i <= 8; i++ ) {
   591 	int j = i-1;
   592     temp = add( shr( LARpp_prev[j], 2 ), shr( LARpp[j], 2 ) );
   593     LARp[j] = add( temp, shr( LARpp[j], 1 ) );
   594   }
   595 }
   596 
   597 
   598 void cparc4( int2 LARp[], int2 LARpp_prev[], int2 LARpp[] )
   599 {
   600   int i;
   601   
   602 /*
   603 #    FOR k_start=40 to k_end = 159.
   604 #    |==== FOR i=1 to 8:
   605 #    |         LARp[i] = LARpp[i];
   606 #    |==== NEXT i:
   607 */
   608   /* k_start=40 to k_end=159 */
   609 	
   610   for ( i = 1; i <= 8; i++ ) {
   611 	int j = i-1;
   612     LARp[j] = LARpp[j];
   613     /* note new LARs saved here for next frame */
   614     LARpp_prev[j] = LARpp[j];
   615   }
   616 
   617 }
   618 
   619 
   620 /*
   621 #  4.2.9.2 Computation of the rp[] from the interpolated LARp[]
   622 #
   623 #  The input of this procedure is the interpolated LARp[1..8] array. The
   624 #  reflection coefficients, rp[i], are used in the analysis filter and in
   625 #  the synthesis filter.
   626 */
   627 
   628 void crp( int2 rp[], int2 LARp[] )
   629 {
   630   int i;
   631 
   632   int2 temp;
   633 
   634 /*
   635 #    |== FOR i=1 to 8:
   636 #    |    temp = abs( LARp[i] );
   637 #    |    IF ( temp < 11059 ) THEN temp = temp << 1;
   638 #    |         ELSE IF ( temp < 20070 ) THEN temp = add( temp, 11059 );
   639 #    |              ELSE temp = add( ( temp >> 2 ), 26112 );
   640 #    |    rp[i] = temp;
   641 #    |    IF ( LARp[i] < 0 ) THEN rp[i] = sub( 0, rp[i] );
   642 #    |== NEXT i:
   643 */
   644   for (i=1; i <= 8; i++) {
   645 	int j = i-1;
   646     temp = abs_s( LARp[j] );
   647     if ( sub( temp, 11059 ) < 0 )
   648       temp = shl( temp, 1 );
   649     else if ( sub( temp, 20070 ) < 0 )
   650       temp = add( temp, 11059 );
   651     else
   652       temp = add( shr( temp, 2 ), 26112 );
   653 
   654     if ( LARp[j] < 0 )
   655       rp[j] = negate( temp );
   656     else
   657       rp[j] = temp;
   658   }
   659 }
   660 
   661 
   662 /*
   663 #  4.2.10. Short term analysis filtering
   664 #
   665 #  This procedure computes the short term residual d[..] to be fed
   666 #  to the RPE-LTP loop from s[..] signal and from the local rp[..]
   667 #  array (quantized reflection coefficients). As the call of this
   668 #  procedure can be done in many ways (see the interpolation of the LAR
   669 #  coefficients), it is assumed that the computation begins with index
   670 #  k_start (for arrays d[..] and s[..]) and stops with index k_end
   671 #  (k_start and k_end are defined in 4.2.9.1): This procedure also need
   672 #  to keep the array u[0..7] in memory for each call.
   673 #
   674 #  Keep the array u[0..7] in memory.
   675 #  Initial value: u[0..7]=0;
   676 */
   677 
   678 void invfil( CGSM610FR_Encoder* aEncoder, int2 d[], int2 s[], int2 rp[], int k_start, int k_end )
   679 {
   680   //ALEX//extern int2 u[];
   681 
   682   int k, i;
   683 
   684   int2 temp;
   685   int2 sav;
   686   int2 di;
   687 /*
   688 #    |== FOR k=k_start to k_end:
   689 #    |    di = s[k];
   690 #    |    sav = di;
   691 #    |==== FOR i=1 to 8:
   692 #    |         temp = add( u[i], mult_r( rp[i], di ) );
   693 #    |         di = add( di, mult_r( rp[i], u[i] ) );
   694 #    |         u[i] = sav;
   695 #    |         sav = temp;
   696 #    |==== NEXT i:
   697 #    |    d[k] = di;
   698 #    |== NEXT k:
   699 */
   700   for ( k = k_start; k <= k_end; k++ ) {
   701     di = s[k];
   702     sav = di;
   703     for ( i = 1; i <= 8; i++ ) {
   704 	  int j = i-1;
   705       temp = add( aEncoder->u[j], mult_r( rp[j], di ) );
   706       di = add( di, mult_r( rp[j], aEncoder->u[j] ) );
   707       aEncoder->u[j] = sav;
   708       sav = temp;
   709     }
   710     d[k] = di;
   711   }
   712 
   713 }
   714 
   715 
   716 /*
   717 #  4.2.11. Calculation of the LTP parameters
   718 #
   719 #  This procedure computes the LTP gain (bc) and the LTP lag (Nc) for
   720 #  the long term analysis filter. This is deone by calculating a maximum
   721 #  of the cross-correlation function between the current sub-segment
   722 #  short term residual signal d[0..39] (output of the short term 
   723 #  analysis filter; for each sub-segment of the RPE-LTP analysis) and the
   724 #  previous reconstructed short term residual signal dp[-120..-1]. A
   725 #  dynamic scaling must be performed to avoid overflow.
   726 # 
   727 #  Initial value: dp[-120..-1]=0;
   728 */
   729 
   730 void ltpcomp( CGSM610FR_Encoder* aEncoder, int2 *Nc, int2 *bc, int2 d[], int k_start )
   731 {
   732   int k, i;
   733 
   734   int2 lambda;
   735   int2 temp;
   736   int2 scal;
   737   int2 dmax;
   738   int4 L_max;
   739   int2 wt[40]; /* scaled residual, original cannot be destroyed */
   740   int4 L_result;
   741   int4 L_power;
   742   int2 R;
   743   int2 S;
   744 /*
   745 # Search of optimum scaling of d[kstart+0..39]
   746 #    dmax = 0;
   747 #    |== FOR k=0 to 39:
   748 #    |    temp = abs( d[k] );
   749 #    |    IF ( temp > dmax ) THEN dmax = temp;
   750 #    |== NEXT k:
   751 */
   752   dmax = 0;
   753   for (k=0; k <= 39; k++) {
   754     temp = abs_s( d[k+k_start] );
   755     if ( sub( temp, dmax ) > 0 )
   756       dmax = temp;
   757   }
   758 /*
   759 #    temp = 0;
   760 #    IF ( dmax == 0 ) THEN scal = 0;
   761 #         ELSE temp = norm( (long)dmax << 16 );
   762 #    IF ( temp > 6 ) THEN scal = 0;
   763 #         ELSE scal = sub( 6, temp );
   764 */  
   765   temp = 0;
   766   if ( dmax == 0 )
   767     scal = 0;
   768   else
   769     temp = norm_s( dmax );
   770 
   771   if ( sub( temp, 6 ) > 0 )
   772     scal = 0;
   773   else
   774     scal = sub( 6, temp );  /* 0 <= scal <= 6 */
   775 /*
   776 # Initialization of a working array wt[0..39]
   777 #    |== FOR k=0 to 39:
   778 #    |    wt[k] = d[k] >> scal;
   779 #    |== NEXT k:
   780 */
   781   for (k=0; k <= 39; k++)
   782     wt[k] = shr( d[k+k_start], scal ); /* scal >= 0 */
   783 /*
   784 # Search for the maximum of crosscorrelation and coding of the LTP lag.
   785 #    L_max = 0;
   786 #    Nc = 40;
   787 # 
   788 #    |== FOR lambda=40 to 120:
   789 #    |    L_result = 0;
   790 #    |==== FOR k=0 to 39:
   791 #    |         L_temp = L_mult( wt[k], dp[k-lambda] );
   792 #    |         L_result = L_add( L_temp, L_result );
   793 #    |==== NEXT k:
   794 #    |    IF ( L_result > L_max ) THEN
   795 #    |                                  |    Nc = lambda;
   796 #    |                                  |    L_max = L_result;
   797 #    |== NEXT lambda:
   798 */
   799   L_max = 0; /* 32 bits maximum */
   800   *Nc = 40; /* index for the maximum of cross correlation */
   801   for ( lambda = 40; lambda <= 120; lambda++ ) {
   802     L_result = 0;
   803     for (k = 0; k <= 39; k++)
   804 	  L_result = L_mac( L_result, wt[k], aEncoder->dp[k-lambda+120] );
   805   /* Borland C++ 3.1 error if -3 (386-instructions) are used.
   806   ** The code makes error (compared to (L_result > L_max)
   807   ** comparison. The problem disapears if the result of L_sub
   808   ** is stored to variable, e.g.
   809   **   if ( ( L_debug = L_sub( L_result, L_max ) ) > 0 ) {
   810   **
   811   ** Problem does not occur when -2 option (only 286
   812   ** instructions are used)
   813   **
   814   ** The problem exist e.g. with GSM full rate test seq01.ib
   815   */
   816     if ( L_sub( L_result, L_max ) > 0 ) {
   817 	 *Nc = lambda;
   818 	 L_max = L_result;
   819     }
   820   }
   821 
   822 /*
   823 # Re-scaling of L-max
   824 #    L_max = L_max >> sub( 6, scal );
   825 */
   826   L_max = L_shr( L_max, sub( 6, scal ) );
   827 /*
   828 # Initialization of a working array wt[0..39]
   829 #    |== FOR k=0 to 39:
   830 #    |    wt[k] = dp[k-Nc] >> 3;
   831 #    |== NEXT k:
   832 */
   833   for (k = 0; k <= 39; k++)
   834     wt[k] = shr( aEncoder->dp[k - *Nc + 120], 3 );
   835 /*
   836 # Compute the power of the reconstructed short term residual signal dp[..]
   837 #    L_power = 0;
   838 #    |== FOR k=0 to 39:
   839 #    |    L_temp = L_mult( wt[k], wt[k] );
   840 #    |    L_power = L_add( L_temp, L_power );
   841 #    |== NEXT k:
   842 */
   843   L_power = 0;
   844   for ( k = 0; k <= 39; k++ )
   845     L_power = L_mac( L_power, wt[k], wt[k] );
   846 /*
   847 # Normalization of L_max  and L_power
   848 #    IF ( L_max <= 0 ) THEN
   849 #                             | bc = 0;
   850 #                             | EXIT; /cont. with 4.2.12/
   851 */
   852   if ( L_max <= 0 ) {
   853     *bc = 0;
   854     return;
   855   }
   856 /*
   857 #    IF ( L_max >= L_power ) THEN
   858 #                             | bc = 3;
   859 #                             | EXIT; /cont. with 4.2.12/
   860 */
   861   if ( L_sub( L_max, L_power ) >= 0 ) {
   862     *bc = 3;
   863     return;
   864   }
   865 /*
   866 #    temp = norm( L_power );
   867 #    R = ( L_max << temp ) >> 16 );
   868 #    S = ( L_power << temp ) >> 16 );
   869 */
   870   temp = norm_l( L_power );
   871   R = extract_h( L_shl( L_max, temp ) );
   872   S = extract_h( L_shl( L_power, temp ) );
   873 /*
   874 # Coding of the LTP gain
   875 #
   876 # Table 4.3a must be used to obtain the level DLB[i] for the
   877 # quantization of the LTP gain b to get the coded version bc.
   878 #
   879 #    |== FOR bc=0 to 2:
   880 #    |    IF ( R <= mult( S, DLB[bc] ) ) THEN EXIT; /cont. with 4.2.12/
   881 #    |== NEXT bc:
   882 #
   883 #    bc = 3;
   884 */
   885   for ( i = 0; i <= 2; i++ ) {
   886     if ( sub( R, mult( S, DLB[i] ) ) <= 0 ) {
   887       *bc = int2 (i);
   888       return;
   889     }
   890   }
   891 
   892   *bc = 3;
   893 
   894 }
   895 
   896 
   897 /*
   898 #  4.2.12. Long term analysis filtering
   899 #
   900 #  In this part, we have to decode the bc parameter to compute the
   901 #  samples of the estimate dpp[0..39]. The decoding of bc needs the use
   902 #  of table 4.3b. The long term residual signal e[0..39] is then
   903 #  calculated to be fed to the RPE encoding section.
   904 */
   905 
   906 void ltpfil( CGSM610FR_Encoder* aEncoder, int2 e[], int2 dpp[], int2 d[], int2 bc, int2 Nc, int k_start )
   907 {
   908   int2 bp;
   909   int k;
   910 
   911 /*
   912 #    Decoding of the coded LTP gain.
   913 #       bp = QLB[bc];
   914 */
   915 	bp = QLB[bc];
   916 /*
   917 # Calculating the array e[0..39] and the array dpp[0..39]
   918 #
   919 #    |== FOR k=0 to 39:
   920 #    |         dpp[k] = mult_r( bp, dp[k-Nc] );
   921 #    |         e[k] = sub( d[k], dpp[k] );
   922 #    |== NEXT k:
   923 */
   924   for ( k = 0; k <= 39; k++ ) {
   925     dpp[k] = mult_r( bp, aEncoder->dp[k - Nc + 120] );
   926     e[k] = sub( d[k+k_start], dpp[k] );
   927   }
   928 }
   929 
   930 
   931 /*
   932 #  4.2.13. Weighting filter
   933 #
   934 #  The coefficients of teh weighting filter are stored in tables (see
   935 #  table 4.4). The following scaling is used:
   936 #
   937 #  H[0..10] = integer( real_H[0..10]*8192 );
   938 */
   939 
   940 void weight( int2 x[], int2 e[] )
   941 {
   942   int k, i;
   943 
   944   int2 wt[50];
   945   int4 L_result;
   946 /*
   947 # Initialization of a temporary working array wt[0..49]
   948 #    |== FOR k=0 to 4:
   949 #    |    wt[k] = 0;
   950 #    |== NEXT k:
   951 #
   952 #    |== FOR k=5 to 44:
   953 #    |    wt[k] = e[k-5];
   954 #    |== NEXT k:
   955 #
   956 #    |== FOR k=45 to 49:
   957 #    |    wt[k] = 0;
   958 #    |== NEXT k:
   959 */
   960   for ( k = 0; k <= 4; k++ ) 
   961     wt[k] = 0;
   962      
   963   for ( k = 5; k <= 44; k++ ) 
   964     wt[k] = e[k-5];
   965      
   966   for ( k = 45; k <= 49; k++ ) 
   967     wt[k] = 0;
   968 /*
   969 # Compute the signal x[0..39]
   970 #    |== FOR k=0 to 39:
   971 #    |    L_result = 8192;
   972 #    |==== FOR i=0 to 10:
   973 #    |         L_temp = L_mult( wt[k+i], H[i] );
   974 #    |         L_result = L_add( L_result, L_temp );
   975 #    |==== NEXT i:
   976 #    |    L_result = L_add( L_result, L_result ); /scaling L_result (x2)/
   977 #    |    L_result = L_add( L_result, L_result ); /scaling L_result (x4)/
   978 #    |    x[k] = (int)( L_result >> 16 );
   979 #    |== NEXT k:
   980 */
   981   for ( k = 0; k <= 39; k++ ) {
   982     L_result = L_deposit_l( 8192 );
   983     for ( i = 0; i <= 10; i++ )
   984       L_result = L_mac( L_result, wt[k+i], H[i] );
   985 
   986     /* scaling L_result (x4) and extract: scaling possible with new shift
   987      * because saturation is added L_shl
   988      *
   989      * L_result = L_add( L_result, L_result );
   990      * L_result = L_add( L_result, L_result );
   991      * x[k] = extract_h( L_result ); 
   992      @ Scaling can be done with L_shift because now shift has saturation
   993      */
   994 
   995     x[k] = extract_h( L_shl( L_result, 2 ) );
   996   }
   997 }
   998 
   999 
  1000 /*
  1001 #  4.2.14. RPE grid selection
  1002 #
  1003 #  The signal x[0..39] is used to select the RPE grid which is
  1004 #  represented by Mc.
  1005 */
  1006 
  1007 int2 gridsel( int2 xM[], int2 x[] )
  1008 {
  1009   int i, k;
  1010 
  1011   int2 temp1;
  1012   int4 L_EM;
  1013   int4 L_result;
  1014   int2 Mc;
  1015 /*
  1016 #    EM = 0;
  1017 #    Mc = 0;
  1018 */
  1019   L_EM = 0;
  1020   Mc = 0;
  1021 /*
  1022 #    |== FOR m=0 to 3:
  1023 #    |    L_result = 0;
  1024 #    |==== FOR k=0 to 12:
  1025 #    |         temp1 = x[i+(3*k)] >> 2;
  1026 #    |         L_temp = L_mult( temp1, temp1 );
  1027 #    |         L_result = L_add( L_temp, L_result );
  1028 #    |==== NEXT i:
  1029 #    |    IF ( L_result > L_max ) THEN
  1030 #    |                                  |    Mc = m;
  1031 #    |                                  |    EM = L_result;
  1032 #    |== NEXT m:
  1033 */  
  1034   for ( i = 0; i <= 3; i++ ) {
  1035     L_result = 0;
  1036     for ( k = 0; k <= 12; k++ ) {
  1037       temp1 = shr( x[i+(3*k)], 2 );
  1038       L_result = L_mac( L_result, temp1, temp1 );
  1039     }
  1040     if ( L_sub( L_result, L_EM ) > 0 ) {
  1041       Mc = int2 (i);
  1042       L_EM = L_result;
  1043     }
  1044   }
  1045 /*
  1046 # Down-sampling by factor 3 to get the selected xM[0..12] RPE sequence
  1047 #    |== FOR i=0 to 12:
  1048 #    |    xM[k] = x[Mc+(3*i)];
  1049 #    |== NEXT i:
  1050 */
  1051   for ( k = 0; k <= 12; k++ )
  1052     xM[k] = x[Mc+(3*k)];
  1053 
  1054   return Mc;
  1055 }
  1056 
  1057 
  1058 /*
  1059 # Compute exponent and mantissa of the decoded version of xmaxc
  1060 #
  1061 # Part of APCM and (subrogram apcm() InvAPCM (iapcm())
  1062 */
  1063 
  1064 void expman( int2 *Exp, int2 *mant, int2 xmaxc )
  1065 {
  1066   int i;
  1067 /*
  1068 # Compute exponent and mantissa of the decoded version of xmaxc.
  1069 #
  1070 #    exp = 0;
  1071 #    IF ( xmaxc > 15 ) THEN exp = sub( ( xmaxc >> 3 ), 1 );
  1072 #    mant = sub( xmaxc, ( exp << 3 ) );
  1073 */
  1074   *Exp = 0;
  1075   if ( sub( xmaxc, 15 ) > 0 )
  1076     *Exp = sub( shr( xmaxc, 3 ), 1 );
  1077   
  1078   *mant = sub( xmaxc, shl( *Exp, 3 ) );
  1079 /*
  1080 # Normalize mantissa 0 <= mant <= 7.
  1081 #    IF ( mant == 0 ) THEN    |    exp = -4;
  1082 #                             |    mant = 15 ;
  1083 #    ELSE | itest = 0;
  1084 #         |== FOR i=0 to 2:
  1085 #         |    IF ( mant > 7 ) THEN itest = 1;
  1086 #         |    IF ( itest == 0 ) THEN mant = add( ( mant << 1 ), 1 );
  1087 #         |    IF ( itest == 0 ) THEN exp = sub( exp, 1 );
  1088 #         |== NEXT i:
  1089 */
  1090   if ( *mant == 0 ) {
  1091     *Exp = -4;
  1092     *mant = 15 ;
  1093   }
  1094   else {
  1095     for ( i = 0; i <= 2; i++ ) {
  1096       if ( sub( *mant, 7 ) > 0 )
  1097 	break;
  1098       else {
  1099 	*mant = add( shl( *mant, 1 ), 1 );
  1100 	*Exp = sub( *Exp, 1 );
  1101       }
  1102     }
  1103   }
  1104 /*
  1105 #    mant = sub( mant, 8 );
  1106 */
  1107   *mant = sub( *mant, 8 );
  1108 }
  1109 
  1110 
  1111 int2 quantize_xmax( int2 xmax )
  1112 {
  1113   int i;
  1114 
  1115   int2 Exp;
  1116   int2 temp;
  1117   int2 itest;
  1118 /*
  1119 # Quantizing and coding of xmax to get xmaxc.
  1120 #    exp = 0;
  1121 #    temp = xmax >> 9;
  1122 #    itest = 0;
  1123 #    |== FOR i=0 to 5:
  1124 #    |    IF ( temp <= 0 ) THEN itest = 1;
  1125 #    |    temp = temp >> 1;
  1126 #    |    IF ( itest == 0 ) THEN exp = add( exp, 1 )  ;
  1127 #    |== NEXT i:
  1128 */
  1129   Exp = 0;
  1130   temp = shr( xmax, 9 );
  1131   itest = 0;
  1132   for ( i = 0; i <= 5; i++ ) {
  1133     if ( temp <= 0 )
  1134       itest = 1;
  1135     temp = shr( temp, 1 );
  1136     if ( itest == 0 )
  1137       Exp = add( Exp, 1 )  ;
  1138   }
  1139 
  1140 /*
  1141 #    temp = add( exp, 5 );
  1142 #    xmaxc = add( ( xmax >> temp ), ( exp << 3 ) );
  1143 */
  1144   temp = add( Exp, 5 );
  1145 
  1146   return ( add( shr( xmax, temp ), shl( Exp, 3 ) ) ); /* xmaxc */
  1147 
  1148 }
  1149 
  1150 
  1151 /*
  1152 #  4.2.15. APCM quantization of the selected RPE sequence
  1153 #
  1154 #  Keep in memory exp and mant for the following inverse APCM quantizer.
  1155 *
  1156 * return unquantzed xmax for SID computation
  1157 */
  1158 
  1159 int2 apcm( int2 *xmaxc, int2 xM[], int2 xMc[], int2 *Exp, int2 *mant )
  1160 {
  1161   int k;
  1162 
  1163   int2 temp;
  1164   int2 temp1;
  1165   int2 temp2;
  1166   int2 temp3;
  1167   int2 xmax;
  1168 /*
  1169 # Find the maximum absolute value of xM[0..12].
  1170 #    xmax = 0;
  1171 #    |== FOR k=0 to 12:
  1172 #    |    temp = abs( xM[k] );
  1173 #    |    IF ( temp > xmax ) THEN xmax = temp;
  1174 #    |== NEXT i:
  1175 */
  1176   xmax = 0;
  1177   for ( k = 0; k <= 12; k++ ) {
  1178     temp = abs_s( xM[k] );
  1179      if ( sub( temp, xmax ) > 0 )
  1180        xmax = temp;
  1181   }
  1182 
  1183   /*
  1184    * quantization of xmax moved to function because it is used
  1185    * also in comfort noise generation
  1186    */
  1187   *xmaxc = quantize_xmax( xmax );
  1188 
  1189   expman( Exp, mant, *xmaxc ); /* compute exp. and mant. */
  1190 /*
  1191 # Quantizing and coding of the xM[0..12] RPE sequence to get the xMc[0..12]
  1192 #
  1193 # This computation uses the fact that the decoded version of xmaxc can
  1194 # be calculated by using the exponent and mantissa part of xmaxc
  1195 # (logarithmic table).
  1196 #
  1197 # So, this method avoids any division and uses only scaling of the RPE
  1198 # samples by a function of the exponent. A direct multiplication by the
  1199 # inverse of the mantissa (NRFAC[0..7] found in table 4.5) gives the 3
  1200 # bit coded version xMc[0..12} of the RPE samples.
  1201 #
  1202 # Direct computation of xMc[0..12] using table 4.5.
  1203 #    temp1 = sub( 6, exp );   /normalization by the exponent/
  1204 #    temp2 = NRFAC[mant];     /see table 4.5 (inverse mantissa)/
  1205 #    |== FOR k=0 to 12:
  1206 #    |    xM[k] = xM[k] << temp1;
  1207 #    |    xM[k] = mult( xM[k], temp2 );
  1208 #    |    xMc[k] = add( ( xM[k] >> 12 ), 4 );     / See note below/
  1209 #    |== NEXT i:
  1210 #
  1211 # NOTE: This equation is used to make all the xMx[i] positive.
  1212 */
  1213   temp1 = sub( 6, *Exp );
  1214   temp2 = NRFAC[*mant];
  1215 
  1216   for ( k = 0; k <= 12; k++ )  {
  1217     temp3 = shl( xM[k], temp1 );
  1218     temp3 = mult( temp3, temp2 );
  1219     xMc[k] = add( shr( temp3, 12 ), 4 );
  1220   }
  1221 
  1222   return xmax;
  1223 }
  1224 
  1225 /*
  1226 #  4.2.16. APCM inverse quantization
  1227 #
  1228 #  This part is for decoding the RPE sequence of coded xMc[0..12] samples
  1229 #  to obtain the xMp[0..12] array. Table 4.6 is used to get the mantissa
  1230 #  of xmaxc (FAC[0..7]).
  1231 */
  1232 
  1233 void iapcm( int2 xMp[], int2 xMc[], int2 Exp, int2 mant )
  1234 {
  1235   //ALEX//extern int2 FAC[];
  1236 
  1237   int k;
  1238 
  1239   int2 temp;
  1240   int2 temp1;
  1241   int2 temp2;
  1242   int2 temp3;
  1243 /*
  1244 #    temp1 = FAC[mant];       /See 4.2.15 for mant/
  1245 #    temp2 = sub( 6, exp );   /See 4.2.15 for exp/
  1246 #    temp3 = 1 << sub( temp2, 1 );
  1247 */
  1248   temp1 = FAC[mant];
  1249   temp2 = sub( 6, Exp );
  1250   temp3 = shl( 1, sub( temp2, 1 ) );
  1251 /*
  1252 #    |== FOR k=0 to 12:
  1253 #    |    temp = sub( ( xMc[k] << 1 ), 7 );  /See note below/
  1254 #    |    temp = temp << 12;
  1255 #    |    temp = mult_r( temp1, temp );
  1256 #    |    temp = add( temp, temp3 );
  1257 #    |    xMp[k] = temp >> temp2;
  1258 #    |== NEXT i:
  1259 #
  1260 # NOTE: This subtraction is used to restore the sign of xMc[i].
  1261 */
  1262   for ( k = 0; k <= 12; k++ ) {
  1263     temp = sub( shl( xMc[k], 1 ), 7 );
  1264     temp = shl( temp, 12 );
  1265     temp = mult_r( temp1, temp );
  1266     temp = add( temp, temp3 );
  1267     xMp[k] = shr( temp, temp2 );
  1268   }
  1269 }
  1270 
  1271 /*
  1272 #  4.2.17. RPE grid positioning
  1273 #
  1274 #  This procedure computes the reconstructed long term residual signal
  1275 #  ep[0..39] for the LTP analysis filter. The inputs are the Mc which is
  1276 #  the grid position selection and the xMp[0..12] decoded RPE samples
  1277 #  which are upsampled by factor of 3 by inserting zero values.
  1278 */
  1279 
  1280 void gridpos( int2 ep[], int2 xMp[], int2 Mc )
  1281 {
  1282   int k;
  1283 /*
  1284 #    |== FOR k=0 to 39:
  1285 #    |    ep[k] = 0;
  1286 #    |== NEXT k:
  1287 */
  1288   for ( k = 0; k <= 39; k++ ) 
  1289     ep[k] = 0;
  1290 /*
  1291 #    |== FOR i=0 to 12:
  1292 #    |    ep[Mc + (3*k)] = xMp[k];
  1293 #    |== NEXT i:
  1294 */
  1295   for ( k = 0; k <= 12; k++ ) 
  1296     ep[Mc + (3*k)] = xMp[k];
  1297 }
  1298 
  1299 
  1300 /*
  1301 #  4.2.18. Update of the reconstructed short term residual signal dp[]
  1302 #
  1303 #  Keep the array dp[-120..-1] in memory for the next sub-segment.
  1304 #  Initial value: dp[-120..-1]=0;
  1305 */
  1306 
  1307 void ltpupd( CGSM610FR_Encoder* aEncoder, int2 dpp[], int2 ep[] )
  1308 {
  1309   int i;
  1310 /*
  1311 #    |== FOR k=0 to 79:
  1312 #    |    dp[-120+k] = dp[-80+k];
  1313 #    |== NEXT k:
  1314 */
  1315   for (i = 0; i <= 79; i++) 
  1316     aEncoder->dp[-120+i+120] = aEncoder->dp[-80+i+120];
  1317 /*
  1318 #    |== FOR k=0 to 39:
  1319 #    |    dp[-40+k] = add( ep[k], dpp[k] );
  1320 #    |== NEXT k:
  1321 */
  1322   for (i = 0; i <= 39; i++) 
  1323     aEncoder->dp[-40+i+120] = add( ep[i], dpp[i] );
  1324 }
  1325 
  1326 
  1327 /*
  1328 #  4.3.2. Long term synthesis filtering
  1329 #
  1330 #  Keep the nrp value for the next sub-segment.
  1331 #  Initial value: nrp=40;
  1332 #
  1333 #  Keep the array drp[-120..-1] for the next sub-segment.
  1334 #  Initial value: drp[-120..-1]=0;
  1335 */
  1336 
  1337 void ltpsyn( CGSM610FR_Decoder* aDecoder, int2 erp[], int2 wt[], int2 bcr, int2 Ncr )
  1338 {
  1339   int k, i;
  1340 
  1341   int2 drpp;
  1342   int2 Nr;
  1343   int2 brp;
  1344 /*
  1345 # Check the limits of Nr
  1346 #    Nr = Ncr;
  1347 #    IF ( Ncr < 40 ) THEN Nr = nrp;
  1348 #    IF ( Ncr > 120 ) THEN Nr = nrp;
  1349 #    nrp = Nr;
  1350 */
  1351   if ( sub( Ncr, 40 ) < 0 )
  1352     Nr = aDecoder->nrp;
  1353   else if ( sub( Ncr, 120 ) > 0 )
  1354     Nr = aDecoder->nrp;
  1355   else
  1356     Nr = Ncr;
  1357 
  1358   aDecoder->nrp = Nr;
  1359 
  1360 /*
  1361 # Decoding of the LTP gain bcr.
  1362 #    brp = QLB[bcr];
  1363 */
  1364   brp = QLB[bcr];
  1365 /*
  1366 # Computation of the reconstructed short term residual signal drp[0..39].
  1367 #    |== FOR k=0 to 39:
  1368 #    |    drpp = mult_r( brp, drp[k-Nr] );
  1369 #    |    drp[k+120] = add( erp[k], drpp );
  1370 #    |== NEXT k:
  1371 */
  1372   for ( k = 0; k <= 39; k++ ) { 
  1373     drpp = mult_r( brp, aDecoder->drp[k-Nr+120] );
  1374     wt[k] = add( erp[k], drpp );
  1375   }
  1376 /*
  1377 # Update of the reconstructed short term residual signal drp[-1..-120]
  1378 #    |== FOR k=0 to 119:
  1379 #    |    drp[-120+k] = drp[-80+k];
  1380 #    |== NEXT k:
  1381 */
  1382 
  1383   for ( i = 0; i < 80; i++ )
  1384     aDecoder->drp[i] = aDecoder->drp[40+i];
  1385 
  1386   for ( i = 0; i < 40; i++ )
  1387     aDecoder->drp[i+80] = wt[i];
  1388 }
  1389 
  1390 
  1391 /*
  1392 #  4.3.4. Short term synthesis filtering section
  1393 #
  1394 #  This procedure uses the drp[0..39] signal and produces the sr[0..159]
  1395 #  signal which is the output of the short term synthesis filter. For
  1396 #  ease of explanation, a temporary array wt[0..159] is used.
  1397 #
  1398 #  Initialization of the array wt[0..159].
  1399 #
  1400 #  For the first sub-segment in a frame:
  1401 #    |== FOR k=0 to 39:
  1402 #    |    wt[k] = drp[k];
  1403 #    |== NEXT k:
  1404 #
  1405 #  For the second sub-segment in a frame:
  1406 #    |== FOR k=0 to 39:
  1407 #    |    wt[40+k] = drp[k];
  1408 #    |== NEXT k:
  1409 #
  1410 #  For the third sub-segment in a frame:
  1411 #    |== FOR k=0 to 39:
  1412 #    |    wt[80+k] = drp[k];
  1413 #    |== NEXT k:
  1414 #
  1415 #  For the fourth sub-segment in a frame:
  1416 #    |== FOR k=0 to 39:
  1417 #    |    wt[120+k] = drp[k];
  1418 #    |== NEXT k:
  1419 #
  1420 #  As the call of the short term synthesis filter procedure can be done
  1421 #  in many ways (see the interpolation of the LAR coefficient), it is
  1422 #  assumed that the computation begins with index k_start (for arrays
  1423 #  wt[..] and sr[..]) and stops with index k_end (k_start and k_end are
  1424 #  defined in 4.2.9.1). The procedure also needs to keep the array
  1425 #  v[0..8] in memory between calls.
  1426 #
  1427 #  Keep the array v[0..8] in memory for the next call.
  1428 #  Initial value: v[0..8]=0;
  1429 */
  1430 
  1431 void synfil( CGSM610FR_Decoder* aDecoder, int2 sr[], int2 wt[], int2 rrp[], int k_start, int k_end )
  1432 {
  1433   int k;
  1434   int i;
  1435 
  1436   int2 sri;
  1437 /*
  1438 #    |== FOR k=k_start to k_end:
  1439 #    |    sri = wt[k];
  1440 #    |==== FOR i=1 to 8:
  1441 #    |         sri = sub( sri, mult_r( rrp[9-i], v[8-i] ) );
  1442 #    |         v[9-i] = add( v[8-i], mult_r( rrp[9-i], sri ) ) ;
  1443 #    |==== NEXT i:
  1444 #    |    sr[k] = sri;
  1445 #    |    v[0] = sri;
  1446 #    |== NEXT k:
  1447 */
  1448   for ( k = k_start; k <= k_end; k++ ) {
  1449     sri = wt[k];
  1450     for ( i = 1; i <= 8; i++ ) {
  1451 	  int j = i+1;
  1452       sri = sub( sri, mult_r( rrp[9-j], aDecoder->v[8-i] ) );
  1453       aDecoder->v[9-i] = add( aDecoder->v[8-i], mult_r( rrp[9-j], sri ) ) ;
  1454     }
  1455     sr[k] = sri;
  1456     aDecoder->v[0] = sri;
  1457   }
  1458 
  1459 }
  1460 
  1461 
  1462 /*
  1463 ** 4.3.5., 4.3.6., 4.3.7. Postprocessing
  1464 **
  1465 ** Combined deemphasis, upscaling and truncation
  1466 */
  1467 void postpr( CGSM610FR_Decoder* aDecoder, int2 srop[], int2 sr[] )
  1468 {
  1469   int k;
  1470 /*
  1471 # 4.3.5. Deemphasis filtering
  1472 #
  1473 # Keep msr in memory for the next frame.
  1474 # Initial value: msr=0;
  1475 */
  1476 /*
  1477 #    |== FOR k=0 to 159:
  1478 #    |    temp = add( sr[k], mult_r( msr, 28180 ) );
  1479 #    |    msr = temp;
  1480 #    |    sro[k] = msr;
  1481 #    |== NEXT k:
  1482 */
  1483 /*
  1484 # 4.3.6 Upscaling of the output signal
  1485 */
  1486 /*
  1487 #    |== FOR k=0 to 159:
  1488 #    |    srop[k] = add( sro[k], sro[k] );
  1489 #    |== NEXT k:
  1490 */
  1491 /*
  1492 # 4.3.7. Truncation of the output variable
  1493 */
  1494 /*
  1495 #    |== FOR k=0 to 159:
  1496 #    |    srop[k] = srop[k] >> 3;
  1497 #    |    srop[k] = srop[k] << 3;
  1498 #    |== NEXT k:
  1499 */
  1500 
  1501   for ( k = 0; k <= 159; k++ ) {
  1502     aDecoder->msr = add( sr[k], mult_r( aDecoder->msr, 28180 ) );
  1503     srop[k] = int2 (shl( aDecoder->msr, 1 ) & 0xfff8);
  1504   }
  1505 }
  1506