Updated our source code header to explicitly mention that we are GPL v2 or
[Rockbox.git] / apps / codecs / libspeex / lsp.c
blob3973554cfbae81fec033037a0ba0681a44d7a860
1 /*---------------------------------------------------------------------------*\
2 Original copyright
3 FILE........: lsp.c
4 AUTHOR......: David Rowe
5 DATE CREATED: 24/2/93
7 Heavily modified by Jean-Marc Valin (c) 2002-2006 (fixed-point,
8 optimizations, additional functions, ...)
10 This file contains functions for converting Linear Prediction
11 Coefficients (LPC) to Line Spectral Pair (LSP) and back. Note that the
12 LSP coefficients are not in radians format but in the x domain of the
13 unit circle.
15 Speex License:
17 Redistribution and use in source and binary forms, with or without
18 modification, are permitted provided that the following conditions
19 are met:
21 - Redistributions of source code must retain the above copyright
22 notice, this list of conditions and the following disclaimer.
24 - Redistributions in binary form must reproduce the above copyright
25 notice, this list of conditions and the following disclaimer in the
26 documentation and/or other materials provided with the distribution.
28 - Neither the name of the Xiph.org Foundation nor the names of its
29 contributors may be used to endorse or promote products derived from
30 this software without specific prior written permission.
32 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
33 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
34 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
35 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
36 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
37 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
38 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
39 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
40 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
41 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
42 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
45 /*---------------------------------------------------------------------------*\
47 Introduction to Line Spectrum Pairs (LSPs)
48 ------------------------------------------
50 LSPs are used to encode the LPC filter coefficients {ak} for
51 transmission over the channel. LSPs have several properties (like
52 less sensitivity to quantisation noise) that make them superior to
53 direct quantisation of {ak}.
55 A(z) is a polynomial of order lpcrdr with {ak} as the coefficients.
57 A(z) is transformed to P(z) and Q(z) (using a substitution and some
58 algebra), to obtain something like:
60 A(z) = 0.5[P(z)(z+z^-1) + Q(z)(z-z^-1)] (1)
62 As you can imagine A(z) has complex zeros all over the z-plane. P(z)
63 and Q(z) have the very neat property of only having zeros _on_ the
64 unit circle. So to find them we take a test point z=exp(jw) and
65 evaluate P (exp(jw)) and Q(exp(jw)) using a grid of points between 0
66 and pi.
68 The zeros (roots) of P(z) also happen to alternate, which is why we
69 swap coefficients as we find roots. So the process of finding the
70 LSP frequencies is basically finding the roots of 5th order
71 polynomials.
73 The root so P(z) and Q(z) occur in symmetrical pairs at +/-w, hence
74 the name Line Spectrum Pairs (LSPs).
76 To convert back to ak we just evaluate (1), "clocking" an impulse
77 thru it lpcrdr times gives us the impulse response of A(z) which is
78 {ak}.
80 \*---------------------------------------------------------------------------*/
82 #ifdef HAVE_CONFIG_H
83 #include "config-speex.h"
84 #endif
86 #include <math.h>
87 #include "lsp.h"
88 #include "stack_alloc.h"
89 #include "math_approx.h"
91 #ifndef M_PI
92 #define M_PI 3.14159265358979323846 /* pi */
93 #endif
95 #ifndef NULL
96 #define NULL 0
97 #endif
99 #ifdef FIXED_POINT
101 #define FREQ_SCALE 16384
103 /*#define ANGLE2X(a) (32768*cos(((a)/8192.)))*/
104 #define ANGLE2X(a) (SHL16(spx_cos(a),2))
106 /*#define X2ANGLE(x) (acos(.00006103515625*(x))*LSP_SCALING)*/
107 #define X2ANGLE(x) (spx_acos(x))
109 #ifdef BFIN_ASM
110 #include "lsp_bfin.h"
111 #endif
113 #else
115 /*#define C1 0.99940307
116 #define C2 -0.49558072
117 #define C3 0.03679168*/
119 #define FREQ_SCALE 1.
120 #define ANGLE2X(a) (spx_cos(a))
121 #define X2ANGLE(x) (acos(x))
123 #endif
126 /*---------------------------------------------------------------------------*\
128 FUNCTION....: cheb_poly_eva()
130 AUTHOR......: David Rowe
131 DATE CREATED: 24/2/93
133 This function evaluates a series of Chebyshev polynomials
135 \*---------------------------------------------------------------------------*/
137 #ifndef SPEEX_DISABLE_ENCODER
139 #ifdef FIXED_POINT
141 #ifndef OVERRIDE_CHEB_POLY_EVA
142 static inline spx_word32_t cheb_poly_eva(
143 spx_word16_t *coef, /* P or Q coefs in Q13 format */
144 spx_word16_t x, /* cos of freq (-1.0 to 1.0) in Q14 format */
145 int m, /* LPC order/2 */
146 char *stack
149 int i;
150 spx_word16_t b0, b1;
151 spx_word32_t sum;
153 /*Prevents overflows*/
154 if (x>16383)
155 x = 16383;
156 if (x<-16383)
157 x = -16383;
159 /* Initialise values */
160 b1=16384;
161 b0=x;
163 /* Evaluate Chebyshev series formulation usin g iterative approach */
164 sum = ADD32(EXTEND32(coef[m]), EXTEND32(MULT16_16_P14(coef[m-1],x)));
165 for(i=2;i<=m;i++)
167 spx_word16_t tmp=b0;
168 b0 = SUB16(MULT16_16_Q13(x,b0), b1);
169 b1 = tmp;
170 sum = ADD32(sum, EXTEND32(MULT16_16_P14(coef[m-i],b0)));
173 return sum;
175 #endif
177 #else
179 static float cheb_poly_eva(spx_word32_t *coef, spx_word16_t x, int m, char *stack)
181 int k;
182 float b0, b1, tmp;
184 /* Initial conditions */
185 b0=0; /* b_(m+1) */
186 b1=0; /* b_(m+2) */
188 x*=2;
190 /* Calculate the b_(k) */
191 for(k=m;k>0;k--)
193 tmp=b0; /* tmp holds the previous value of b0 */
194 b0=x*b0-b1+coef[m-k]; /* b0 holds its new value based on b0 and b1 */
195 b1=tmp; /* b1 holds the previous value of b0 */
198 return(-b1+.5*x*b0+coef[m]);
200 #endif
202 /*---------------------------------------------------------------------------*\
204 FUNCTION....: lpc_to_lsp()
206 AUTHOR......: David Rowe
207 DATE CREATED: 24/2/93
209 This function converts LPC coefficients to LSP
210 coefficients.
212 \*---------------------------------------------------------------------------*/
214 #ifdef FIXED_POINT
215 #define SIGN_CHANGE(a,b) (((a)&0x70000000)^((b)&0x70000000)||(b==0))
216 #else
217 #define SIGN_CHANGE(a,b) (((a)*(b))<0.0)
218 #endif
221 int lpc_to_lsp (spx_coef_t *a,int lpcrdr,spx_lsp_t *freq,int nb,spx_word16_t delta, char *stack)
222 /* float *a lpc coefficients */
223 /* int lpcrdr order of LPC coefficients (10) */
224 /* float *freq LSP frequencies in the x domain */
225 /* int nb number of sub-intervals (4) */
226 /* float delta grid spacing interval (0.02) */
230 spx_word16_t temp_xr,xl,xr,xm=0;
231 spx_word32_t psuml,psumr,psumm,temp_psumr/*,temp_qsumr*/;
232 int i,j,m,flag,k;
233 VARDECL(spx_word32_t *Q); /* ptrs for memory allocation */
234 VARDECL(spx_word32_t *P);
235 VARDECL(spx_word16_t *Q16); /* ptrs for memory allocation */
236 VARDECL(spx_word16_t *P16);
237 spx_word32_t *px; /* ptrs of respective P'(z) & Q'(z) */
238 spx_word32_t *qx;
239 spx_word32_t *p;
240 spx_word32_t *q;
241 spx_word16_t *pt; /* ptr used for cheb_poly_eval()
242 whether P' or Q' */
243 int roots=0; /* DR 8/2/94: number of roots found */
244 flag = 1; /* program is searching for a root when,
245 1 else has found one */
246 m = lpcrdr/2; /* order of P'(z) & Q'(z) polynomials */
248 /* Allocate memory space for polynomials */
249 ALLOC(Q, (m+1), spx_word32_t);
250 ALLOC(P, (m+1), spx_word32_t);
252 /* determine P'(z)'s and Q'(z)'s coefficients where
253 P'(z) = P(z)/(1 + z^(-1)) and Q'(z) = Q(z)/(1-z^(-1)) */
255 px = P; /* initialise ptrs */
256 qx = Q;
257 p = px;
258 q = qx;
260 #ifdef FIXED_POINT
261 *px++ = LPC_SCALING;
262 *qx++ = LPC_SCALING;
263 for(i=0;i<m;i++){
264 *px++ = SUB32(ADD32(EXTEND32(a[i]),EXTEND32(a[lpcrdr-i-1])), *p++);
265 *qx++ = ADD32(SUB32(EXTEND32(a[i]),EXTEND32(a[lpcrdr-i-1])), *q++);
267 px = P;
268 qx = Q;
269 for(i=0;i<m;i++)
271 /*if (fabs(*px)>=32768)
272 speex_warning_int("px", *px);
273 if (fabs(*qx)>=32768)
274 speex_warning_int("qx", *qx);*/
275 *px = PSHR32(*px,2);
276 *qx = PSHR32(*qx,2);
277 px++;
278 qx++;
280 /* The reason for this lies in the way cheb_poly_eva() is implemented for fixed-point */
281 P[m] = PSHR32(P[m],3);
282 Q[m] = PSHR32(Q[m],3);
283 #else
284 *px++ = LPC_SCALING;
285 *qx++ = LPC_SCALING;
286 for(i=0;i<m;i++){
287 *px++ = (a[i]+a[lpcrdr-1-i]) - *p++;
288 *qx++ = (a[i]-a[lpcrdr-1-i]) + *q++;
290 px = P;
291 qx = Q;
292 for(i=0;i<m;i++){
293 *px = 2**px;
294 *qx = 2**qx;
295 px++;
296 qx++;
298 #endif
300 px = P; /* re-initialise ptrs */
301 qx = Q;
303 /* now that we have computed P and Q convert to 16 bits to
304 speed up cheb_poly_eval */
306 ALLOC(P16, m+1, spx_word16_t);
307 ALLOC(Q16, m+1, spx_word16_t);
309 for (i=0;i<m+1;i++)
311 P16[i] = P[i];
312 Q16[i] = Q[i];
315 /* Search for a zero in P'(z) polynomial first and then alternate to Q'(z).
316 Keep alternating between the two polynomials as each zero is found */
318 xr = 0; /* initialise xr to zero */
319 xl = FREQ_SCALE; /* start at point xl = 1 */
321 for(j=0;j<lpcrdr;j++){
322 if(j&1) /* determines whether P' or Q' is eval. */
323 pt = Q16;
324 else
325 pt = P16;
327 psuml = cheb_poly_eva(pt,xl,m,stack); /* evals poly. at xl */
328 flag = 1;
329 while(flag && (xr >= -FREQ_SCALE)){
330 spx_word16_t dd;
331 /* Modified by JMV to provide smaller steps around x=+-1 */
332 #ifdef FIXED_POINT
333 dd = MULT16_16_Q15(delta,SUB16(FREQ_SCALE, MULT16_16_Q14(MULT16_16_Q14(xl,xl),14000)));
334 if (psuml<512 && psuml>-512)
335 dd = PSHR16(dd,1);
336 #else
337 dd=delta*(1-.9*xl*xl);
338 if (fabs(psuml)<.2)
339 dd *= .5;
340 #endif
341 xr = SUB16(xl, dd); /* interval spacing */
342 psumr = cheb_poly_eva(pt,xr,m,stack);/* poly(xl-delta_x) */
343 temp_psumr = psumr;
344 temp_xr = xr;
346 /* if no sign change increment xr and re-evaluate poly(xr). Repeat til
347 sign change.
348 if a sign change has occurred the interval is bisected and then
349 checked again for a sign change which determines in which
350 interval the zero lies in.
351 If there is no sign change between poly(xm) and poly(xl) set interval
352 between xm and xr else set interval between xl and xr and repeat till
353 root is located within the specified limits */
355 if(SIGN_CHANGE(psumr,psuml))
357 roots++;
359 psumm=psuml;
360 for(k=0;k<=nb;k++){
361 #ifdef FIXED_POINT
362 xm = ADD16(PSHR16(xl,1),PSHR16(xr,1)); /* bisect the interval */
363 #else
364 xm = .5*(xl+xr); /* bisect the interval */
365 #endif
366 psumm=cheb_poly_eva(pt,xm,m,stack);
367 /*if(psumm*psuml>0.)*/
368 if(!SIGN_CHANGE(psumm,psuml))
370 psuml=psumm;
371 xl=xm;
372 } else {
373 psumr=psumm;
374 xr=xm;
378 /* once zero is found, reset initial interval to xr */
379 freq[j] = X2ANGLE(xm);
380 xl = xm;
381 flag = 0; /* reset flag for next search */
383 else{
384 psuml=temp_psumr;
385 xl=temp_xr;
389 return(roots);
392 #endif /* SPEEX_DISABLE_ENCODER */
394 /*---------------------------------------------------------------------------*\
396 FUNCTION....: lsp_to_lpc()
398 AUTHOR......: David Rowe
399 DATE CREATED: 24/2/93
401 Converts LSP coefficients to LPC coefficients.
403 \*---------------------------------------------------------------------------*/
405 #ifdef FIXED_POINT
407 void lsp_to_lpc(spx_lsp_t *freq,spx_coef_t *ak,int lpcrdr, char *stack)
408 /* float *freq array of LSP frequencies in the x domain */
409 /* float *ak array of LPC coefficients */
410 /* int lpcrdr order of LPC coefficients */
412 int i,j;
413 spx_word32_t xout1,xout2,xin;
414 spx_word32_t mult, a;
415 VARDECL(spx_word16_t *freqn);
416 VARDECL(spx_word32_t **xp);
417 VARDECL(spx_word32_t *xpmem);
418 VARDECL(spx_word32_t **xq);
419 VARDECL(spx_word32_t *xqmem);
420 int m = lpcrdr>>1;
424 Reconstruct P(z) and Q(z) by cascading second order polynomials
425 in form 1 - 2cos(w)z(-1) + z(-2), where w is the LSP frequency.
426 In the time domain this is:
428 y(n) = x(n) - 2cos(w)x(n-1) + x(n-2)
430 This is what the ALLOCS below are trying to do:
432 int xp[m+1][lpcrdr+1+2]; // P matrix in QIMP
433 int xq[m+1][lpcrdr+1+2]; // Q matrix in QIMP
435 These matrices store the output of each stage on each row. The
436 final (m-th) row has the output of the final (m-th) cascaded
437 2nd order filter. The first row is the impulse input to the
438 system (not written as it is known).
440 The version below takes advantage of the fact that a lot of the
441 outputs are zero or known, for example if we put an inpulse
442 into the first section the "clock" it 10 times only the first 3
443 outputs samples are non-zero (it's an FIR filter).
446 ALLOC(xp, (m+1), spx_word32_t*);
447 ALLOC(xpmem, (m+1)*(lpcrdr+1+2), spx_word32_t);
449 ALLOC(xq, (m+1), spx_word32_t*);
450 ALLOC(xqmem, (m+1)*(lpcrdr+1+2), spx_word32_t);
452 for(i=0; i<=m; i++) {
453 xp[i] = xpmem + i*(lpcrdr+1+2);
454 xq[i] = xqmem + i*(lpcrdr+1+2);
457 /* work out 2cos terms in Q14 */
459 ALLOC(freqn, lpcrdr, spx_word16_t);
460 for (i=0;i<lpcrdr;i++)
461 freqn[i] = ANGLE2X(freq[i]);
463 #define QIMP 21 /* scaling for impulse */
465 xin = SHL32(EXTEND32(1), (QIMP-1)); /* 0.5 in QIMP format */
467 /* first col and last non-zero values of each row are trivial */
469 for(i=0;i<=m;i++) {
470 xp[i][1] = 0;
471 xp[i][2] = xin;
472 xp[i][2+2*i] = xin;
473 xq[i][1] = 0;
474 xq[i][2] = xin;
475 xq[i][2+2*i] = xin;
478 /* 2nd row (first output row) is trivial */
480 xp[1][3] = -MULT16_32_Q14(freqn[0],xp[0][2]);
481 xq[1][3] = -MULT16_32_Q14(freqn[1],xq[0][2]);
483 xout1 = xout2 = 0;
485 /* now generate remaining rows */
487 for(i=1;i<m;i++) {
489 for(j=1;j<2*(i+1)-1;j++) {
490 mult = MULT16_32_Q14(freqn[2*i],xp[i][j+1]);
491 xp[i+1][j+2] = ADD32(SUB32(xp[i][j+2], mult), xp[i][j]);
492 mult = MULT16_32_Q14(freqn[2*i+1],xq[i][j+1]);
493 xq[i+1][j+2] = ADD32(SUB32(xq[i][j+2], mult), xq[i][j]);
496 /* for last col xp[i][j+2] = xq[i][j+2] = 0 */
498 mult = MULT16_32_Q14(freqn[2*i],xp[i][j+1]);
499 xp[i+1][j+2] = SUB32(xp[i][j], mult);
500 mult = MULT16_32_Q14(freqn[2*i+1],xq[i][j+1]);
501 xq[i+1][j+2] = SUB32(xq[i][j], mult);
504 /* process last row to extra a{k} */
506 for(j=1;j<=lpcrdr;j++) {
507 int shift = QIMP-13;
509 /* final filter sections */
510 a = PSHR32(xp[m][j+2] + xout1 + xq[m][j+2] - xout2, shift);
511 xout1 = xp[m][j+2];
512 xout2 = xq[m][j+2];
514 /* hard limit ak's to +/- 32767 */
516 if (a < -32767) a = -32767;
517 if (a > 32767) a = 32767;
518 ak[j-1] = (short)a;
524 #else
526 void lsp_to_lpc(spx_lsp_t *freq,spx_coef_t *ak,int lpcrdr, char *stack)
527 /* float *freq array of LSP frequencies in the x domain */
528 /* float *ak array of LPC coefficients */
529 /* int lpcrdr order of LPC coefficients */
533 int i,j;
534 float xout1,xout2,xin1,xin2;
535 VARDECL(float *Wp);
536 float *pw,*n1,*n2,*n3,*n4=NULL;
537 VARDECL(float *x_freq);
538 int m = lpcrdr>>1;
540 ALLOC(Wp, 4*m+2, float);
541 pw = Wp;
543 /* initialise contents of array */
545 for(i=0;i<=4*m+1;i++){ /* set contents of buffer to 0 */
546 *pw++ = 0.0;
549 /* Set pointers up */
551 pw = Wp;
552 xin1 = 1.0;
553 xin2 = 1.0;
555 ALLOC(x_freq, lpcrdr, float);
556 for (i=0;i<lpcrdr;i++)
557 x_freq[i] = ANGLE2X(freq[i]);
559 /* reconstruct P(z) and Q(z) by cascading second order
560 polynomials in form 1 - 2xz(-1) +z(-2), where x is the
561 LSP coefficient */
563 for(j=0;j<=lpcrdr;j++){
564 int i2=0;
565 for(i=0;i<m;i++,i2+=2){
566 n1 = pw+(i*4);
567 n2 = n1 + 1;
568 n3 = n2 + 1;
569 n4 = n3 + 1;
570 xout1 = xin1 - 2.f*x_freq[i2] * *n1 + *n2;
571 xout2 = xin2 - 2.f*x_freq[i2+1] * *n3 + *n4;
572 *n2 = *n1;
573 *n4 = *n3;
574 *n1 = xin1;
575 *n3 = xin2;
576 xin1 = xout1;
577 xin2 = xout2;
579 xout1 = xin1 + *(n4+1);
580 xout2 = xin2 - *(n4+2);
581 if (j>0)
582 ak[j-1] = (xout1 + xout2)*0.5f;
583 *(n4+1) = xin1;
584 *(n4+2) = xin2;
586 xin1 = 0.0;
587 xin2 = 0.0;
591 #endif
594 #ifdef FIXED_POINT
596 /*Makes sure the LSPs are stable*/
597 void lsp_enforce_margin(spx_lsp_t *lsp, int len, spx_word16_t margin)
599 int i;
600 spx_word16_t m = margin;
601 spx_word16_t m2 = 25736-margin;
603 if (lsp[0]<m)
604 lsp[0]=m;
605 if (lsp[len-1]>m2)
606 lsp[len-1]=m2;
607 for (i=1;i<len-1;i++)
609 if (lsp[i]<lsp[i-1]+m)
610 lsp[i]=lsp[i-1]+m;
612 if (lsp[i]>lsp[i+1]-m)
613 lsp[i]= SHR16(lsp[i],1) + SHR16(lsp[i+1]-m,1);
618 void lsp_interpolate(spx_lsp_t *old_lsp, spx_lsp_t *new_lsp, spx_lsp_t *interp_lsp, int len, int subframe, int nb_subframes)
620 int i;
621 spx_word16_t tmp = DIV32_16(SHL32(EXTEND32(1 + subframe),14),nb_subframes);
622 spx_word16_t tmp2 = 16384-tmp;
623 for (i=0;i<len;i++)
625 interp_lsp[i] = MULT16_16_P14(tmp2,old_lsp[i]) + MULT16_16_P14(tmp,new_lsp[i]);
629 #else
631 /*Makes sure the LSPs are stable*/
632 void lsp_enforce_margin(spx_lsp_t *lsp, int len, spx_word16_t margin)
634 int i;
635 if (lsp[0]<LSP_SCALING*margin)
636 lsp[0]=LSP_SCALING*margin;
637 if (lsp[len-1]>LSP_SCALING*(M_PI-margin))
638 lsp[len-1]=LSP_SCALING*(M_PI-margin);
639 for (i=1;i<len-1;i++)
641 if (lsp[i]<lsp[i-1]+LSP_SCALING*margin)
642 lsp[i]=lsp[i-1]+LSP_SCALING*margin;
644 if (lsp[i]>lsp[i+1]-LSP_SCALING*margin)
645 lsp[i]= .5f* (lsp[i] + lsp[i+1]-LSP_SCALING*margin);
650 void lsp_interpolate(spx_lsp_t *old_lsp, spx_lsp_t *new_lsp, spx_lsp_t *interp_lsp, int len, int subframe, int nb_subframes)
652 int i;
653 float tmp = (1.0f + subframe)/nb_subframes;
654 for (i=0;i<len;i++)
656 interp_lsp[i] = (1-tmp)*old_lsp[i] + tmp*new_lsp[i];
660 #endif