1 /********************************************************************
3 * THIS FILE IS PART OF THE Ogg Vorbis SOFTWARE CODEC SOURCE CODE. *
4 * USE, DISTRIBUTION AND REPRODUCTION OF THIS SOURCE IS GOVERNED BY *
5 * THE GNU PUBLIC LICENSE 2, WHICH IS INCLUDED WITH THIS SOURCE. *
6 * PLEASE READ THESE TERMS DISTRIBUTING. *
8 * THE OggSQUISH SOURCE CODE IS (C) COPYRIGHT 1994-2000 *
9 * by Monty <monty@xiph.org> and The XIPHOPHORUS Company *
10 * http://www.xiph.org/ *
12 ********************************************************************
14 function: LSP (also called LSF) conversion routines
15 last mod: $Id: lsp.c,v 1.10 2000/10/12 03:12:53 xiphmont Exp $
17 The LSP generation code is taken (with minimal modification) from
18 "On the Computation of the LSP Frequencies" by Joseph Rothweiler
19 <rothwlr@altavista.net>, available at:
21 http://www2.xtdl.com/~rothwlr/lsfpaper/lsfpage.html
23 ********************************************************************/
25 /* Note that the lpc-lsp conversion finds the roots of polynomial with
26 an iterative root polisher (CACM algorithm 283). It *is* possible
27 to confuse this algorithm into not converging; that should only
28 happen with absurdly closely spaced roots (very sharp peaks in the
29 LPC f response) which in turn should be impossible in our use of
30 the code. If this *does* happen anyway, it's a bug in the floor
31 finder; find the cause of the confusion (probably a single bin
32 spike or accidental near-float-limit resolution problems) and
44 /* three possible LSP to f curve functions; the exact computation
45 (float), a lookup based float implementation, and an integer
46 implementation. The float lookup is likely the optimal choice on
47 any machine with an FPU. The integer implementation is *not* fixed
48 point (due to the need for a large dynamic range and thus a
49 seperately tracked exponent) and thus much more complex than the
50 relatively simple float implementations. It's mostly for future
51 work on a fully fixed point implementation for processors like the
54 /* undefine both for the 'old' but more precise implementation */
59 #include "lookup.c" /* catch this in the build system; we #include for
60 compilers (like gcc) that can't inline across
63 /* side effect: changes *lsp to cosines of lsp */
64 void vorbis_lsp_to_curve(float *curve,int *map,int n,int ln,float *lsp,int m,
65 float amp,float ampoffset){
68 for(i=0;i<m;i++)lsp[i]=vorbis_coslook(lsp[i]);
76 float w=vorbis_coslook(wdel*k);
78 for(j=0;j<m;j+=2) p *= lsp[j]-w;
79 for(j=1;j<m;j+=2) q *= lsp[j]-w;
81 q=frexp(p*p*(1.+w)+q*q*(1.-w),&qexp);
82 q=vorbis_fromdBlook(amp*
84 vorbis_invsq2explook(qexp+m)-
88 while(map[i]==k)curve[i++]=q;
95 #include "lookup.c" /* catch this in the build system; we #include for
96 compilers (like gcc) that can't inline across
99 static int MLOOP_1[64]={
100 0,10,11,11, 12,12,12,12, 13,13,13,13, 13,13,13,13,
101 14,14,14,14, 14,14,14,14, 14,14,14,14, 14,14,14,14,
102 15,15,15,15, 15,15,15,15, 15,15,15,15, 15,15,15,15,
103 15,15,15,15, 15,15,15,15, 15,15,15,15, 15,15,15,15,
106 static int MLOOP_2[64]={
107 0,4,5,5, 6,6,6,6, 7,7,7,7, 7,7,7,7,
108 8,8,8,8, 8,8,8,8, 8,8,8,8, 8,8,8,8,
109 9,9,9,9, 9,9,9,9, 9,9,9,9, 9,9,9,9,
110 9,9,9,9, 9,9,9,9, 9,9,9,9, 9,9,9,9,
113 static int MLOOP_3[8]={0,1,2,2,3,3,3,3};
116 /* side effect: changes *lsp to cosines of lsp */
117 void vorbis_lsp_to_curve(float *curve,int *map,int n,int ln,float *lsp,int m,
118 float amp,float ampoffset){
122 /* set up for using all int later */
124 int ampoffseti=rint(ampoffset*4096.);
125 int ampi=rint(amp*16.);
126 long *ilsp=alloca(m*sizeof(long));
127 for(i=0;i<m;i++)ilsp[i]=vorbis_coslook_i(lsp[i]/M_PI*65536.+.5);
132 unsigned long pi=46341; /* 2**-.5 in 0.16 */
133 unsigned long qi=46341;
135 long wi=vorbis_coslook_i(k*65536/ln);
137 pi*=labs(ilsp[0]-wi);
138 qi*=labs(ilsp[1]-wi);
141 if(!(shift=MLOOP_1[(pi|qi)>>25]))
142 if(!(shift=MLOOP_2[(pi|qi)>>19]))
143 shift=MLOOP_3[(pi|qi)>>16];
144 pi=(pi>>shift)*labs(ilsp[j]-wi);
145 qi=(qi>>shift)*labs(ilsp[j+1]-wi);
148 if(!(shift=MLOOP_1[(pi|qi)>>25]))
149 if(!(shift=MLOOP_2[(pi|qi)>>19]))
150 shift=MLOOP_3[(pi|qi)>>16];
155 /* pi,qi normalized collectively, both tracked using qexp */
157 /* p*=p(1-w), q*=q(1+w), let normalization drift because it isn't
158 worth tracking step by step */
169 /* we've let the normalization drift because it wasn't important;
170 however, for the lookup, things must be normalized again. We
171 need at most one right shift or a number of left shifts */
173 if(qi&0xffff0000){ /* checks for 1.xxxxxxxxxxxxxxxx */
176 while(qi && !(qi&0x8000)){ /* checks for 0.0xxxxxxxxxxxxxxx or less*/
180 amp=vorbis_fromdBlook_i(ampi* /* n.4 */
181 vorbis_invsqlook_i(qi,qexp)-
183 ampoffseti); /* 8.12[0] */
186 while(map[++i]==k)curve[i]=amp;
192 /* old, nonoptimized but simple version for any poor sap who needs to
193 figure out what the hell this code does, or wants the other tiny
194 fraction of a dB precision */
196 /* side effect: changes *lsp to cosines of lsp */
197 void vorbis_lsp_to_curve(float *curve,int *map,int n,int ln,float *lsp,int m,
198 float amp,float ampoffset){
201 for(i=0;i<m;i++)lsp[i]=2*cos(lsp[i]);
208 float w=2*cos(wdel*k);
215 q=fromdB(amp/sqrt(p+q)-ampoffset);
218 while(map[++i]==k)curve[i]=q;
225 static void cheby(float *g, int ord) {
229 for(i=2; i<= ord; i++) {
230 for(j=ord; j >= i; j--) {
237 static int comp(const void *a,const void *b){
238 if(*(float *)a<*(float *)b)
244 /* This is one of those 'mathemeticians should not write code' kind of
245 cases. Newton's method of polishing roots is straightforward
246 enough... except in those cases where it just fails in the real
247 world. In our case below, we're worried about a local mini/maxima
248 shooting a root estimation off to infinity, or the new estimation
249 chaotically oscillating about convergence (shouldn't actually be a
250 problem in our usage.
252 Maehly's modification (zero suppression, to prevent two tenative
253 roots from collapsing to the same actual root) similarly can
254 temporarily shoot a root off toward infinity. It would come
255 back... if it were not for the fact that machine representation has
256 limited dynamic range and resolution. This too is guarded by
259 Last problem is convergence criteria; we don't know what a 'double'
260 is on our hardware/compiler, and the convergence limit is bounded
261 by roundoff noise. So, we hack convergence:
263 Require at most 1e-6 mean squared error for all zeroes. When
264 converging, start the clock ticking at 1e-6; limit our polishing to
265 as many more iterations as took us to get this far, 100 max.
267 Past max iters, quit when MSE is no longer decreasing *or* we go
268 below ~1e-20 MSE, whichever happens first. */
270 static void Newton_Raphson_Maehly(float *a,int ord,float *r){
271 int i, k, count=0, maxiter=0;
272 double error=1.,besterror=1.;
273 double *root=alloca(ord*sizeof(double));
275 for(i=0; i<ord;i++) root[i] = 2.0 * (i+0.5) / ord - 1.0;
280 for(i=0; i<ord; i++) { /* Update each point. */
281 double ac=0.,pp=0.,delta;
282 double rooti=root[i];
284 for(k=ord-1; k>= 0; k--) {
288 if (k != i) ac += 1./(rooti - root[k]);
294 /* don't allow the correction to scream off into infinity if we
295 happened to polish right at a local mini/maximum */
297 if(delta<-3)delta=-3;
298 if(delta>3.)delta=3.; /* 3 is not a random choice; it's large
299 enough to make sure the first pass
300 can't accidentally limit two poles to
301 the same value in a fatal nonelastic
305 error += delta*delta;
308 if(maxiter && count>maxiter && error>=besterror)break;
310 /* anything to help out the polisher; converge using doubles */
311 if(!count || error<besterror){
312 for(i=0; i<ord; i++) r[i]=root[i];
314 if(error<1.e-6){ /* rough minimum criteria */
316 if(maxiter>100)maxiter=100;
323 /* Replaced the original bubble sort with a real sort. With your
324 help, we can eliminate the bubble sort in our lifetime. --Monty */
326 qsort(r,ord,sizeof(float),comp);
330 /* Convert lpc coefficients to lsp coefficients */
331 void vorbis_lpc_to_lsp(float *lpc,float *lsp,int m){
333 float *g1=alloca(sizeof(float)*(order2+1));
334 float *g2=alloca(sizeof(float)*(order2+1));
335 float *g1r=alloca(sizeof(float)*(order2+1));
336 float *g2r=alloca(sizeof(float)*(order2+1));
339 /* Compute the lengths of the x polynomials. */
340 /* Compute the first half of K & R F1 & F2 polynomials. */
341 /* Compute half of the symmetric and antisymmetric polynomials. */
342 /* Remove the roots at +1 and -1. */
345 for(i=0;i<order2;i++) g1[order2-i-1] = lpc[i]+lpc[m-i-1];
347 for(i=0;i<order2;i++) g2[order2-i-1] = lpc[i]-lpc[m-i-1];
349 for(i=0; i<order2;i++) g1[order2-i-1] -= g1[order2-i];
350 for(i=0; i<order2;i++) g2[order2-i-1] += g2[order2-i];
352 /* Convert into polynomials in cos(alpha) */
356 /* Find the roots of the 2 even polynomials.*/
358 Newton_Raphson_Maehly(g1,order2,g1r);
359 Newton_Raphson_Maehly(g2,order2,g2r);
362 lsp[i] = acos(g1r[i/2]);
363 lsp[i+1] = acos(g2r[i/2]);