2 * The simplest AC3 encoder
3 * Copyright (c) 2000 Fabrice Bellard.
5 * This library is free software; you can redistribute it and/or
6 * modify it under the terms of the GNU Lesser General Public
7 * License as published by the Free Software Foundation; either
8 * version 2 of the License, or (at your option) any later version.
10 * This library is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * Lesser General Public License for more details.
15 * You should have received a copy of the GNU Lesser General Public
16 * License along with this library; if not, write to the Free Software
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
22 * The simplest AC3 encoder.
25 //#define DEBUG_BITALLOC
30 typedef struct AC3EncodeContext {
36 unsigned int sample_rate;
38 unsigned int frame_size_min; /* minimum frame size in case rounding is necessary */
39 unsigned int frame_size; /* current frame size in words */
41 unsigned int frmsizecod;
42 unsigned int fscod; /* frequency */
46 short last_samples[AC3_MAX_CHANNELS][256];
47 unsigned int chbwcod[AC3_MAX_CHANNELS];
48 int nb_coefs[AC3_MAX_CHANNELS];
50 /* bitrate allocation control */
51 int sgaincod, sdecaycod, fdecaycod, dbkneecod, floorcod;
52 AC3BitAllocParameters bit_alloc;
54 int fgaincod[AC3_MAX_CHANNELS];
55 int fsnroffst[AC3_MAX_CHANNELS];
56 /* mantissa encoding */
57 int mant1_cnt, mant2_cnt, mant4_cnt;
63 #define N (1 << MDCT_NBITS)
65 /* new exponents are sent if their Norm 1 exceed this number */
66 #define EXP_DIFF_THRESHOLD 1000
68 static void fft_init(int ln);
69 static void ac3_crc_init(void);
71 static inline int16_t fix15(float a)
74 v = (int)(a * (float)(1 << 15));
82 static inline int calc_lowcomp1(int a, int b0, int b1)
84 if ((b0 + 256) == b1) {
93 static inline int calc_lowcomp(int a, int b0, int b1, int bin)
96 if ((b0 + 256) == b1) {
102 } else if (bin < 20) {
103 if ((b0 + 256) == b1) {
105 } else if (b0 > b1) {
116 /* AC3 bit allocation. The algorithm is the one described in the AC3
118 void ac3_parametric_bit_allocation(AC3BitAllocParameters *s, uint8_t *bap,
119 int8_t *exp, int start, int end,
120 int snroffset, int fgain, int is_lfe,
121 int deltbae,int deltnseg,
122 uint8_t *deltoffst, uint8_t *deltlen, uint8_t *deltba)
124 int bin,i,j,k,end1,v,v1,bndstrt,bndend,lowcomp,begin;
125 int fastleak,slowleak,address,tmp;
126 int16_t psd[256]; /* scaled exponents */
127 int16_t bndpsd[50]; /* interpolated exponents */
128 int16_t excite[50]; /* excitation */
129 int16_t mask[50]; /* masking value */
131 /* exponent mapping to PSD */
132 for(bin=start;bin<end;bin++) {
133 psd[bin]=(3072 - (exp[bin] << 7));
136 /* PSD integration */
143 if (end1 > end) end1=end;
144 for(i=j;i<end1;i++) {
151 if (adr > 255) adr=255;
155 if (adr > 255) adr=255;
162 } while (end > bndtab[k]);
164 /* excitation function */
165 bndstrt = masktab[start];
166 bndend = masktab[end-1] + 1;
170 lowcomp = calc_lowcomp1(lowcomp, bndpsd[0], bndpsd[1]) ;
171 excite[0] = bndpsd[0] - fgain - lowcomp ;
172 lowcomp = calc_lowcomp1(lowcomp, bndpsd[1], bndpsd[2]) ;
173 excite[1] = bndpsd[1] - fgain - lowcomp ;
175 for (bin = 2; bin < 7; bin++) {
176 if (!(is_lfe && bin == 6))
177 lowcomp = calc_lowcomp1(lowcomp, bndpsd[bin], bndpsd[bin+1]) ;
178 fastleak = bndpsd[bin] - fgain ;
179 slowleak = bndpsd[bin] - s->sgain ;
180 excite[bin] = fastleak - lowcomp ;
181 if (!(is_lfe && bin == 6)) {
182 if (bndpsd[bin] <= bndpsd[bin+1]) {
190 if (end1 > 22) end1=22;
192 for (bin = begin; bin < end1; bin++) {
193 if (!(is_lfe && bin == 6))
194 lowcomp = calc_lowcomp(lowcomp, bndpsd[bin], bndpsd[bin+1], bin) ;
196 fastleak -= s->fdecay ;
197 v = bndpsd[bin] - fgain;
198 if (fastleak < v) fastleak = v;
200 slowleak -= s->sdecay ;
201 v = bndpsd[bin] - s->sgain;
202 if (slowleak < v) slowleak = v;
204 v=fastleak - lowcomp;
205 if (slowleak > v) v=slowleak;
211 /* coupling channel */
214 fastleak = (s->cplfleak << 8) + 768;
215 slowleak = (s->cplsleak << 8) + 768;
218 for (bin = begin; bin < bndend; bin++) {
219 fastleak -= s->fdecay ;
220 v = bndpsd[bin] - fgain;
221 if (fastleak < v) fastleak = v;
222 slowleak -= s->sdecay ;
223 v = bndpsd[bin] - s->sgain;
224 if (slowleak < v) slowleak = v;
227 if (slowleak > v) v = slowleak;
231 /* compute masking curve */
233 for (bin = bndstrt; bin < bndend; bin++) {
235 tmp = s->dbknee - bndpsd[bin];
239 v=hth[bin >> s->halfratecod][s->fscod];
244 /* delta bit allocation */
246 if (deltbae == 0 || deltbae == 1) {
247 int band, seg, delta;
249 for (seg = 0; seg < deltnseg; seg++) {
250 band += deltoffst[seg] ;
251 if (deltba[seg] >= 4) {
252 delta = (deltba[seg] - 3) << 7;
254 delta = (deltba[seg] - 4) << 7;
256 for (k = 0; k < deltlen[seg]; k++) {
257 mask[band] += delta ;
263 /* compute bit allocation */
275 end1=bndtab[j] + bndsz[j];
276 if (end1 > end) end1=end;
278 for (k = i; k < end1; k++) {
279 address = (psd[i] - v) >> 5 ;
280 if (address < 0) address=0;
281 else if (address > 63) address=63;
282 bap[i] = baptab[address];
285 } while (end > bndtab[j++]) ;
288 typedef struct IComplex {
292 static void fft_init(int ln)
299 for(i=0;i<(n/2);i++) {
300 alpha = 2 * M_PI * (float)i / (float)n;
301 costab[i] = fix15(cos(alpha));
302 sintab[i] = fix15(sin(alpha));
308 m |= ((i >> j) & 1) << (ln-j-1);
315 #define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
322 pre = (bx + ax) >> 1;\
323 pim = (by + ay) >> 1;\
324 qre = (bx - ax) >> 1;\
325 qim = (by - ay) >> 1;\
328 #define MUL16(a,b) ((a) * (b))
330 #define CMUL(pre, pim, are, aim, bre, bim) \
332 pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\
333 pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\
337 /* do a 2^n point complex fft on 2^ln points. */
338 static void fft(IComplex *z, int ln)
342 register IComplex *p,*q;
364 BF(p[0].re, p[0].im, p[1].re, p[1].im,
365 p[0].re, p[0].im, p[1].re, p[1].im);
374 BF(p[0].re, p[0].im, p[2].re, p[2].im,
375 p[0].re, p[0].im, p[2].re, p[2].im);
376 BF(p[1].re, p[1].im, p[3].re, p[3].im,
377 p[1].re, p[1].im, p[3].im, -p[3].re);
389 for (j = 0; j < nblocks; ++j) {
391 BF(p->re, p->im, q->re, q->im,
392 p->re, p->im, q->re, q->im);
396 for(l = nblocks; l < np2; l += nblocks) {
397 CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
398 BF(p->re, p->im, q->re, q->im,
399 p->re, p->im, tmp_re, tmp_im);
406 nblocks = nblocks >> 1;
407 nloops = nloops << 1;
408 } while (nblocks != 0);
411 /* do a 512 point mdct */
412 static void mdct512(int32_t *out, int16_t *in)
414 int i, re, im, re1, im1;
418 /* shift to simplify computations */
420 rot[i] = -in[i + 3*N/4];
422 rot[i] = in[i - N/4];
426 re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1;
427 im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1;
428 CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
431 fft(x, MDCT_NBITS - 2);
437 CMUL(re1, im1, re, im, xsin1[i], xcos1[i]);
439 out[N/2-1-2*i] = re1;
443 /* XXX: use another norm ? */
444 static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n)
449 sum += abs(exp1[i] - exp2[i]);
454 static void compute_exp_strategy(uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
455 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
461 /* estimate if the exponent variation & decide if they should be
462 reused in the next frame */
463 exp_strategy[0][ch] = EXP_NEW;
464 for(i=1;i<NB_BLOCKS;i++) {
465 exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2);
467 av_log(NULL, AV_LOG_DEBUG, "exp_diff=%d\n", exp_diff);
469 if (exp_diff > EXP_DIFF_THRESHOLD)
470 exp_strategy[i][ch] = EXP_NEW;
472 exp_strategy[i][ch] = EXP_REUSE;
477 /* now select the encoding strategy type : if exponents are often
478 recoded, we use a coarse encoding */
480 while (i < NB_BLOCKS) {
482 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE)
486 exp_strategy[i][ch] = EXP_D45;
490 exp_strategy[i][ch] = EXP_D25;
493 exp_strategy[i][ch] = EXP_D15;
500 /* set exp[i] to min(exp[i], exp1[i]) */
501 static void exponent_min(uint8_t exp[N/2], uint8_t exp1[N/2], int n)
506 if (exp1[i] < exp[i])
511 /* update the exponents so that they are the ones the decoder will
512 decode. Return the number of bits used to code the exponents */
513 static int encode_exp(uint8_t encoded_exp[N/2],
518 int group_size, nb_groups, i, j, k, exp_min;
521 switch(exp_strategy) {
533 nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3;
535 /* for each group, compute the minimum exponent */
536 exp1[0] = exp[0]; /* DC exponent is handled separately */
538 for(i=1;i<=nb_groups;i++) {
540 assert(exp_min >= 0 && exp_min <= 24);
541 for(j=1;j<group_size;j++) {
542 if (exp[k+j] < exp_min)
549 /* constraint for DC exponent */
553 /* Decrease the delta between each groups to within 2
554 * so that they can be differentially encoded */
555 for (i=1;i<=nb_groups;i++)
556 exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2);
557 for (i=nb_groups-1;i>=0;i--)
558 exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2);
560 /* now we have the exponent values the decoder will see */
561 encoded_exp[0] = exp1[0];
563 for(i=1;i<=nb_groups;i++) {
564 for(j=0;j<group_size;j++) {
565 encoded_exp[k+j] = exp1[i];
571 av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy);
572 for(i=0;i<=nb_groups * group_size;i++) {
573 av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]);
575 av_log(NULL, AV_LOG_DEBUG, "\n");
578 return 4 + (nb_groups / 3) * 7;
581 /* return the size in bits taken by the mantissa */
582 static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs)
587 for(i=0;i<nb_coefs;i++) {
594 /* 3 mantissa in 5 bits */
595 if (s->mant1_cnt == 0)
597 if (++s->mant1_cnt == 3)
601 /* 3 mantissa in 7 bits */
602 if (s->mant2_cnt == 0)
604 if (++s->mant2_cnt == 3)
611 /* 2 mantissa in 7 bits */
612 if (s->mant4_cnt == 0)
614 if (++s->mant4_cnt == 2)
632 static int bit_alloc(AC3EncodeContext *s,
633 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
634 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
635 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
636 int frame_bits, int csnroffst, int fsnroffst)
641 for(i=0;i<NB_BLOCKS;i++) {
645 for(ch=0;ch<s->nb_all_channels;ch++) {
646 ac3_parametric_bit_allocation(&s->bit_alloc,
647 bap[i][ch], (int8_t *)encoded_exp[i][ch],
649 (((csnroffst-15) << 4) +
651 fgaintab[s->fgaincod[ch]],
652 ch == s->lfe_channel,
653 2, 0, NULL, NULL, NULL);
654 frame_bits += compute_mantissa_size(s, bap[i][ch],
659 printf("csnr=%d fsnr=%d frame_bits=%d diff=%d\n",
660 csnroffst, fsnroffst, frame_bits,
661 16 * s->frame_size - ((frame_bits + 7) & ~7));
663 return 16 * s->frame_size - frame_bits;
668 static int compute_bit_allocation(AC3EncodeContext *s,
669 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
670 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
671 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
675 int csnroffst, fsnroffst;
676 uint8_t bap1[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
677 static int frame_bits_inc[8] = { 0, 0, 2, 2, 2, 4, 2, 4 };
679 /* init default parameters */
685 for(ch=0;ch<s->nb_all_channels;ch++)
688 /* compute real values */
689 s->bit_alloc.fscod = s->fscod;
690 s->bit_alloc.halfratecod = s->halfratecod;
691 s->bit_alloc.sdecay = sdecaytab[s->sdecaycod] >> s->halfratecod;
692 s->bit_alloc.fdecay = fdecaytab[s->fdecaycod] >> s->halfratecod;
693 s->bit_alloc.sgain = sgaintab[s->sgaincod];
694 s->bit_alloc.dbknee = dbkneetab[s->dbkneecod];
695 s->bit_alloc.floor = floortab[s->floorcod];
699 // if (s->acmod == 2)
701 frame_bits += frame_bits_inc[s->acmod];
704 for(i=0;i<NB_BLOCKS;i++) {
705 frame_bits += s->nb_channels * 2 + 2; /* blksw * c, dithflag * c, dynrnge, cplstre */
707 frame_bits++; /* rematstr */
708 frame_bits += 2 * s->nb_channels; /* chexpstr[2] * c */
710 frame_bits++; /* lfeexpstr */
711 for(ch=0;ch<s->nb_channels;ch++) {
712 if (exp_strategy[i][ch] != EXP_REUSE)
713 frame_bits += 6 + 2; /* chbwcod[6], gainrng[2] */
715 frame_bits++; /* baie */
716 frame_bits++; /* snr */
717 frame_bits += 2; /* delta / skip */
719 frame_bits++; /* cplinu for block 0 */
721 /* sdcycod[2], fdcycod[2], sgaincod[2], dbpbcod[2], floorcod[3] */
723 /* (fsnoffset[4] + fgaincod[4]) * c */
724 frame_bits += 2*4 + 3 + 6 + s->nb_all_channels * (4 + 3);
726 /* auxdatae, crcrsv */
732 /* now the big work begins : do the bit allocation. Modify the snr
733 offset until we can pack everything in the requested frame size */
735 csnroffst = s->csnroffst;
736 while (csnroffst >= 0 &&
737 bit_alloc(s, bap, encoded_exp, exp_strategy, frame_bits, csnroffst, 0) < 0)
738 csnroffst -= SNR_INC1;
740 av_log(NULL, AV_LOG_ERROR, "Yack, Error !!!\n");
743 while ((csnroffst + SNR_INC1) <= 63 &&
744 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
745 csnroffst + SNR_INC1, 0) >= 0) {
746 csnroffst += SNR_INC1;
747 memcpy(bap, bap1, sizeof(bap1));
749 while ((csnroffst + 1) <= 63 &&
750 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits, csnroffst + 1, 0) >= 0) {
752 memcpy(bap, bap1, sizeof(bap1));
756 while ((fsnroffst + SNR_INC1) <= 15 &&
757 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
758 csnroffst, fsnroffst + SNR_INC1) >= 0) {
759 fsnroffst += SNR_INC1;
760 memcpy(bap, bap1, sizeof(bap1));
762 while ((fsnroffst + 1) <= 15 &&
763 bit_alloc(s, bap1, encoded_exp, exp_strategy, frame_bits,
764 csnroffst, fsnroffst + 1) >= 0) {
766 memcpy(bap, bap1, sizeof(bap1));
769 s->csnroffst = csnroffst;
770 for(ch=0;ch<s->nb_all_channels;ch++)
771 s->fsnroffst[ch] = fsnroffst;
772 #if defined(DEBUG_BITALLOC)
777 for(ch=0;ch<s->nb_all_channels;ch++) {
778 printf("Block #%d Ch%d:\n", i, ch);
780 for(j=0;j<s->nb_coefs[ch];j++) {
781 printf("%d ",bap[i][ch][j]);
791 void ac3_common_init(void)
794 /* compute bndtab and masktab from bandsz */
800 for(j=0;j<v;j++) masktab[k++]=i;
807 static int AC3_encode_init(AVCodecContext *avctx)
809 int freq = avctx->sample_rate;
810 int bitrate = avctx->bit_rate;
811 int channels = avctx->channels;
812 AC3EncodeContext *s = avctx->priv_data;
815 static const uint8_t acmod_defs[6] = {
819 0x06, /* L R SL SR */
820 0x07, /* L C R SL SR */
821 0x07, /* L C R SL SR (+LFE) */
824 avctx->frame_size = AC3_FRAME_SIZE;
826 /* number of channels */
827 if (channels < 1 || channels > 6)
829 s->acmod = acmod_defs[channels - 1];
830 s->lfe = (channels == 6) ? 1 : 0;
831 s->nb_all_channels = channels;
832 s->nb_channels = channels > 5 ? 5 : channels;
833 s->lfe_channel = s->lfe ? 5 : -1;
838 if ((ac3_freqs[j] >> i) == freq)
843 s->sample_rate = freq;
846 s->bsid = 8 + s->halfratecod;
847 s->bsmod = 0; /* complete main audio service */
849 /* bitrate & frame size */
852 if ((ac3_bitratetab[i] >> s->halfratecod) == bitrate)
857 s->bit_rate = bitrate;
858 s->frmsizecod = i << 1;
859 s->frame_size_min = (bitrate * 1000 * AC3_FRAME_SIZE) / (freq * 16);
860 /* for now we do not handle fractional sizes */
861 s->frame_size = s->frame_size_min;
863 /* bit allocation init */
864 for(ch=0;ch<s->nb_channels;ch++) {
865 /* bandwidth for each channel */
866 /* XXX: should compute the bandwidth according to the frame
867 size, so that we avoid anoying high freq artefacts */
868 s->chbwcod[ch] = 50; /* sample bandwidth as mpeg audio layer 2 table 0 */
869 s->nb_coefs[ch] = ((s->chbwcod[ch] + 12) * 3) + 37;
872 s->nb_coefs[s->lfe_channel] = 7; /* fixed */
874 /* initial snr offset */
880 fft_init(MDCT_NBITS - 2);
882 alpha = 2 * M_PI * (i + 1.0 / 8.0) / (float)N;
883 xcos1[i] = fix15(-cos(alpha));
884 xsin1[i] = fix15(-sin(alpha));
889 avctx->coded_frame= avcodec_alloc_frame();
890 avctx->coded_frame->key_frame= 1;
895 /* output the AC3 frame header */
896 static void output_frame_header(AC3EncodeContext *s, unsigned char *frame)
898 init_put_bits(&s->pb, frame, AC3_MAX_CODED_FRAME_SIZE);
900 put_bits(&s->pb, 16, 0x0b77); /* frame header */
901 put_bits(&s->pb, 16, 0); /* crc1: will be filled later */
902 put_bits(&s->pb, 2, s->fscod);
903 put_bits(&s->pb, 6, s->frmsizecod + (s->frame_size - s->frame_size_min));
904 put_bits(&s->pb, 5, s->bsid);
905 put_bits(&s->pb, 3, s->bsmod);
906 put_bits(&s->pb, 3, s->acmod);
907 if ((s->acmod & 0x01) && s->acmod != 0x01)
908 put_bits(&s->pb, 2, 1); /* XXX -4.5 dB */
910 put_bits(&s->pb, 2, 1); /* XXX -6 dB */
911 if (s->acmod == 0x02)
912 put_bits(&s->pb, 2, 0); /* surround not indicated */
913 put_bits(&s->pb, 1, s->lfe); /* LFE */
914 put_bits(&s->pb, 5, 31); /* dialog norm: -31 db */
915 put_bits(&s->pb, 1, 0); /* no compression control word */
916 put_bits(&s->pb, 1, 0); /* no lang code */
917 put_bits(&s->pb, 1, 0); /* no audio production info */
918 put_bits(&s->pb, 1, 0); /* no copyright */
919 put_bits(&s->pb, 1, 1); /* original bitstream */
920 put_bits(&s->pb, 1, 0); /* no time code 1 */
921 put_bits(&s->pb, 1, 0); /* no time code 2 */
922 put_bits(&s->pb, 1, 0); /* no addtional bit stream info */
925 /* symetric quantization on 'levels' levels */
926 static inline int sym_quant(int c, int e, int levels)
931 v = (levels * (c << e)) >> 24;
933 v = (levels >> 1) + v;
935 v = (levels * ((-c) << e)) >> 24;
937 v = (levels >> 1) - v;
939 assert (v >= 0 && v < levels);
943 /* asymetric quantization on 2^qbits levels */
944 static inline int asym_quant(int c, int e, int qbits)
948 lshift = e + qbits - 24;
955 m = (1 << (qbits-1));
959 return v & ((1 << qbits)-1);
962 /* Output one audio block. There are NB_BLOCKS audio blocks in one AC3
964 static void output_audio_block(AC3EncodeContext *s,
965 uint8_t exp_strategy[AC3_MAX_CHANNELS],
966 uint8_t encoded_exp[AC3_MAX_CHANNELS][N/2],
967 uint8_t bap[AC3_MAX_CHANNELS][N/2],
968 int32_t mdct_coefs[AC3_MAX_CHANNELS][N/2],
969 int8_t global_exp[AC3_MAX_CHANNELS],
972 int ch, nb_groups, group_size, i, baie, rbnd;
974 uint16_t qmant[AC3_MAX_CHANNELS][N/2];
976 int mant1_cnt, mant2_cnt, mant4_cnt;
977 uint16_t *qmant1_ptr, *qmant2_ptr, *qmant4_ptr;
978 int delta0, delta1, delta2;
980 for(ch=0;ch<s->nb_channels;ch++)
981 put_bits(&s->pb, 1, 0); /* 512 point MDCT */
982 for(ch=0;ch<s->nb_channels;ch++)
983 put_bits(&s->pb, 1, 1); /* no dither */
984 put_bits(&s->pb, 1, 0); /* no dynamic range */
985 if (block_num == 0) {
986 /* for block 0, even if no coupling, we must say it. This is a
988 put_bits(&s->pb, 1, 1); /* coupling strategy present */
989 put_bits(&s->pb, 1, 0); /* no coupling strategy */
991 put_bits(&s->pb, 1, 0); /* no new coupling strategy */
998 /* first block must define rematrixing (rematstr) */
999 put_bits(&s->pb, 1, 1);
1001 /* dummy rematrixing rematflg(1:4)=0 */
1002 for (rbnd=0;rbnd<4;rbnd++)
1003 put_bits(&s->pb, 1, 0);
1007 /* no matrixing (but should be used in the future) */
1008 put_bits(&s->pb, 1, 0);
1014 static int count = 0;
1015 av_log(NULL, AV_LOG_DEBUG, "Block #%d (%d)\n", block_num, count++);
1018 /* exponent strategy */
1019 for(ch=0;ch<s->nb_channels;ch++) {
1020 put_bits(&s->pb, 2, exp_strategy[ch]);
1024 put_bits(&s->pb, 1, exp_strategy[s->lfe_channel]);
1027 for(ch=0;ch<s->nb_channels;ch++) {
1028 if (exp_strategy[ch] != EXP_REUSE)
1029 put_bits(&s->pb, 6, s->chbwcod[ch]);
1033 for (ch = 0; ch < s->nb_all_channels; ch++) {
1034 switch(exp_strategy[ch]) {
1048 nb_groups = (s->nb_coefs[ch] + (group_size * 3) - 4) / (3 * group_size);
1049 p = encoded_exp[ch];
1051 /* first exponent */
1053 put_bits(&s->pb, 4, exp1);
1055 /* next ones are delta encoded */
1056 for(i=0;i<nb_groups;i++) {
1057 /* merge three delta in one code */
1061 delta0 = exp1 - exp0 + 2;
1066 delta1 = exp1 - exp0 + 2;
1071 delta2 = exp1 - exp0 + 2;
1073 put_bits(&s->pb, 7, ((delta0 * 5 + delta1) * 5) + delta2);
1076 if (ch != s->lfe_channel)
1077 put_bits(&s->pb, 2, 0); /* no gain range info */
1080 /* bit allocation info */
1081 baie = (block_num == 0);
1082 put_bits(&s->pb, 1, baie);
1084 put_bits(&s->pb, 2, s->sdecaycod);
1085 put_bits(&s->pb, 2, s->fdecaycod);
1086 put_bits(&s->pb, 2, s->sgaincod);
1087 put_bits(&s->pb, 2, s->dbkneecod);
1088 put_bits(&s->pb, 3, s->floorcod);
1092 put_bits(&s->pb, 1, baie); /* always present with bai */
1094 put_bits(&s->pb, 6, s->csnroffst);
1095 for(ch=0;ch<s->nb_all_channels;ch++) {
1096 put_bits(&s->pb, 4, s->fsnroffst[ch]);
1097 put_bits(&s->pb, 3, s->fgaincod[ch]);
1101 put_bits(&s->pb, 1, 0); /* no delta bit allocation */
1102 put_bits(&s->pb, 1, 0); /* no data to skip */
1104 /* mantissa encoding : we use two passes to handle the grouping. A
1105 one pass method may be faster, but it would necessitate to
1106 modify the output stream. */
1108 /* first pass: quantize */
1109 mant1_cnt = mant2_cnt = mant4_cnt = 0;
1110 qmant1_ptr = qmant2_ptr = qmant4_ptr = NULL;
1112 for (ch = 0; ch < s->nb_all_channels; ch++) {
1115 for(i=0;i<s->nb_coefs[ch];i++) {
1116 c = mdct_coefs[ch][i];
1117 e = encoded_exp[ch][i] - global_exp[ch];
1124 v = sym_quant(c, e, 3);
1127 qmant1_ptr = &qmant[ch][i];
1132 *qmant1_ptr += 3 * v;
1144 v = sym_quant(c, e, 5);
1147 qmant2_ptr = &qmant[ch][i];
1152 *qmant2_ptr += 5 * v;
1164 v = sym_quant(c, e, 7);
1167 v = sym_quant(c, e, 11);
1170 qmant4_ptr = &qmant[ch][i];
1182 v = sym_quant(c, e, 15);
1185 v = asym_quant(c, e, 14);
1188 v = asym_quant(c, e, 16);
1191 v = asym_quant(c, e, b - 1);
1198 /* second pass : output the values */
1199 for (ch = 0; ch < s->nb_all_channels; ch++) {
1202 for(i=0;i<s->nb_coefs[ch];i++) {
1210 put_bits(&s->pb, 5, q);
1214 put_bits(&s->pb, 7, q);
1217 put_bits(&s->pb, 3, q);
1221 put_bits(&s->pb, 7, q);
1224 put_bits(&s->pb, 14, q);
1227 put_bits(&s->pb, 16, q);
1230 put_bits(&s->pb, b - 1, q);
1237 /* compute the ac3 crc */
1239 #define CRC16_POLY ((1 << 0) | (1 << 2) | (1 << 15) | (1 << 16))
1241 static void ac3_crc_init(void)
1243 unsigned int c, n, k;
1245 for(n=0;n<256;n++) {
1247 for (k = 0; k < 8; k++) {
1249 c = ((c << 1) & 0xffff) ^ (CRC16_POLY & 0xffff);
1257 static unsigned int ac3_crc(uint8_t *data, int n, unsigned int crc)
1261 crc = (crc_table[data[i] ^ (crc >> 8)] ^ (crc << 8)) & 0xffff;
1266 static unsigned int mul_poly(unsigned int a, unsigned int b, unsigned int poly)
1282 static unsigned int pow_poly(unsigned int a, unsigned int n, unsigned int poly)
1288 r = mul_poly(r, a, poly);
1289 a = mul_poly(a, a, poly);
1296 /* compute log2(max(abs(tab[]))) */
1297 static int log2_tab(int16_t *tab, int n)
1308 static void lshift_tab(int16_t *tab, int n, int lshift)
1316 } else if (lshift < 0) {
1324 /* fill the end of the frame and compute the two crcs */
1325 static int output_frame_end(AC3EncodeContext *s)
1327 int frame_size, frame_size_58, n, crc1, crc2, crc_inv;
1330 frame_size = s->frame_size; /* frame size in words */
1331 /* align to 8 bits */
1332 flush_put_bits(&s->pb);
1333 /* add zero bytes to reach the frame size */
1335 n = 2 * s->frame_size - (pbBufPtr(&s->pb) - frame) - 2;
1338 memset(pbBufPtr(&s->pb), 0, n);
1340 /* Now we must compute both crcs : this is not so easy for crc1
1341 because it is at the beginning of the data... */
1342 frame_size_58 = (frame_size >> 1) + (frame_size >> 3);
1343 crc1 = ac3_crc(frame + 4, (2 * frame_size_58) - 4, 0);
1344 /* XXX: could precompute crc_inv */
1345 crc_inv = pow_poly((CRC16_POLY >> 1), (16 * frame_size_58) - 16, CRC16_POLY);
1346 crc1 = mul_poly(crc_inv, crc1, CRC16_POLY);
1347 frame[2] = crc1 >> 8;
1350 crc2 = ac3_crc(frame + 2 * frame_size_58, (frame_size - frame_size_58) * 2 - 2, 0);
1351 frame[2*frame_size - 2] = crc2 >> 8;
1352 frame[2*frame_size - 1] = crc2;
1354 // printf("n=%d frame_size=%d\n", n, frame_size);
1355 return frame_size * 2;
1358 static int AC3_encode_frame(AVCodecContext *avctx,
1359 unsigned char *frame, int buf_size, void *data)
1361 AC3EncodeContext *s = avctx->priv_data;
1362 short *samples = data;
1364 int16_t input_samples[N];
1365 int32_t mdct_coef[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1366 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1367 uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS];
1368 uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1369 uint8_t bap[NB_BLOCKS][AC3_MAX_CHANNELS][N/2];
1370 int8_t exp_samples[NB_BLOCKS][AC3_MAX_CHANNELS];
1374 for(ch=0;ch<s->nb_all_channels;ch++) {
1375 /* fixed mdct to the six sub blocks & exponent computation */
1376 for(i=0;i<NB_BLOCKS;i++) {
1380 /* compute input samples */
1381 memcpy(input_samples, s->last_samples[ch], N/2 * sizeof(int16_t));
1382 sinc = s->nb_all_channels;
1383 sptr = samples + (sinc * (N/2) * i) + ch;
1384 for(j=0;j<N/2;j++) {
1386 input_samples[j + N/2] = v;
1387 s->last_samples[ch][j] = v;
1391 /* apply the MDCT window */
1392 for(j=0;j<N/2;j++) {
1393 input_samples[j] = MUL16(input_samples[j],
1394 ac3_window[j]) >> 15;
1395 input_samples[N-j-1] = MUL16(input_samples[N-j-1],
1396 ac3_window[j]) >> 15;
1399 /* Normalize the samples to use the maximum available
1401 v = 14 - log2_tab(input_samples, N);
1404 exp_samples[i][ch] = v - 8;
1405 lshift_tab(input_samples, N, v);
1408 mdct512(mdct_coef[i][ch], input_samples);
1410 /* compute "exponents". We take into account the
1411 normalization there */
1412 for(j=0;j<N/2;j++) {
1414 v = abs(mdct_coef[i][ch][j]);
1418 e = 23 - av_log2(v) + exp_samples[i][ch];
1421 mdct_coef[i][ch][j] = 0;
1428 compute_exp_strategy(exp_strategy, exp, ch, ch == s->lfe_channel);
1430 /* compute the exponents as the decoder will see them. The
1431 EXP_REUSE case must be handled carefully : we select the
1432 min of the exponents */
1434 while (i < NB_BLOCKS) {
1436 while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE) {
1437 exponent_min(exp[i][ch], exp[j][ch], s->nb_coefs[ch]);
1440 frame_bits += encode_exp(encoded_exp[i][ch],
1441 exp[i][ch], s->nb_coefs[ch],
1442 exp_strategy[i][ch]);
1443 /* copy encoded exponents for reuse case */
1444 for(k=i+1;k<j;k++) {
1445 memcpy(encoded_exp[k][ch], encoded_exp[i][ch],
1446 s->nb_coefs[ch] * sizeof(uint8_t));
1452 compute_bit_allocation(s, bap, encoded_exp, exp_strategy, frame_bits);
1453 /* everything is known... let's output the frame */
1454 output_frame_header(s, frame);
1456 for(i=0;i<NB_BLOCKS;i++) {
1457 output_audio_block(s, exp_strategy[i], encoded_exp[i],
1458 bap[i], mdct_coef[i], exp_samples[i], i);
1460 return output_frame_end(s);
1463 static int AC3_encode_close(AVCodecContext *avctx)
1465 av_freep(&avctx->coded_frame);
1470 /*************************************************************************/
1477 IComplex in[FN], in1[FN];
1479 float sum_re, sum_im, a;
1484 in[i].re = random() % 65535 - 32767;
1485 in[i].im = random() % 65535 - 32767;
1495 a = -2 * M_PI * (n * k) / FN;
1496 sum_re += in1[n].re * cos(a) - in1[n].im * sin(a);
1497 sum_im += in1[n].re * sin(a) + in1[n].im * cos(a);
1499 printf("%3d: %6d,%6d %6.0f,%6.0f\n",
1500 k, in[k].re, in[k].im, sum_re / FN, sum_im / FN);
1504 void mdct_test(void)
1507 int32_t output[N/2];
1510 float s, a, err, e, emax;
1514 input[i] = (random() % 65535 - 32767) * 9 / 10;
1515 input1[i] = input[i];
1518 mdct512(output, input);
1521 for(k=0;k<N/2;k++) {
1524 a = (2*M_PI*(2*n+1+N/2)*(2*k+1) / (4 * N));
1525 s += input1[n] * cos(a);
1527 output1[k] = -2 * s / N;
1532 for(i=0;i<N/2;i++) {
1533 printf("%3d: %7d %7.0f\n", i, output[i], output1[i]);
1534 e = output[i] - output1[i];
1539 printf("err2=%f emax=%f\n", err / (N/2), emax);
1544 AC3EncodeContext ctx;
1545 unsigned char frame[AC3_MAX_CODED_FRAME_SIZE];
1546 short samples[AC3_FRAME_SIZE];
1549 AC3_encode_init(&ctx, 44100, 64000, 1);
1554 for(i=0;i<AC3_FRAME_SIZE;i++)
1555 samples[i] = (int)(sin(2*M_PI*i*1000.0/44100) * 10000);
1556 ret = AC3_encode_frame(&ctx, frame, samples);
1557 printf("ret=%d\n", ret);
1561 AVCodec ac3_encoder = {
1565 sizeof(AC3EncodeContext),