3 * This code was developed as part of Google Summer of Code 2006.
4 * E-AC-3 support was added as part of Google Summer of Code 2007.
6 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com).
7 * Copyright (c) 2007-2008 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
8 * Copyright (c) 2007 Justin Ruggles <justin.ruggles@gmail.com>
10 * Portions of this code are derived from liba52
11 * http://liba52.sourceforge.net
12 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
13 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
15 * This file is part of FFmpeg.
17 * FFmpeg is free software; you can redistribute it and/or
18 * modify it under the terms of the GNU General Public
19 * License as published by the Free Software Foundation; either
20 * version 2 of the License, or (at your option) any later version.
22 * FFmpeg is distributed in the hope that it will be useful,
23 * but WITHOUT ANY WARRANTY; without even the implied warranty of
24 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
25 * General Public License for more details.
27 * You should have received a copy of the GNU General Public
28 * License along with FFmpeg; if not, write to the Free Software
29 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
37 #include "libavutil/crc.h"
38 #include "ac3_parser.h"
40 #include "ac3dec_data.h"
42 /** Large enough for maximum possible frame size when the specification limit is ignored */
43 #define AC3_FRAME_BUFFER_SIZE 32768
46 * table for ungrouping 3 values in 7 bits.
47 * used for exponents and bap=2 mantissas
49 static uint8_t ungroup_3_in_7_bits_tab[128][3];
52 /** tables for ungrouping mantissas */
53 static int b1_mantissas[32][3];
54 static int b2_mantissas[128][3];
55 static int b3_mantissas[8];
56 static int b4_mantissas[128][2];
57 static int b5_mantissas[16];
60 * Quantization table: levels for symmetric. bits for asymmetric.
61 * reference: Table 7.18 Mapping of bap to Quantizer
63 static const uint8_t quantization_tab[16] = {
65 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
68 /** dynamic range table. converts codes to scale factors. */
69 static float dynamic_range_tab[256];
71 /** Adjustments in dB gain */
72 #define LEVEL_PLUS_3DB 1.4142135623730950
73 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
74 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
75 #define LEVEL_MINUS_3DB 0.7071067811865476
76 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
77 #define LEVEL_MINUS_6DB 0.5000000000000000
78 #define LEVEL_MINUS_9DB 0.3535533905932738
79 #define LEVEL_ZERO 0.0000000000000000
80 #define LEVEL_ONE 1.0000000000000000
82 static const float gain_levels[9] = {
86 LEVEL_MINUS_1POINT5DB,
88 LEVEL_MINUS_4POINT5DB,
95 * Table for center mix levels
96 * reference: Section 5.4.2.4 cmixlev
98 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
101 * Table for surround mix levels
102 * reference: Section 5.4.2.5 surmixlev
104 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
107 * Table for default stereo downmixing coefficients
108 * reference: Section 7.8.2 Downmixing Into Two Channels
110 static const uint8_t ac3_default_coeffs[8][5][2] = {
111 { { 2, 7 }, { 7, 2 }, },
113 { { 2, 7 }, { 7, 2 }, },
114 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
115 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
116 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
117 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
118 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
122 * Symmetrical Dequantization
123 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
124 * Tables 7.19 to 7.23
127 symmetric_dequant(int code, int levels)
129 return ((code - (levels >> 1)) << 24) / levels;
133 * Initialize tables at runtime.
135 static av_cold void ac3_tables_init(void)
139 /* generate table for ungrouping 3 values in 7 bits
140 reference: Section 7.1.3 Exponent Decoding */
141 for(i=0; i<128; i++) {
142 ungroup_3_in_7_bits_tab[i][0] = i / 25;
143 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
144 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
147 /* generate grouped mantissa tables
148 reference: Section 7.3.5 Ungrouping of Mantissas */
149 for(i=0; i<32; i++) {
150 /* bap=1 mantissas */
151 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
152 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
153 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
155 for(i=0; i<128; i++) {
156 /* bap=2 mantissas */
157 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
158 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
159 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
161 /* bap=4 mantissas */
162 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
163 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
165 /* generate ungrouped mantissa tables
166 reference: Tables 7.21 and 7.23 */
168 /* bap=3 mantissas */
169 b3_mantissas[i] = symmetric_dequant(i, 7);
171 for(i=0; i<15; i++) {
172 /* bap=5 mantissas */
173 b5_mantissas[i] = symmetric_dequant(i, 15);
176 /* generate dynamic range table
177 reference: Section 7.7.1 Dynamic Range Control */
178 for(i=0; i<256; i++) {
179 int v = (i >> 5) - ((i >> 7) << 3) - 5;
180 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
186 * AVCodec initialization
188 static av_cold int ac3_decode_init(AVCodecContext *avctx)
190 AC3DecodeContext *s = avctx->priv_data;
195 ff_mdct_init(&s->imdct_256, 8, 1);
196 ff_mdct_init(&s->imdct_512, 9, 1);
197 ff_kbd_window_init(s->window, 5.0, 256);
198 dsputil_init(&s->dsp, avctx);
199 av_lfg_init(&s->dith_state, 0);
201 /* set bias values for float to int16 conversion */
202 if(s->dsp.float_to_int16_interleave == ff_float_to_int16_interleave_c) {
203 s->add_bias = 385.0f;
207 s->mul_bias = 32767.0f;
210 /* allow downmixing to stereo or mono */
211 if (avctx->channels > 0 && avctx->request_channels > 0 &&
212 avctx->request_channels < avctx->channels &&
213 avctx->request_channels <= 2) {
214 avctx->channels = avctx->request_channels;
218 /* allocate context input buffer */
219 if (avctx->error_recognition >= FF_ER_CAREFUL) {
220 s->input_buffer = av_mallocz(AC3_FRAME_BUFFER_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
221 if (!s->input_buffer)
222 return AVERROR_NOMEM;
225 avctx->sample_fmt = SAMPLE_FMT_S16;
230 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
231 * GetBitContext within AC3DecodeContext must point to
232 * the start of the synchronized AC-3 bitstream.
234 static int ac3_parse_header(AC3DecodeContext *s)
236 GetBitContext *gbc = &s->gbc;
239 /* read the rest of the bsi. read twice for dual mono mode. */
240 i = !(s->channel_mode);
242 skip_bits(gbc, 5); // skip dialog normalization
244 skip_bits(gbc, 8); //skip compression
246 skip_bits(gbc, 8); //skip language code
248 skip_bits(gbc, 7); //skip audio production information
251 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
253 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
254 TODO: read & use the xbsi1 downmix levels */
256 skip_bits(gbc, 14); //skip timecode1 / xbsi1
258 skip_bits(gbc, 14); //skip timecode2 / xbsi2
260 /* skip additional bitstream info */
261 if (get_bits1(gbc)) {
262 i = get_bits(gbc, 6);
272 * Common function to parse AC-3 or E-AC-3 frame header
274 static int parse_frame_header(AC3DecodeContext *s)
279 err = ff_ac3_parse_header(&s->gbc, &hdr);
283 /* get decoding parameters from header info */
284 s->bit_alloc_params.sr_code = hdr.sr_code;
285 s->channel_mode = hdr.channel_mode;
286 s->lfe_on = hdr.lfe_on;
287 s->bit_alloc_params.sr_shift = hdr.sr_shift;
288 s->sample_rate = hdr.sample_rate;
289 s->bit_rate = hdr.bit_rate;
290 s->channels = hdr.channels;
291 s->fbw_channels = s->channels - s->lfe_on;
292 s->lfe_ch = s->fbw_channels + 1;
293 s->frame_size = hdr.frame_size;
294 s->center_mix_level = hdr.center_mix_level;
295 s->surround_mix_level = hdr.surround_mix_level;
296 s->num_blocks = hdr.num_blocks;
297 s->frame_type = hdr.frame_type;
298 s->substreamid = hdr.substreamid;
301 s->start_freq[s->lfe_ch] = 0;
302 s->end_freq[s->lfe_ch] = 7;
303 s->num_exp_groups[s->lfe_ch] = 2;
304 s->channel_in_cpl[s->lfe_ch] = 0;
307 if (hdr.bitstream_id <= 10) {
309 s->snr_offset_strategy = 2;
310 s->block_switch_syntax = 1;
311 s->dither_flag_syntax = 1;
312 s->bit_allocation_syntax = 1;
313 s->fast_gain_syntax = 0;
314 s->first_cpl_leak = 0;
317 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
318 return ac3_parse_header(s);
321 return ff_eac3_parse_header(s);
326 * Set stereo downmixing coefficients based on frame header info.
327 * reference: Section 7.8.2 Downmixing Into Two Channels
329 static void set_downmix_coeffs(AC3DecodeContext *s)
332 float cmix = gain_levels[center_levels[s->center_mix_level]];
333 float smix = gain_levels[surround_levels[s->surround_mix_level]];
336 for(i=0; i<s->fbw_channels; i++) {
337 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
338 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
340 if(s->channel_mode > 1 && s->channel_mode & 1) {
341 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
343 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
344 int nf = s->channel_mode - 2;
345 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
347 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
348 int nf = s->channel_mode - 4;
349 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
354 for(i=0; i<s->fbw_channels; i++) {
355 norm0 += s->downmix_coeffs[i][0];
356 norm1 += s->downmix_coeffs[i][1];
358 norm0 = 1.0f / norm0;
359 norm1 = 1.0f / norm1;
360 for(i=0; i<s->fbw_channels; i++) {
361 s->downmix_coeffs[i][0] *= norm0;
362 s->downmix_coeffs[i][1] *= norm1;
365 if(s->output_mode == AC3_CHMODE_MONO) {
366 for(i=0; i<s->fbw_channels; i++)
367 s->downmix_coeffs[i][0] = (s->downmix_coeffs[i][0] + s->downmix_coeffs[i][1]) * LEVEL_MINUS_3DB;
372 * Decode the grouped exponents according to exponent strategy.
373 * reference: Section 7.1.3 Exponent Decoding
375 static int decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
376 uint8_t absexp, int8_t *dexps)
378 int i, j, grp, group_size;
383 group_size = exp_strategy + (exp_strategy == EXP_D45);
384 for(grp=0,i=0; grp<ngrps; grp++) {
385 expacc = get_bits(gbc, 7);
386 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
387 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
388 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
391 /* convert to absolute exps and expand groups */
393 for(i=0,j=0; i<ngrps*3; i++) {
394 prevexp += dexp[i] - 2;
395 if (prevexp < 0 || prevexp > 24)
397 switch (group_size) {
398 case 4: dexps[j++] = prevexp;
399 dexps[j++] = prevexp;
400 case 2: dexps[j++] = prevexp;
401 case 1: dexps[j++] = prevexp;
408 * Generate transform coefficients for each coupled channel in the coupling
409 * range using the coupling coefficients and coupling coordinates.
410 * reference: Section 7.4.3 Coupling Coordinate Format
412 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
414 int i, j, ch, bnd, subbnd;
417 i = s->start_freq[CPL_CH];
418 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
421 for(j=0; j<12; j++) {
422 for(ch=1; ch<=s->fbw_channels; ch++) {
423 if(s->channel_in_cpl[ch]) {
424 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
425 if (ch == 2 && s->phase_flags[bnd])
426 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
431 } while(s->cpl_band_struct[subbnd]);
436 * Grouped mantissas for 3-level 5-level and 11-level quantization
448 * Decode the transform coefficients for a particular channel
449 * reference: Section 7.3 Quantization and Decoding of Mantissas
451 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
453 GetBitContext *gbc = &s->gbc;
454 int i, gcode, tbap, start, end;
459 exps = s->dexps[ch_index];
460 bap = s->bap[ch_index];
461 coeffs = s->fixed_coeffs[ch_index];
462 start = s->start_freq[ch_index];
463 end = s->end_freq[ch_index];
465 for (i = start; i < end; i++) {
469 coeffs[i] = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
474 gcode = get_bits(gbc, 5);
475 m->b1_mant[0] = b1_mantissas[gcode][0];
476 m->b1_mant[1] = b1_mantissas[gcode][1];
477 m->b1_mant[2] = b1_mantissas[gcode][2];
480 coeffs[i] = m->b1_mant[m->b1ptr++];
485 gcode = get_bits(gbc, 7);
486 m->b2_mant[0] = b2_mantissas[gcode][0];
487 m->b2_mant[1] = b2_mantissas[gcode][1];
488 m->b2_mant[2] = b2_mantissas[gcode][2];
491 coeffs[i] = m->b2_mant[m->b2ptr++];
495 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
500 gcode = get_bits(gbc, 7);
501 m->b4_mant[0] = b4_mantissas[gcode][0];
502 m->b4_mant[1] = b4_mantissas[gcode][1];
505 coeffs[i] = m->b4_mant[m->b4ptr++];
509 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
513 /* asymmetric dequantization */
514 int qlevel = quantization_tab[tbap];
515 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
519 coeffs[i] >>= exps[i];
524 * Remove random dithering from coefficients with zero-bit mantissas
525 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
527 static void remove_dithering(AC3DecodeContext *s) {
533 for(ch=1; ch<=s->fbw_channels; ch++) {
534 if(!s->dither_flag[ch]) {
535 coeffs = s->fixed_coeffs[ch];
537 if(s->channel_in_cpl[ch])
538 end = s->start_freq[CPL_CH];
540 end = s->end_freq[ch];
541 for(i=0; i<end; i++) {
545 if(s->channel_in_cpl[ch]) {
546 bap = s->bap[CPL_CH];
547 for(; i<s->end_freq[CPL_CH]; i++) {
556 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
559 if (!s->channel_uses_aht[ch]) {
560 ac3_decode_transform_coeffs_ch(s, ch, m);
562 /* if AHT is used, mantissas for all blocks are encoded in the first
563 block of the frame. */
566 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
567 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
568 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
574 * Decode the transform coefficients.
576 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
582 m.b1ptr = m.b2ptr = m.b4ptr = 3;
584 for (ch = 1; ch <= s->channels; ch++) {
585 /* transform coefficients for full-bandwidth channel */
586 decode_transform_coeffs_ch(s, blk, ch, &m);
587 /* tranform coefficients for coupling channel come right after the
588 coefficients for the first coupled channel*/
589 if (s->channel_in_cpl[ch]) {
591 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
592 calc_transform_coeffs_cpl(s);
595 end = s->end_freq[CPL_CH];
597 end = s->end_freq[ch];
600 s->fixed_coeffs[ch][end] = 0;
604 /* zero the dithered coefficients for appropriate channels */
609 * Stereo rematrixing.
610 * reference: Section 7.5.4 Rematrixing : Decoding Technique
612 static void do_rematrixing(AC3DecodeContext *s)
618 end = FFMIN(s->end_freq[1], s->end_freq[2]);
620 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
621 if(s->rematrixing_flags[bnd]) {
622 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
623 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
624 tmp0 = s->fixed_coeffs[1][i];
625 tmp1 = s->fixed_coeffs[2][i];
626 s->fixed_coeffs[1][i] = tmp0 + tmp1;
627 s->fixed_coeffs[2][i] = tmp0 - tmp1;
634 * Inverse MDCT Transform.
635 * Convert frequency domain coefficients to time-domain audio samples.
636 * reference: Section 7.9.4 Transformation Equations
638 static inline void do_imdct(AC3DecodeContext *s, int channels)
641 float add_bias = s->add_bias;
642 if(s->out_channels==1 && channels>1)
643 add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
645 for (ch=1; ch<=channels; ch++) {
646 if (s->block_switch[ch]) {
648 float *x = s->tmp_output+128;
650 x[i] = s->transform_coeffs[ch][2*i];
651 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
652 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
654 x[i] = s->transform_coeffs[ch][2*i+1];
655 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
657 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
658 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
659 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
665 * Downmix the output to mono or stereo.
667 void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
672 for(i=0; i<len; i++) {
674 for(j=0; j<in_ch; j++) {
675 v0 += samples[j][i] * matrix[j][0];
676 v1 += samples[j][i] * matrix[j][1];
681 } else if(out_ch == 1) {
682 for(i=0; i<len; i++) {
684 for(j=0; j<in_ch; j++)
685 v0 += samples[j][i] * matrix[j][0];
692 * Upmix delay samples from stereo to original channel layout.
694 static void ac3_upmix_delay(AC3DecodeContext *s)
696 int channel_data_size = sizeof(s->delay[0]);
697 switch(s->channel_mode) {
698 case AC3_CHMODE_DUALMONO:
699 case AC3_CHMODE_STEREO:
700 /* upmix mono to stereo */
701 memcpy(s->delay[1], s->delay[0], channel_data_size);
703 case AC3_CHMODE_2F2R:
704 memset(s->delay[3], 0, channel_data_size);
705 case AC3_CHMODE_2F1R:
706 memset(s->delay[2], 0, channel_data_size);
708 case AC3_CHMODE_3F2R:
709 memset(s->delay[4], 0, channel_data_size);
710 case AC3_CHMODE_3F1R:
711 memset(s->delay[3], 0, channel_data_size);
713 memcpy(s->delay[2], s->delay[1], channel_data_size);
714 memset(s->delay[1], 0, channel_data_size);
720 * Decode band structure for coupling, spectral extension, or enhanced coupling.
721 * @param[in] gbc bit reader context
722 * @param[in] blk block number
723 * @param[in] eac3 flag to indicate E-AC-3
724 * @param[in] ecpl flag to indicate enhanced coupling
725 * @param[in] start_subband subband number for start of range
726 * @param[in] end_subband subband number for end of range
727 * @param[in] default_band_struct default band structure table
728 * @param[out] band_struct decoded band structure
729 * @param[out] num_subbands number of subbands (optionally NULL)
730 * @param[out] num_bands number of bands (optionally NULL)
731 * @param[out] band_sizes array containing the number of bins in each band (optionally NULL)
733 static void decode_band_structure(GetBitContext *gbc, int blk, int eac3,
734 int ecpl, int start_subband, int end_subband,
735 const uint8_t *default_band_struct,
736 uint8_t *band_struct, int *num_subbands,
737 int *num_bands, uint8_t *band_sizes)
739 int subbnd, bnd, n_subbands, n_bands=0;
742 n_subbands = end_subband - start_subband;
744 /* decode band structure from bitstream or use default */
745 if (!eac3 || get_bits1(gbc)) {
746 for (subbnd = 0; subbnd < n_subbands - 1; subbnd++) {
747 band_struct[subbnd] = get_bits1(gbc);
751 &default_band_struct[start_subband+1],
754 band_struct[n_subbands-1] = 0;
756 /* calculate number of bands and band sizes based on band structure.
757 note that the first 4 subbands in enhanced coupling span only 6 bins
759 if (num_bands || band_sizes ) {
760 n_bands = n_subbands;
761 bnd_sz[0] = ecpl ? 6 : 12;
762 for (bnd = 0, subbnd = 1; subbnd < n_subbands; subbnd++) {
763 int subbnd_size = (ecpl && subbnd < 4) ? 6 : 12;
764 if (band_struct[subbnd-1]) {
766 bnd_sz[bnd] += subbnd_size;
768 bnd_sz[++bnd] = subbnd_size;
773 /* set optional output params */
775 *num_subbands = n_subbands;
777 *num_bands = n_bands;
779 memcpy(band_sizes, bnd_sz, n_bands);
783 * Decode a single audio block from the AC-3 bitstream.
785 static int decode_audio_block(AC3DecodeContext *s, int blk)
787 int fbw_channels = s->fbw_channels;
788 int channel_mode = s->channel_mode;
790 int different_transforms;
793 GetBitContext *gbc = &s->gbc;
794 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
796 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
798 /* block switch flags */
799 different_transforms = 0;
800 if (s->block_switch_syntax) {
801 for (ch = 1; ch <= fbw_channels; ch++) {
802 s->block_switch[ch] = get_bits1(gbc);
803 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
804 different_transforms = 1;
808 /* dithering flags */
809 if (s->dither_flag_syntax) {
810 for (ch = 1; ch <= fbw_channels; ch++) {
811 s->dither_flag[ch] = get_bits1(gbc);
816 i = !(s->channel_mode);
819 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
820 s->avctx->drc_scale)+1.0;
821 } else if(blk == 0) {
822 s->dynamic_range[i] = 1.0f;
826 /* spectral extension strategy */
827 if (s->eac3 && (!blk || get_bits1(gbc))) {
828 if (get_bits1(gbc)) {
829 av_log_missing_feature(s->avctx, "Spectral extension", 1);
832 /* TODO: parse spectral extension strategy info */
835 /* TODO: spectral extension coordinates */
837 /* coupling strategy */
838 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
839 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
841 s->cpl_in_use[blk] = get_bits1(gbc);
842 if (s->cpl_in_use[blk]) {
843 /* coupling in use */
844 int cpl_start_subband, cpl_end_subband;
846 if (channel_mode < AC3_CHMODE_STEREO) {
847 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
851 /* check for enhanced coupling */
852 if (s->eac3 && get_bits1(gbc)) {
853 /* TODO: parse enhanced coupling strategy info */
854 av_log_missing_feature(s->avctx, "Enhanced coupling", 1);
858 /* determine which channels are coupled */
859 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
860 s->channel_in_cpl[1] = 1;
861 s->channel_in_cpl[2] = 1;
863 for (ch = 1; ch <= fbw_channels; ch++)
864 s->channel_in_cpl[ch] = get_bits1(gbc);
867 /* phase flags in use */
868 if (channel_mode == AC3_CHMODE_STEREO)
869 s->phase_flags_in_use = get_bits1(gbc);
871 /* coupling frequency range */
872 /* TODO: modify coupling end freq if spectral extension is used */
873 cpl_start_subband = get_bits(gbc, 4);
874 cpl_end_subband = get_bits(gbc, 4) + 3;
875 s->num_cpl_subbands = cpl_end_subband - cpl_start_subband;
876 if (s->num_cpl_subbands < 0) {
877 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d > %d)\n",
878 cpl_start_subband, cpl_end_subband);
881 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
882 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
884 decode_band_structure(gbc, blk, s->eac3, 0,
885 cpl_start_subband, cpl_end_subband,
886 ff_eac3_default_cpl_band_struct,
887 s->cpl_band_struct, &s->num_cpl_subbands,
888 &s->num_cpl_bands, NULL);
890 /* coupling not in use */
891 for (ch = 1; ch <= fbw_channels; ch++) {
892 s->channel_in_cpl[ch] = 0;
893 s->first_cpl_coords[ch] = 1;
895 s->first_cpl_leak = s->eac3;
896 s->phase_flags_in_use = 0;
898 } else if (!s->eac3) {
900 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
903 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
906 cpl_in_use = s->cpl_in_use[blk];
908 /* coupling coordinates */
910 int cpl_coords_exist = 0;
912 for (ch = 1; ch <= fbw_channels; ch++) {
913 if (s->channel_in_cpl[ch]) {
914 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
915 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
916 s->first_cpl_coords[ch] = 0;
917 cpl_coords_exist = 1;
918 master_cpl_coord = 3 * get_bits(gbc, 2);
919 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
920 cpl_coord_exp = get_bits(gbc, 4);
921 cpl_coord_mant = get_bits(gbc, 4);
922 if (cpl_coord_exp == 15)
923 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
925 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
926 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
929 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
933 /* channel not in coupling */
934 s->first_cpl_coords[ch] = 1;
938 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
939 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
940 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
945 /* stereo rematrixing strategy and band structure */
946 if (channel_mode == AC3_CHMODE_STEREO) {
947 if ((s->eac3 && !blk) || get_bits1(gbc)) {
948 s->num_rematrixing_bands = 4;
949 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
950 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
951 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
952 s->rematrixing_flags[bnd] = get_bits1(gbc);
954 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
959 /* exponent strategies for each channel */
960 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
962 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
963 if(s->exp_strategy[blk][ch] != EXP_REUSE)
964 bit_alloc_stages[ch] = 3;
967 /* channel bandwidth */
968 for (ch = 1; ch <= fbw_channels; ch++) {
969 s->start_freq[ch] = 0;
970 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
972 int prev = s->end_freq[ch];
973 if (s->channel_in_cpl[ch])
974 s->end_freq[ch] = s->start_freq[CPL_CH];
976 int bandwidth_code = get_bits(gbc, 6);
977 if (bandwidth_code > 60) {
978 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60\n", bandwidth_code);
981 s->end_freq[ch] = bandwidth_code * 3 + 73;
983 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
984 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
985 if(blk > 0 && s->end_freq[ch] != prev)
986 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
989 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
990 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
991 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
994 /* decode exponents for each channel */
995 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
996 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
997 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
998 if (decode_exponents(gbc, s->exp_strategy[blk][ch],
999 s->num_exp_groups[ch], s->dexps[ch][0],
1000 &s->dexps[ch][s->start_freq[ch]+!!ch])) {
1001 av_log(s->avctx, AV_LOG_ERROR, "exponent out-of-range\n");
1004 if(ch != CPL_CH && ch != s->lfe_ch)
1005 skip_bits(gbc, 2); /* skip gainrng */
1009 /* bit allocation information */
1010 if (s->bit_allocation_syntax) {
1011 if (get_bits1(gbc)) {
1012 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1013 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
1014 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
1015 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
1016 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
1017 for(ch=!cpl_in_use; ch<=s->channels; ch++)
1018 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1020 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
1025 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
1026 if(!s->eac3 || !blk){
1027 if(s->snr_offset_strategy && get_bits1(gbc)) {
1030 csnr = (get_bits(gbc, 6) - 15) << 4;
1031 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
1033 if (ch == i || s->snr_offset_strategy == 2)
1034 snr = (csnr + get_bits(gbc, 4)) << 2;
1035 /* run at least last bit allocation stage if snr offset changes */
1036 if(blk && s->snr_offset[ch] != snr) {
1037 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
1039 s->snr_offset[ch] = snr;
1041 /* fast gain (normal AC-3 only) */
1043 int prev = s->fast_gain[ch];
1044 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1045 /* run last 2 bit allocation stages if fast gain changes */
1046 if(blk && prev != s->fast_gain[ch])
1047 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1050 } else if (!s->eac3 && !blk) {
1051 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
1056 /* fast gain (E-AC-3 only) */
1057 if (s->fast_gain_syntax && get_bits1(gbc)) {
1058 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
1059 int prev = s->fast_gain[ch];
1060 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1061 /* run last 2 bit allocation stages if fast gain changes */
1062 if(blk && prev != s->fast_gain[ch])
1063 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1065 } else if (s->eac3 && !blk) {
1066 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1067 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1070 /* E-AC-3 to AC-3 converter SNR offset */
1071 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1072 skip_bits(gbc, 10); // skip converter snr offset
1075 /* coupling leak information */
1077 if (s->first_cpl_leak || get_bits1(gbc)) {
1078 int fl = get_bits(gbc, 3);
1079 int sl = get_bits(gbc, 3);
1080 /* run last 2 bit allocation stages for coupling channel if
1081 coupling leak changes */
1082 if(blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1083 sl != s->bit_alloc_params.cpl_slow_leak)) {
1084 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1086 s->bit_alloc_params.cpl_fast_leak = fl;
1087 s->bit_alloc_params.cpl_slow_leak = sl;
1088 } else if (!s->eac3 && !blk) {
1089 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1092 s->first_cpl_leak = 0;
1095 /* delta bit allocation information */
1096 if (s->dba_syntax && get_bits1(gbc)) {
1097 /* delta bit allocation exists (strategy) */
1098 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1099 s->dba_mode[ch] = get_bits(gbc, 2);
1100 if (s->dba_mode[ch] == DBA_RESERVED) {
1101 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1104 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1106 /* channel delta offset, len and bit allocation */
1107 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1108 if (s->dba_mode[ch] == DBA_NEW) {
1109 s->dba_nsegs[ch] = get_bits(gbc, 3);
1110 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1111 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1112 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1113 s->dba_values[ch][seg] = get_bits(gbc, 3);
1115 /* run last 2 bit allocation stages if new dba values */
1116 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1119 } else if(blk == 0) {
1120 for(ch=0; ch<=s->channels; ch++) {
1121 s->dba_mode[ch] = DBA_NONE;
1125 /* Bit allocation */
1126 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
1127 if(bit_alloc_stages[ch] > 2) {
1128 /* Exponent mapping into PSD and PSD integration */
1129 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1130 s->start_freq[ch], s->end_freq[ch],
1131 s->psd[ch], s->band_psd[ch]);
1133 if(bit_alloc_stages[ch] > 1) {
1134 /* Compute excitation function, Compute masking curve, and
1135 Apply delta bit allocation */
1136 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1137 s->start_freq[ch], s->end_freq[ch],
1138 s->fast_gain[ch], (ch == s->lfe_ch),
1139 s->dba_mode[ch], s->dba_nsegs[ch],
1140 s->dba_offsets[ch], s->dba_lengths[ch],
1141 s->dba_values[ch], s->mask[ch]);
1143 if(bit_alloc_stages[ch] > 0) {
1144 /* Compute bit allocation */
1145 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1146 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1147 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1148 s->start_freq[ch], s->end_freq[ch],
1150 s->bit_alloc_params.floor,
1151 bap_tab, s->bap[ch]);
1155 /* unused dummy data */
1156 if (s->skip_syntax && get_bits1(gbc)) {
1157 int skipl = get_bits(gbc, 9);
1162 /* unpack the transform coefficients
1163 this also uncouples channels if coupling is in use. */
1164 decode_transform_coeffs(s, blk);
1166 /* TODO: generate enhanced coupling coordinates and uncouple */
1168 /* TODO: apply spectral extension */
1170 /* recover coefficients if rematrixing is in use */
1171 if(s->channel_mode == AC3_CHMODE_STEREO)
1174 /* apply scaling to coefficients (headroom, dynrng) */
1175 for(ch=1; ch<=s->channels; ch++) {
1176 float gain = s->mul_bias / 4194304.0f;
1177 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1178 gain *= s->dynamic_range[ch-1];
1180 gain *= s->dynamic_range[0];
1182 s->dsp.int32_to_float_fmul_scalar(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1185 /* downmix and MDCT. order depends on whether block switching is used for
1186 any channel in this block. this is because coefficients for the long
1187 and short transforms cannot be mixed. */
1188 downmix_output = s->channels != s->out_channels &&
1189 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1190 s->fbw_channels == s->out_channels);
1191 if(different_transforms) {
1192 /* the delay samples have already been downmixed, so we upmix the delay
1193 samples in order to reconstruct all channels before downmixing. */
1199 do_imdct(s, s->channels);
1201 if(downmix_output) {
1202 s->dsp.ac3_downmix(s->output, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1205 if(downmix_output) {
1206 s->dsp.ac3_downmix(s->transform_coeffs+1, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1209 if(downmix_output && !s->downmixed) {
1211 s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, s->fbw_channels, 128);
1214 do_imdct(s, s->out_channels);
1221 * Decode a single AC-3 frame.
1223 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1224 const uint8_t *buf, int buf_size)
1226 AC3DecodeContext *s = avctx->priv_data;
1227 int16_t *out_samples = (int16_t *)data;
1230 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1231 if (s->input_buffer) {
1232 /* copy input buffer to decoder context to avoid reading past the end
1233 of the buffer, which can be caused by a damaged input stream. */
1234 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1235 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1237 init_get_bits(&s->gbc, buf, buf_size * 8);
1240 /* parse the syncinfo */
1242 err = parse_frame_header(s);
1244 /* check that reported frame size fits in input buffer */
1245 if(s->frame_size > buf_size) {
1246 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1247 err = AC3_PARSE_ERROR_FRAME_SIZE;
1250 /* check for crc mismatch */
1251 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_recognition >= FF_ER_CAREFUL) {
1252 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1253 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1254 err = AC3_PARSE_ERROR_CRC;
1258 if(err && err != AC3_PARSE_ERROR_CRC) {
1260 case AC3_PARSE_ERROR_SYNC:
1261 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1263 case AC3_PARSE_ERROR_BSID:
1264 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1266 case AC3_PARSE_ERROR_SAMPLE_RATE:
1267 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1269 case AC3_PARSE_ERROR_FRAME_SIZE:
1270 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1272 case AC3_PARSE_ERROR_FRAME_TYPE:
1273 /* skip frame if CRC is ok. otherwise use error concealment. */
1274 /* TODO: add support for substreams and dependent frames */
1275 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1276 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1277 return s->frame_size;
1279 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1283 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1288 /* if frame is ok, set audio parameters */
1290 avctx->sample_rate = s->sample_rate;
1291 avctx->bit_rate = s->bit_rate;
1293 /* channel config */
1294 s->out_channels = s->channels;
1295 s->output_mode = s->channel_mode;
1297 s->output_mode |= AC3_OUTPUT_LFEON;
1298 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1299 avctx->request_channels < s->channels) {
1300 s->out_channels = avctx->request_channels;
1301 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1303 avctx->channels = s->out_channels;
1305 /* set downmixing coefficients if needed */
1306 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1307 s->fbw_channels == s->out_channels)) {
1308 set_downmix_coeffs(s);
1310 } else if (!s->out_channels) {
1311 s->out_channels = avctx->channels;
1312 if(s->out_channels < s->channels)
1313 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1316 /* decode the audio blocks */
1317 for (blk = 0; blk < s->num_blocks; blk++) {
1318 const float *output[s->out_channels];
1319 if (!err && decode_audio_block(s, blk)) {
1320 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1323 for (ch = 0; ch < s->out_channels; ch++)
1324 output[ch] = s->output[ch];
1325 s->dsp.float_to_int16_interleave(out_samples, output, 256, s->out_channels);
1326 out_samples += 256 * s->out_channels;
1328 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1329 return s->frame_size;
1333 * Uninitialize the AC-3 decoder.
1335 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1337 AC3DecodeContext *s = avctx->priv_data;
1338 ff_mdct_end(&s->imdct_512);
1339 ff_mdct_end(&s->imdct_256);
1341 av_freep(&s->input_buffer);
1346 AVCodec ac3_decoder = {
1348 .type = CODEC_TYPE_AUDIO,
1350 .priv_data_size = sizeof (AC3DecodeContext),
1351 .init = ac3_decode_init,
1352 .close = ac3_decode_end,
1353 .decode = ac3_decode_frame,
1354 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1357 AVCodec eac3_decoder = {
1359 .type = CODEC_TYPE_AUDIO,
1360 .id = CODEC_ID_EAC3,
1361 .priv_data_size = sizeof (AC3DecodeContext),
1362 .init = ac3_decode_init,
1363 .close = ac3_decode_end,
1364 .decode = ac3_decode_frame,
1365 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),