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 void 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; i<ngrps*3; i++) {
394 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
395 for(j=0; j<group_size; j++) {
396 dexps[(i*group_size)+j] = prevexp;
402 * Generate transform coefficients for each coupled channel in the coupling
403 * range using the coupling coefficients and coupling coordinates.
404 * reference: Section 7.4.3 Coupling Coordinate Format
406 static void calc_transform_coeffs_cpl(AC3DecodeContext *s)
408 int i, j, ch, bnd, subbnd;
411 i = s->start_freq[CPL_CH];
412 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
415 for(j=0; j<12; j++) {
416 for(ch=1; ch<=s->fbw_channels; ch++) {
417 if(s->channel_in_cpl[ch]) {
418 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
419 if (ch == 2 && s->phase_flags[bnd])
420 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
425 } while(s->cpl_band_struct[subbnd]);
430 * Grouped mantissas for 3-level 5-level and 11-level quantization
442 * Decode the transform coefficients for a particular channel
443 * reference: Section 7.3 Quantization and Decoding of Mantissas
445 static void ac3_decode_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
447 GetBitContext *gbc = &s->gbc;
448 int i, gcode, tbap, start, end;
453 exps = s->dexps[ch_index];
454 bap = s->bap[ch_index];
455 coeffs = s->fixed_coeffs[ch_index];
456 start = s->start_freq[ch_index];
457 end = s->end_freq[ch_index];
459 for (i = start; i < end; i++) {
463 coeffs[i] = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
468 gcode = get_bits(gbc, 5);
469 m->b1_mant[0] = b1_mantissas[gcode][0];
470 m->b1_mant[1] = b1_mantissas[gcode][1];
471 m->b1_mant[2] = b1_mantissas[gcode][2];
474 coeffs[i] = m->b1_mant[m->b1ptr++];
479 gcode = get_bits(gbc, 7);
480 m->b2_mant[0] = b2_mantissas[gcode][0];
481 m->b2_mant[1] = b2_mantissas[gcode][1];
482 m->b2_mant[2] = b2_mantissas[gcode][2];
485 coeffs[i] = m->b2_mant[m->b2ptr++];
489 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
494 gcode = get_bits(gbc, 7);
495 m->b4_mant[0] = b4_mantissas[gcode][0];
496 m->b4_mant[1] = b4_mantissas[gcode][1];
499 coeffs[i] = m->b4_mant[m->b4ptr++];
503 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
507 /* asymmetric dequantization */
508 int qlevel = quantization_tab[tbap];
509 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
513 coeffs[i] >>= exps[i];
518 * Remove random dithering from coefficients with zero-bit mantissas
519 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
521 static void remove_dithering(AC3DecodeContext *s) {
527 for(ch=1; ch<=s->fbw_channels; ch++) {
528 if(!s->dither_flag[ch]) {
529 coeffs = s->fixed_coeffs[ch];
531 if(s->channel_in_cpl[ch])
532 end = s->start_freq[CPL_CH];
534 end = s->end_freq[ch];
535 for(i=0; i<end; i++) {
539 if(s->channel_in_cpl[ch]) {
540 bap = s->bap[CPL_CH];
541 for(; i<s->end_freq[CPL_CH]; i++) {
550 static void decode_transform_coeffs_ch(AC3DecodeContext *s, int blk, int ch,
553 if (!s->channel_uses_aht[ch]) {
554 ac3_decode_transform_coeffs_ch(s, ch, m);
556 /* if AHT is used, mantissas for all blocks are encoded in the first
557 block of the frame. */
560 ff_eac3_decode_transform_coeffs_aht_ch(s, ch);
561 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
562 s->fixed_coeffs[ch][bin] = s->pre_mantissa[ch][bin][blk] >> s->dexps[ch][bin];
568 * Decode the transform coefficients.
570 static void decode_transform_coeffs(AC3DecodeContext *s, int blk)
576 m.b1ptr = m.b2ptr = m.b4ptr = 3;
578 for (ch = 1; ch <= s->channels; ch++) {
579 /* transform coefficients for full-bandwidth channel */
580 decode_transform_coeffs_ch(s, blk, ch, &m);
581 /* tranform coefficients for coupling channel come right after the
582 coefficients for the first coupled channel*/
583 if (s->channel_in_cpl[ch]) {
585 decode_transform_coeffs_ch(s, blk, CPL_CH, &m);
586 calc_transform_coeffs_cpl(s);
589 end = s->end_freq[CPL_CH];
591 end = s->end_freq[ch];
594 s->fixed_coeffs[ch][end] = 0;
598 /* zero the dithered coefficients for appropriate channels */
603 * Stereo rematrixing.
604 * reference: Section 7.5.4 Rematrixing : Decoding Technique
606 static void do_rematrixing(AC3DecodeContext *s)
612 end = FFMIN(s->end_freq[1], s->end_freq[2]);
614 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
615 if(s->rematrixing_flags[bnd]) {
616 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
617 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
618 tmp0 = s->fixed_coeffs[1][i];
619 tmp1 = s->fixed_coeffs[2][i];
620 s->fixed_coeffs[1][i] = tmp0 + tmp1;
621 s->fixed_coeffs[2][i] = tmp0 - tmp1;
628 * Inverse MDCT Transform.
629 * Convert frequency domain coefficients to time-domain audio samples.
630 * reference: Section 7.9.4 Transformation Equations
632 static inline void do_imdct(AC3DecodeContext *s, int channels)
635 float add_bias = s->add_bias;
636 if(s->out_channels==1 && channels>1)
637 add_bias *= LEVEL_MINUS_3DB; // compensate for the gain in downmix
639 for (ch=1; ch<=channels; ch++) {
640 if (s->block_switch[ch]) {
642 float *x = s->tmp_output+128;
644 x[i] = s->transform_coeffs[ch][2*i];
645 ff_imdct_half(&s->imdct_256, s->tmp_output, x);
646 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
648 x[i] = s->transform_coeffs[ch][2*i+1];
649 ff_imdct_half(&s->imdct_256, s->delay[ch-1], x);
651 ff_imdct_half(&s->imdct_512, s->tmp_output, s->transform_coeffs[ch]);
652 s->dsp.vector_fmul_window(s->output[ch-1], s->delay[ch-1], s->tmp_output, s->window, add_bias, 128);
653 memcpy(s->delay[ch-1], s->tmp_output+128, 128*sizeof(float));
659 * Downmix the output to mono or stereo.
661 void ff_ac3_downmix_c(float (*samples)[256], float (*matrix)[2], int out_ch, int in_ch, int len)
666 for(i=0; i<len; i++) {
668 for(j=0; j<in_ch; j++) {
669 v0 += samples[j][i] * matrix[j][0];
670 v1 += samples[j][i] * matrix[j][1];
675 } else if(out_ch == 1) {
676 for(i=0; i<len; i++) {
678 for(j=0; j<in_ch; j++)
679 v0 += samples[j][i] * matrix[j][0];
686 * Upmix delay samples from stereo to original channel layout.
688 static void ac3_upmix_delay(AC3DecodeContext *s)
690 int channel_data_size = sizeof(s->delay[0]);
691 switch(s->channel_mode) {
692 case AC3_CHMODE_DUALMONO:
693 case AC3_CHMODE_STEREO:
694 /* upmix mono to stereo */
695 memcpy(s->delay[1], s->delay[0], channel_data_size);
697 case AC3_CHMODE_2F2R:
698 memset(s->delay[3], 0, channel_data_size);
699 case AC3_CHMODE_2F1R:
700 memset(s->delay[2], 0, channel_data_size);
702 case AC3_CHMODE_3F2R:
703 memset(s->delay[4], 0, channel_data_size);
704 case AC3_CHMODE_3F1R:
705 memset(s->delay[3], 0, channel_data_size);
707 memcpy(s->delay[2], s->delay[1], channel_data_size);
708 memset(s->delay[1], 0, channel_data_size);
714 * Decode a single audio block from the AC-3 bitstream.
716 static int decode_audio_block(AC3DecodeContext *s, int blk)
718 int fbw_channels = s->fbw_channels;
719 int channel_mode = s->channel_mode;
721 int different_transforms;
724 GetBitContext *gbc = &s->gbc;
725 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
727 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
729 /* block switch flags */
730 different_transforms = 0;
731 if (s->block_switch_syntax) {
732 for (ch = 1; ch <= fbw_channels; ch++) {
733 s->block_switch[ch] = get_bits1(gbc);
734 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
735 different_transforms = 1;
739 /* dithering flags */
740 if (s->dither_flag_syntax) {
741 for (ch = 1; ch <= fbw_channels; ch++) {
742 s->dither_flag[ch] = get_bits1(gbc);
747 i = !(s->channel_mode);
750 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
751 s->avctx->drc_scale)+1.0;
752 } else if(blk == 0) {
753 s->dynamic_range[i] = 1.0f;
757 /* spectral extension strategy */
758 if (s->eac3 && (!blk || get_bits1(gbc))) {
759 if (get_bits1(gbc)) {
760 av_log_missing_feature(s->avctx, "Spectral extension", 1);
763 /* TODO: parse spectral extension strategy info */
766 /* TODO: spectral extension coordinates */
768 /* coupling strategy */
769 if (s->eac3 ? s->cpl_strategy_exists[blk] : get_bits1(gbc)) {
770 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
772 s->cpl_in_use[blk] = get_bits1(gbc);
773 if (s->cpl_in_use[blk]) {
774 /* coupling in use */
775 int cpl_start_subband, cpl_end_subband;
777 if (channel_mode < AC3_CHMODE_STEREO) {
778 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
782 /* check for enhanced coupling */
783 if (s->eac3 && get_bits1(gbc)) {
784 /* TODO: parse enhanced coupling strategy info */
785 av_log_missing_feature(s->avctx, "Enhanced coupling", 1);
789 /* determine which channels are coupled */
790 if (s->eac3 && s->channel_mode == AC3_CHMODE_STEREO) {
791 s->channel_in_cpl[1] = 1;
792 s->channel_in_cpl[2] = 1;
794 for (ch = 1; ch <= fbw_channels; ch++)
795 s->channel_in_cpl[ch] = get_bits1(gbc);
798 /* phase flags in use */
799 if (channel_mode == AC3_CHMODE_STEREO)
800 s->phase_flags_in_use = get_bits1(gbc);
802 /* coupling frequency range */
803 /* TODO: modify coupling end freq if spectral extension is used */
804 cpl_start_subband = get_bits(gbc, 4);
805 cpl_end_subband = get_bits(gbc, 4) + 3;
806 s->num_cpl_subbands = cpl_end_subband - cpl_start_subband;
807 if (s->num_cpl_subbands < 0) {
808 av_log(s->avctx, AV_LOG_ERROR, "invalid coupling range (%d > %d)\n",
809 cpl_start_subband, cpl_end_subband);
812 s->start_freq[CPL_CH] = cpl_start_subband * 12 + 37;
813 s->end_freq[CPL_CH] = cpl_end_subband * 12 + 37;
815 /* coupling band structure */
816 s->num_cpl_bands = s->num_cpl_subbands;
817 if (!s->eac3 || get_bits1(gbc)) {
818 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
819 s->cpl_band_struct[bnd] = get_bits1(gbc);
822 memcpy(s->cpl_band_struct,
823 &ff_eac3_default_cpl_band_struct[cpl_start_subband+1],
824 s->num_cpl_subbands-1);
826 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
828 /* calculate number of coupling bands based on band structure */
829 for (bnd = 0; bnd < s->num_cpl_subbands-1; bnd++) {
830 s->num_cpl_bands -= s->cpl_band_struct[bnd];
833 /* coupling not in use */
834 for (ch = 1; ch <= fbw_channels; ch++) {
835 s->channel_in_cpl[ch] = 0;
836 s->first_cpl_coords[ch] = 1;
838 s->first_cpl_leak = s->eac3;
839 s->phase_flags_in_use = 0;
841 } else if (!s->eac3) {
843 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
846 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
849 cpl_in_use = s->cpl_in_use[blk];
851 /* coupling coordinates */
853 int cpl_coords_exist = 0;
855 for (ch = 1; ch <= fbw_channels; ch++) {
856 if (s->channel_in_cpl[ch]) {
857 if ((s->eac3 && s->first_cpl_coords[ch]) || get_bits1(gbc)) {
858 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
859 s->first_cpl_coords[ch] = 0;
860 cpl_coords_exist = 1;
861 master_cpl_coord = 3 * get_bits(gbc, 2);
862 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
863 cpl_coord_exp = get_bits(gbc, 4);
864 cpl_coord_mant = get_bits(gbc, 4);
865 if (cpl_coord_exp == 15)
866 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
868 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
869 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
872 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
876 /* channel not in coupling */
877 s->first_cpl_coords[ch] = 1;
881 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
882 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
883 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
888 /* stereo rematrixing strategy and band structure */
889 if (channel_mode == AC3_CHMODE_STEREO) {
890 if ((s->eac3 && !blk) || get_bits1(gbc)) {
891 s->num_rematrixing_bands = 4;
892 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
893 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
894 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
895 s->rematrixing_flags[bnd] = get_bits1(gbc);
897 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
902 /* exponent strategies for each channel */
903 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
905 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
906 if(s->exp_strategy[blk][ch] != EXP_REUSE)
907 bit_alloc_stages[ch] = 3;
910 /* channel bandwidth */
911 for (ch = 1; ch <= fbw_channels; ch++) {
912 s->start_freq[ch] = 0;
913 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
915 int prev = s->end_freq[ch];
916 if (s->channel_in_cpl[ch])
917 s->end_freq[ch] = s->start_freq[CPL_CH];
919 int bandwidth_code = get_bits(gbc, 6);
920 if (bandwidth_code > 60) {
921 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
924 s->end_freq[ch] = bandwidth_code * 3 + 73;
926 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
927 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
928 if(blk > 0 && s->end_freq[ch] != prev)
929 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
932 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
933 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
934 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
937 /* decode exponents for each channel */
938 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
939 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
940 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
941 decode_exponents(gbc, s->exp_strategy[blk][ch],
942 s->num_exp_groups[ch], s->dexps[ch][0],
943 &s->dexps[ch][s->start_freq[ch]+!!ch]);
944 if(ch != CPL_CH && ch != s->lfe_ch)
945 skip_bits(gbc, 2); /* skip gainrng */
949 /* bit allocation information */
950 if (s->bit_allocation_syntax) {
951 if (get_bits1(gbc)) {
952 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
953 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
954 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
955 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
956 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
957 for(ch=!cpl_in_use; ch<=s->channels; ch++)
958 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
960 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
965 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
966 if(!s->eac3 || !blk){
967 if(s->snr_offset_strategy && get_bits1(gbc)) {
970 csnr = (get_bits(gbc, 6) - 15) << 4;
971 for (i = ch = !cpl_in_use; ch <= s->channels; ch++) {
973 if (ch == i || s->snr_offset_strategy == 2)
974 snr = (csnr + get_bits(gbc, 4)) << 2;
975 /* run at least last bit allocation stage if snr offset changes */
976 if(blk && s->snr_offset[ch] != snr) {
977 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 1);
979 s->snr_offset[ch] = snr;
981 /* fast gain (normal AC-3 only) */
983 int prev = s->fast_gain[ch];
984 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
985 /* run last 2 bit allocation stages if fast gain changes */
986 if(blk && prev != s->fast_gain[ch])
987 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
990 } else if (!s->eac3 && !blk) {
991 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
996 /* fast gain (E-AC-3 only) */
997 if (s->fast_gain_syntax && get_bits1(gbc)) {
998 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
999 int prev = s->fast_gain[ch];
1000 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
1001 /* run last 2 bit allocation stages if fast gain changes */
1002 if(blk && prev != s->fast_gain[ch])
1003 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1005 } else if (s->eac3 && !blk) {
1006 for (ch = !cpl_in_use; ch <= s->channels; ch++)
1007 s->fast_gain[ch] = ff_ac3_fast_gain_tab[4];
1010 /* E-AC-3 to AC-3 converter SNR offset */
1011 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && get_bits1(gbc)) {
1012 skip_bits(gbc, 10); // skip converter snr offset
1015 /* coupling leak information */
1017 if (s->first_cpl_leak || get_bits1(gbc)) {
1018 int fl = get_bits(gbc, 3);
1019 int sl = get_bits(gbc, 3);
1020 /* run last 2 bit allocation stages for coupling channel if
1021 coupling leak changes */
1022 if(blk && (fl != s->bit_alloc_params.cpl_fast_leak ||
1023 sl != s->bit_alloc_params.cpl_slow_leak)) {
1024 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
1026 s->bit_alloc_params.cpl_fast_leak = fl;
1027 s->bit_alloc_params.cpl_slow_leak = sl;
1028 } else if (!s->eac3 && !blk) {
1029 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1032 s->first_cpl_leak = 0;
1035 /* delta bit allocation information */
1036 if (s->dba_syntax && get_bits1(gbc)) {
1037 /* delta bit allocation exists (strategy) */
1038 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1039 s->dba_mode[ch] = get_bits(gbc, 2);
1040 if (s->dba_mode[ch] == DBA_RESERVED) {
1041 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1044 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1046 /* channel delta offset, len and bit allocation */
1047 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
1048 if (s->dba_mode[ch] == DBA_NEW) {
1049 s->dba_nsegs[ch] = get_bits(gbc, 3);
1050 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1051 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1052 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1053 s->dba_values[ch][seg] = get_bits(gbc, 3);
1055 /* run last 2 bit allocation stages if new dba values */
1056 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1059 } else if(blk == 0) {
1060 for(ch=0; ch<=s->channels; ch++) {
1061 s->dba_mode[ch] = DBA_NONE;
1065 /* Bit allocation */
1066 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
1067 if(bit_alloc_stages[ch] > 2) {
1068 /* Exponent mapping into PSD and PSD integration */
1069 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1070 s->start_freq[ch], s->end_freq[ch],
1071 s->psd[ch], s->band_psd[ch]);
1073 if(bit_alloc_stages[ch] > 1) {
1074 /* Compute excitation function, Compute masking curve, and
1075 Apply delta bit allocation */
1076 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1077 s->start_freq[ch], s->end_freq[ch],
1078 s->fast_gain[ch], (ch == s->lfe_ch),
1079 s->dba_mode[ch], s->dba_nsegs[ch],
1080 s->dba_offsets[ch], s->dba_lengths[ch],
1081 s->dba_values[ch], s->mask[ch]);
1083 if(bit_alloc_stages[ch] > 0) {
1084 /* Compute bit allocation */
1085 const uint8_t *bap_tab = s->channel_uses_aht[ch] ?
1086 ff_eac3_hebap_tab : ff_ac3_bap_tab;
1087 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1088 s->start_freq[ch], s->end_freq[ch],
1090 s->bit_alloc_params.floor,
1091 bap_tab, s->bap[ch]);
1095 /* unused dummy data */
1096 if (s->skip_syntax && get_bits1(gbc)) {
1097 int skipl = get_bits(gbc, 9);
1102 /* unpack the transform coefficients
1103 this also uncouples channels if coupling is in use. */
1104 decode_transform_coeffs(s, blk);
1106 /* TODO: generate enhanced coupling coordinates and uncouple */
1108 /* TODO: apply spectral extension */
1110 /* recover coefficients if rematrixing is in use */
1111 if(s->channel_mode == AC3_CHMODE_STEREO)
1114 /* apply scaling to coefficients (headroom, dynrng) */
1115 for(ch=1; ch<=s->channels; ch++) {
1116 float gain = s->mul_bias / 4194304.0f;
1117 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1118 gain *= s->dynamic_range[ch-1];
1120 gain *= s->dynamic_range[0];
1122 s->dsp.int32_to_float_fmul_scalar(s->transform_coeffs[ch], s->fixed_coeffs[ch], gain, 256);
1125 /* downmix and MDCT. order depends on whether block switching is used for
1126 any channel in this block. this is because coefficients for the long
1127 and short transforms cannot be mixed. */
1128 downmix_output = s->channels != s->out_channels &&
1129 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1130 s->fbw_channels == s->out_channels);
1131 if(different_transforms) {
1132 /* the delay samples have already been downmixed, so we upmix the delay
1133 samples in order to reconstruct all channels before downmixing. */
1139 do_imdct(s, s->channels);
1141 if(downmix_output) {
1142 s->dsp.ac3_downmix(s->output, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1145 if(downmix_output) {
1146 s->dsp.ac3_downmix(s->transform_coeffs+1, s->downmix_coeffs, s->out_channels, s->fbw_channels, 256);
1149 if(downmix_output && !s->downmixed) {
1151 s->dsp.ac3_downmix(s->delay, s->downmix_coeffs, s->out_channels, s->fbw_channels, 128);
1154 do_imdct(s, s->out_channels);
1161 * Decode a single AC-3 frame.
1163 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1164 const uint8_t *buf, int buf_size)
1166 AC3DecodeContext *s = avctx->priv_data;
1167 int16_t *out_samples = (int16_t *)data;
1170 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1171 if (s->input_buffer) {
1172 /* copy input buffer to decoder context to avoid reading past the end
1173 of the buffer, which can be caused by a damaged input stream. */
1174 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_FRAME_BUFFER_SIZE));
1175 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1177 init_get_bits(&s->gbc, buf, buf_size * 8);
1180 /* parse the syncinfo */
1182 err = parse_frame_header(s);
1184 /* check that reported frame size fits in input buffer */
1185 if(s->frame_size > buf_size) {
1186 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1187 err = AC3_PARSE_ERROR_FRAME_SIZE;
1190 /* check for crc mismatch */
1191 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_recognition >= FF_ER_CAREFUL) {
1192 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1193 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1194 err = AC3_PARSE_ERROR_CRC;
1198 if(err && err != AC3_PARSE_ERROR_CRC) {
1200 case AC3_PARSE_ERROR_SYNC:
1201 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1203 case AC3_PARSE_ERROR_BSID:
1204 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1206 case AC3_PARSE_ERROR_SAMPLE_RATE:
1207 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1209 case AC3_PARSE_ERROR_FRAME_SIZE:
1210 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1212 case AC3_PARSE_ERROR_FRAME_TYPE:
1213 /* skip frame if CRC is ok. otherwise use error concealment. */
1214 /* TODO: add support for substreams and dependent frames */
1215 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1216 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1217 return s->frame_size;
1219 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1223 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1228 /* if frame is ok, set audio parameters */
1230 avctx->sample_rate = s->sample_rate;
1231 avctx->bit_rate = s->bit_rate;
1233 /* channel config */
1234 s->out_channels = s->channels;
1235 s->output_mode = s->channel_mode;
1237 s->output_mode |= AC3_OUTPUT_LFEON;
1238 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1239 avctx->request_channels < s->channels) {
1240 s->out_channels = avctx->request_channels;
1241 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1243 avctx->channels = s->out_channels;
1245 /* set downmixing coefficients if needed */
1246 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1247 s->fbw_channels == s->out_channels)) {
1248 set_downmix_coeffs(s);
1250 } else if (!s->out_channels) {
1251 s->out_channels = avctx->channels;
1252 if(s->out_channels < s->channels)
1253 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1256 /* decode the audio blocks */
1257 for (blk = 0; blk < s->num_blocks; blk++) {
1258 const float *output[s->out_channels];
1259 if (!err && decode_audio_block(s, blk)) {
1260 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1262 for (ch = 0; ch < s->out_channels; ch++)
1263 output[ch] = s->output[ch];
1264 s->dsp.float_to_int16_interleave(out_samples, output, 256, s->out_channels);
1265 out_samples += 256 * s->out_channels;
1267 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1268 return s->frame_size;
1272 * Uninitialize the AC-3 decoder.
1274 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1276 AC3DecodeContext *s = avctx->priv_data;
1277 ff_mdct_end(&s->imdct_512);
1278 ff_mdct_end(&s->imdct_256);
1280 av_freep(&s->input_buffer);
1285 AVCodec ac3_decoder = {
1287 .type = CODEC_TYPE_AUDIO,
1289 .priv_data_size = sizeof (AC3DecodeContext),
1290 .init = ac3_decode_init,
1291 .close = ac3_decode_end,
1292 .decode = ac3_decode_frame,
1293 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52A (AC-3)"),
1296 AVCodec eac3_decoder = {
1298 .type = CODEC_TYPE_AUDIO,
1299 .id = CODEC_ID_EAC3,
1300 .priv_data_size = sizeof (AC3DecodeContext),
1301 .init = ac3_decode_init,
1302 .close = ac3_decode_end,
1303 .decode = ac3_decode_frame,
1304 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52B (AC-3, E-AC-3)"),