3 * This code is developed as part of Google Summer of Code 2006 Program.
5 * Copyright (c) 2006 Kartikey Mahendra BHATT (bhattkm at gmail dot com).
6 * Copyright (c) 2007 Justin Ruggles
8 * Portions of this code are derived from liba52
9 * http://liba52.sourceforge.net
10 * Copyright (C) 2000-2003 Michel Lespinasse <walken@zoy.org>
11 * Copyright (C) 1999-2000 Aaron Holtzman <aholtzma@ess.engr.uvic.ca>
13 * This file is part of FFmpeg.
15 * FFmpeg is free software; you can redistribute it and/or
16 * modify it under the terms of the GNU General Public
17 * License as published by the Free Software Foundation; either
18 * version 2 of the License, or (at your option) any later version.
20 * FFmpeg is distributed in the hope that it will be useful,
21 * but WITHOUT ANY WARRANTY; without even the implied warranty of
22 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
23 * General Public License for more details.
25 * You should have received a copy of the GNU General Public
26 * License along with FFmpeg; if not, write to the Free Software
27 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
35 #include "libavutil/crc.h"
36 #include "libavutil/random.h"
38 #include "ac3_parser.h"
39 #include "bitstream.h"
42 #include "ac3dec_data.h"
44 /** Maximum possible frame size when the specification limit is ignored */
45 #define AC3_MAX_FRAME_SIZE 21695
47 /** table for grouping exponents */
48 static uint8_t exp_ungroup_tab[128][3];
51 /** tables for ungrouping mantissas */
52 static int b1_mantissas[32][3];
53 static int b2_mantissas[128][3];
54 static int b3_mantissas[8];
55 static int b4_mantissas[128][2];
56 static int b5_mantissas[16];
59 * Quantization table: levels for symmetric. bits for asymmetric.
60 * reference: Table 7.18 Mapping of bap to Quantizer
62 static const uint8_t quantization_tab[16] = {
64 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
67 /** dynamic range table. converts codes to scale factors. */
68 static float dynamic_range_tab[256];
70 /** Adjustments in dB gain */
71 #define LEVEL_PLUS_3DB 1.4142135623730950
72 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
73 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
74 #define LEVEL_MINUS_3DB 0.7071067811865476
75 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
76 #define LEVEL_MINUS_6DB 0.5000000000000000
77 #define LEVEL_MINUS_9DB 0.3535533905932738
78 #define LEVEL_ZERO 0.0000000000000000
79 #define LEVEL_ONE 1.0000000000000000
81 static const float gain_levels[9] = {
85 LEVEL_MINUS_1POINT5DB,
87 LEVEL_MINUS_4POINT5DB,
94 * Table for center mix levels
95 * reference: Section 5.4.2.4 cmixlev
97 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
100 * Table for surround mix levels
101 * reference: Section 5.4.2.5 surmixlev
103 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
106 * Table for default stereo downmixing coefficients
107 * reference: Section 7.8.2 Downmixing Into Two Channels
109 static const uint8_t ac3_default_coeffs[8][5][2] = {
110 { { 2, 7 }, { 7, 2 }, },
112 { { 2, 7 }, { 7, 2 }, },
113 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
114 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
115 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
116 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
117 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
121 * Symmetrical Dequantization
122 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
123 * Tables 7.19 to 7.23
126 symmetric_dequant(int code, int levels)
128 return ((code - (levels >> 1)) << 24) / levels;
132 * Initialize tables at runtime.
134 static av_cold void ac3_tables_init(void)
138 /* generate grouped mantissa tables
139 reference: Section 7.3.5 Ungrouping of Mantissas */
140 for(i=0; i<32; i++) {
141 /* bap=1 mantissas */
142 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
143 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
144 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
146 for(i=0; i<128; i++) {
147 /* bap=2 mantissas */
148 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
149 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
150 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
152 /* bap=4 mantissas */
153 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
154 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
156 /* generate ungrouped mantissa tables
157 reference: Tables 7.21 and 7.23 */
159 /* bap=3 mantissas */
160 b3_mantissas[i] = symmetric_dequant(i, 7);
162 for(i=0; i<15; i++) {
163 /* bap=5 mantissas */
164 b5_mantissas[i] = symmetric_dequant(i, 15);
167 /* generate dynamic range table
168 reference: Section 7.7.1 Dynamic Range Control */
169 for(i=0; i<256; i++) {
170 int v = (i >> 5) - ((i >> 7) << 3) - 5;
171 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
174 /* generate exponent tables
175 reference: Section 7.1.3 Exponent Decoding */
176 for(i=0; i<128; i++) {
177 exp_ungroup_tab[i][0] = i / 25;
178 exp_ungroup_tab[i][1] = (i % 25) / 5;
179 exp_ungroup_tab[i][2] = (i % 25) % 5;
185 * AVCodec initialization
187 static av_cold int ac3_decode_init(AVCodecContext *avctx)
189 AC3DecodeContext *s = avctx->priv_data;
194 ff_mdct_init(&s->imdct_256, 8, 1);
195 ff_mdct_init(&s->imdct_512, 9, 1);
196 ff_kbd_window_init(s->window, 5.0, 256);
197 dsputil_init(&s->dsp, avctx);
198 av_init_random(0, &s->dith_state);
200 /* set bias values for float to int16 conversion */
201 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
202 s->add_bias = 385.0f;
206 s->mul_bias = 32767.0f;
209 /* allow downmixing to stereo or mono */
210 if (avctx->channels > 0 && avctx->request_channels > 0 &&
211 avctx->request_channels < avctx->channels &&
212 avctx->request_channels <= 2) {
213 avctx->channels = avctx->request_channels;
217 /* allocate context input buffer */
218 if (avctx->error_resilience >= FF_ER_CAREFUL) {
219 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
220 if (!s->input_buffer)
221 return AVERROR_NOMEM;
224 avctx->sample_fmt = SAMPLE_FMT_S16;
229 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
230 * GetBitContext within AC3DecodeContext must point to
231 * start of the synchronized ac3 bitstream.
233 static int ac3_parse_header(AC3DecodeContext *s)
235 GetBitContext *gbc = &s->gbc;
238 /* read the rest of the bsi. read twice for dual mono mode. */
239 i = !(s->channel_mode);
241 skip_bits(gbc, 5); // skip dialog normalization
243 skip_bits(gbc, 8); //skip compression
245 skip_bits(gbc, 8); //skip language code
247 skip_bits(gbc, 7); //skip audio production information
250 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
252 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
253 TODO: read & use the xbsi1 downmix levels */
255 skip_bits(gbc, 14); //skip timecode1 / xbsi1
257 skip_bits(gbc, 14); //skip timecode2 / xbsi2
259 /* skip additional bitstream info */
260 if (get_bits1(gbc)) {
261 i = get_bits(gbc, 6);
271 * Common function to parse AC3 or E-AC3 frame header
273 static int parse_frame_header(AC3DecodeContext *s)
276 GetBitContext *gbc = &s->gbc;
279 err = ff_ac3_parse_header(gbc, &hdr);
283 if(hdr.bitstream_id > 10)
284 return AC3_PARSE_ERROR_BSID;
286 /* get decoding parameters from header info */
287 s->bit_alloc_params.sr_code = hdr.sr_code;
288 s->channel_mode = hdr.channel_mode;
289 s->lfe_on = hdr.lfe_on;
290 s->bit_alloc_params.sr_shift = hdr.sr_shift;
291 s->sample_rate = hdr.sample_rate;
292 s->bit_rate = hdr.bit_rate;
293 s->channels = hdr.channels;
294 s->fbw_channels = s->channels - s->lfe_on;
295 s->lfe_ch = s->fbw_channels + 1;
296 s->frame_size = hdr.frame_size;
297 s->center_mix_level = hdr.center_mix_level;
298 s->surround_mix_level = hdr.surround_mix_level;
299 s->num_blocks = hdr.num_blocks;
300 s->frame_type = hdr.frame_type;
301 s->substreamid = hdr.substreamid;
304 s->start_freq[s->lfe_ch] = 0;
305 s->end_freq[s->lfe_ch] = 7;
306 s->num_exp_groups[s->lfe_ch] = 2;
307 s->channel_in_cpl[s->lfe_ch] = 0;
310 return ac3_parse_header(s);
314 * Set stereo downmixing coefficients based on frame header info.
315 * reference: Section 7.8.2 Downmixing Into Two Channels
317 static void set_downmix_coeffs(AC3DecodeContext *s)
320 float cmix = gain_levels[center_levels[s->center_mix_level]];
321 float smix = gain_levels[surround_levels[s->surround_mix_level]];
323 for(i=0; i<s->fbw_channels; i++) {
324 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
325 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
327 if(s->channel_mode > 1 && s->channel_mode & 1) {
328 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
330 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
331 int nf = s->channel_mode - 2;
332 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
334 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
335 int nf = s->channel_mode - 4;
336 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
339 /* calculate adjustment needed for each channel to avoid clipping */
340 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
341 for(i=0; i<s->fbw_channels; i++) {
342 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
343 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
345 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
346 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
350 * Decode the grouped exponents according to exponent strategy.
351 * reference: Section 7.1.3 Exponent Decoding
353 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
354 uint8_t absexp, int8_t *dexps)
356 int i, j, grp, group_size;
361 group_size = exp_strategy + (exp_strategy == EXP_D45);
362 for(grp=0,i=0; grp<ngrps; grp++) {
363 expacc = get_bits(gbc, 7);
364 dexp[i++] = exp_ungroup_tab[expacc][0];
365 dexp[i++] = exp_ungroup_tab[expacc][1];
366 dexp[i++] = exp_ungroup_tab[expacc][2];
369 /* convert to absolute exps and expand groups */
371 for(i=0; i<ngrps*3; i++) {
372 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
373 for(j=0; j<group_size; j++) {
374 dexps[(i*group_size)+j] = prevexp;
380 * Generate transform coefficients for each coupled channel in the coupling
381 * range using the coupling coefficients and coupling coordinates.
382 * reference: Section 7.4.3 Coupling Coordinate Format
384 static void uncouple_channels(AC3DecodeContext *s)
386 int i, j, ch, bnd, subbnd;
389 i = s->start_freq[CPL_CH];
390 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
393 for(j=0; j<12; j++) {
394 for(ch=1; ch<=s->fbw_channels; ch++) {
395 if(s->channel_in_cpl[ch]) {
396 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
397 if (ch == 2 && s->phase_flags[bnd])
398 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
403 } while(s->cpl_band_struct[subbnd]);
408 * Grouped mantissas for 3-level 5-level and 11-level quantization
420 * Get the transform coefficients for a particular channel
421 * reference: Section 7.3 Quantization and Decoding of Mantissas
423 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
425 GetBitContext *gbc = &s->gbc;
426 int i, gcode, tbap, start, end;
431 exps = s->dexps[ch_index];
432 bap = s->bap[ch_index];
433 coeffs = s->fixed_coeffs[ch_index];
434 start = s->start_freq[ch_index];
435 end = s->end_freq[ch_index];
437 for (i = start; i < end; i++) {
441 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 0x400000;
446 gcode = get_bits(gbc, 5);
447 m->b1_mant[0] = b1_mantissas[gcode][0];
448 m->b1_mant[1] = b1_mantissas[gcode][1];
449 m->b1_mant[2] = b1_mantissas[gcode][2];
452 coeffs[i] = m->b1_mant[m->b1ptr++];
457 gcode = get_bits(gbc, 7);
458 m->b2_mant[0] = b2_mantissas[gcode][0];
459 m->b2_mant[1] = b2_mantissas[gcode][1];
460 m->b2_mant[2] = b2_mantissas[gcode][2];
463 coeffs[i] = m->b2_mant[m->b2ptr++];
467 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
472 gcode = get_bits(gbc, 7);
473 m->b4_mant[0] = b4_mantissas[gcode][0];
474 m->b4_mant[1] = b4_mantissas[gcode][1];
477 coeffs[i] = m->b4_mant[m->b4ptr++];
481 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
485 /* asymmetric dequantization */
486 int qlevel = quantization_tab[tbap];
487 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
491 coeffs[i] >>= exps[i];
496 * Remove random dithering from coefficients with zero-bit mantissas
497 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
499 static void remove_dithering(AC3DecodeContext *s) {
505 for(ch=1; ch<=s->fbw_channels; ch++) {
506 if(!s->dither_flag[ch]) {
507 coeffs = s->fixed_coeffs[ch];
509 if(s->channel_in_cpl[ch])
510 end = s->start_freq[CPL_CH];
512 end = s->end_freq[ch];
513 for(i=0; i<end; i++) {
517 if(s->channel_in_cpl[ch]) {
518 bap = s->bap[CPL_CH];
519 for(; i<s->end_freq[CPL_CH]; i++) {
529 * Get the transform coefficients.
531 static void get_transform_coeffs(AC3DecodeContext *s)
537 m.b1ptr = m.b2ptr = m.b4ptr = 3;
539 for (ch = 1; ch <= s->channels; ch++) {
540 /* transform coefficients for full-bandwidth channel */
541 get_transform_coeffs_ch(s, ch, &m);
542 /* tranform coefficients for coupling channel come right after the
543 coefficients for the first coupled channel*/
544 if (s->channel_in_cpl[ch]) {
546 get_transform_coeffs_ch(s, CPL_CH, &m);
547 uncouple_channels(s);
550 end = s->end_freq[CPL_CH];
552 end = s->end_freq[ch];
555 s->fixed_coeffs[ch][end] = 0;
559 /* if any channel doesn't use dithering, zero appropriate coefficients */
565 * Stereo rematrixing.
566 * reference: Section 7.5.4 Rematrixing : Decoding Technique
568 static void do_rematrixing(AC3DecodeContext *s)
574 end = FFMIN(s->end_freq[1], s->end_freq[2]);
576 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
577 if(s->rematrixing_flags[bnd]) {
578 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
579 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
580 tmp0 = s->fixed_coeffs[1][i];
581 tmp1 = s->fixed_coeffs[2][i];
582 s->fixed_coeffs[1][i] = tmp0 + tmp1;
583 s->fixed_coeffs[2][i] = tmp0 - tmp1;
590 * Perform the 256-point IMDCT
592 static void do_imdct_256(AC3DecodeContext *s, int chindex)
595 DECLARE_ALIGNED_16(float, x[128]);
597 float *o_ptr = s->tmp_output;
600 /* de-interleave coefficients */
601 for(k=0; k<128; k++) {
602 x[k] = s->transform_coeffs[chindex][2*k+i];
605 /* run standard IMDCT */
606 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
608 /* reverse the post-rotation & reordering from standard IMDCT */
609 for(k=0; k<32; k++) {
610 z[i][32+k].re = -o_ptr[128+2*k];
611 z[i][32+k].im = -o_ptr[2*k];
612 z[i][31-k].re = o_ptr[2*k+1];
613 z[i][31-k].im = o_ptr[128+2*k+1];
617 /* apply AC-3 post-rotation & reordering */
618 for(k=0; k<64; k++) {
619 o_ptr[ 2*k ] = -z[0][ k].im;
620 o_ptr[ 2*k+1] = z[0][63-k].re;
621 o_ptr[128+2*k ] = -z[0][ k].re;
622 o_ptr[128+2*k+1] = z[0][63-k].im;
623 o_ptr[256+2*k ] = -z[1][ k].re;
624 o_ptr[256+2*k+1] = z[1][63-k].im;
625 o_ptr[384+2*k ] = z[1][ k].im;
626 o_ptr[384+2*k+1] = -z[1][63-k].re;
631 * Inverse MDCT Transform.
632 * Convert frequency domain coefficients to time-domain audio samples.
633 * reference: Section 7.9.4 Transformation Equations
635 static inline void do_imdct(AC3DecodeContext *s, int channels)
639 for (ch=1; ch<=channels; ch++) {
640 if (s->block_switch[ch]) {
643 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
644 s->transform_coeffs[ch], s->tmp_imdct);
646 /* For the first half of the block, apply the window, add the delay
647 from the previous block, and send to output */
648 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
649 s->window, s->delay[ch-1], 0, 256, 1);
650 /* For the second half of the block, apply the window and store the
651 samples to delay, to be combined with the next block */
652 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
658 * Downmix the output to mono or stereo.
660 static void ac3_downmix(AC3DecodeContext *s,
661 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
666 for(i=0; i<256; i++) {
668 for(j=0; j<s->fbw_channels; j++) {
669 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
670 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
672 v0 *= s->downmix_coeff_adjust[0];
673 v1 *= s->downmix_coeff_adjust[1];
674 if(s->output_mode == AC3_CHMODE_MONO) {
675 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
676 } else if(s->output_mode == AC3_CHMODE_STEREO) {
677 samples[ ch_offset][i] = v0;
678 samples[1+ch_offset][i] = v1;
684 * Upmix delay samples from stereo to original channel layout.
686 static void ac3_upmix_delay(AC3DecodeContext *s)
688 int channel_data_size = sizeof(s->delay[0]);
689 switch(s->channel_mode) {
690 case AC3_CHMODE_DUALMONO:
691 case AC3_CHMODE_STEREO:
692 /* upmix mono to stereo */
693 memcpy(s->delay[1], s->delay[0], channel_data_size);
695 case AC3_CHMODE_2F2R:
696 memset(s->delay[3], 0, channel_data_size);
697 case AC3_CHMODE_2F1R:
698 memset(s->delay[2], 0, channel_data_size);
700 case AC3_CHMODE_3F2R:
701 memset(s->delay[4], 0, channel_data_size);
702 case AC3_CHMODE_3F1R:
703 memset(s->delay[3], 0, channel_data_size);
705 memcpy(s->delay[2], s->delay[1], channel_data_size);
706 memset(s->delay[1], 0, channel_data_size);
712 * Parse an audio block from AC-3 bitstream.
714 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
716 int fbw_channels = s->fbw_channels;
717 int channel_mode = s->channel_mode;
719 int different_transforms;
722 GetBitContext *gbc = &s->gbc;
723 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
725 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
727 /* block switch flags */
728 different_transforms = 0;
729 for (ch = 1; ch <= fbw_channels; ch++) {
730 s->block_switch[ch] = get_bits1(gbc);
731 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
732 different_transforms = 1;
735 /* dithering flags */
737 for (ch = 1; ch <= fbw_channels; ch++) {
738 s->dither_flag[ch] = get_bits1(gbc);
739 if(!s->dither_flag[ch])
744 i = !(s->channel_mode);
747 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
748 s->avctx->drc_scale)+1.0;
749 } else if(blk == 0) {
750 s->dynamic_range[i] = 1.0f;
754 /* coupling strategy */
755 if (get_bits1(gbc)) {
756 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
757 s->cpl_in_use[blk] = get_bits1(gbc);
758 if (s->cpl_in_use[blk]) {
759 /* coupling in use */
760 int cpl_begin_freq, cpl_end_freq;
762 if (channel_mode < AC3_CHMODE_STEREO) {
763 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
767 /* determine which channels are coupled */
768 for (ch = 1; ch <= fbw_channels; ch++)
769 s->channel_in_cpl[ch] = get_bits1(gbc);
771 /* phase flags in use */
772 if (channel_mode == AC3_CHMODE_STEREO)
773 s->phase_flags_in_use = get_bits1(gbc);
775 /* coupling frequency range and band structure */
776 cpl_begin_freq = get_bits(gbc, 4);
777 cpl_end_freq = get_bits(gbc, 4);
778 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
779 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
782 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
783 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
784 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
785 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
786 if (get_bits1(gbc)) {
787 s->cpl_band_struct[bnd] = 1;
791 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
793 /* coupling not in use */
794 for (ch = 1; ch <= fbw_channels; ch++)
795 s->channel_in_cpl[ch] = 0;
798 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
801 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
803 cpl_in_use = s->cpl_in_use[blk];
805 /* coupling coordinates */
807 int cpl_coords_exist = 0;
809 for (ch = 1; ch <= fbw_channels; ch++) {
810 if (s->channel_in_cpl[ch]) {
811 if (get_bits1(gbc)) {
812 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
813 cpl_coords_exist = 1;
814 master_cpl_coord = 3 * get_bits(gbc, 2);
815 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
816 cpl_coord_exp = get_bits(gbc, 4);
817 cpl_coord_mant = get_bits(gbc, 4);
818 if (cpl_coord_exp == 15)
819 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
821 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
822 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
825 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
831 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
832 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
833 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
838 /* stereo rematrixing strategy and band structure */
839 if (channel_mode == AC3_CHMODE_STEREO) {
840 if (get_bits1(gbc)) {
841 s->num_rematrixing_bands = 4;
842 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
843 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
844 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
845 s->rematrixing_flags[bnd] = get_bits1(gbc);
847 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
852 /* exponent strategies for each channel */
853 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
854 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
855 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
856 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
857 if(s->exp_strategy[blk][ch] != EXP_REUSE)
858 bit_alloc_stages[ch] = 3;
861 /* channel bandwidth */
862 for (ch = 1; ch <= fbw_channels; ch++) {
863 s->start_freq[ch] = 0;
864 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
866 int prev = s->end_freq[ch];
867 if (s->channel_in_cpl[ch])
868 s->end_freq[ch] = s->start_freq[CPL_CH];
870 int bandwidth_code = get_bits(gbc, 6);
871 if (bandwidth_code > 60) {
872 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
875 s->end_freq[ch] = bandwidth_code * 3 + 73;
877 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
878 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
879 if(blk > 0 && s->end_freq[ch] != prev)
880 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
883 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
884 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
885 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
888 /* decode exponents for each channel */
889 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
890 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
891 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
892 decode_exponents(gbc, s->exp_strategy[blk][ch],
893 s->num_exp_groups[ch], s->dexps[ch][0],
894 &s->dexps[ch][s->start_freq[ch]+!!ch]);
895 if(ch != CPL_CH && ch != s->lfe_ch)
896 skip_bits(gbc, 2); /* skip gainrng */
900 /* bit allocation information */
901 if (get_bits1(gbc)) {
902 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
903 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
904 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
905 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
906 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
907 for(ch=!cpl_in_use; ch<=s->channels; ch++)
908 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
910 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
914 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
915 if (get_bits1(gbc)) {
917 csnr = (get_bits(gbc, 6) - 15) << 4;
918 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
919 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
920 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
922 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
924 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
928 /* coupling leak information */
930 if (get_bits1(gbc)) {
931 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
932 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
933 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
935 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
940 /* delta bit allocation information */
941 if (get_bits1(gbc)) {
942 /* delta bit allocation exists (strategy) */
943 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
944 s->dba_mode[ch] = get_bits(gbc, 2);
945 if (s->dba_mode[ch] == DBA_RESERVED) {
946 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
949 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
951 /* channel delta offset, len and bit allocation */
952 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
953 if (s->dba_mode[ch] == DBA_NEW) {
954 s->dba_nsegs[ch] = get_bits(gbc, 3);
955 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
956 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
957 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
958 s->dba_values[ch][seg] = get_bits(gbc, 3);
960 /* run last 2 bit allocation stages if new dba values */
961 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
964 } else if(blk == 0) {
965 for(ch=0; ch<=s->channels; ch++) {
966 s->dba_mode[ch] = DBA_NONE;
971 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
972 if(bit_alloc_stages[ch] > 2) {
973 /* Exponent mapping into PSD and PSD integration */
974 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
975 s->start_freq[ch], s->end_freq[ch],
976 s->psd[ch], s->band_psd[ch]);
978 if(bit_alloc_stages[ch] > 1) {
979 /* Compute excitation function, Compute masking curve, and
980 Apply delta bit allocation */
981 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
982 s->start_freq[ch], s->end_freq[ch],
983 s->fast_gain[ch], (ch == s->lfe_ch),
984 s->dba_mode[ch], s->dba_nsegs[ch],
985 s->dba_offsets[ch], s->dba_lengths[ch],
986 s->dba_values[ch], s->mask[ch]);
988 if(bit_alloc_stages[ch] > 0) {
989 /* Compute bit allocation */
990 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
991 s->start_freq[ch], s->end_freq[ch],
993 s->bit_alloc_params.floor,
994 ff_ac3_bap_tab, s->bap[ch]);
998 /* unused dummy data */
999 if (get_bits1(gbc)) {
1000 int skipl = get_bits(gbc, 9);
1005 /* unpack the transform coefficients
1006 this also uncouples channels if coupling is in use. */
1007 get_transform_coeffs(s);
1009 /* recover coefficients if rematrixing is in use */
1010 if(s->channel_mode == AC3_CHMODE_STEREO)
1013 /* apply scaling to coefficients (headroom, dynrng) */
1014 for(ch=1; ch<=s->channels; ch++) {
1015 float gain = s->mul_bias / 4194304.0f;
1016 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1017 gain *= s->dynamic_range[ch-1];
1019 gain *= s->dynamic_range[0];
1021 for(i=0; i<256; i++) {
1022 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1026 /* downmix and MDCT. order depends on whether block switching is used for
1027 any channel in this block. this is because coefficients for the long
1028 and short transforms cannot be mixed. */
1029 downmix_output = s->channels != s->out_channels &&
1030 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1031 s->fbw_channels == s->out_channels);
1032 if(different_transforms) {
1033 /* the delay samples have already been downmixed, so we upmix the delay
1034 samples in order to reconstruct all channels before downmixing. */
1040 do_imdct(s, s->channels);
1042 if(downmix_output) {
1043 ac3_downmix(s, s->output, 0);
1046 if(downmix_output) {
1047 ac3_downmix(s, s->transform_coeffs, 1);
1052 ac3_downmix(s, s->delay, 0);
1055 do_imdct(s, s->out_channels);
1058 /* convert float to 16-bit integer */
1059 for(ch=0; ch<s->out_channels; ch++) {
1060 for(i=0; i<256; i++) {
1061 s->output[ch][i] += s->add_bias;
1063 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1070 * Decode a single AC-3 frame.
1072 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1073 const uint8_t *buf, int buf_size)
1075 AC3DecodeContext *s = avctx->priv_data;
1076 int16_t *out_samples = (int16_t *)data;
1077 int i, blk, ch, err;
1079 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1080 if (s->input_buffer) {
1081 /* copy input buffer to decoder context to avoid reading past the end
1082 of the buffer, which can be caused by a damaged input stream. */
1083 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1084 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1086 init_get_bits(&s->gbc, buf, buf_size * 8);
1089 /* parse the syncinfo */
1091 err = parse_frame_header(s);
1093 /* check that reported frame size fits in input buffer */
1094 if(s->frame_size > buf_size) {
1095 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1096 err = AC3_PARSE_ERROR_FRAME_SIZE;
1099 /* check for crc mismatch */
1100 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1101 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1102 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1103 err = AC3_PARSE_ERROR_CRC;
1107 if(err && err != AC3_PARSE_ERROR_CRC) {
1109 case AC3_PARSE_ERROR_SYNC:
1110 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1112 case AC3_PARSE_ERROR_BSID:
1113 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1115 case AC3_PARSE_ERROR_SAMPLE_RATE:
1116 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1118 case AC3_PARSE_ERROR_FRAME_SIZE:
1119 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1121 case AC3_PARSE_ERROR_FRAME_TYPE:
1122 /* skip frame if CRC is ok. otherwise use error concealment. */
1123 /* TODO: add support for substreams and dependent frames */
1124 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1125 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1126 return s->frame_size;
1128 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1132 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1137 /* if frame is ok, set audio parameters */
1139 avctx->sample_rate = s->sample_rate;
1140 avctx->bit_rate = s->bit_rate;
1142 /* channel config */
1143 s->out_channels = s->channels;
1144 s->output_mode = s->channel_mode;
1146 s->output_mode |= AC3_OUTPUT_LFEON;
1147 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1148 avctx->request_channels < s->channels) {
1149 s->out_channels = avctx->request_channels;
1150 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1152 avctx->channels = s->out_channels;
1154 /* set downmixing coefficients if needed */
1155 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1156 s->fbw_channels == s->out_channels)) {
1157 set_downmix_coeffs(s);
1159 } else if (!s->out_channels) {
1160 s->out_channels = avctx->channels;
1161 if(s->out_channels < s->channels)
1162 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1165 /* parse the audio blocks */
1166 for (blk = 0; blk < s->num_blocks; blk++) {
1167 if (!err && ac3_parse_audio_block(s, blk)) {
1168 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1171 /* interleave output samples */
1172 for (i = 0; i < 256; i++)
1173 for (ch = 0; ch < s->out_channels; ch++)
1174 *(out_samples++) = s->int_output[ch][i];
1176 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1177 return s->frame_size;
1181 * Uninitialize the AC-3 decoder.
1183 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1185 AC3DecodeContext *s = avctx->priv_data;
1186 ff_mdct_end(&s->imdct_512);
1187 ff_mdct_end(&s->imdct_256);
1189 av_freep(&s->input_buffer);
1194 AVCodec ac3_decoder = {
1196 .type = CODEC_TYPE_AUDIO,
1198 .priv_data_size = sizeof (AC3DecodeContext),
1199 .init = ac3_decode_init,
1200 .close = ac3_decode_end,
1201 .decode = ac3_decode_frame,
1202 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),