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
48 * table for ungrouping 3 values in 7 bits.
49 * used for exponents and bap=2 mantissas
51 static uint8_t ungroup_3_in_7_bits_tab[128][3];
54 /** tables for ungrouping mantissas */
55 static int b1_mantissas[32][3];
56 static int b2_mantissas[128][3];
57 static int b3_mantissas[8];
58 static int b4_mantissas[128][2];
59 static int b5_mantissas[16];
62 * Quantization table: levels for symmetric. bits for asymmetric.
63 * reference: Table 7.18 Mapping of bap to Quantizer
65 static const uint8_t quantization_tab[16] = {
67 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
70 /** dynamic range table. converts codes to scale factors. */
71 static float dynamic_range_tab[256];
73 /** Adjustments in dB gain */
74 #define LEVEL_PLUS_3DB 1.4142135623730950
75 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
76 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
77 #define LEVEL_MINUS_3DB 0.7071067811865476
78 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
79 #define LEVEL_MINUS_6DB 0.5000000000000000
80 #define LEVEL_MINUS_9DB 0.3535533905932738
81 #define LEVEL_ZERO 0.0000000000000000
82 #define LEVEL_ONE 1.0000000000000000
84 static const float gain_levels[9] = {
88 LEVEL_MINUS_1POINT5DB,
90 LEVEL_MINUS_4POINT5DB,
97 * Table for center mix levels
98 * reference: Section 5.4.2.4 cmixlev
100 static const uint8_t center_levels[4] = { 4, 5, 6, 5 };
103 * Table for surround mix levels
104 * reference: Section 5.4.2.5 surmixlev
106 static const uint8_t surround_levels[4] = { 4, 6, 7, 6 };
109 * Table for default stereo downmixing coefficients
110 * reference: Section 7.8.2 Downmixing Into Two Channels
112 static const uint8_t ac3_default_coeffs[8][5][2] = {
113 { { 2, 7 }, { 7, 2 }, },
115 { { 2, 7 }, { 7, 2 }, },
116 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
117 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
118 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
119 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
120 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
124 * Symmetrical Dequantization
125 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
126 * Tables 7.19 to 7.23
129 symmetric_dequant(int code, int levels)
131 return ((code - (levels >> 1)) << 24) / levels;
135 * Initialize tables at runtime.
137 static av_cold void ac3_tables_init(void)
141 /* generate table for ungrouping 3 values in 7 bits
142 reference: Section 7.1.3 Exponent Decoding */
143 for(i=0; i<128; i++) {
144 ungroup_3_in_7_bits_tab[i][0] = i / 25;
145 ungroup_3_in_7_bits_tab[i][1] = (i % 25) / 5;
146 ungroup_3_in_7_bits_tab[i][2] = (i % 25) % 5;
149 /* generate grouped mantissa tables
150 reference: Section 7.3.5 Ungrouping of Mantissas */
151 for(i=0; i<32; i++) {
152 /* bap=1 mantissas */
153 b1_mantissas[i][0] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][0], 3);
154 b1_mantissas[i][1] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][1], 3);
155 b1_mantissas[i][2] = symmetric_dequant(ff_ac3_ungroup_3_in_5_bits_tab[i][2], 3);
157 for(i=0; i<128; i++) {
158 /* bap=2 mantissas */
159 b2_mantissas[i][0] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][0], 5);
160 b2_mantissas[i][1] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][1], 5);
161 b2_mantissas[i][2] = symmetric_dequant(ungroup_3_in_7_bits_tab[i][2], 5);
163 /* bap=4 mantissas */
164 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
165 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
167 /* generate ungrouped mantissa tables
168 reference: Tables 7.21 and 7.23 */
170 /* bap=3 mantissas */
171 b3_mantissas[i] = symmetric_dequant(i, 7);
173 for(i=0; i<15; i++) {
174 /* bap=5 mantissas */
175 b5_mantissas[i] = symmetric_dequant(i, 15);
178 /* generate dynamic range table
179 reference: Section 7.7.1 Dynamic Range Control */
180 for(i=0; i<256; i++) {
181 int v = (i >> 5) - ((i >> 7) << 3) - 5;
182 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
188 * AVCodec initialization
190 static av_cold int ac3_decode_init(AVCodecContext *avctx)
192 AC3DecodeContext *s = avctx->priv_data;
197 ff_mdct_init(&s->imdct_256, 8, 1);
198 ff_mdct_init(&s->imdct_512, 9, 1);
199 ff_kbd_window_init(s->window, 5.0, 256);
200 dsputil_init(&s->dsp, avctx);
201 av_init_random(0, &s->dith_state);
203 /* set bias values for float to int16 conversion */
204 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
205 s->add_bias = 385.0f;
209 s->mul_bias = 32767.0f;
212 /* allow downmixing to stereo or mono */
213 if (avctx->channels > 0 && avctx->request_channels > 0 &&
214 avctx->request_channels < avctx->channels &&
215 avctx->request_channels <= 2) {
216 avctx->channels = avctx->request_channels;
220 /* allocate context input buffer */
221 if (avctx->error_resilience >= FF_ER_CAREFUL) {
222 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
223 if (!s->input_buffer)
224 return AVERROR_NOMEM;
227 avctx->sample_fmt = SAMPLE_FMT_S16;
232 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
233 * GetBitContext within AC3DecodeContext must point to
234 * the start of the synchronized AC-3 bitstream.
236 static int ac3_parse_header(AC3DecodeContext *s)
238 GetBitContext *gbc = &s->gbc;
241 /* read the rest of the bsi. read twice for dual mono mode. */
242 i = !(s->channel_mode);
244 skip_bits(gbc, 5); // skip dialog normalization
246 skip_bits(gbc, 8); //skip compression
248 skip_bits(gbc, 8); //skip language code
250 skip_bits(gbc, 7); //skip audio production information
253 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
255 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
256 TODO: read & use the xbsi1 downmix levels */
258 skip_bits(gbc, 14); //skip timecode1 / xbsi1
260 skip_bits(gbc, 14); //skip timecode2 / xbsi2
262 /* skip additional bitstream info */
263 if (get_bits1(gbc)) {
264 i = get_bits(gbc, 6);
274 * Common function to parse AC-3 or E-AC-3 frame header
276 static int parse_frame_header(AC3DecodeContext *s)
279 GetBitContext *gbc = &s->gbc;
282 err = ff_ac3_parse_header(gbc, &hdr);
286 if(hdr.bitstream_id > 10)
287 return AC3_PARSE_ERROR_BSID;
289 /* get decoding parameters from header info */
290 s->bit_alloc_params.sr_code = hdr.sr_code;
291 s->channel_mode = hdr.channel_mode;
292 s->lfe_on = hdr.lfe_on;
293 s->bit_alloc_params.sr_shift = hdr.sr_shift;
294 s->sample_rate = hdr.sample_rate;
295 s->bit_rate = hdr.bit_rate;
296 s->channels = hdr.channels;
297 s->fbw_channels = s->channels - s->lfe_on;
298 s->lfe_ch = s->fbw_channels + 1;
299 s->frame_size = hdr.frame_size;
300 s->center_mix_level = hdr.center_mix_level;
301 s->surround_mix_level = hdr.surround_mix_level;
302 s->num_blocks = hdr.num_blocks;
303 s->frame_type = hdr.frame_type;
304 s->substreamid = hdr.substreamid;
307 s->start_freq[s->lfe_ch] = 0;
308 s->end_freq[s->lfe_ch] = 7;
309 s->num_exp_groups[s->lfe_ch] = 2;
310 s->channel_in_cpl[s->lfe_ch] = 0;
313 return ac3_parse_header(s);
317 * Set stereo downmixing coefficients based on frame header info.
318 * reference: Section 7.8.2 Downmixing Into Two Channels
320 static void set_downmix_coeffs(AC3DecodeContext *s)
323 float cmix = gain_levels[center_levels[s->center_mix_level]];
324 float smix = gain_levels[surround_levels[s->surround_mix_level]];
326 for(i=0; i<s->fbw_channels; i++) {
327 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
328 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
330 if(s->channel_mode > 1 && s->channel_mode & 1) {
331 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
333 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
334 int nf = s->channel_mode - 2;
335 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
337 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
338 int nf = s->channel_mode - 4;
339 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
342 /* calculate adjustment needed for each channel to avoid clipping */
343 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
344 for(i=0; i<s->fbw_channels; i++) {
345 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
346 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
348 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
349 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
353 * Decode the grouped exponents according to exponent strategy.
354 * reference: Section 7.1.3 Exponent Decoding
356 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
357 uint8_t absexp, int8_t *dexps)
359 int i, j, grp, group_size;
364 group_size = exp_strategy + (exp_strategy == EXP_D45);
365 for(grp=0,i=0; grp<ngrps; grp++) {
366 expacc = get_bits(gbc, 7);
367 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
368 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
369 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
372 /* convert to absolute exps and expand groups */
374 for(i=0; i<ngrps*3; i++) {
375 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
376 for(j=0; j<group_size; j++) {
377 dexps[(i*group_size)+j] = prevexp;
383 * Generate transform coefficients for each coupled channel in the coupling
384 * range using the coupling coefficients and coupling coordinates.
385 * reference: Section 7.4.3 Coupling Coordinate Format
387 static void uncouple_channels(AC3DecodeContext *s)
389 int i, j, ch, bnd, subbnd;
392 i = s->start_freq[CPL_CH];
393 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
396 for(j=0; j<12; j++) {
397 for(ch=1; ch<=s->fbw_channels; ch++) {
398 if(s->channel_in_cpl[ch]) {
399 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
400 if (ch == 2 && s->phase_flags[bnd])
401 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
406 } while(s->cpl_band_struct[subbnd]);
411 * Grouped mantissas for 3-level 5-level and 11-level quantization
423 * Get the transform coefficients for a particular channel
424 * reference: Section 7.3 Quantization and Decoding of Mantissas
426 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
428 GetBitContext *gbc = &s->gbc;
429 int i, gcode, tbap, start, end;
434 exps = s->dexps[ch_index];
435 bap = s->bap[ch_index];
436 coeffs = s->fixed_coeffs[ch_index];
437 start = s->start_freq[ch_index];
438 end = s->end_freq[ch_index];
440 for (i = start; i < end; i++) {
444 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 0x400000;
449 gcode = get_bits(gbc, 5);
450 m->b1_mant[0] = b1_mantissas[gcode][0];
451 m->b1_mant[1] = b1_mantissas[gcode][1];
452 m->b1_mant[2] = b1_mantissas[gcode][2];
455 coeffs[i] = m->b1_mant[m->b1ptr++];
460 gcode = get_bits(gbc, 7);
461 m->b2_mant[0] = b2_mantissas[gcode][0];
462 m->b2_mant[1] = b2_mantissas[gcode][1];
463 m->b2_mant[2] = b2_mantissas[gcode][2];
466 coeffs[i] = m->b2_mant[m->b2ptr++];
470 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
475 gcode = get_bits(gbc, 7);
476 m->b4_mant[0] = b4_mantissas[gcode][0];
477 m->b4_mant[1] = b4_mantissas[gcode][1];
480 coeffs[i] = m->b4_mant[m->b4ptr++];
484 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
488 /* asymmetric dequantization */
489 int qlevel = quantization_tab[tbap];
490 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
494 coeffs[i] >>= exps[i];
499 * Remove random dithering from coefficients with zero-bit mantissas
500 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
502 static void remove_dithering(AC3DecodeContext *s) {
508 for(ch=1; ch<=s->fbw_channels; ch++) {
509 if(!s->dither_flag[ch]) {
510 coeffs = s->fixed_coeffs[ch];
512 if(s->channel_in_cpl[ch])
513 end = s->start_freq[CPL_CH];
515 end = s->end_freq[ch];
516 for(i=0; i<end; i++) {
520 if(s->channel_in_cpl[ch]) {
521 bap = s->bap[CPL_CH];
522 for(; i<s->end_freq[CPL_CH]; i++) {
532 * Get the transform coefficients.
534 static void get_transform_coeffs(AC3DecodeContext *s)
540 m.b1ptr = m.b2ptr = m.b4ptr = 3;
542 for (ch = 1; ch <= s->channels; ch++) {
543 /* transform coefficients for full-bandwidth channel */
544 get_transform_coeffs_ch(s, ch, &m);
545 /* tranform coefficients for coupling channel come right after the
546 coefficients for the first coupled channel*/
547 if (s->channel_in_cpl[ch]) {
549 get_transform_coeffs_ch(s, CPL_CH, &m);
550 uncouple_channels(s);
553 end = s->end_freq[CPL_CH];
555 end = s->end_freq[ch];
558 s->fixed_coeffs[ch][end] = 0;
562 /* if any channel doesn't use dithering, zero appropriate coefficients */
568 * Stereo rematrixing.
569 * reference: Section 7.5.4 Rematrixing : Decoding Technique
571 static void do_rematrixing(AC3DecodeContext *s)
577 end = FFMIN(s->end_freq[1], s->end_freq[2]);
579 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
580 if(s->rematrixing_flags[bnd]) {
581 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
582 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
583 tmp0 = s->fixed_coeffs[1][i];
584 tmp1 = s->fixed_coeffs[2][i];
585 s->fixed_coeffs[1][i] = tmp0 + tmp1;
586 s->fixed_coeffs[2][i] = tmp0 - tmp1;
593 * Perform the 256-point IMDCT
595 static void do_imdct_256(AC3DecodeContext *s, int chindex)
598 DECLARE_ALIGNED_16(float, x[128]);
600 float *o_ptr = s->tmp_output;
603 /* de-interleave coefficients */
604 for(k=0; k<128; k++) {
605 x[k] = s->transform_coeffs[chindex][2*k+i];
608 /* run standard IMDCT */
609 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
611 /* reverse the post-rotation & reordering from standard IMDCT */
612 for(k=0; k<32; k++) {
613 z[i][32+k].re = -o_ptr[128+2*k];
614 z[i][32+k].im = -o_ptr[2*k];
615 z[i][31-k].re = o_ptr[2*k+1];
616 z[i][31-k].im = o_ptr[128+2*k+1];
620 /* apply AC-3 post-rotation & reordering */
621 for(k=0; k<64; k++) {
622 o_ptr[ 2*k ] = -z[0][ k].im;
623 o_ptr[ 2*k+1] = z[0][63-k].re;
624 o_ptr[128+2*k ] = -z[0][ k].re;
625 o_ptr[128+2*k+1] = z[0][63-k].im;
626 o_ptr[256+2*k ] = -z[1][ k].re;
627 o_ptr[256+2*k+1] = z[1][63-k].im;
628 o_ptr[384+2*k ] = z[1][ k].im;
629 o_ptr[384+2*k+1] = -z[1][63-k].re;
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)
642 for (ch=1; ch<=channels; ch++) {
643 if (s->block_switch[ch]) {
646 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
647 s->transform_coeffs[ch], s->tmp_imdct);
649 /* For the first half of the block, apply the window, add the delay
650 from the previous block, and send to output */
651 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
652 s->window, s->delay[ch-1], 0, 256, 1);
653 /* For the second half of the block, apply the window and store the
654 samples to delay, to be combined with the next block */
655 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
661 * Downmix the output to mono or stereo.
663 static void ac3_downmix(AC3DecodeContext *s,
664 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
669 for(i=0; i<256; i++) {
671 for(j=0; j<s->fbw_channels; j++) {
672 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
673 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
675 v0 *= s->downmix_coeff_adjust[0];
676 v1 *= s->downmix_coeff_adjust[1];
677 if(s->output_mode == AC3_CHMODE_MONO) {
678 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
679 } else if(s->output_mode == AC3_CHMODE_STEREO) {
680 samples[ ch_offset][i] = v0;
681 samples[1+ch_offset][i] = v1;
687 * Upmix delay samples from stereo to original channel layout.
689 static void ac3_upmix_delay(AC3DecodeContext *s)
691 int channel_data_size = sizeof(s->delay[0]);
692 switch(s->channel_mode) {
693 case AC3_CHMODE_DUALMONO:
694 case AC3_CHMODE_STEREO:
695 /* upmix mono to stereo */
696 memcpy(s->delay[1], s->delay[0], channel_data_size);
698 case AC3_CHMODE_2F2R:
699 memset(s->delay[3], 0, channel_data_size);
700 case AC3_CHMODE_2F1R:
701 memset(s->delay[2], 0, channel_data_size);
703 case AC3_CHMODE_3F2R:
704 memset(s->delay[4], 0, channel_data_size);
705 case AC3_CHMODE_3F1R:
706 memset(s->delay[3], 0, channel_data_size);
708 memcpy(s->delay[2], s->delay[1], channel_data_size);
709 memset(s->delay[1], 0, channel_data_size);
715 * Parse an audio block from AC-3 bitstream.
717 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
719 int fbw_channels = s->fbw_channels;
720 int channel_mode = s->channel_mode;
722 int different_transforms;
725 GetBitContext *gbc = &s->gbc;
726 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
728 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
730 /* block switch flags */
731 different_transforms = 0;
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;
738 /* dithering flags */
740 for (ch = 1; ch <= fbw_channels; ch++) {
741 s->dither_flag[ch] = get_bits1(gbc);
742 if(!s->dither_flag[ch])
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 /* coupling strategy */
758 if (get_bits1(gbc)) {
759 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
760 s->cpl_in_use[blk] = get_bits1(gbc);
761 if (s->cpl_in_use[blk]) {
762 /* coupling in use */
763 int cpl_begin_freq, cpl_end_freq;
765 if (channel_mode < AC3_CHMODE_STEREO) {
766 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
770 /* determine which channels are coupled */
771 for (ch = 1; ch <= fbw_channels; ch++)
772 s->channel_in_cpl[ch] = get_bits1(gbc);
774 /* phase flags in use */
775 if (channel_mode == AC3_CHMODE_STEREO)
776 s->phase_flags_in_use = get_bits1(gbc);
778 /* coupling frequency range and band structure */
779 cpl_begin_freq = get_bits(gbc, 4);
780 cpl_end_freq = get_bits(gbc, 4);
781 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
782 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
785 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
786 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
787 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
788 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
789 if (get_bits1(gbc)) {
790 s->cpl_band_struct[bnd] = 1;
794 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
796 /* coupling not in use */
797 for (ch = 1; ch <= fbw_channels; ch++)
798 s->channel_in_cpl[ch] = 0;
801 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
804 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
806 cpl_in_use = s->cpl_in_use[blk];
808 /* coupling coordinates */
810 int cpl_coords_exist = 0;
812 for (ch = 1; ch <= fbw_channels; ch++) {
813 if (s->channel_in_cpl[ch]) {
814 if (get_bits1(gbc)) {
815 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
816 cpl_coords_exist = 1;
817 master_cpl_coord = 3 * get_bits(gbc, 2);
818 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
819 cpl_coord_exp = get_bits(gbc, 4);
820 cpl_coord_mant = get_bits(gbc, 4);
821 if (cpl_coord_exp == 15)
822 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
824 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
825 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
828 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
834 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
835 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
836 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
841 /* stereo rematrixing strategy and band structure */
842 if (channel_mode == AC3_CHMODE_STEREO) {
843 if (get_bits1(gbc)) {
844 s->num_rematrixing_bands = 4;
845 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
846 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
847 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
848 s->rematrixing_flags[bnd] = get_bits1(gbc);
850 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
855 /* exponent strategies for each channel */
856 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
857 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
858 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
859 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
860 if(s->exp_strategy[blk][ch] != EXP_REUSE)
861 bit_alloc_stages[ch] = 3;
864 /* channel bandwidth */
865 for (ch = 1; ch <= fbw_channels; ch++) {
866 s->start_freq[ch] = 0;
867 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
869 int prev = s->end_freq[ch];
870 if (s->channel_in_cpl[ch])
871 s->end_freq[ch] = s->start_freq[CPL_CH];
873 int bandwidth_code = get_bits(gbc, 6);
874 if (bandwidth_code > 60) {
875 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
878 s->end_freq[ch] = bandwidth_code * 3 + 73;
880 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
881 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
882 if(blk > 0 && s->end_freq[ch] != prev)
883 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
886 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
887 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
888 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
891 /* decode exponents for each channel */
892 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
893 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
894 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
895 decode_exponents(gbc, s->exp_strategy[blk][ch],
896 s->num_exp_groups[ch], s->dexps[ch][0],
897 &s->dexps[ch][s->start_freq[ch]+!!ch]);
898 if(ch != CPL_CH && ch != s->lfe_ch)
899 skip_bits(gbc, 2); /* skip gainrng */
903 /* bit allocation information */
904 if (get_bits1(gbc)) {
905 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
906 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
907 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
908 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
909 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
910 for(ch=!cpl_in_use; ch<=s->channels; ch++)
911 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
913 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
917 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
918 if (get_bits1(gbc)) {
920 csnr = (get_bits(gbc, 6) - 15) << 4;
921 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
922 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
923 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
925 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
927 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
931 /* coupling leak information */
933 if (get_bits1(gbc)) {
934 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
935 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
936 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
938 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
943 /* delta bit allocation information */
944 if (get_bits1(gbc)) {
945 /* delta bit allocation exists (strategy) */
946 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
947 s->dba_mode[ch] = get_bits(gbc, 2);
948 if (s->dba_mode[ch] == DBA_RESERVED) {
949 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
952 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
954 /* channel delta offset, len and bit allocation */
955 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
956 if (s->dba_mode[ch] == DBA_NEW) {
957 s->dba_nsegs[ch] = get_bits(gbc, 3);
958 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
959 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
960 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
961 s->dba_values[ch][seg] = get_bits(gbc, 3);
963 /* run last 2 bit allocation stages if new dba values */
964 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
967 } else if(blk == 0) {
968 for(ch=0; ch<=s->channels; ch++) {
969 s->dba_mode[ch] = DBA_NONE;
974 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
975 if(bit_alloc_stages[ch] > 2) {
976 /* Exponent mapping into PSD and PSD integration */
977 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
978 s->start_freq[ch], s->end_freq[ch],
979 s->psd[ch], s->band_psd[ch]);
981 if(bit_alloc_stages[ch] > 1) {
982 /* Compute excitation function, Compute masking curve, and
983 Apply delta bit allocation */
984 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
985 s->start_freq[ch], s->end_freq[ch],
986 s->fast_gain[ch], (ch == s->lfe_ch),
987 s->dba_mode[ch], s->dba_nsegs[ch],
988 s->dba_offsets[ch], s->dba_lengths[ch],
989 s->dba_values[ch], s->mask[ch]);
991 if(bit_alloc_stages[ch] > 0) {
992 /* Compute bit allocation */
993 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
994 s->start_freq[ch], s->end_freq[ch],
996 s->bit_alloc_params.floor,
997 ff_ac3_bap_tab, s->bap[ch]);
1001 /* unused dummy data */
1002 if (get_bits1(gbc)) {
1003 int skipl = get_bits(gbc, 9);
1008 /* unpack the transform coefficients
1009 this also uncouples channels if coupling is in use. */
1010 get_transform_coeffs(s);
1012 /* recover coefficients if rematrixing is in use */
1013 if(s->channel_mode == AC3_CHMODE_STEREO)
1016 /* apply scaling to coefficients (headroom, dynrng) */
1017 for(ch=1; ch<=s->channels; ch++) {
1018 float gain = s->mul_bias / 4194304.0f;
1019 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1020 gain *= s->dynamic_range[ch-1];
1022 gain *= s->dynamic_range[0];
1024 for(i=0; i<256; i++) {
1025 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1029 /* downmix and MDCT. order depends on whether block switching is used for
1030 any channel in this block. this is because coefficients for the long
1031 and short transforms cannot be mixed. */
1032 downmix_output = s->channels != s->out_channels &&
1033 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1034 s->fbw_channels == s->out_channels);
1035 if(different_transforms) {
1036 /* the delay samples have already been downmixed, so we upmix the delay
1037 samples in order to reconstruct all channels before downmixing. */
1043 do_imdct(s, s->channels);
1045 if(downmix_output) {
1046 ac3_downmix(s, s->output, 0);
1049 if(downmix_output) {
1050 ac3_downmix(s, s->transform_coeffs, 1);
1055 ac3_downmix(s, s->delay, 0);
1058 do_imdct(s, s->out_channels);
1061 /* convert float to 16-bit integer */
1062 for(ch=0; ch<s->out_channels; ch++) {
1063 for(i=0; i<256; i++) {
1064 s->output[ch][i] += s->add_bias;
1066 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1073 * Decode a single AC-3 frame.
1075 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1076 const uint8_t *buf, int buf_size)
1078 AC3DecodeContext *s = avctx->priv_data;
1079 int16_t *out_samples = (int16_t *)data;
1080 int i, blk, ch, err;
1082 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1083 if (s->input_buffer) {
1084 /* copy input buffer to decoder context to avoid reading past the end
1085 of the buffer, which can be caused by a damaged input stream. */
1086 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1087 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1089 init_get_bits(&s->gbc, buf, buf_size * 8);
1092 /* parse the syncinfo */
1094 err = parse_frame_header(s);
1096 /* check that reported frame size fits in input buffer */
1097 if(s->frame_size > buf_size) {
1098 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1099 err = AC3_PARSE_ERROR_FRAME_SIZE;
1102 /* check for crc mismatch */
1103 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1104 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1105 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1106 err = AC3_PARSE_ERROR_CRC;
1110 if(err && err != AC3_PARSE_ERROR_CRC) {
1112 case AC3_PARSE_ERROR_SYNC:
1113 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1115 case AC3_PARSE_ERROR_BSID:
1116 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1118 case AC3_PARSE_ERROR_SAMPLE_RATE:
1119 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1121 case AC3_PARSE_ERROR_FRAME_SIZE:
1122 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1124 case AC3_PARSE_ERROR_FRAME_TYPE:
1125 /* skip frame if CRC is ok. otherwise use error concealment. */
1126 /* TODO: add support for substreams and dependent frames */
1127 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1128 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1129 return s->frame_size;
1131 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1135 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1140 /* if frame is ok, set audio parameters */
1142 avctx->sample_rate = s->sample_rate;
1143 avctx->bit_rate = s->bit_rate;
1145 /* channel config */
1146 s->out_channels = s->channels;
1147 s->output_mode = s->channel_mode;
1149 s->output_mode |= AC3_OUTPUT_LFEON;
1150 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1151 avctx->request_channels < s->channels) {
1152 s->out_channels = avctx->request_channels;
1153 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1155 avctx->channels = s->out_channels;
1157 /* set downmixing coefficients if needed */
1158 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1159 s->fbw_channels == s->out_channels)) {
1160 set_downmix_coeffs(s);
1162 } else if (!s->out_channels) {
1163 s->out_channels = avctx->channels;
1164 if(s->out_channels < s->channels)
1165 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1168 /* parse the audio blocks */
1169 for (blk = 0; blk < s->num_blocks; blk++) {
1170 if (!err && ac3_parse_audio_block(s, blk)) {
1171 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1174 /* interleave output samples */
1175 for (i = 0; i < 256; i++)
1176 for (ch = 0; ch < s->out_channels; ch++)
1177 *(out_samples++) = s->int_output[ch][i];
1179 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1180 return s->frame_size;
1184 * Uninitialize the AC-3 decoder.
1186 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1188 AC3DecodeContext *s = avctx->priv_data;
1189 ff_mdct_end(&s->imdct_512);
1190 ff_mdct_end(&s->imdct_256);
1192 av_freep(&s->input_buffer);
1197 AVCodec ac3_decoder = {
1199 .type = CODEC_TYPE_AUDIO,
1201 .priv_data_size = sizeof (AC3DecodeContext),
1202 .init = ac3_decode_init,
1203 .close = ac3_decode_end,
1204 .decode = ac3_decode_frame,
1205 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),