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)
281 err = ff_ac3_parse_header(&s->gbc, &hdr);
285 /* get decoding parameters from header info */
286 s->bit_alloc_params.sr_code = hdr.sr_code;
287 s->channel_mode = hdr.channel_mode;
288 s->lfe_on = hdr.lfe_on;
289 s->bit_alloc_params.sr_shift = hdr.sr_shift;
290 s->sample_rate = hdr.sample_rate;
291 s->bit_rate = hdr.bit_rate;
292 s->channels = hdr.channels;
293 s->fbw_channels = s->channels - s->lfe_on;
294 s->lfe_ch = s->fbw_channels + 1;
295 s->frame_size = hdr.frame_size;
296 s->center_mix_level = hdr.center_mix_level;
297 s->surround_mix_level = hdr.surround_mix_level;
298 s->num_blocks = hdr.num_blocks;
299 s->frame_type = hdr.frame_type;
300 s->substreamid = hdr.substreamid;
303 s->start_freq[s->lfe_ch] = 0;
304 s->end_freq[s->lfe_ch] = 7;
305 s->num_exp_groups[s->lfe_ch] = 2;
306 s->channel_in_cpl[s->lfe_ch] = 0;
309 if(hdr.bitstream_id > 10)
310 return AC3_PARSE_ERROR_BSID;
312 return ac3_parse_header(s);
316 * Set stereo downmixing coefficients based on frame header info.
317 * reference: Section 7.8.2 Downmixing Into Two Channels
319 static void set_downmix_coeffs(AC3DecodeContext *s)
322 float cmix = gain_levels[center_levels[s->center_mix_level]];
323 float smix = gain_levels[surround_levels[s->surround_mix_level]];
325 for(i=0; i<s->fbw_channels; i++) {
326 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
327 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
329 if(s->channel_mode > 1 && s->channel_mode & 1) {
330 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
332 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
333 int nf = s->channel_mode - 2;
334 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
336 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
337 int nf = s->channel_mode - 4;
338 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
341 /* calculate adjustment needed for each channel to avoid clipping */
342 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
343 for(i=0; i<s->fbw_channels; i++) {
344 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
345 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
347 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
348 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
352 * Decode the grouped exponents according to exponent strategy.
353 * reference: Section 7.1.3 Exponent Decoding
355 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
356 uint8_t absexp, int8_t *dexps)
358 int i, j, grp, group_size;
363 group_size = exp_strategy + (exp_strategy == EXP_D45);
364 for(grp=0,i=0; grp<ngrps; grp++) {
365 expacc = get_bits(gbc, 7);
366 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][0];
367 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][1];
368 dexp[i++] = ungroup_3_in_7_bits_tab[expacc][2];
371 /* convert to absolute exps and expand groups */
373 for(i=0; i<ngrps*3; i++) {
374 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
375 for(j=0; j<group_size; j++) {
376 dexps[(i*group_size)+j] = prevexp;
382 * Generate transform coefficients for each coupled channel in the coupling
383 * range using the coupling coefficients and coupling coordinates.
384 * reference: Section 7.4.3 Coupling Coordinate Format
386 static void uncouple_channels(AC3DecodeContext *s)
388 int i, j, ch, bnd, subbnd;
391 i = s->start_freq[CPL_CH];
392 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
395 for(j=0; j<12; j++) {
396 for(ch=1; ch<=s->fbw_channels; ch++) {
397 if(s->channel_in_cpl[ch]) {
398 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
399 if (ch == 2 && s->phase_flags[bnd])
400 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
405 } while(s->cpl_band_struct[subbnd]);
410 * Grouped mantissas for 3-level 5-level and 11-level quantization
422 * Get the transform coefficients for a particular channel
423 * reference: Section 7.3 Quantization and Decoding of Mantissas
425 static void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
427 GetBitContext *gbc = &s->gbc;
428 int i, gcode, tbap, start, end;
433 exps = s->dexps[ch_index];
434 bap = s->bap[ch_index];
435 coeffs = s->fixed_coeffs[ch_index];
436 start = s->start_freq[ch_index];
437 end = s->end_freq[ch_index];
439 for (i = start; i < end; i++) {
443 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 0x400000;
448 gcode = get_bits(gbc, 5);
449 m->b1_mant[0] = b1_mantissas[gcode][0];
450 m->b1_mant[1] = b1_mantissas[gcode][1];
451 m->b1_mant[2] = b1_mantissas[gcode][2];
454 coeffs[i] = m->b1_mant[m->b1ptr++];
459 gcode = get_bits(gbc, 7);
460 m->b2_mant[0] = b2_mantissas[gcode][0];
461 m->b2_mant[1] = b2_mantissas[gcode][1];
462 m->b2_mant[2] = b2_mantissas[gcode][2];
465 coeffs[i] = m->b2_mant[m->b2ptr++];
469 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
474 gcode = get_bits(gbc, 7);
475 m->b4_mant[0] = b4_mantissas[gcode][0];
476 m->b4_mant[1] = b4_mantissas[gcode][1];
479 coeffs[i] = m->b4_mant[m->b4ptr++];
483 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
487 /* asymmetric dequantization */
488 int qlevel = quantization_tab[tbap];
489 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
493 coeffs[i] >>= exps[i];
498 * Remove random dithering from coefficients with zero-bit mantissas
499 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
501 static void remove_dithering(AC3DecodeContext *s) {
507 for(ch=1; ch<=s->fbw_channels; ch++) {
508 if(!s->dither_flag[ch]) {
509 coeffs = s->fixed_coeffs[ch];
511 if(s->channel_in_cpl[ch])
512 end = s->start_freq[CPL_CH];
514 end = s->end_freq[ch];
515 for(i=0; i<end; i++) {
519 if(s->channel_in_cpl[ch]) {
520 bap = s->bap[CPL_CH];
521 for(; i<s->end_freq[CPL_CH]; i++) {
531 * Get the transform coefficients.
533 static void get_transform_coeffs(AC3DecodeContext *s)
539 m.b1ptr = m.b2ptr = m.b4ptr = 3;
541 for (ch = 1; ch <= s->channels; ch++) {
542 /* transform coefficients for full-bandwidth channel */
543 get_transform_coeffs_ch(s, ch, &m);
544 /* tranform coefficients for coupling channel come right after the
545 coefficients for the first coupled channel*/
546 if (s->channel_in_cpl[ch]) {
548 get_transform_coeffs_ch(s, CPL_CH, &m);
549 uncouple_channels(s);
552 end = s->end_freq[CPL_CH];
554 end = s->end_freq[ch];
557 s->fixed_coeffs[ch][end] = 0;
561 /* if any channel doesn't use dithering, zero appropriate coefficients */
567 * Stereo rematrixing.
568 * reference: Section 7.5.4 Rematrixing : Decoding Technique
570 static void do_rematrixing(AC3DecodeContext *s)
576 end = FFMIN(s->end_freq[1], s->end_freq[2]);
578 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
579 if(s->rematrixing_flags[bnd]) {
580 bndend = FFMIN(end, ff_ac3_rematrix_band_tab[bnd+1]);
581 for(i=ff_ac3_rematrix_band_tab[bnd]; i<bndend; i++) {
582 tmp0 = s->fixed_coeffs[1][i];
583 tmp1 = s->fixed_coeffs[2][i];
584 s->fixed_coeffs[1][i] = tmp0 + tmp1;
585 s->fixed_coeffs[2][i] = tmp0 - tmp1;
592 * Perform the 256-point IMDCT
594 static void do_imdct_256(AC3DecodeContext *s, int chindex)
597 DECLARE_ALIGNED_16(float, x[128]);
599 float *o_ptr = s->tmp_output;
602 /* de-interleave coefficients */
603 for(k=0; k<128; k++) {
604 x[k] = s->transform_coeffs[chindex][2*k+i];
607 /* run standard IMDCT */
608 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
610 /* reverse the post-rotation & reordering from standard IMDCT */
611 for(k=0; k<32; k++) {
612 z[i][32+k].re = -o_ptr[128+2*k];
613 z[i][32+k].im = -o_ptr[2*k];
614 z[i][31-k].re = o_ptr[2*k+1];
615 z[i][31-k].im = o_ptr[128+2*k+1];
619 /* apply AC-3 post-rotation & reordering */
620 for(k=0; k<64; k++) {
621 o_ptr[ 2*k ] = -z[0][ k].im;
622 o_ptr[ 2*k+1] = z[0][63-k].re;
623 o_ptr[128+2*k ] = -z[0][ k].re;
624 o_ptr[128+2*k+1] = z[0][63-k].im;
625 o_ptr[256+2*k ] = -z[1][ k].re;
626 o_ptr[256+2*k+1] = z[1][63-k].im;
627 o_ptr[384+2*k ] = z[1][ k].im;
628 o_ptr[384+2*k+1] = -z[1][63-k].re;
633 * Inverse MDCT Transform.
634 * Convert frequency domain coefficients to time-domain audio samples.
635 * reference: Section 7.9.4 Transformation Equations
637 static inline void do_imdct(AC3DecodeContext *s, int channels)
641 for (ch=1; ch<=channels; ch++) {
642 if (s->block_switch[ch]) {
645 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
646 s->transform_coeffs[ch], s->tmp_imdct);
648 /* For the first half of the block, apply the window, add the delay
649 from the previous block, and send to output */
650 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
651 s->window, s->delay[ch-1], 0, 256, 1);
652 /* For the second half of the block, apply the window and store the
653 samples to delay, to be combined with the next block */
654 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
660 * Downmix the output to mono or stereo.
662 static void ac3_downmix(AC3DecodeContext *s,
663 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
668 for(i=0; i<256; i++) {
670 for(j=0; j<s->fbw_channels; j++) {
671 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
672 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
674 v0 *= s->downmix_coeff_adjust[0];
675 v1 *= s->downmix_coeff_adjust[1];
676 if(s->output_mode == AC3_CHMODE_MONO) {
677 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
678 } else if(s->output_mode == AC3_CHMODE_STEREO) {
679 samples[ ch_offset][i] = v0;
680 samples[1+ch_offset][i] = v1;
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 for (ch = 1; ch <= fbw_channels; ch++) {
732 s->block_switch[ch] = get_bits1(gbc);
733 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
734 different_transforms = 1;
737 /* dithering flags */
739 for (ch = 1; ch <= fbw_channels; ch++) {
740 s->dither_flag[ch] = get_bits1(gbc);
741 if(!s->dither_flag[ch])
746 i = !(s->channel_mode);
749 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
750 s->avctx->drc_scale)+1.0;
751 } else if(blk == 0) {
752 s->dynamic_range[i] = 1.0f;
756 /* coupling strategy */
757 if (get_bits1(gbc)) {
758 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
759 s->cpl_in_use[blk] = get_bits1(gbc);
760 if (s->cpl_in_use[blk]) {
761 /* coupling in use */
762 int cpl_begin_freq, cpl_end_freq;
764 if (channel_mode < AC3_CHMODE_STEREO) {
765 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
769 /* determine which channels are coupled */
770 for (ch = 1; ch <= fbw_channels; ch++)
771 s->channel_in_cpl[ch] = get_bits1(gbc);
773 /* phase flags in use */
774 if (channel_mode == AC3_CHMODE_STEREO)
775 s->phase_flags_in_use = get_bits1(gbc);
777 /* coupling frequency range and band structure */
778 cpl_begin_freq = get_bits(gbc, 4);
779 cpl_end_freq = get_bits(gbc, 4);
780 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
781 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
784 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
785 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
786 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
787 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
788 if (get_bits1(gbc)) {
789 s->cpl_band_struct[bnd] = 1;
793 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
795 /* coupling not in use */
796 for (ch = 1; ch <= fbw_channels; ch++)
797 s->channel_in_cpl[ch] = 0;
800 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
803 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
805 cpl_in_use = s->cpl_in_use[blk];
807 /* coupling coordinates */
809 int cpl_coords_exist = 0;
811 for (ch = 1; ch <= fbw_channels; ch++) {
812 if (s->channel_in_cpl[ch]) {
813 if (get_bits1(gbc)) {
814 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
815 cpl_coords_exist = 1;
816 master_cpl_coord = 3 * get_bits(gbc, 2);
817 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
818 cpl_coord_exp = get_bits(gbc, 4);
819 cpl_coord_mant = get_bits(gbc, 4);
820 if (cpl_coord_exp == 15)
821 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
823 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
824 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
827 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
833 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
834 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
835 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
840 /* stereo rematrixing strategy and band structure */
841 if (channel_mode == AC3_CHMODE_STEREO) {
842 if (get_bits1(gbc)) {
843 s->num_rematrixing_bands = 4;
844 if(cpl_in_use && s->start_freq[CPL_CH] <= 61)
845 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
846 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
847 s->rematrixing_flags[bnd] = get_bits1(gbc);
849 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
854 /* exponent strategies for each channel */
855 s->exp_strategy[blk][CPL_CH] = EXP_REUSE;
856 s->exp_strategy[blk][s->lfe_ch] = EXP_REUSE;
857 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
858 s->exp_strategy[blk][ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
859 if(s->exp_strategy[blk][ch] != EXP_REUSE)
860 bit_alloc_stages[ch] = 3;
863 /* channel bandwidth */
864 for (ch = 1; ch <= fbw_channels; ch++) {
865 s->start_freq[ch] = 0;
866 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
868 int prev = s->end_freq[ch];
869 if (s->channel_in_cpl[ch])
870 s->end_freq[ch] = s->start_freq[CPL_CH];
872 int bandwidth_code = get_bits(gbc, 6);
873 if (bandwidth_code > 60) {
874 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
877 s->end_freq[ch] = bandwidth_code * 3 + 73;
879 group_size = 3 << (s->exp_strategy[blk][ch] - 1);
880 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
881 if(blk > 0 && s->end_freq[ch] != prev)
882 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
885 if (cpl_in_use && s->exp_strategy[blk][CPL_CH] != EXP_REUSE) {
886 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
887 (3 << (s->exp_strategy[blk][CPL_CH] - 1));
890 /* decode exponents for each channel */
891 for (ch = !cpl_in_use; ch <= s->channels; ch++) {
892 if (s->exp_strategy[blk][ch] != EXP_REUSE) {
893 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
894 decode_exponents(gbc, s->exp_strategy[blk][ch],
895 s->num_exp_groups[ch], s->dexps[ch][0],
896 &s->dexps[ch][s->start_freq[ch]+!!ch]);
897 if(ch != CPL_CH && ch != s->lfe_ch)
898 skip_bits(gbc, 2); /* skip gainrng */
902 /* bit allocation information */
903 if (get_bits1(gbc)) {
904 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
905 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
906 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
907 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
908 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
909 for(ch=!cpl_in_use; ch<=s->channels; ch++)
910 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
912 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
916 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
917 if (get_bits1(gbc)) {
919 csnr = (get_bits(gbc, 6) - 15) << 4;
920 for (ch = !cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
921 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
922 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
924 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
926 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
930 /* coupling leak information */
932 if (get_bits1(gbc)) {
933 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
934 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
935 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
937 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
942 /* delta bit allocation information */
943 if (get_bits1(gbc)) {
944 /* delta bit allocation exists (strategy) */
945 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
946 s->dba_mode[ch] = get_bits(gbc, 2);
947 if (s->dba_mode[ch] == DBA_RESERVED) {
948 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
951 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
953 /* channel delta offset, len and bit allocation */
954 for (ch = !cpl_in_use; ch <= fbw_channels; ch++) {
955 if (s->dba_mode[ch] == DBA_NEW) {
956 s->dba_nsegs[ch] = get_bits(gbc, 3);
957 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
958 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
959 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
960 s->dba_values[ch][seg] = get_bits(gbc, 3);
962 /* run last 2 bit allocation stages if new dba values */
963 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
966 } else if(blk == 0) {
967 for(ch=0; ch<=s->channels; ch++) {
968 s->dba_mode[ch] = DBA_NONE;
973 for(ch=!cpl_in_use; ch<=s->channels; ch++) {
974 if(bit_alloc_stages[ch] > 2) {
975 /* Exponent mapping into PSD and PSD integration */
976 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
977 s->start_freq[ch], s->end_freq[ch],
978 s->psd[ch], s->band_psd[ch]);
980 if(bit_alloc_stages[ch] > 1) {
981 /* Compute excitation function, Compute masking curve, and
982 Apply delta bit allocation */
983 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
984 s->start_freq[ch], s->end_freq[ch],
985 s->fast_gain[ch], (ch == s->lfe_ch),
986 s->dba_mode[ch], s->dba_nsegs[ch],
987 s->dba_offsets[ch], s->dba_lengths[ch],
988 s->dba_values[ch], s->mask[ch]);
990 if(bit_alloc_stages[ch] > 0) {
991 /* Compute bit allocation */
992 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
993 s->start_freq[ch], s->end_freq[ch],
995 s->bit_alloc_params.floor,
996 ff_ac3_bap_tab, s->bap[ch]);
1000 /* unused dummy data */
1001 if (get_bits1(gbc)) {
1002 int skipl = get_bits(gbc, 9);
1007 /* unpack the transform coefficients
1008 this also uncouples channels if coupling is in use. */
1009 get_transform_coeffs(s);
1011 /* recover coefficients if rematrixing is in use */
1012 if(s->channel_mode == AC3_CHMODE_STEREO)
1015 /* apply scaling to coefficients (headroom, dynrng) */
1016 for(ch=1; ch<=s->channels; ch++) {
1017 float gain = s->mul_bias / 4194304.0f;
1018 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1019 gain *= s->dynamic_range[ch-1];
1021 gain *= s->dynamic_range[0];
1023 for(i=0; i<256; i++) {
1024 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1028 /* downmix and MDCT. order depends on whether block switching is used for
1029 any channel in this block. this is because coefficients for the long
1030 and short transforms cannot be mixed. */
1031 downmix_output = s->channels != s->out_channels &&
1032 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1033 s->fbw_channels == s->out_channels);
1034 if(different_transforms) {
1035 /* the delay samples have already been downmixed, so we upmix the delay
1036 samples in order to reconstruct all channels before downmixing. */
1042 do_imdct(s, s->channels);
1044 if(downmix_output) {
1045 ac3_downmix(s, s->output, 0);
1048 if(downmix_output) {
1049 ac3_downmix(s, s->transform_coeffs, 1);
1054 ac3_downmix(s, s->delay, 0);
1057 do_imdct(s, s->out_channels);
1060 /* convert float to 16-bit integer */
1061 for(ch=0; ch<s->out_channels; ch++) {
1062 for(i=0; i<256; i++) {
1063 s->output[ch][i] += s->add_bias;
1065 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1072 * Decode a single AC-3 frame.
1074 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1075 const uint8_t *buf, int buf_size)
1077 AC3DecodeContext *s = avctx->priv_data;
1078 int16_t *out_samples = (int16_t *)data;
1079 int i, blk, ch, err;
1081 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1082 if (s->input_buffer) {
1083 /* copy input buffer to decoder context to avoid reading past the end
1084 of the buffer, which can be caused by a damaged input stream. */
1085 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1086 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1088 init_get_bits(&s->gbc, buf, buf_size * 8);
1091 /* parse the syncinfo */
1093 err = parse_frame_header(s);
1095 /* check that reported frame size fits in input buffer */
1096 if(s->frame_size > buf_size) {
1097 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1098 err = AC3_PARSE_ERROR_FRAME_SIZE;
1101 /* check for crc mismatch */
1102 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1103 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1104 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1105 err = AC3_PARSE_ERROR_CRC;
1109 if(err && err != AC3_PARSE_ERROR_CRC) {
1111 case AC3_PARSE_ERROR_SYNC:
1112 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1114 case AC3_PARSE_ERROR_BSID:
1115 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1117 case AC3_PARSE_ERROR_SAMPLE_RATE:
1118 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1120 case AC3_PARSE_ERROR_FRAME_SIZE:
1121 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1123 case AC3_PARSE_ERROR_FRAME_TYPE:
1124 /* skip frame if CRC is ok. otherwise use error concealment. */
1125 /* TODO: add support for substreams and dependent frames */
1126 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1127 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1128 return s->frame_size;
1130 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1134 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1139 /* if frame is ok, set audio parameters */
1141 avctx->sample_rate = s->sample_rate;
1142 avctx->bit_rate = s->bit_rate;
1144 /* channel config */
1145 s->out_channels = s->channels;
1146 s->output_mode = s->channel_mode;
1148 s->output_mode |= AC3_OUTPUT_LFEON;
1149 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1150 avctx->request_channels < s->channels) {
1151 s->out_channels = avctx->request_channels;
1152 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1154 avctx->channels = s->out_channels;
1156 /* set downmixing coefficients if needed */
1157 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1158 s->fbw_channels == s->out_channels)) {
1159 set_downmix_coeffs(s);
1161 } else if (!s->out_channels) {
1162 s->out_channels = avctx->channels;
1163 if(s->out_channels < s->channels)
1164 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1167 /* decode the audio blocks */
1168 for (blk = 0; blk < s->num_blocks; blk++) {
1169 if (!err && decode_audio_block(s, blk)) {
1170 av_log(avctx, AV_LOG_ERROR, "error decoding the audio block\n");
1173 /* interleave output samples */
1174 for (i = 0; i < 256; i++)
1175 for (ch = 0; ch < s->out_channels; ch++)
1176 *(out_samples++) = s->int_output[ch][i];
1178 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1179 return s->frame_size;
1183 * Uninitialize the AC-3 decoder.
1185 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1187 AC3DecodeContext *s = avctx->priv_data;
1188 ff_mdct_end(&s->imdct_512);
1189 ff_mdct_end(&s->imdct_256);
1191 av_freep(&s->input_buffer);
1196 AVCodec ac3_decoder = {
1198 .type = CODEC_TYPE_AUDIO,
1200 .priv_data_size = sizeof (AC3DecodeContext),
1201 .init = ac3_decode_init,
1202 .close = ac3_decode_end,
1203 .decode = ac3_decode_frame,
1204 .long_name = NULL_IF_CONFIG_SMALL("ATSC A/52 / AC-3"),