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
36 #include "ac3_parser.h"
37 #include "bitstream.h"
43 * Table of bin locations for rematrixing bands
44 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
46 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
48 /** table for grouping exponents */
49 static uint8_t exp_ungroup_tab[128][3];
52 /** tables for ungrouping mantissas */
53 static int b1_mantissas[32][3];
54 static int b2_mantissas[128][3];
55 static int b3_mantissas[8];
56 static int b4_mantissas[128][2];
57 static int b5_mantissas[16];
60 * Quantization table: levels for symmetric. bits for asymmetric.
61 * reference: Table 7.18 Mapping of bap to Quantizer
63 static const uint8_t quantization_tab[16] = {
65 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
68 /** dynamic range table. converts codes to scale factors. */
69 static float dynamic_range_tab[256];
71 /** Adjustments in dB gain */
72 #define LEVEL_MINUS_3DB 0.7071067811865476
73 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
74 #define LEVEL_MINUS_6DB 0.5000000000000000
75 #define LEVEL_MINUS_9DB 0.3535533905932738
76 #define LEVEL_ZERO 0.0000000000000000
77 #define LEVEL_ONE 1.0000000000000000
79 static const float gain_levels[6] = {
83 LEVEL_MINUS_4POINT5DB,
89 * Table for center mix levels
90 * reference: Section 5.4.2.4 cmixlev
92 static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
95 * Table for surround mix levels
96 * reference: Section 5.4.2.5 surmixlev
98 static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
101 * Table for default stereo downmixing coefficients
102 * reference: Section 7.8.2 Downmixing Into Two Channels
104 static const uint8_t ac3_default_coeffs[8][5][2] = {
105 { { 1, 0 }, { 0, 1 }, },
107 { { 1, 0 }, { 0, 1 }, },
108 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
109 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
110 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
111 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
112 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
115 /* override ac3.h to include coupling channel */
116 #undef AC3_MAX_CHANNELS
117 #define AC3_MAX_CHANNELS 7
120 #define AC3_OUTPUT_LFEON 8
123 int channel_mode; ///< channel mode (acmod)
124 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
125 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
126 int dither_all; ///< true if all channels are dithered
127 int cpl_in_use; ///< coupling in use
128 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
129 int phase_flags_in_use; ///< phase flags in use
130 int phase_flags[18]; ///< phase flags
131 int cpl_band_struct[18]; ///< coupling band structure
132 int num_rematrixing_bands; ///< number of rematrixing bands
133 int rematrixing_flags[4]; ///< rematrixing flags
134 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
135 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
136 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
137 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
138 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
139 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
140 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
141 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
143 int sample_rate; ///< sample frequency, in Hz
144 int bit_rate; ///< stream bit rate, in bits-per-second
145 int frame_size; ///< current frame size, in bytes
147 int channels; ///< number of total channels
148 int fbw_channels; ///< number of full-bandwidth channels
149 int lfe_on; ///< lfe channel in use
150 int lfe_ch; ///< index of LFE channel
151 int output_mode; ///< output channel configuration
152 int out_channels; ///< number of output channels
154 int center_mix_level; ///< Center mix level index
155 int surround_mix_level; ///< Surround mix level index
156 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
157 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
158 float dynamic_range[2]; ///< dynamic range
159 int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
160 int num_cpl_bands; ///< number of coupling bands
161 int num_cpl_subbands; ///< number of coupling sub bands
162 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
163 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
164 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
166 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
167 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
168 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
169 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
170 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
172 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
173 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
174 int downmixed; ///< indicates if coeffs are currently downmixed
177 MDCTContext imdct_512; ///< for 512 sample IMDCT
178 MDCTContext imdct_256; ///< for 256 sample IMDCT
179 DSPContext dsp; ///< for optimization
180 float add_bias; ///< offset for float_to_int16 conversion
181 float mul_bias; ///< scaling for float_to_int16 conversion
183 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
184 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
185 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
186 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
187 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
188 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
191 GetBitContext gbc; ///< bitstream reader
192 AVRandomState dith_state; ///< for dither generation
193 AVCodecContext *avctx; ///< parent context
197 * Symmetrical Dequantization
198 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
199 * Tables 7.19 to 7.23
202 symmetric_dequant(int code, int levels)
204 return ((code - (levels >> 1)) << 24) / levels;
208 * Initialize tables at runtime.
210 static void ac3_tables_init(void)
214 /* generate grouped mantissa tables
215 reference: Section 7.3.5 Ungrouping of Mantissas */
216 for(i=0; i<32; i++) {
217 /* bap=1 mantissas */
218 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
219 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
220 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
222 for(i=0; i<128; i++) {
223 /* bap=2 mantissas */
224 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
225 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
226 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
228 /* bap=4 mantissas */
229 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
230 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
232 /* generate ungrouped mantissa tables
233 reference: Tables 7.21 and 7.23 */
235 /* bap=3 mantissas */
236 b3_mantissas[i] = symmetric_dequant(i, 7);
238 for(i=0; i<15; i++) {
239 /* bap=5 mantissas */
240 b5_mantissas[i] = symmetric_dequant(i, 15);
243 /* generate dynamic range table
244 reference: Section 7.7.1 Dynamic Range Control */
245 for(i=0; i<256; i++) {
246 int v = (i >> 5) - ((i >> 7) << 3) - 5;
247 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
250 /* generate exponent tables
251 reference: Section 7.1.3 Exponent Decoding */
252 for(i=0; i<128; i++) {
253 exp_ungroup_tab[i][0] = i / 25;
254 exp_ungroup_tab[i][1] = (i % 25) / 5;
255 exp_ungroup_tab[i][2] = (i % 25) % 5;
261 * AVCodec initialization
263 static int ac3_decode_init(AVCodecContext *avctx)
265 AC3DecodeContext *s = avctx->priv_data;
270 ff_mdct_init(&s->imdct_256, 8, 1);
271 ff_mdct_init(&s->imdct_512, 9, 1);
272 ff_kbd_window_init(s->window, 5.0, 256);
273 dsputil_init(&s->dsp, avctx);
274 av_init_random(0, &s->dith_state);
276 /* set bias values for float to int16 conversion */
277 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
278 s->add_bias = 385.0f;
282 s->mul_bias = 32767.0f;
285 /* allow downmixing to stereo or mono */
286 if (avctx->channels > 0 && avctx->request_channels > 0 &&
287 avctx->request_channels < avctx->channels &&
288 avctx->request_channels <= 2) {
289 avctx->channels = avctx->request_channels;
297 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
298 * GetBitContext within AC3DecodeContext must point to
299 * start of the synchronized ac3 bitstream.
301 static int ac3_parse_header(AC3DecodeContext *s)
304 GetBitContext *gbc = &s->gbc;
307 err = ff_ac3_parse_header(gbc->buffer, &hdr);
311 if(hdr.bitstream_id > 10)
312 return AC3_PARSE_ERROR_BSID;
314 /* get decoding parameters from header info */
315 s->bit_alloc_params.sr_code = hdr.sr_code;
316 s->channel_mode = hdr.channel_mode;
317 s->lfe_on = hdr.lfe_on;
318 s->bit_alloc_params.sr_shift = hdr.sr_shift;
319 s->sample_rate = hdr.sample_rate;
320 s->bit_rate = hdr.bit_rate;
321 s->channels = hdr.channels;
322 s->fbw_channels = s->channels - s->lfe_on;
323 s->lfe_ch = s->fbw_channels + 1;
324 s->frame_size = hdr.frame_size;
326 /* set default output to all source channels */
327 s->out_channels = s->channels;
328 s->output_mode = s->channel_mode;
330 s->output_mode |= AC3_OUTPUT_LFEON;
332 /* set default mix levels */
333 s->center_mix_level = 3; // -4.5dB
334 s->surround_mix_level = 4; // -6.0dB
336 /* skip over portion of header which has already been read */
337 skip_bits(gbc, 16); // skip the sync_word
338 skip_bits(gbc, 16); // skip crc1
339 skip_bits(gbc, 8); // skip fscod and frmsizecod
340 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
341 if(s->channel_mode == AC3_CHMODE_STEREO) {
342 skip_bits(gbc, 2); // skip dsurmod
344 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
345 s->center_mix_level = center_levels[get_bits(gbc, 2)];
346 if(s->channel_mode & 4)
347 s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
349 skip_bits1(gbc); // skip lfeon
351 /* read the rest of the bsi. read twice for dual mono mode. */
352 i = !(s->channel_mode);
354 skip_bits(gbc, 5); // skip dialog normalization
356 skip_bits(gbc, 8); //skip compression
358 skip_bits(gbc, 8); //skip language code
360 skip_bits(gbc, 7); //skip audio production information
363 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
365 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
366 TODO: read & use the xbsi1 downmix levels */
368 skip_bits(gbc, 14); //skip timecode1 / xbsi1
370 skip_bits(gbc, 14); //skip timecode2 / xbsi2
372 /* skip additional bitstream info */
373 if (get_bits1(gbc)) {
374 i = get_bits(gbc, 6);
384 * Set stereo downmixing coefficients based on frame header info.
385 * reference: Section 7.8.2 Downmixing Into Two Channels
387 static void set_downmix_coeffs(AC3DecodeContext *s)
390 float cmix = gain_levels[s->center_mix_level];
391 float smix = gain_levels[s->surround_mix_level];
393 for(i=0; i<s->fbw_channels; i++) {
394 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
395 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
397 if(s->channel_mode > 1 && s->channel_mode & 1) {
398 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
400 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
401 int nf = s->channel_mode - 2;
402 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
404 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
405 int nf = s->channel_mode - 4;
406 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
409 /* calculate adjustment needed for each channel to avoid clipping */
410 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
411 for(i=0; i<s->fbw_channels; i++) {
412 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
413 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
415 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
416 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
420 * Decode the grouped exponents according to exponent strategy.
421 * reference: Section 7.1.3 Exponent Decoding
423 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
424 uint8_t absexp, int8_t *dexps)
426 int i, j, grp, group_size;
431 group_size = exp_strategy + (exp_strategy == EXP_D45);
432 for(grp=0,i=0; grp<ngrps; grp++) {
433 expacc = get_bits(gbc, 7);
434 dexp[i++] = exp_ungroup_tab[expacc][0];
435 dexp[i++] = exp_ungroup_tab[expacc][1];
436 dexp[i++] = exp_ungroup_tab[expacc][2];
439 /* convert to absolute exps and expand groups */
441 for(i=0; i<ngrps*3; i++) {
442 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
443 for(j=0; j<group_size; j++) {
444 dexps[(i*group_size)+j] = prevexp;
450 * Generate transform coefficients for each coupled channel in the coupling
451 * range using the coupling coefficients and coupling coordinates.
452 * reference: Section 7.4.3 Coupling Coordinate Format
454 static void uncouple_channels(AC3DecodeContext *s)
456 int i, j, ch, bnd, subbnd;
459 i = s->start_freq[CPL_CH];
460 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
463 for(j=0; j<12; j++) {
464 for(ch=1; ch<=s->fbw_channels; ch++) {
465 if(s->channel_in_cpl[ch]) {
466 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
467 if (ch == 2 && s->phase_flags[bnd])
468 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
473 } while(s->cpl_band_struct[subbnd]);
478 * Grouped mantissas for 3-level 5-level and 11-level quantization
490 * Get the transform coefficients for a particular channel
491 * reference: Section 7.3 Quantization and Decoding of Mantissas
493 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
495 GetBitContext *gbc = &s->gbc;
496 int i, gcode, tbap, start, end;
501 exps = s->dexps[ch_index];
502 bap = s->bap[ch_index];
503 coeffs = s->fixed_coeffs[ch_index];
504 start = s->start_freq[ch_index];
505 end = s->end_freq[ch_index];
507 for (i = start; i < end; i++) {
511 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
516 gcode = get_bits(gbc, 5);
517 m->b1_mant[0] = b1_mantissas[gcode][0];
518 m->b1_mant[1] = b1_mantissas[gcode][1];
519 m->b1_mant[2] = b1_mantissas[gcode][2];
522 coeffs[i] = m->b1_mant[m->b1ptr++];
527 gcode = get_bits(gbc, 7);
528 m->b2_mant[0] = b2_mantissas[gcode][0];
529 m->b2_mant[1] = b2_mantissas[gcode][1];
530 m->b2_mant[2] = b2_mantissas[gcode][2];
533 coeffs[i] = m->b2_mant[m->b2ptr++];
537 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
542 gcode = get_bits(gbc, 7);
543 m->b4_mant[0] = b4_mantissas[gcode][0];
544 m->b4_mant[1] = b4_mantissas[gcode][1];
547 coeffs[i] = m->b4_mant[m->b4ptr++];
551 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
555 /* asymmetric dequantization */
556 int qlevel = quantization_tab[tbap];
557 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
561 coeffs[i] >>= exps[i];
568 * Remove random dithering from coefficients with zero-bit mantissas
569 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
571 static void remove_dithering(AC3DecodeContext *s) {
577 for(ch=1; ch<=s->fbw_channels; ch++) {
578 if(!s->dither_flag[ch]) {
579 coeffs = s->fixed_coeffs[ch];
581 if(s->channel_in_cpl[ch])
582 end = s->start_freq[CPL_CH];
584 end = s->end_freq[ch];
585 for(i=0; i<end; i++) {
589 if(s->channel_in_cpl[ch]) {
590 bap = s->bap[CPL_CH];
591 for(; i<s->end_freq[CPL_CH]; i++) {
601 * Get the transform coefficients.
603 static int get_transform_coeffs(AC3DecodeContext *s)
609 m.b1ptr = m.b2ptr = m.b4ptr = 3;
611 for (ch = 1; ch <= s->channels; ch++) {
612 /* transform coefficients for full-bandwidth channel */
613 if (get_transform_coeffs_ch(s, ch, &m))
615 /* tranform coefficients for coupling channel come right after the
616 coefficients for the first coupled channel*/
617 if (s->channel_in_cpl[ch]) {
619 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
620 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
623 uncouple_channels(s);
626 end = s->end_freq[CPL_CH];
628 end = s->end_freq[ch];
631 s->transform_coeffs[ch][end] = 0;
635 /* if any channel doesn't use dithering, zero appropriate coefficients */
643 * Stereo rematrixing.
644 * reference: Section 7.5.4 Rematrixing : Decoding Technique
646 static void do_rematrixing(AC3DecodeContext *s)
652 end = FFMIN(s->end_freq[1], s->end_freq[2]);
654 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
655 if(s->rematrixing_flags[bnd]) {
656 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
657 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
658 tmp0 = s->fixed_coeffs[1][i];
659 tmp1 = s->fixed_coeffs[2][i];
660 s->fixed_coeffs[1][i] = tmp0 + tmp1;
661 s->fixed_coeffs[2][i] = tmp0 - tmp1;
668 * Perform the 256-point IMDCT
670 static void do_imdct_256(AC3DecodeContext *s, int chindex)
673 DECLARE_ALIGNED_16(float, x[128]);
675 float *o_ptr = s->tmp_output;
678 /* de-interleave coefficients */
679 for(k=0; k<128; k++) {
680 x[k] = s->transform_coeffs[chindex][2*k+i];
683 /* run standard IMDCT */
684 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
686 /* reverse the post-rotation & reordering from standard IMDCT */
687 for(k=0; k<32; k++) {
688 z[i][32+k].re = -o_ptr[128+2*k];
689 z[i][32+k].im = -o_ptr[2*k];
690 z[i][31-k].re = o_ptr[2*k+1];
691 z[i][31-k].im = o_ptr[128+2*k+1];
695 /* apply AC-3 post-rotation & reordering */
696 for(k=0; k<64; k++) {
697 o_ptr[ 2*k ] = -z[0][ k].im;
698 o_ptr[ 2*k+1] = z[0][63-k].re;
699 o_ptr[128+2*k ] = -z[0][ k].re;
700 o_ptr[128+2*k+1] = z[0][63-k].im;
701 o_ptr[256+2*k ] = -z[1][ k].re;
702 o_ptr[256+2*k+1] = z[1][63-k].im;
703 o_ptr[384+2*k ] = z[1][ k].im;
704 o_ptr[384+2*k+1] = -z[1][63-k].re;
709 * Inverse MDCT Transform.
710 * Convert frequency domain coefficients to time-domain audio samples.
711 * reference: Section 7.9.4 Transformation Equations
713 static inline void do_imdct(AC3DecodeContext *s, int channels)
717 for (ch=1; ch<=channels; ch++) {
718 if (s->block_switch[ch]) {
721 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
722 s->transform_coeffs[ch], s->tmp_imdct);
724 /* For the first half of the block, apply the window, add the delay
725 from the previous block, and send to output */
726 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
727 s->window, s->delay[ch-1], 0, 256, 1);
728 /* For the second half of the block, apply the window and store the
729 samples to delay, to be combined with the next block */
730 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
736 * Downmix the output to mono or stereo.
738 static void ac3_downmix(AC3DecodeContext *s,
739 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
744 for(i=0; i<256; i++) {
746 for(j=0; j<s->fbw_channels; j++) {
747 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
748 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
750 v0 *= s->downmix_coeff_adjust[0];
751 v1 *= s->downmix_coeff_adjust[1];
752 if(s->output_mode == AC3_CHMODE_MONO) {
753 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
754 } else if(s->output_mode == AC3_CHMODE_STEREO) {
755 samples[ ch_offset][i] = v0;
756 samples[1+ch_offset][i] = v1;
762 * Upmix delay samples from stereo to original channel layout.
764 static void ac3_upmix_delay(AC3DecodeContext *s)
766 int channel_data_size = sizeof(s->delay[0]);
767 switch(s->channel_mode) {
768 case AC3_CHMODE_DUALMONO:
769 case AC3_CHMODE_STEREO:
770 /* upmix mono to stereo */
771 memcpy(s->delay[1], s->delay[0], channel_data_size);
773 case AC3_CHMODE_2F2R:
774 memset(s->delay[3], 0, channel_data_size);
775 case AC3_CHMODE_2F1R:
776 memset(s->delay[2], 0, channel_data_size);
778 case AC3_CHMODE_3F2R:
779 memset(s->delay[4], 0, channel_data_size);
780 case AC3_CHMODE_3F1R:
781 memset(s->delay[3], 0, channel_data_size);
783 memcpy(s->delay[2], s->delay[1], channel_data_size);
784 memset(s->delay[1], 0, channel_data_size);
790 * Parse an audio block from AC-3 bitstream.
792 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
794 int fbw_channels = s->fbw_channels;
795 int channel_mode = s->channel_mode;
797 int different_transforms;
799 GetBitContext *gbc = &s->gbc;
800 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
802 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
804 /* block switch flags */
805 different_transforms = 0;
806 for (ch = 1; ch <= fbw_channels; ch++) {
807 s->block_switch[ch] = get_bits1(gbc);
808 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
809 different_transforms = 1;
812 /* dithering flags */
814 for (ch = 1; ch <= fbw_channels; ch++) {
815 s->dither_flag[ch] = get_bits1(gbc);
816 if(!s->dither_flag[ch])
821 i = !(s->channel_mode);
824 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
825 s->avctx->drc_scale)+1.0;
826 } else if(blk == 0) {
827 s->dynamic_range[i] = 1.0f;
831 /* coupling strategy */
832 if (get_bits1(gbc)) {
833 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
834 s->cpl_in_use = get_bits1(gbc);
836 /* coupling in use */
837 int cpl_begin_freq, cpl_end_freq;
839 /* determine which channels are coupled */
840 for (ch = 1; ch <= fbw_channels; ch++)
841 s->channel_in_cpl[ch] = get_bits1(gbc);
843 /* phase flags in use */
844 if (channel_mode == AC3_CHMODE_STEREO)
845 s->phase_flags_in_use = get_bits1(gbc);
847 /* coupling frequency range and band structure */
848 cpl_begin_freq = get_bits(gbc, 4);
849 cpl_end_freq = get_bits(gbc, 4);
850 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
851 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
854 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
855 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
856 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
857 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
858 if (get_bits1(gbc)) {
859 s->cpl_band_struct[bnd] = 1;
863 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
865 /* coupling not in use */
866 for (ch = 1; ch <= fbw_channels; ch++)
867 s->channel_in_cpl[ch] = 0;
871 /* coupling coordinates */
873 int cpl_coords_exist = 0;
875 for (ch = 1; ch <= fbw_channels; ch++) {
876 if (s->channel_in_cpl[ch]) {
877 if (get_bits1(gbc)) {
878 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
879 cpl_coords_exist = 1;
880 master_cpl_coord = 3 * get_bits(gbc, 2);
881 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
882 cpl_coord_exp = get_bits(gbc, 4);
883 cpl_coord_mant = get_bits(gbc, 4);
884 if (cpl_coord_exp == 15)
885 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
887 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
888 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
894 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
895 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
896 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
901 /* stereo rematrixing strategy and band structure */
902 if (channel_mode == AC3_CHMODE_STEREO) {
903 if (get_bits1(gbc)) {
904 s->num_rematrixing_bands = 4;
905 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
906 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
907 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
908 s->rematrixing_flags[bnd] = get_bits1(gbc);
912 /* exponent strategies for each channel */
913 s->exp_strategy[CPL_CH] = EXP_REUSE;
914 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
915 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
917 s->exp_strategy[ch] = get_bits(gbc, 1);
919 s->exp_strategy[ch] = get_bits(gbc, 2);
920 if(s->exp_strategy[ch] != EXP_REUSE)
921 bit_alloc_stages[ch] = 3;
924 /* channel bandwidth */
925 for (ch = 1; ch <= fbw_channels; ch++) {
926 s->start_freq[ch] = 0;
927 if (s->exp_strategy[ch] != EXP_REUSE) {
928 int prev = s->end_freq[ch];
929 if (s->channel_in_cpl[ch])
930 s->end_freq[ch] = s->start_freq[CPL_CH];
932 int bandwidth_code = get_bits(gbc, 6);
933 if (bandwidth_code > 60) {
934 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
937 s->end_freq[ch] = bandwidth_code * 3 + 73;
939 if(blk > 0 && s->end_freq[ch] != prev)
940 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
943 s->start_freq[s->lfe_ch] = 0;
944 s->end_freq[s->lfe_ch] = 7;
946 /* decode exponents for each channel */
947 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
948 if (s->exp_strategy[ch] != EXP_REUSE) {
949 int group_size, num_groups;
950 group_size = 3 << (s->exp_strategy[ch] - 1);
952 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
953 else if(ch == s->lfe_ch)
956 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
957 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
958 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
959 &s->dexps[ch][s->start_freq[ch]+!!ch]);
960 if(ch != CPL_CH && ch != s->lfe_ch)
961 skip_bits(gbc, 2); /* skip gainrng */
965 /* bit allocation information */
966 if (get_bits1(gbc)) {
967 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
968 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
969 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
970 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
971 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
972 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
973 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
977 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
978 if (get_bits1(gbc)) {
980 csnr = (get_bits(gbc, 6) - 15) << 4;
981 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
982 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
983 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
985 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
988 /* coupling leak information */
989 if (s->cpl_in_use && get_bits1(gbc)) {
990 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
991 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
992 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
995 /* delta bit allocation information */
996 if (get_bits1(gbc)) {
997 /* delta bit allocation exists (strategy) */
998 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
999 s->dba_mode[ch] = get_bits(gbc, 2);
1000 if (s->dba_mode[ch] == DBA_RESERVED) {
1001 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1004 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1006 /* channel delta offset, len and bit allocation */
1007 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1008 if (s->dba_mode[ch] == DBA_NEW) {
1009 s->dba_nsegs[ch] = get_bits(gbc, 3);
1010 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1011 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1012 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1013 s->dba_values[ch][seg] = get_bits(gbc, 3);
1017 } else if(blk == 0) {
1018 for(ch=0; ch<=s->channels; ch++) {
1019 s->dba_mode[ch] = DBA_NONE;
1023 /* Bit allocation */
1024 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1025 if(bit_alloc_stages[ch] > 2) {
1026 /* Exponent mapping into PSD and PSD integration */
1027 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1028 s->start_freq[ch], s->end_freq[ch],
1029 s->psd[ch], s->band_psd[ch]);
1031 if(bit_alloc_stages[ch] > 1) {
1032 /* Compute excitation function, Compute masking curve, and
1033 Apply delta bit allocation */
1034 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1035 s->start_freq[ch], s->end_freq[ch],
1036 s->fast_gain[ch], (ch == s->lfe_ch),
1037 s->dba_mode[ch], s->dba_nsegs[ch],
1038 s->dba_offsets[ch], s->dba_lengths[ch],
1039 s->dba_values[ch], s->mask[ch]);
1041 if(bit_alloc_stages[ch] > 0) {
1042 /* Compute bit allocation */
1043 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1044 s->start_freq[ch], s->end_freq[ch],
1046 s->bit_alloc_params.floor,
1051 /* unused dummy data */
1052 if (get_bits1(gbc)) {
1053 int skipl = get_bits(gbc, 9);
1058 /* unpack the transform coefficients
1059 this also uncouples channels if coupling is in use. */
1060 if (get_transform_coeffs(s)) {
1061 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1065 /* recover coefficients if rematrixing is in use */
1066 if(s->channel_mode == AC3_CHMODE_STEREO)
1069 /* apply scaling to coefficients (headroom, dynrng) */
1070 for(ch=1; ch<=s->channels; ch++) {
1071 float gain = s->mul_bias / 4194304.0f;
1072 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1073 gain *= s->dynamic_range[ch-1];
1075 gain *= s->dynamic_range[0];
1077 for(i=0; i<256; i++) {
1078 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1082 /* downmix and MDCT. order depends on whether block switching is used for
1083 any channel in this block. this is because coefficients for the long
1084 and short transforms cannot be mixed. */
1085 downmix_output = s->channels != s->out_channels &&
1086 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1087 s->fbw_channels == s->out_channels);
1088 if(different_transforms) {
1089 /* the delay samples have already been downmixed, so we upmix the delay
1090 samples in order to reconstruct all channels before downmixing. */
1096 do_imdct(s, s->channels);
1098 if(downmix_output) {
1099 ac3_downmix(s, s->output, 0);
1102 if(downmix_output) {
1103 ac3_downmix(s, s->transform_coeffs, 1);
1108 ac3_downmix(s, s->delay, 0);
1111 do_imdct(s, s->out_channels);
1114 /* convert float to 16-bit integer */
1115 for(ch=0; ch<s->out_channels; ch++) {
1116 for(i=0; i<256; i++) {
1117 s->output[ch][i] += s->add_bias;
1119 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1126 * Decode a single AC-3 frame.
1128 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1129 const uint8_t *buf, int buf_size)
1131 AC3DecodeContext *s = avctx->priv_data;
1132 int16_t *out_samples = (int16_t *)data;
1133 int i, blk, ch, err;
1135 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1136 init_get_bits(&s->gbc, buf, buf_size * 8);
1138 /* parse the syncinfo */
1139 err = ac3_parse_header(s);
1142 case AC3_PARSE_ERROR_SYNC:
1143 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1145 case AC3_PARSE_ERROR_BSID:
1146 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1148 case AC3_PARSE_ERROR_SAMPLE_RATE:
1149 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1151 case AC3_PARSE_ERROR_FRAME_SIZE:
1152 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1155 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1161 /* check that reported frame size fits in input buffer */
1162 if(s->frame_size > buf_size) {
1163 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1167 /* check for crc mismatch */
1168 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1169 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1170 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1173 /* TODO: error concealment */
1176 avctx->sample_rate = s->sample_rate;
1177 avctx->bit_rate = s->bit_rate;
1179 /* channel config */
1180 s->out_channels = s->channels;
1181 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1182 avctx->request_channels < s->channels) {
1183 s->out_channels = avctx->request_channels;
1184 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1186 avctx->channels = s->out_channels;
1188 /* set downmixing coefficients if needed */
1189 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1190 s->fbw_channels == s->out_channels)) {
1191 set_downmix_coeffs(s);
1194 /* parse the audio blocks */
1195 for (blk = 0; blk < NB_BLOCKS; blk++) {
1196 if (ac3_parse_audio_block(s, blk)) {
1197 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1199 return s->frame_size;
1201 for (i = 0; i < 256; i++)
1202 for (ch = 0; ch < s->out_channels; ch++)
1203 *(out_samples++) = s->int_output[ch][i];
1205 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1206 return s->frame_size;
1210 * Uninitialize the AC-3 decoder.
1212 static int ac3_decode_end(AVCodecContext *avctx)
1214 AC3DecodeContext *s = avctx->priv_data;
1215 ff_mdct_end(&s->imdct_512);
1216 ff_mdct_end(&s->imdct_256);
1221 AVCodec ac3_decoder = {
1223 .type = CODEC_TYPE_AUDIO,
1225 .priv_data_size = sizeof (AC3DecodeContext),
1226 .init = ac3_decode_init,
1227 .close = ac3_decode_end,
1228 .decode = ac3_decode_frame,