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 };
49 * table for exponent to scale_factor mapping
50 * scale_factors[i] = 2 ^ -i
52 static float scale_factors[25];
54 /** table for grouping exponents */
55 static uint8_t exp_ungroup_tab[128][3];
58 /** tables for ungrouping mantissas */
59 static float b1_mantissas[32][3];
60 static float b2_mantissas[128][3];
61 static float b3_mantissas[8];
62 static float b4_mantissas[128][2];
63 static float b5_mantissas[16];
66 * Quantization table: levels for symmetric. bits for asymmetric.
67 * reference: Table 7.18 Mapping of bap to Quantizer
69 static const uint8_t quantization_tab[16] = {
71 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
74 /** dynamic range table. converts codes to scale factors. */
75 static float dynamic_range_tab[256];
77 /** Adjustments in dB gain */
78 #define LEVEL_MINUS_3DB 0.7071067811865476
79 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
80 #define LEVEL_MINUS_6DB 0.5000000000000000
81 #define LEVEL_MINUS_9DB 0.3535533905932738
82 #define LEVEL_ZERO 0.0000000000000000
83 #define LEVEL_ONE 1.0000000000000000
85 static const float gain_levels[6] = {
89 LEVEL_MINUS_4POINT5DB,
95 * Table for center mix levels
96 * reference: Section 5.4.2.4 cmixlev
98 static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
101 * Table for surround mix levels
102 * reference: Section 5.4.2.5 surmixlev
104 static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
107 * Table for default stereo downmixing coefficients
108 * reference: Section 7.8.2 Downmixing Into Two Channels
110 static const uint8_t ac3_default_coeffs[8][5][2] = {
111 { { 1, 0 }, { 0, 1 }, },
113 { { 1, 0 }, { 0, 1 }, },
114 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
115 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
116 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
117 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
118 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
121 /* override ac3.h to include coupling channel */
122 #undef AC3_MAX_CHANNELS
123 #define AC3_MAX_CHANNELS 7
126 #define AC3_OUTPUT_LFEON 8
129 int channel_mode; ///< channel mode (acmod)
130 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
131 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
132 int dither_all; ///< true if all channels are dithered
133 int cpl_in_use; ///< coupling in use
134 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
135 int phase_flags_in_use; ///< phase flags in use
136 int phase_flags[18]; ///< phase flags
137 int cpl_band_struct[18]; ///< coupling band structure
138 int rematrixing_strategy; ///< rematrixing strategy
139 int num_rematrixing_bands; ///< number of rematrixing bands
140 int rematrixing_flags[4]; ///< rematrixing flags
141 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
142 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
143 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
144 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
145 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
146 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
147 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
148 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
150 int sample_rate; ///< sample frequency, in Hz
151 int bit_rate; ///< stream bit rate, in bits-per-second
152 int frame_size; ///< current frame size, in bytes
154 int channels; ///< number of total channels
155 int fbw_channels; ///< number of full-bandwidth channels
156 int lfe_on; ///< lfe channel in use
157 int lfe_ch; ///< index of LFE channel
158 int output_mode; ///< output channel configuration
159 int out_channels; ///< number of output channels
161 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
162 float dynamic_range[2]; ///< dynamic range
163 float cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
164 int num_cpl_bands; ///< number of coupling bands
165 int num_cpl_subbands; ///< number of coupling sub bands
166 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
167 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
168 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
170 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
171 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
172 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
173 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
174 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
176 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
179 MDCTContext imdct_512; ///< for 512 sample IMDCT
180 MDCTContext imdct_256; ///< for 256 sample IMDCT
181 DSPContext dsp; ///< for optimization
182 float add_bias; ///< offset for float_to_int16 conversion
183 float mul_bias; ///< scaling for float_to_int16 conversion
185 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]); ///< output after imdct transform and windowing
186 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
187 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]); ///< delay - added to the next block
188 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
189 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
190 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
193 GetBitContext gbc; ///< bitstream reader
194 AVRandomState dith_state; ///< for dither generation
195 AVCodecContext *avctx; ///< parent context
199 * Generate a Kaiser-Bessel Derived Window.
201 static void ac3_window_init(float *window)
204 double sum = 0.0, bessel, tmp;
205 double local_window[256];
206 double alpha2 = (5.0 * M_PI / 256.0) * (5.0 * M_PI / 256.0);
208 for (i = 0; i < 256; i++) {
209 tmp = i * (256 - i) * alpha2;
211 for (j = 100; j > 0; j--) /* default to 100 iterations */
212 bessel = bessel * tmp / (j * j) + 1;
214 local_window[i] = sum;
218 for (i = 0; i < 256; i++)
219 window[i] = sqrt(local_window[i] / sum);
223 * Symmetrical Dequantization
224 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
225 * Tables 7.19 to 7.23
228 symmetric_dequant(int code, int levels)
230 return (code - (levels >> 1)) * (2.0f / levels);
234 * Initialize tables at runtime.
236 static void ac3_tables_init(void)
240 /* generate grouped mantissa tables
241 reference: Section 7.3.5 Ungrouping of Mantissas */
242 for(i=0; i<32; i++) {
243 /* bap=1 mantissas */
244 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
245 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
246 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
248 for(i=0; i<128; i++) {
249 /* bap=2 mantissas */
250 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
251 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
252 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
254 /* bap=4 mantissas */
255 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
256 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
258 /* generate ungrouped mantissa tables
259 reference: Tables 7.21 and 7.23 */
261 /* bap=3 mantissas */
262 b3_mantissas[i] = symmetric_dequant(i, 7);
264 for(i=0; i<15; i++) {
265 /* bap=5 mantissas */
266 b5_mantissas[i] = symmetric_dequant(i, 15);
269 /* generate dynamic range table
270 reference: Section 7.7.1 Dynamic Range Control */
271 for(i=0; i<256; i++) {
272 int v = (i >> 5) - ((i >> 7) << 3) - 5;
273 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
276 /* generate scale factors for exponents and asymmetrical dequantization
277 reference: Section 7.3.2 Expansion of Mantissas for Asymmetric Quantization */
278 for (i = 0; i < 25; i++)
279 scale_factors[i] = pow(2.0, -i);
281 /* generate exponent tables
282 reference: Section 7.1.3 Exponent Decoding */
283 for(i=0; i<128; i++) {
284 exp_ungroup_tab[i][0] = i / 25;
285 exp_ungroup_tab[i][1] = (i % 25) / 5;
286 exp_ungroup_tab[i][2] = (i % 25) % 5;
292 * AVCodec initialization
294 static int ac3_decode_init(AVCodecContext *avctx)
296 AC3DecodeContext *s = avctx->priv_data;
301 ff_mdct_init(&s->imdct_256, 8, 1);
302 ff_mdct_init(&s->imdct_512, 9, 1);
303 ac3_window_init(s->window);
304 dsputil_init(&s->dsp, avctx);
305 av_init_random(0, &s->dith_state);
307 /* set bias values for float to int16 conversion */
308 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
309 s->add_bias = 385.0f;
313 s->mul_bias = 32767.0f;
316 /* allow downmixing to stereo or mono */
317 if (avctx->channels > 0 && avctx->request_channels > 0 &&
318 avctx->request_channels < avctx->channels &&
319 avctx->request_channels <= 2) {
320 avctx->channels = avctx->request_channels;
327 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
328 * GetBitContext within AC3DecodeContext must point to
329 * start of the synchronized ac3 bitstream.
331 static int ac3_parse_header(AC3DecodeContext *s)
334 GetBitContext *gbc = &s->gbc;
335 float center_mix_level, surround_mix_level;
338 err = ff_ac3_parse_header(gbc->buffer, &hdr);
342 if(hdr.bitstream_id > 10)
343 return AC3_PARSE_ERROR_BSID;
345 /* get decoding parameters from header info */
346 s->bit_alloc_params.sr_code = hdr.sr_code;
347 s->channel_mode = hdr.channel_mode;
348 s->lfe_on = hdr.lfe_on;
349 s->bit_alloc_params.sr_shift = hdr.sr_shift;
350 s->sample_rate = hdr.sample_rate;
351 s->bit_rate = hdr.bit_rate;
352 s->channels = hdr.channels;
353 s->fbw_channels = s->channels - s->lfe_on;
354 s->lfe_ch = s->fbw_channels + 1;
355 s->frame_size = hdr.frame_size;
357 /* set default output to all source channels */
358 s->out_channels = s->channels;
359 s->output_mode = s->channel_mode;
361 s->output_mode |= AC3_OUTPUT_LFEON;
363 /* skip over portion of header which has already been read */
364 skip_bits(gbc, 16); // skip the sync_word
365 skip_bits(gbc, 16); // skip crc1
366 skip_bits(gbc, 8); // skip fscod and frmsizecod
367 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
368 if(s->channel_mode == AC3_CHMODE_STEREO) {
369 skip_bits(gbc, 2); // skip dsurmod
371 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
372 center_mix_level = gain_levels[center_levels[get_bits(gbc, 2)]];
373 if(s->channel_mode & 4)
374 surround_mix_level = gain_levels[surround_levels[get_bits(gbc, 2)]];
376 skip_bits1(gbc); // skip lfeon
378 /* read the rest of the bsi. read twice for dual mono mode. */
379 i = !(s->channel_mode);
381 skip_bits(gbc, 5); // skip dialog normalization
383 skip_bits(gbc, 8); //skip compression
385 skip_bits(gbc, 8); //skip language code
387 skip_bits(gbc, 7); //skip audio production information
390 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
392 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
393 TODO: read & use the xbsi1 downmix levels */
395 skip_bits(gbc, 14); //skip timecode1 / xbsi1
397 skip_bits(gbc, 14); //skip timecode2 / xbsi2
399 /* skip additional bitstream info */
400 if (get_bits1(gbc)) {
401 i = get_bits(gbc, 6);
407 /* set stereo downmixing coefficients
408 reference: Section 7.8.2 Downmixing Into Two Channels */
409 for(i=0; i<s->fbw_channels; i++) {
410 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
411 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
413 if(s->channel_mode > 1 && s->channel_mode & 1) {
414 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = center_mix_level;
416 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
417 int nf = s->channel_mode - 2;
418 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = surround_mix_level * LEVEL_MINUS_3DB;
420 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
421 int nf = s->channel_mode - 4;
422 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = surround_mix_level;
429 * Decode the grouped exponents according to exponent strategy.
430 * reference: Section 7.1.3 Exponent Decoding
432 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
433 uint8_t absexp, int8_t *dexps)
435 int i, j, grp, group_size;
440 group_size = exp_strategy + (exp_strategy == EXP_D45);
441 for(grp=0,i=0; grp<ngrps; grp++) {
442 expacc = get_bits(gbc, 7);
443 dexp[i++] = exp_ungroup_tab[expacc][0];
444 dexp[i++] = exp_ungroup_tab[expacc][1];
445 dexp[i++] = exp_ungroup_tab[expacc][2];
448 /* convert to absolute exps and expand groups */
450 for(i=0; i<ngrps*3; i++) {
451 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
452 for(j=0; j<group_size; j++) {
453 dexps[(i*group_size)+j] = prevexp;
459 * Generate transform coefficients for each coupled channel in the coupling
460 * range using the coupling coefficients and coupling coordinates.
461 * reference: Section 7.4.3 Coupling Coordinate Format
463 static void uncouple_channels(AC3DecodeContext *s)
465 int i, j, ch, bnd, subbnd;
468 i = s->start_freq[CPL_CH];
469 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
472 for(j=0; j<12; j++) {
473 for(ch=1; ch<=s->fbw_channels; ch++) {
474 if(s->channel_in_cpl[ch]) {
475 s->transform_coeffs[ch][i] = s->transform_coeffs[CPL_CH][i] * s->cpl_coords[ch][bnd] * 8.0f;
476 if (ch == 2 && s->phase_flags[bnd])
477 s->transform_coeffs[ch][i] = -s->transform_coeffs[ch][i];
482 } while(s->cpl_band_struct[subbnd]);
487 * Grouped mantissas for 3-level 5-level and 11-level quantization
499 * Get the transform coefficients for a particular channel
500 * reference: Section 7.3 Quantization and Decoding of Mantissas
502 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
504 GetBitContext *gbc = &s->gbc;
505 int i, gcode, tbap, start, end;
510 exps = s->dexps[ch_index];
511 bap = s->bap[ch_index];
512 coeffs = s->transform_coeffs[ch_index];
513 start = s->start_freq[ch_index];
514 end = s->end_freq[ch_index];
516 for (i = start; i < end; i++) {
520 coeffs[i] = ((av_random(&s->dith_state) & 0xFFFF) / 65535.0f) - 0.5f;
525 gcode = get_bits(gbc, 5);
526 m->b1_mant[0] = b1_mantissas[gcode][0];
527 m->b1_mant[1] = b1_mantissas[gcode][1];
528 m->b1_mant[2] = b1_mantissas[gcode][2];
531 coeffs[i] = m->b1_mant[m->b1ptr++];
536 gcode = get_bits(gbc, 7);
537 m->b2_mant[0] = b2_mantissas[gcode][0];
538 m->b2_mant[1] = b2_mantissas[gcode][1];
539 m->b2_mant[2] = b2_mantissas[gcode][2];
542 coeffs[i] = m->b2_mant[m->b2ptr++];
546 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
551 gcode = get_bits(gbc, 7);
552 m->b4_mant[0] = b4_mantissas[gcode][0];
553 m->b4_mant[1] = b4_mantissas[gcode][1];
556 coeffs[i] = m->b4_mant[m->b4ptr++];
560 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
564 /* asymmetric dequantization */
565 coeffs[i] = get_sbits(gbc, quantization_tab[tbap]) * scale_factors[quantization_tab[tbap]-1];
568 coeffs[i] *= scale_factors[exps[i]];
575 * Remove random dithering from coefficients with zero-bit mantissas
576 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
578 static void remove_dithering(AC3DecodeContext *s) {
584 for(ch=1; ch<=s->fbw_channels; ch++) {
585 if(!s->dither_flag[ch]) {
586 coeffs = s->transform_coeffs[ch];
588 if(s->channel_in_cpl[ch])
589 end = s->start_freq[CPL_CH];
591 end = s->end_freq[ch];
592 for(i=0; i<end; i++) {
596 if(s->channel_in_cpl[ch]) {
597 bap = s->bap[CPL_CH];
598 for(; i<s->end_freq[CPL_CH]; i++) {
608 * Get the transform coefficients.
610 static int get_transform_coeffs(AC3DecodeContext *s)
616 m.b1ptr = m.b2ptr = m.b4ptr = 3;
618 for (ch = 1; ch <= s->channels; ch++) {
619 /* transform coefficients for full-bandwidth channel */
620 if (get_transform_coeffs_ch(s, ch, &m))
622 /* tranform coefficients for coupling channel come right after the
623 coefficients for the first coupled channel*/
624 if (s->channel_in_cpl[ch]) {
626 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
627 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
630 uncouple_channels(s);
633 end = s->end_freq[CPL_CH];
635 end = s->end_freq[ch];
638 s->transform_coeffs[ch][end] = 0;
642 /* if any channel doesn't use dithering, zero appropriate coefficients */
650 * Stereo rematrixing.
651 * reference: Section 7.5.4 Rematrixing : Decoding Technique
653 static void do_rematrixing(AC3DecodeContext *s)
659 end = FFMIN(s->end_freq[1], s->end_freq[2]);
661 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
662 if(s->rematrixing_flags[bnd]) {
663 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
664 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
665 tmp0 = s->transform_coeffs[1][i];
666 tmp1 = s->transform_coeffs[2][i];
667 s->transform_coeffs[1][i] = tmp0 + tmp1;
668 s->transform_coeffs[2][i] = tmp0 - tmp1;
675 * Perform the 256-point IMDCT
677 static void do_imdct_256(AC3DecodeContext *s, int chindex)
680 DECLARE_ALIGNED_16(float, x[128]);
682 float *o_ptr = s->tmp_output;
685 /* de-interleave coefficients */
686 for(k=0; k<128; k++) {
687 x[k] = s->transform_coeffs[chindex][2*k+i];
690 /* run standard IMDCT */
691 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
693 /* reverse the post-rotation & reordering from standard IMDCT */
694 for(k=0; k<32; k++) {
695 z[i][32+k].re = -o_ptr[128+2*k];
696 z[i][32+k].im = -o_ptr[2*k];
697 z[i][31-k].re = o_ptr[2*k+1];
698 z[i][31-k].im = o_ptr[128+2*k+1];
702 /* apply AC-3 post-rotation & reordering */
703 for(k=0; k<64; k++) {
704 o_ptr[ 2*k ] = -z[0][ k].im;
705 o_ptr[ 2*k+1] = z[0][63-k].re;
706 o_ptr[128+2*k ] = -z[0][ k].re;
707 o_ptr[128+2*k+1] = z[0][63-k].im;
708 o_ptr[256+2*k ] = -z[1][ k].re;
709 o_ptr[256+2*k+1] = z[1][63-k].im;
710 o_ptr[384+2*k ] = z[1][ k].im;
711 o_ptr[384+2*k+1] = -z[1][63-k].re;
716 * Inverse MDCT Transform.
717 * Convert frequency domain coefficients to time-domain audio samples.
718 * reference: Section 7.9.4 Transformation Equations
720 static inline void do_imdct(AC3DecodeContext *s)
725 /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
726 channels = s->fbw_channels;
727 if(s->output_mode & AC3_OUTPUT_LFEON)
730 for (ch=1; ch<=channels; ch++) {
731 if (s->block_switch[ch]) {
734 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
735 s->transform_coeffs[ch], s->tmp_imdct);
737 /* For the first half of the block, apply the window, add the delay
738 from the previous block, and send to output */
739 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
740 s->window, s->delay[ch-1], 0, 256, 1);
741 /* For the second half of the block, apply the window and store the
742 samples to delay, to be combined with the next block */
743 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
749 * Downmix the output to mono or stereo.
751 static void ac3_downmix(AC3DecodeContext *s)
754 float v0, v1, s0, s1;
756 for(i=0; i<256; i++) {
757 v0 = v1 = s0 = s1 = 0.0f;
758 for(j=0; j<s->fbw_channels; j++) {
759 v0 += s->output[j][i] * s->downmix_coeffs[j][0];
760 v1 += s->output[j][i] * s->downmix_coeffs[j][1];
761 s0 += s->downmix_coeffs[j][0];
762 s1 += s->downmix_coeffs[j][1];
766 if(s->output_mode == AC3_CHMODE_MONO) {
767 s->output[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
768 } else if(s->output_mode == AC3_CHMODE_STEREO) {
769 s->output[0][i] = v0;
770 s->output[1][i] = v1;
776 * Parse an audio block from AC-3 bitstream.
778 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
780 int fbw_channels = s->fbw_channels;
781 int channel_mode = s->channel_mode;
783 GetBitContext *gbc = &s->gbc;
784 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
786 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
788 /* block switch flags */
789 for (ch = 1; ch <= fbw_channels; ch++)
790 s->block_switch[ch] = get_bits1(gbc);
792 /* dithering flags */
794 for (ch = 1; ch <= fbw_channels; ch++) {
795 s->dither_flag[ch] = get_bits1(gbc);
796 if(!s->dither_flag[ch])
801 i = !(s->channel_mode);
804 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
805 s->avctx->drc_scale)+1.0;
806 } else if(blk == 0) {
807 s->dynamic_range[i] = 1.0f;
811 /* coupling strategy */
812 if (get_bits1(gbc)) {
813 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
814 s->cpl_in_use = get_bits1(gbc);
816 /* coupling in use */
817 int cpl_begin_freq, cpl_end_freq;
819 /* determine which channels are coupled */
820 for (ch = 1; ch <= fbw_channels; ch++)
821 s->channel_in_cpl[ch] = get_bits1(gbc);
823 /* phase flags in use */
824 if (channel_mode == AC3_CHMODE_STEREO)
825 s->phase_flags_in_use = get_bits1(gbc);
827 /* coupling frequency range and band structure */
828 cpl_begin_freq = get_bits(gbc, 4);
829 cpl_end_freq = get_bits(gbc, 4);
830 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
831 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
834 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
835 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
836 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
837 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
838 if (get_bits1(gbc)) {
839 s->cpl_band_struct[bnd] = 1;
844 /* coupling not in use */
845 for (ch = 1; ch <= fbw_channels; ch++)
846 s->channel_in_cpl[ch] = 0;
850 /* coupling coordinates */
852 int cpl_coords_exist = 0;
854 for (ch = 1; ch <= fbw_channels; ch++) {
855 if (s->channel_in_cpl[ch]) {
856 if (get_bits1(gbc)) {
857 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
858 cpl_coords_exist = 1;
859 master_cpl_coord = 3 * get_bits(gbc, 2);
860 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
861 cpl_coord_exp = get_bits(gbc, 4);
862 cpl_coord_mant = get_bits(gbc, 4);
863 if (cpl_coord_exp == 15)
864 s->cpl_coords[ch][bnd] = cpl_coord_mant / 16.0f;
866 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16.0f) / 32.0f;
867 s->cpl_coords[ch][bnd] *= scale_factors[cpl_coord_exp + master_cpl_coord];
873 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
874 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
875 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
880 /* stereo rematrixing strategy and band structure */
881 if (channel_mode == AC3_CHMODE_STEREO) {
882 s->rematrixing_strategy = get_bits1(gbc);
883 if (s->rematrixing_strategy) {
884 s->num_rematrixing_bands = 4;
885 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
886 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
887 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
888 s->rematrixing_flags[bnd] = get_bits1(gbc);
892 /* exponent strategies for each channel */
893 s->exp_strategy[CPL_CH] = EXP_REUSE;
894 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
895 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
897 s->exp_strategy[ch] = get_bits(gbc, 1);
899 s->exp_strategy[ch] = get_bits(gbc, 2);
900 if(s->exp_strategy[ch] != EXP_REUSE)
901 bit_alloc_stages[ch] = 3;
904 /* channel bandwidth */
905 for (ch = 1; ch <= fbw_channels; ch++) {
906 s->start_freq[ch] = 0;
907 if (s->exp_strategy[ch] != EXP_REUSE) {
908 int prev = s->end_freq[ch];
909 if (s->channel_in_cpl[ch])
910 s->end_freq[ch] = s->start_freq[CPL_CH];
912 int bandwidth_code = get_bits(gbc, 6);
913 if (bandwidth_code > 60) {
914 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
917 s->end_freq[ch] = bandwidth_code * 3 + 73;
919 if(blk > 0 && s->end_freq[ch] != prev)
920 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
923 s->start_freq[s->lfe_ch] = 0;
924 s->end_freq[s->lfe_ch] = 7;
926 /* decode exponents for each channel */
927 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
928 if (s->exp_strategy[ch] != EXP_REUSE) {
929 int group_size, num_groups;
930 group_size = 3 << (s->exp_strategy[ch] - 1);
932 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
933 else if(ch == s->lfe_ch)
936 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
937 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
938 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
939 &s->dexps[ch][s->start_freq[ch]+!!ch]);
940 if(ch != CPL_CH && ch != s->lfe_ch)
941 skip_bits(gbc, 2); /* skip gainrng */
945 /* bit allocation information */
946 if (get_bits1(gbc)) {
947 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
948 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
949 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
950 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
951 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
952 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
953 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
957 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
958 if (get_bits1(gbc)) {
960 csnr = (get_bits(gbc, 6) - 15) << 4;
961 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
962 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
963 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
965 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
968 /* coupling leak information */
969 if (s->cpl_in_use && get_bits1(gbc)) {
970 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
971 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
972 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
975 /* delta bit allocation information */
976 if (get_bits1(gbc)) {
977 /* delta bit allocation exists (strategy) */
978 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
979 s->dba_mode[ch] = get_bits(gbc, 2);
980 if (s->dba_mode[ch] == DBA_RESERVED) {
981 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
984 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
986 /* channel delta offset, len and bit allocation */
987 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
988 if (s->dba_mode[ch] == DBA_NEW) {
989 s->dba_nsegs[ch] = get_bits(gbc, 3);
990 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
991 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
992 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
993 s->dba_values[ch][seg] = get_bits(gbc, 3);
997 } else if(blk == 0) {
998 for(ch=0; ch<=s->channels; ch++) {
999 s->dba_mode[ch] = DBA_NONE;
1003 /* Bit allocation */
1004 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1005 if(bit_alloc_stages[ch] > 2) {
1006 /* Exponent mapping into PSD and PSD integration */
1007 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1008 s->start_freq[ch], s->end_freq[ch],
1009 s->psd[ch], s->band_psd[ch]);
1011 if(bit_alloc_stages[ch] > 1) {
1012 /* Compute excitation function, Compute masking curve, and
1013 Apply delta bit allocation */
1014 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1015 s->start_freq[ch], s->end_freq[ch],
1016 s->fast_gain[ch], (ch == s->lfe_ch),
1017 s->dba_mode[ch], s->dba_nsegs[ch],
1018 s->dba_offsets[ch], s->dba_lengths[ch],
1019 s->dba_values[ch], s->mask[ch]);
1021 if(bit_alloc_stages[ch] > 0) {
1022 /* Compute bit allocation */
1023 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1024 s->start_freq[ch], s->end_freq[ch],
1026 s->bit_alloc_params.floor,
1031 /* unused dummy data */
1032 if (get_bits1(gbc)) {
1033 int skipl = get_bits(gbc, 9);
1038 /* unpack the transform coefficients
1039 this also uncouples channels if coupling is in use. */
1040 if (get_transform_coeffs(s)) {
1041 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1045 /* recover coefficients if rematrixing is in use */
1046 if(s->channel_mode == AC3_CHMODE_STEREO)
1049 /* apply scaling to coefficients (headroom, dynrng) */
1050 for(ch=1; ch<=s->channels; ch++) {
1051 float gain = 2.0f * s->mul_bias;
1052 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1053 gain *= s->dynamic_range[ch-1];
1055 gain *= s->dynamic_range[0];
1057 for(i=0; i<s->end_freq[ch]; i++) {
1058 s->transform_coeffs[ch][i] *= gain;
1064 /* downmix output if needed */
1065 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1066 s->fbw_channels == s->out_channels)) {
1070 /* convert float to 16-bit integer */
1071 for(ch=0; ch<s->out_channels; ch++) {
1072 for(i=0; i<256; i++) {
1073 s->output[ch][i] += s->add_bias;
1075 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1082 * Decode a single AC-3 frame.
1084 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
1086 AC3DecodeContext *s = avctx->priv_data;
1087 int16_t *out_samples = (int16_t *)data;
1088 int i, blk, ch, err;
1090 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1091 init_get_bits(&s->gbc, buf, buf_size * 8);
1093 /* parse the syncinfo */
1094 err = ac3_parse_header(s);
1097 case AC3_PARSE_ERROR_SYNC:
1098 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1100 case AC3_PARSE_ERROR_BSID:
1101 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1103 case AC3_PARSE_ERROR_SAMPLE_RATE:
1104 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1106 case AC3_PARSE_ERROR_FRAME_SIZE:
1107 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1110 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1116 /* check that reported frame size fits in input buffer */
1117 if(s->frame_size > buf_size) {
1118 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1122 /* check for crc mismatch */
1123 if(avctx->error_resilience > 0) {
1124 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1125 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1128 /* TODO: error concealment */
1131 avctx->sample_rate = s->sample_rate;
1132 avctx->bit_rate = s->bit_rate;
1134 /* channel config */
1135 s->out_channels = s->channels;
1136 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1137 avctx->request_channels < s->channels) {
1138 s->out_channels = avctx->request_channels;
1139 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1141 avctx->channels = s->out_channels;
1143 /* parse the audio blocks */
1144 for (blk = 0; blk < NB_BLOCKS; blk++) {
1145 if (ac3_parse_audio_block(s, blk)) {
1146 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1148 return s->frame_size;
1150 for (i = 0; i < 256; i++)
1151 for (ch = 0; ch < s->out_channels; ch++)
1152 *(out_samples++) = s->int_output[ch][i];
1154 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1155 return s->frame_size;
1159 * Uninitialize the AC-3 decoder.
1161 static int ac3_decode_end(AVCodecContext *avctx)
1163 AC3DecodeContext *s = avctx->priv_data;
1164 ff_mdct_end(&s->imdct_512);
1165 ff_mdct_end(&s->imdct_256);
1170 AVCodec ac3_decoder = {
1172 .type = CODEC_TYPE_AUDIO,
1174 .priv_data_size = sizeof (AC3DecodeContext),
1175 .init = ac3_decode_init,
1176 .close = ac3_decode_end,
1177 .decode = ac3_decode_frame,