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_sum[2]; ///< sum of downmix coeffs for each output channel
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
176 MDCTContext imdct_512; ///< for 512 sample IMDCT
177 MDCTContext imdct_256; ///< for 256 sample IMDCT
178 DSPContext dsp; ///< for optimization
179 float add_bias; ///< offset for float_to_int16 conversion
180 float mul_bias; ///< scaling for float_to_int16 conversion
182 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]); ///< output after imdct transform and windowing
183 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
184 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]); ///< delay - added to the next block
185 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
186 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
187 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
190 GetBitContext gbc; ///< bitstream reader
191 AVRandomState dith_state; ///< for dither generation
192 AVCodecContext *avctx; ///< parent context
196 * Symmetrical Dequantization
197 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
198 * Tables 7.19 to 7.23
201 symmetric_dequant(int code, int levels)
203 return ((code - (levels >> 1)) << 24) / levels;
207 * Initialize tables at runtime.
209 static void ac3_tables_init(void)
213 /* generate grouped mantissa tables
214 reference: Section 7.3.5 Ungrouping of Mantissas */
215 for(i=0; i<32; i++) {
216 /* bap=1 mantissas */
217 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
218 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
219 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
221 for(i=0; i<128; i++) {
222 /* bap=2 mantissas */
223 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
224 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
225 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
227 /* bap=4 mantissas */
228 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
229 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
231 /* generate ungrouped mantissa tables
232 reference: Tables 7.21 and 7.23 */
234 /* bap=3 mantissas */
235 b3_mantissas[i] = symmetric_dequant(i, 7);
237 for(i=0; i<15; i++) {
238 /* bap=5 mantissas */
239 b5_mantissas[i] = symmetric_dequant(i, 15);
242 /* generate dynamic range table
243 reference: Section 7.7.1 Dynamic Range Control */
244 for(i=0; i<256; i++) {
245 int v = (i >> 5) - ((i >> 7) << 3) - 5;
246 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
249 /* generate exponent tables
250 reference: Section 7.1.3 Exponent Decoding */
251 for(i=0; i<128; i++) {
252 exp_ungroup_tab[i][0] = i / 25;
253 exp_ungroup_tab[i][1] = (i % 25) / 5;
254 exp_ungroup_tab[i][2] = (i % 25) % 5;
260 * AVCodec initialization
262 static int ac3_decode_init(AVCodecContext *avctx)
264 AC3DecodeContext *s = avctx->priv_data;
269 ff_mdct_init(&s->imdct_256, 8, 1);
270 ff_mdct_init(&s->imdct_512, 9, 1);
271 ff_kbd_window_init(s->window, 5.0, 256);
272 dsputil_init(&s->dsp, avctx);
273 av_init_random(0, &s->dith_state);
275 /* set bias values for float to int16 conversion */
276 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
277 s->add_bias = 385.0f;
281 s->mul_bias = 32767.0f;
284 /* allow downmixing to stereo or mono */
285 if (avctx->channels > 0 && avctx->request_channels > 0 &&
286 avctx->request_channels < avctx->channels &&
287 avctx->request_channels <= 2) {
288 avctx->channels = avctx->request_channels;
295 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
296 * GetBitContext within AC3DecodeContext must point to
297 * start of the synchronized ac3 bitstream.
299 static int ac3_parse_header(AC3DecodeContext *s)
302 GetBitContext *gbc = &s->gbc;
305 err = ff_ac3_parse_header(gbc->buffer, &hdr);
309 if(hdr.bitstream_id > 10)
310 return AC3_PARSE_ERROR_BSID;
312 /* get decoding parameters from header info */
313 s->bit_alloc_params.sr_code = hdr.sr_code;
314 s->channel_mode = hdr.channel_mode;
315 s->lfe_on = hdr.lfe_on;
316 s->bit_alloc_params.sr_shift = hdr.sr_shift;
317 s->sample_rate = hdr.sample_rate;
318 s->bit_rate = hdr.bit_rate;
319 s->channels = hdr.channels;
320 s->fbw_channels = s->channels - s->lfe_on;
321 s->lfe_ch = s->fbw_channels + 1;
322 s->frame_size = hdr.frame_size;
324 /* set default output to all source channels */
325 s->out_channels = s->channels;
326 s->output_mode = s->channel_mode;
328 s->output_mode |= AC3_OUTPUT_LFEON;
330 /* set default mix levels */
331 s->center_mix_level = 3; // -4.5dB
332 s->surround_mix_level = 4; // -6.0dB
334 /* skip over portion of header which has already been read */
335 skip_bits(gbc, 16); // skip the sync_word
336 skip_bits(gbc, 16); // skip crc1
337 skip_bits(gbc, 8); // skip fscod and frmsizecod
338 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
339 if(s->channel_mode == AC3_CHMODE_STEREO) {
340 skip_bits(gbc, 2); // skip dsurmod
342 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
343 s->center_mix_level = center_levels[get_bits(gbc, 2)];
344 if(s->channel_mode & 4)
345 s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
347 skip_bits1(gbc); // skip lfeon
349 /* read the rest of the bsi. read twice for dual mono mode. */
350 i = !(s->channel_mode);
352 skip_bits(gbc, 5); // skip dialog normalization
354 skip_bits(gbc, 8); //skip compression
356 skip_bits(gbc, 8); //skip language code
358 skip_bits(gbc, 7); //skip audio production information
361 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
363 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
364 TODO: read & use the xbsi1 downmix levels */
366 skip_bits(gbc, 14); //skip timecode1 / xbsi1
368 skip_bits(gbc, 14); //skip timecode2 / xbsi2
370 /* skip additional bitstream info */
371 if (get_bits1(gbc)) {
372 i = get_bits(gbc, 6);
382 * Set stereo downmixing coefficients based on frame header info.
383 * reference: Section 7.8.2 Downmixing Into Two Channels
385 static void set_downmix_coeffs(AC3DecodeContext *s)
388 float cmix = gain_levels[s->center_mix_level];
389 float smix = gain_levels[s->surround_mix_level];
391 for(i=0; i<s->fbw_channels; i++) {
392 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
393 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
395 if(s->channel_mode > 1 && s->channel_mode & 1) {
396 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
398 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
399 int nf = s->channel_mode - 2;
400 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
402 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
403 int nf = s->channel_mode - 4;
404 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
407 s->downmix_coeff_sum[0] = s->downmix_coeff_sum[1] = 0.0f;
408 for(i=0; i<s->fbw_channels; i++) {
409 s->downmix_coeff_sum[0] += s->downmix_coeffs[i][0];
410 s->downmix_coeff_sum[1] += s->downmix_coeffs[i][1];
415 * Decode the grouped exponents according to exponent strategy.
416 * reference: Section 7.1.3 Exponent Decoding
418 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
419 uint8_t absexp, int8_t *dexps)
421 int i, j, grp, group_size;
426 group_size = exp_strategy + (exp_strategy == EXP_D45);
427 for(grp=0,i=0; grp<ngrps; grp++) {
428 expacc = get_bits(gbc, 7);
429 dexp[i++] = exp_ungroup_tab[expacc][0];
430 dexp[i++] = exp_ungroup_tab[expacc][1];
431 dexp[i++] = exp_ungroup_tab[expacc][2];
434 /* convert to absolute exps and expand groups */
436 for(i=0; i<ngrps*3; i++) {
437 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
438 for(j=0; j<group_size; j++) {
439 dexps[(i*group_size)+j] = prevexp;
445 * Generate transform coefficients for each coupled channel in the coupling
446 * range using the coupling coefficients and coupling coordinates.
447 * reference: Section 7.4.3 Coupling Coordinate Format
449 static void uncouple_channels(AC3DecodeContext *s)
451 int i, j, ch, bnd, subbnd;
454 i = s->start_freq[CPL_CH];
455 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
458 for(j=0; j<12; j++) {
459 for(ch=1; ch<=s->fbw_channels; ch++) {
460 if(s->channel_in_cpl[ch]) {
461 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
462 if (ch == 2 && s->phase_flags[bnd])
463 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
468 } while(s->cpl_band_struct[subbnd]);
473 * Grouped mantissas for 3-level 5-level and 11-level quantization
485 * Get the transform coefficients for a particular channel
486 * reference: Section 7.3 Quantization and Decoding of Mantissas
488 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
490 GetBitContext *gbc = &s->gbc;
491 int i, gcode, tbap, start, end;
496 exps = s->dexps[ch_index];
497 bap = s->bap[ch_index];
498 coeffs = s->fixed_coeffs[ch_index];
499 start = s->start_freq[ch_index];
500 end = s->end_freq[ch_index];
502 for (i = start; i < end; i++) {
506 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
511 gcode = get_bits(gbc, 5);
512 m->b1_mant[0] = b1_mantissas[gcode][0];
513 m->b1_mant[1] = b1_mantissas[gcode][1];
514 m->b1_mant[2] = b1_mantissas[gcode][2];
517 coeffs[i] = m->b1_mant[m->b1ptr++];
522 gcode = get_bits(gbc, 7);
523 m->b2_mant[0] = b2_mantissas[gcode][0];
524 m->b2_mant[1] = b2_mantissas[gcode][1];
525 m->b2_mant[2] = b2_mantissas[gcode][2];
528 coeffs[i] = m->b2_mant[m->b2ptr++];
532 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
537 gcode = get_bits(gbc, 7);
538 m->b4_mant[0] = b4_mantissas[gcode][0];
539 m->b4_mant[1] = b4_mantissas[gcode][1];
542 coeffs[i] = m->b4_mant[m->b4ptr++];
546 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
550 /* asymmetric dequantization */
551 int qlevel = quantization_tab[tbap];
552 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
556 coeffs[i] >>= exps[i];
563 * Remove random dithering from coefficients with zero-bit mantissas
564 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
566 static void remove_dithering(AC3DecodeContext *s) {
572 for(ch=1; ch<=s->fbw_channels; ch++) {
573 if(!s->dither_flag[ch]) {
574 coeffs = s->fixed_coeffs[ch];
576 if(s->channel_in_cpl[ch])
577 end = s->start_freq[CPL_CH];
579 end = s->end_freq[ch];
580 for(i=0; i<end; i++) {
584 if(s->channel_in_cpl[ch]) {
585 bap = s->bap[CPL_CH];
586 for(; i<s->end_freq[CPL_CH]; i++) {
596 * Get the transform coefficients.
598 static int get_transform_coeffs(AC3DecodeContext *s)
604 m.b1ptr = m.b2ptr = m.b4ptr = 3;
606 for (ch = 1; ch <= s->channels; ch++) {
607 /* transform coefficients for full-bandwidth channel */
608 if (get_transform_coeffs_ch(s, ch, &m))
610 /* tranform coefficients for coupling channel come right after the
611 coefficients for the first coupled channel*/
612 if (s->channel_in_cpl[ch]) {
614 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
615 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
618 uncouple_channels(s);
621 end = s->end_freq[CPL_CH];
623 end = s->end_freq[ch];
626 s->transform_coeffs[ch][end] = 0;
630 /* if any channel doesn't use dithering, zero appropriate coefficients */
638 * Stereo rematrixing.
639 * reference: Section 7.5.4 Rematrixing : Decoding Technique
641 static void do_rematrixing(AC3DecodeContext *s)
647 end = FFMIN(s->end_freq[1], s->end_freq[2]);
649 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
650 if(s->rematrixing_flags[bnd]) {
651 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
652 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
653 tmp0 = s->fixed_coeffs[1][i];
654 tmp1 = s->fixed_coeffs[2][i];
655 s->fixed_coeffs[1][i] = tmp0 + tmp1;
656 s->fixed_coeffs[2][i] = tmp0 - tmp1;
663 * Perform the 256-point IMDCT
665 static void do_imdct_256(AC3DecodeContext *s, int chindex)
668 DECLARE_ALIGNED_16(float, x[128]);
670 float *o_ptr = s->tmp_output;
673 /* de-interleave coefficients */
674 for(k=0; k<128; k++) {
675 x[k] = s->transform_coeffs[chindex][2*k+i];
678 /* run standard IMDCT */
679 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
681 /* reverse the post-rotation & reordering from standard IMDCT */
682 for(k=0; k<32; k++) {
683 z[i][32+k].re = -o_ptr[128+2*k];
684 z[i][32+k].im = -o_ptr[2*k];
685 z[i][31-k].re = o_ptr[2*k+1];
686 z[i][31-k].im = o_ptr[128+2*k+1];
690 /* apply AC-3 post-rotation & reordering */
691 for(k=0; k<64; k++) {
692 o_ptr[ 2*k ] = -z[0][ k].im;
693 o_ptr[ 2*k+1] = z[0][63-k].re;
694 o_ptr[128+2*k ] = -z[0][ k].re;
695 o_ptr[128+2*k+1] = z[0][63-k].im;
696 o_ptr[256+2*k ] = -z[1][ k].re;
697 o_ptr[256+2*k+1] = z[1][63-k].im;
698 o_ptr[384+2*k ] = z[1][ k].im;
699 o_ptr[384+2*k+1] = -z[1][63-k].re;
704 * Inverse MDCT Transform.
705 * Convert frequency domain coefficients to time-domain audio samples.
706 * reference: Section 7.9.4 Transformation Equations
708 static inline void do_imdct(AC3DecodeContext *s)
713 /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
714 channels = s->fbw_channels;
715 if(s->output_mode & AC3_OUTPUT_LFEON)
718 for (ch=1; ch<=channels; ch++) {
719 if (s->block_switch[ch]) {
722 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
723 s->transform_coeffs[ch], s->tmp_imdct);
725 /* For the first half of the block, apply the window, add the delay
726 from the previous block, and send to output */
727 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
728 s->window, s->delay[ch-1], 0, 256, 1);
729 /* For the second half of the block, apply the window and store the
730 samples to delay, to be combined with the next block */
731 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
737 * Downmix the output to mono or stereo.
739 static void ac3_downmix(AC3DecodeContext *s)
744 for(i=0; i<256; i++) {
746 for(j=0; j<s->fbw_channels; j++) {
747 v0 += s->output[j][i] * s->downmix_coeffs[j][0];
748 v1 += s->output[j][i] * s->downmix_coeffs[j][1];
750 v0 /= s->downmix_coeff_sum[0];
751 v1 /= s->downmix_coeff_sum[1];
752 if(s->output_mode == AC3_CHMODE_MONO) {
753 s->output[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
754 } else if(s->output_mode == AC3_CHMODE_STEREO) {
755 s->output[0][i] = v0;
756 s->output[1][i] = v1;
762 * Parse an audio block from AC-3 bitstream.
764 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
766 int fbw_channels = s->fbw_channels;
767 int channel_mode = s->channel_mode;
769 GetBitContext *gbc = &s->gbc;
770 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
772 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
774 /* block switch flags */
775 for (ch = 1; ch <= fbw_channels; ch++)
776 s->block_switch[ch] = get_bits1(gbc);
778 /* dithering flags */
780 for (ch = 1; ch <= fbw_channels; ch++) {
781 s->dither_flag[ch] = get_bits1(gbc);
782 if(!s->dither_flag[ch])
787 i = !(s->channel_mode);
790 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
791 s->avctx->drc_scale)+1.0;
792 } else if(blk == 0) {
793 s->dynamic_range[i] = 1.0f;
797 /* coupling strategy */
798 if (get_bits1(gbc)) {
799 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
800 s->cpl_in_use = get_bits1(gbc);
802 /* coupling in use */
803 int cpl_begin_freq, cpl_end_freq;
805 /* determine which channels are coupled */
806 for (ch = 1; ch <= fbw_channels; ch++)
807 s->channel_in_cpl[ch] = get_bits1(gbc);
809 /* phase flags in use */
810 if (channel_mode == AC3_CHMODE_STEREO)
811 s->phase_flags_in_use = get_bits1(gbc);
813 /* coupling frequency range and band structure */
814 cpl_begin_freq = get_bits(gbc, 4);
815 cpl_end_freq = get_bits(gbc, 4);
816 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
817 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
820 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
821 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
822 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
823 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
824 if (get_bits1(gbc)) {
825 s->cpl_band_struct[bnd] = 1;
829 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
831 /* coupling not in use */
832 for (ch = 1; ch <= fbw_channels; ch++)
833 s->channel_in_cpl[ch] = 0;
837 /* coupling coordinates */
839 int cpl_coords_exist = 0;
841 for (ch = 1; ch <= fbw_channels; ch++) {
842 if (s->channel_in_cpl[ch]) {
843 if (get_bits1(gbc)) {
844 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
845 cpl_coords_exist = 1;
846 master_cpl_coord = 3 * get_bits(gbc, 2);
847 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
848 cpl_coord_exp = get_bits(gbc, 4);
849 cpl_coord_mant = get_bits(gbc, 4);
850 if (cpl_coord_exp == 15)
851 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
853 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
854 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
860 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
861 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
862 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
867 /* stereo rematrixing strategy and band structure */
868 if (channel_mode == AC3_CHMODE_STEREO) {
869 if (get_bits1(gbc)) {
870 s->num_rematrixing_bands = 4;
871 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
872 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
873 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
874 s->rematrixing_flags[bnd] = get_bits1(gbc);
878 /* exponent strategies for each channel */
879 s->exp_strategy[CPL_CH] = EXP_REUSE;
880 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
881 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
883 s->exp_strategy[ch] = get_bits(gbc, 1);
885 s->exp_strategy[ch] = get_bits(gbc, 2);
886 if(s->exp_strategy[ch] != EXP_REUSE)
887 bit_alloc_stages[ch] = 3;
890 /* channel bandwidth */
891 for (ch = 1; ch <= fbw_channels; ch++) {
892 s->start_freq[ch] = 0;
893 if (s->exp_strategy[ch] != EXP_REUSE) {
894 int prev = s->end_freq[ch];
895 if (s->channel_in_cpl[ch])
896 s->end_freq[ch] = s->start_freq[CPL_CH];
898 int bandwidth_code = get_bits(gbc, 6);
899 if (bandwidth_code > 60) {
900 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
903 s->end_freq[ch] = bandwidth_code * 3 + 73;
905 if(blk > 0 && s->end_freq[ch] != prev)
906 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
909 s->start_freq[s->lfe_ch] = 0;
910 s->end_freq[s->lfe_ch] = 7;
912 /* decode exponents for each channel */
913 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
914 if (s->exp_strategy[ch] != EXP_REUSE) {
915 int group_size, num_groups;
916 group_size = 3 << (s->exp_strategy[ch] - 1);
918 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
919 else if(ch == s->lfe_ch)
922 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
923 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
924 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
925 &s->dexps[ch][s->start_freq[ch]+!!ch]);
926 if(ch != CPL_CH && ch != s->lfe_ch)
927 skip_bits(gbc, 2); /* skip gainrng */
931 /* bit allocation information */
932 if (get_bits1(gbc)) {
933 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
934 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
935 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
936 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
937 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
938 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
939 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
943 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
944 if (get_bits1(gbc)) {
946 csnr = (get_bits(gbc, 6) - 15) << 4;
947 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
948 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
949 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
951 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
954 /* coupling leak information */
955 if (s->cpl_in_use && get_bits1(gbc)) {
956 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
957 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
958 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
961 /* delta bit allocation information */
962 if (get_bits1(gbc)) {
963 /* delta bit allocation exists (strategy) */
964 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
965 s->dba_mode[ch] = get_bits(gbc, 2);
966 if (s->dba_mode[ch] == DBA_RESERVED) {
967 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
970 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
972 /* channel delta offset, len and bit allocation */
973 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
974 if (s->dba_mode[ch] == DBA_NEW) {
975 s->dba_nsegs[ch] = get_bits(gbc, 3);
976 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
977 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
978 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
979 s->dba_values[ch][seg] = get_bits(gbc, 3);
983 } else if(blk == 0) {
984 for(ch=0; ch<=s->channels; ch++) {
985 s->dba_mode[ch] = DBA_NONE;
990 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
991 if(bit_alloc_stages[ch] > 2) {
992 /* Exponent mapping into PSD and PSD integration */
993 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
994 s->start_freq[ch], s->end_freq[ch],
995 s->psd[ch], s->band_psd[ch]);
997 if(bit_alloc_stages[ch] > 1) {
998 /* Compute excitation function, Compute masking curve, and
999 Apply delta bit allocation */
1000 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1001 s->start_freq[ch], s->end_freq[ch],
1002 s->fast_gain[ch], (ch == s->lfe_ch),
1003 s->dba_mode[ch], s->dba_nsegs[ch],
1004 s->dba_offsets[ch], s->dba_lengths[ch],
1005 s->dba_values[ch], s->mask[ch]);
1007 if(bit_alloc_stages[ch] > 0) {
1008 /* Compute bit allocation */
1009 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1010 s->start_freq[ch], s->end_freq[ch],
1012 s->bit_alloc_params.floor,
1017 /* unused dummy data */
1018 if (get_bits1(gbc)) {
1019 int skipl = get_bits(gbc, 9);
1024 /* unpack the transform coefficients
1025 this also uncouples channels if coupling is in use. */
1026 if (get_transform_coeffs(s)) {
1027 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1031 /* recover coefficients if rematrixing is in use */
1032 if(s->channel_mode == AC3_CHMODE_STEREO)
1035 /* apply scaling to coefficients (headroom, dynrng) */
1036 for(ch=1; ch<=s->channels; ch++) {
1037 float gain = s->mul_bias / 4194304.0f;
1038 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1039 gain *= s->dynamic_range[ch-1];
1041 gain *= s->dynamic_range[0];
1043 for(i=0; i<256; i++) {
1044 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1050 /* downmix output if needed */
1051 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1052 s->fbw_channels == s->out_channels)) {
1056 /* convert float to 16-bit integer */
1057 for(ch=0; ch<s->out_channels; ch++) {
1058 for(i=0; i<256; i++) {
1059 s->output[ch][i] += s->add_bias;
1061 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1068 * Decode a single AC-3 frame.
1070 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
1072 AC3DecodeContext *s = avctx->priv_data;
1073 int16_t *out_samples = (int16_t *)data;
1074 int i, blk, ch, err;
1076 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1077 init_get_bits(&s->gbc, buf, buf_size * 8);
1079 /* parse the syncinfo */
1080 err = ac3_parse_header(s);
1083 case AC3_PARSE_ERROR_SYNC:
1084 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1086 case AC3_PARSE_ERROR_BSID:
1087 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1089 case AC3_PARSE_ERROR_SAMPLE_RATE:
1090 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1092 case AC3_PARSE_ERROR_FRAME_SIZE:
1093 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1096 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1102 /* check that reported frame size fits in input buffer */
1103 if(s->frame_size > buf_size) {
1104 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1108 /* check for crc mismatch */
1109 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1110 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1111 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1114 /* TODO: error concealment */
1117 avctx->sample_rate = s->sample_rate;
1118 avctx->bit_rate = s->bit_rate;
1120 /* channel config */
1121 s->out_channels = s->channels;
1122 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1123 avctx->request_channels < s->channels) {
1124 s->out_channels = avctx->request_channels;
1125 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1127 avctx->channels = s->out_channels;
1129 /* set downmixing coefficients if needed */
1130 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1131 s->fbw_channels == s->out_channels)) {
1132 set_downmix_coeffs(s);
1135 /* parse the audio blocks */
1136 for (blk = 0; blk < NB_BLOCKS; blk++) {
1137 if (ac3_parse_audio_block(s, blk)) {
1138 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1140 return s->frame_size;
1142 for (i = 0; i < 256; i++)
1143 for (ch = 0; ch < s->out_channels; ch++)
1144 *(out_samples++) = s->int_output[ch][i];
1146 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1147 return s->frame_size;
1151 * Uninitialize the AC-3 decoder.
1153 static int ac3_decode_end(AVCodecContext *avctx)
1155 AC3DecodeContext *s = avctx->priv_data;
1156 ff_mdct_end(&s->imdct_512);
1157 ff_mdct_end(&s->imdct_256);
1162 AVCodec ac3_decoder = {
1164 .type = CODEC_TYPE_AUDIO,
1166 .priv_data_size = sizeof (AC3DecodeContext),
1167 .init = ac3_decode_init,
1168 .close = ac3_decode_end,
1169 .decode = ac3_decode_frame,