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 num_rematrixing_bands; ///< number of rematrixing bands
139 int rematrixing_flags[4]; ///< rematrixing flags
140 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
141 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
142 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
143 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
144 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
145 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
146 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
147 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
149 int sample_rate; ///< sample frequency, in Hz
150 int bit_rate; ///< stream bit rate, in bits-per-second
151 int frame_size; ///< current frame size, in bytes
153 int channels; ///< number of total channels
154 int fbw_channels; ///< number of full-bandwidth channels
155 int lfe_on; ///< lfe channel in use
156 int lfe_ch; ///< index of LFE channel
157 int output_mode; ///< output channel configuration
158 int out_channels; ///< number of output channels
160 int center_mix_level; ///< Center mix level index
161 int surround_mix_level; ///< Surround mix level index
162 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
163 float dynamic_range[2]; ///< dynamic range
164 float cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
165 int num_cpl_bands; ///< number of coupling bands
166 int num_cpl_subbands; ///< number of coupling sub bands
167 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
168 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
169 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
171 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
172 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
173 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
174 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
175 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
177 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
180 MDCTContext imdct_512; ///< for 512 sample IMDCT
181 MDCTContext imdct_256; ///< for 256 sample IMDCT
182 DSPContext dsp; ///< for optimization
183 float add_bias; ///< offset for float_to_int16 conversion
184 float mul_bias; ///< scaling for float_to_int16 conversion
186 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]); ///< output after imdct transform and windowing
187 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
188 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]); ///< delay - added to the next block
189 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
190 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
191 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
194 GetBitContext gbc; ///< bitstream reader
195 AVRandomState dith_state; ///< for dither generation
196 AVCodecContext *avctx; ///< parent context
200 * Generate a Kaiser-Bessel Derived Window.
202 static void ac3_window_init(float *window)
205 double sum = 0.0, bessel, tmp;
206 double local_window[256];
207 double alpha2 = (5.0 * M_PI / 256.0) * (5.0 * M_PI / 256.0);
209 for (i = 0; i < 256; i++) {
210 tmp = i * (256 - i) * alpha2;
212 for (j = 100; j > 0; j--) /* default to 100 iterations */
213 bessel = bessel * tmp / (j * j) + 1;
215 local_window[i] = sum;
219 for (i = 0; i < 256; i++)
220 window[i] = sqrt(local_window[i] / sum);
224 * Symmetrical Dequantization
225 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
226 * Tables 7.19 to 7.23
229 symmetric_dequant(int code, int levels)
231 return (code - (levels >> 1)) * (2.0f / levels);
235 * Initialize tables at runtime.
237 static void ac3_tables_init(void)
241 /* generate grouped mantissa tables
242 reference: Section 7.3.5 Ungrouping of Mantissas */
243 for(i=0; i<32; i++) {
244 /* bap=1 mantissas */
245 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
246 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
247 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
249 for(i=0; i<128; i++) {
250 /* bap=2 mantissas */
251 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
252 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
253 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
255 /* bap=4 mantissas */
256 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
257 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
259 /* generate ungrouped mantissa tables
260 reference: Tables 7.21 and 7.23 */
262 /* bap=3 mantissas */
263 b3_mantissas[i] = symmetric_dequant(i, 7);
265 for(i=0; i<15; i++) {
266 /* bap=5 mantissas */
267 b5_mantissas[i] = symmetric_dequant(i, 15);
270 /* generate dynamic range table
271 reference: Section 7.7.1 Dynamic Range Control */
272 for(i=0; i<256; i++) {
273 int v = (i >> 5) - ((i >> 7) << 3) - 5;
274 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
277 /* generate scale factors for exponents and asymmetrical dequantization
278 reference: Section 7.3.2 Expansion of Mantissas for Asymmetric Quantization */
279 for (i = 0; i < 25; i++)
280 scale_factors[i] = pow(2.0, -i);
282 /* generate exponent tables
283 reference: Section 7.1.3 Exponent Decoding */
284 for(i=0; i<128; i++) {
285 exp_ungroup_tab[i][0] = i / 25;
286 exp_ungroup_tab[i][1] = (i % 25) / 5;
287 exp_ungroup_tab[i][2] = (i % 25) % 5;
293 * AVCodec initialization
295 static int ac3_decode_init(AVCodecContext *avctx)
297 AC3DecodeContext *s = avctx->priv_data;
302 ff_mdct_init(&s->imdct_256, 8, 1);
303 ff_mdct_init(&s->imdct_512, 9, 1);
304 ac3_window_init(s->window);
305 dsputil_init(&s->dsp, avctx);
306 av_init_random(0, &s->dith_state);
308 /* set bias values for float to int16 conversion */
309 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
310 s->add_bias = 385.0f;
314 s->mul_bias = 32767.0f;
317 /* allow downmixing to stereo or mono */
318 if (avctx->channels > 0 && avctx->request_channels > 0 &&
319 avctx->request_channels < avctx->channels &&
320 avctx->request_channels <= 2) {
321 avctx->channels = avctx->request_channels;
328 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
329 * GetBitContext within AC3DecodeContext must point to
330 * start of the synchronized ac3 bitstream.
332 static int ac3_parse_header(AC3DecodeContext *s)
335 GetBitContext *gbc = &s->gbc;
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 /* set default mix levels */
364 s->center_mix_level = 3; // -4.5dB
365 s->surround_mix_level = 4; // -6.0dB
367 /* skip over portion of header which has already been read */
368 skip_bits(gbc, 16); // skip the sync_word
369 skip_bits(gbc, 16); // skip crc1
370 skip_bits(gbc, 8); // skip fscod and frmsizecod
371 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
372 if(s->channel_mode == AC3_CHMODE_STEREO) {
373 skip_bits(gbc, 2); // skip dsurmod
375 if((s->channel_mode & 1) && s->channel_mode != AC3_CHMODE_MONO)
376 s->center_mix_level = center_levels[get_bits(gbc, 2)];
377 if(s->channel_mode & 4)
378 s->surround_mix_level = surround_levels[get_bits(gbc, 2)];
380 skip_bits1(gbc); // skip lfeon
382 /* read the rest of the bsi. read twice for dual mono mode. */
383 i = !(s->channel_mode);
385 skip_bits(gbc, 5); // skip dialog normalization
387 skip_bits(gbc, 8); //skip compression
389 skip_bits(gbc, 8); //skip language code
391 skip_bits(gbc, 7); //skip audio production information
394 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
396 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
397 TODO: read & use the xbsi1 downmix levels */
399 skip_bits(gbc, 14); //skip timecode1 / xbsi1
401 skip_bits(gbc, 14); //skip timecode2 / xbsi2
403 /* skip additional bitstream info */
404 if (get_bits1(gbc)) {
405 i = get_bits(gbc, 6);
415 * Set stereo downmixing coefficients based on frame header info.
416 * reference: Section 7.8.2 Downmixing Into Two Channels
418 static void set_downmix_coeffs(AC3DecodeContext *s)
421 float cmix = gain_levels[s->center_mix_level];
422 float smix = gain_levels[s->surround_mix_level];
424 for(i=0; i<s->fbw_channels; i++) {
425 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
426 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
428 if(s->channel_mode > 1 && s->channel_mode & 1) {
429 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
431 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
432 int nf = s->channel_mode - 2;
433 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
435 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
436 int nf = s->channel_mode - 4;
437 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
442 * Decode the grouped exponents according to exponent strategy.
443 * reference: Section 7.1.3 Exponent Decoding
445 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
446 uint8_t absexp, int8_t *dexps)
448 int i, j, grp, group_size;
453 group_size = exp_strategy + (exp_strategy == EXP_D45);
454 for(grp=0,i=0; grp<ngrps; grp++) {
455 expacc = get_bits(gbc, 7);
456 dexp[i++] = exp_ungroup_tab[expacc][0];
457 dexp[i++] = exp_ungroup_tab[expacc][1];
458 dexp[i++] = exp_ungroup_tab[expacc][2];
461 /* convert to absolute exps and expand groups */
463 for(i=0; i<ngrps*3; i++) {
464 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
465 for(j=0; j<group_size; j++) {
466 dexps[(i*group_size)+j] = prevexp;
472 * Generate transform coefficients for each coupled channel in the coupling
473 * range using the coupling coefficients and coupling coordinates.
474 * reference: Section 7.4.3 Coupling Coordinate Format
476 static void uncouple_channels(AC3DecodeContext *s)
478 int i, j, ch, bnd, subbnd;
481 i = s->start_freq[CPL_CH];
482 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
485 for(j=0; j<12; j++) {
486 for(ch=1; ch<=s->fbw_channels; ch++) {
487 if(s->channel_in_cpl[ch]) {
488 s->transform_coeffs[ch][i] = s->transform_coeffs[CPL_CH][i] * s->cpl_coords[ch][bnd] * 8.0f;
489 if (ch == 2 && s->phase_flags[bnd])
490 s->transform_coeffs[ch][i] = -s->transform_coeffs[ch][i];
495 } while(s->cpl_band_struct[subbnd]);
500 * Grouped mantissas for 3-level 5-level and 11-level quantization
512 * Get the transform coefficients for a particular channel
513 * reference: Section 7.3 Quantization and Decoding of Mantissas
515 static int get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
517 GetBitContext *gbc = &s->gbc;
518 int i, gcode, tbap, start, end;
523 exps = s->dexps[ch_index];
524 bap = s->bap[ch_index];
525 coeffs = s->transform_coeffs[ch_index];
526 start = s->start_freq[ch_index];
527 end = s->end_freq[ch_index];
529 for (i = start; i < end; i++) {
533 coeffs[i] = ((av_random(&s->dith_state) & 0xFFFF) / 65535.0f) - 0.5f;
538 gcode = get_bits(gbc, 5);
539 m->b1_mant[0] = b1_mantissas[gcode][0];
540 m->b1_mant[1] = b1_mantissas[gcode][1];
541 m->b1_mant[2] = b1_mantissas[gcode][2];
544 coeffs[i] = m->b1_mant[m->b1ptr++];
549 gcode = get_bits(gbc, 7);
550 m->b2_mant[0] = b2_mantissas[gcode][0];
551 m->b2_mant[1] = b2_mantissas[gcode][1];
552 m->b2_mant[2] = b2_mantissas[gcode][2];
555 coeffs[i] = m->b2_mant[m->b2ptr++];
559 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
564 gcode = get_bits(gbc, 7);
565 m->b4_mant[0] = b4_mantissas[gcode][0];
566 m->b4_mant[1] = b4_mantissas[gcode][1];
569 coeffs[i] = m->b4_mant[m->b4ptr++];
573 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
577 /* asymmetric dequantization */
578 coeffs[i] = get_sbits(gbc, quantization_tab[tbap]) * scale_factors[quantization_tab[tbap]-1];
581 coeffs[i] *= scale_factors[exps[i]];
588 * Remove random dithering from coefficients with zero-bit mantissas
589 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
591 static void remove_dithering(AC3DecodeContext *s) {
597 for(ch=1; ch<=s->fbw_channels; ch++) {
598 if(!s->dither_flag[ch]) {
599 coeffs = s->transform_coeffs[ch];
601 if(s->channel_in_cpl[ch])
602 end = s->start_freq[CPL_CH];
604 end = s->end_freq[ch];
605 for(i=0; i<end; i++) {
609 if(s->channel_in_cpl[ch]) {
610 bap = s->bap[CPL_CH];
611 for(; i<s->end_freq[CPL_CH]; i++) {
621 * Get the transform coefficients.
623 static int get_transform_coeffs(AC3DecodeContext *s)
629 m.b1ptr = m.b2ptr = m.b4ptr = 3;
631 for (ch = 1; ch <= s->channels; ch++) {
632 /* transform coefficients for full-bandwidth channel */
633 if (get_transform_coeffs_ch(s, ch, &m))
635 /* tranform coefficients for coupling channel come right after the
636 coefficients for the first coupled channel*/
637 if (s->channel_in_cpl[ch]) {
639 if (get_transform_coeffs_ch(s, CPL_CH, &m)) {
640 av_log(s->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
643 uncouple_channels(s);
646 end = s->end_freq[CPL_CH];
648 end = s->end_freq[ch];
651 s->transform_coeffs[ch][end] = 0;
655 /* if any channel doesn't use dithering, zero appropriate coefficients */
663 * Stereo rematrixing.
664 * reference: Section 7.5.4 Rematrixing : Decoding Technique
666 static void do_rematrixing(AC3DecodeContext *s)
672 end = FFMIN(s->end_freq[1], s->end_freq[2]);
674 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
675 if(s->rematrixing_flags[bnd]) {
676 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
677 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
678 tmp0 = s->transform_coeffs[1][i];
679 tmp1 = s->transform_coeffs[2][i];
680 s->transform_coeffs[1][i] = tmp0 + tmp1;
681 s->transform_coeffs[2][i] = tmp0 - tmp1;
688 * Perform the 256-point IMDCT
690 static void do_imdct_256(AC3DecodeContext *s, int chindex)
693 DECLARE_ALIGNED_16(float, x[128]);
695 float *o_ptr = s->tmp_output;
698 /* de-interleave coefficients */
699 for(k=0; k<128; k++) {
700 x[k] = s->transform_coeffs[chindex][2*k+i];
703 /* run standard IMDCT */
704 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
706 /* reverse the post-rotation & reordering from standard IMDCT */
707 for(k=0; k<32; k++) {
708 z[i][32+k].re = -o_ptr[128+2*k];
709 z[i][32+k].im = -o_ptr[2*k];
710 z[i][31-k].re = o_ptr[2*k+1];
711 z[i][31-k].im = o_ptr[128+2*k+1];
715 /* apply AC-3 post-rotation & reordering */
716 for(k=0; k<64; k++) {
717 o_ptr[ 2*k ] = -z[0][ k].im;
718 o_ptr[ 2*k+1] = z[0][63-k].re;
719 o_ptr[128+2*k ] = -z[0][ k].re;
720 o_ptr[128+2*k+1] = z[0][63-k].im;
721 o_ptr[256+2*k ] = -z[1][ k].re;
722 o_ptr[256+2*k+1] = z[1][63-k].im;
723 o_ptr[384+2*k ] = z[1][ k].im;
724 o_ptr[384+2*k+1] = -z[1][63-k].re;
729 * Inverse MDCT Transform.
730 * Convert frequency domain coefficients to time-domain audio samples.
731 * reference: Section 7.9.4 Transformation Equations
733 static inline void do_imdct(AC3DecodeContext *s)
738 /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
739 channels = s->fbw_channels;
740 if(s->output_mode & AC3_OUTPUT_LFEON)
743 for (ch=1; ch<=channels; ch++) {
744 if (s->block_switch[ch]) {
747 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
748 s->transform_coeffs[ch], s->tmp_imdct);
750 /* For the first half of the block, apply the window, add the delay
751 from the previous block, and send to output */
752 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
753 s->window, s->delay[ch-1], 0, 256, 1);
754 /* For the second half of the block, apply the window and store the
755 samples to delay, to be combined with the next block */
756 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
762 * Downmix the output to mono or stereo.
764 static void ac3_downmix(AC3DecodeContext *s)
767 float v0, v1, s0, s1;
769 for(i=0; i<256; i++) {
770 v0 = v1 = s0 = s1 = 0.0f;
771 for(j=0; j<s->fbw_channels; j++) {
772 v0 += s->output[j][i] * s->downmix_coeffs[j][0];
773 v1 += s->output[j][i] * s->downmix_coeffs[j][1];
774 s0 += s->downmix_coeffs[j][0];
775 s1 += s->downmix_coeffs[j][1];
779 if(s->output_mode == AC3_CHMODE_MONO) {
780 s->output[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
781 } else if(s->output_mode == AC3_CHMODE_STEREO) {
782 s->output[0][i] = v0;
783 s->output[1][i] = v1;
789 * Parse an audio block from AC-3 bitstream.
791 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
793 int fbw_channels = s->fbw_channels;
794 int channel_mode = s->channel_mode;
796 GetBitContext *gbc = &s->gbc;
797 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
799 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
801 /* block switch flags */
802 for (ch = 1; ch <= fbw_channels; ch++)
803 s->block_switch[ch] = get_bits1(gbc);
805 /* dithering flags */
807 for (ch = 1; ch <= fbw_channels; ch++) {
808 s->dither_flag[ch] = get_bits1(gbc);
809 if(!s->dither_flag[ch])
814 i = !(s->channel_mode);
817 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
818 s->avctx->drc_scale)+1.0;
819 } else if(blk == 0) {
820 s->dynamic_range[i] = 1.0f;
824 /* coupling strategy */
825 if (get_bits1(gbc)) {
826 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
827 s->cpl_in_use = get_bits1(gbc);
829 /* coupling in use */
830 int cpl_begin_freq, cpl_end_freq;
832 /* determine which channels are coupled */
833 for (ch = 1; ch <= fbw_channels; ch++)
834 s->channel_in_cpl[ch] = get_bits1(gbc);
836 /* phase flags in use */
837 if (channel_mode == AC3_CHMODE_STEREO)
838 s->phase_flags_in_use = get_bits1(gbc);
840 /* coupling frequency range and band structure */
841 cpl_begin_freq = get_bits(gbc, 4);
842 cpl_end_freq = get_bits(gbc, 4);
843 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
844 av_log(s->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
847 s->num_cpl_bands = s->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
848 s->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
849 s->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
850 for (bnd = 0; bnd < s->num_cpl_subbands - 1; bnd++) {
851 if (get_bits1(gbc)) {
852 s->cpl_band_struct[bnd] = 1;
856 s->cpl_band_struct[s->num_cpl_subbands-1] = 0;
858 /* coupling not in use */
859 for (ch = 1; ch <= fbw_channels; ch++)
860 s->channel_in_cpl[ch] = 0;
864 /* coupling coordinates */
866 int cpl_coords_exist = 0;
868 for (ch = 1; ch <= fbw_channels; ch++) {
869 if (s->channel_in_cpl[ch]) {
870 if (get_bits1(gbc)) {
871 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
872 cpl_coords_exist = 1;
873 master_cpl_coord = 3 * get_bits(gbc, 2);
874 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
875 cpl_coord_exp = get_bits(gbc, 4);
876 cpl_coord_mant = get_bits(gbc, 4);
877 if (cpl_coord_exp == 15)
878 s->cpl_coords[ch][bnd] = cpl_coord_mant / 16.0f;
880 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16.0f) / 32.0f;
881 s->cpl_coords[ch][bnd] *= scale_factors[cpl_coord_exp + master_cpl_coord];
887 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
888 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
889 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
894 /* stereo rematrixing strategy and band structure */
895 if (channel_mode == AC3_CHMODE_STEREO) {
896 if (get_bits1(gbc)) {
897 s->num_rematrixing_bands = 4;
898 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
899 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
900 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
901 s->rematrixing_flags[bnd] = get_bits1(gbc);
905 /* exponent strategies for each channel */
906 s->exp_strategy[CPL_CH] = EXP_REUSE;
907 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
908 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
910 s->exp_strategy[ch] = get_bits(gbc, 1);
912 s->exp_strategy[ch] = get_bits(gbc, 2);
913 if(s->exp_strategy[ch] != EXP_REUSE)
914 bit_alloc_stages[ch] = 3;
917 /* channel bandwidth */
918 for (ch = 1; ch <= fbw_channels; ch++) {
919 s->start_freq[ch] = 0;
920 if (s->exp_strategy[ch] != EXP_REUSE) {
921 int prev = s->end_freq[ch];
922 if (s->channel_in_cpl[ch])
923 s->end_freq[ch] = s->start_freq[CPL_CH];
925 int bandwidth_code = get_bits(gbc, 6);
926 if (bandwidth_code > 60) {
927 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
930 s->end_freq[ch] = bandwidth_code * 3 + 73;
932 if(blk > 0 && s->end_freq[ch] != prev)
933 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
936 s->start_freq[s->lfe_ch] = 0;
937 s->end_freq[s->lfe_ch] = 7;
939 /* decode exponents for each channel */
940 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
941 if (s->exp_strategy[ch] != EXP_REUSE) {
942 int group_size, num_groups;
943 group_size = 3 << (s->exp_strategy[ch] - 1);
945 num_groups = (s->end_freq[ch] - s->start_freq[ch]) / group_size;
946 else if(ch == s->lfe_ch)
949 num_groups = (s->end_freq[ch] + group_size - 4) / group_size;
950 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
951 decode_exponents(gbc, s->exp_strategy[ch], num_groups, s->dexps[ch][0],
952 &s->dexps[ch][s->start_freq[ch]+!!ch]);
953 if(ch != CPL_CH && ch != s->lfe_ch)
954 skip_bits(gbc, 2); /* skip gainrng */
958 /* bit allocation information */
959 if (get_bits1(gbc)) {
960 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
961 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
962 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
963 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
964 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
965 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
966 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
970 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
971 if (get_bits1(gbc)) {
973 csnr = (get_bits(gbc, 6) - 15) << 4;
974 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
975 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
976 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
978 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
981 /* coupling leak information */
982 if (s->cpl_in_use && get_bits1(gbc)) {
983 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
984 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
985 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
988 /* delta bit allocation information */
989 if (get_bits1(gbc)) {
990 /* delta bit allocation exists (strategy) */
991 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
992 s->dba_mode[ch] = get_bits(gbc, 2);
993 if (s->dba_mode[ch] == DBA_RESERVED) {
994 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
997 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
999 /* channel delta offset, len and bit allocation */
1000 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1001 if (s->dba_mode[ch] == DBA_NEW) {
1002 s->dba_nsegs[ch] = get_bits(gbc, 3);
1003 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1004 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1005 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1006 s->dba_values[ch][seg] = get_bits(gbc, 3);
1010 } else if(blk == 0) {
1011 for(ch=0; ch<=s->channels; ch++) {
1012 s->dba_mode[ch] = DBA_NONE;
1016 /* Bit allocation */
1017 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1018 if(bit_alloc_stages[ch] > 2) {
1019 /* Exponent mapping into PSD and PSD integration */
1020 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1021 s->start_freq[ch], s->end_freq[ch],
1022 s->psd[ch], s->band_psd[ch]);
1024 if(bit_alloc_stages[ch] > 1) {
1025 /* Compute excitation function, Compute masking curve, and
1026 Apply delta bit allocation */
1027 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1028 s->start_freq[ch], s->end_freq[ch],
1029 s->fast_gain[ch], (ch == s->lfe_ch),
1030 s->dba_mode[ch], s->dba_nsegs[ch],
1031 s->dba_offsets[ch], s->dba_lengths[ch],
1032 s->dba_values[ch], s->mask[ch]);
1034 if(bit_alloc_stages[ch] > 0) {
1035 /* Compute bit allocation */
1036 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1037 s->start_freq[ch], s->end_freq[ch],
1039 s->bit_alloc_params.floor,
1044 /* unused dummy data */
1045 if (get_bits1(gbc)) {
1046 int skipl = get_bits(gbc, 9);
1051 /* unpack the transform coefficients
1052 this also uncouples channels if coupling is in use. */
1053 if (get_transform_coeffs(s)) {
1054 av_log(s->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1058 /* recover coefficients if rematrixing is in use */
1059 if(s->channel_mode == AC3_CHMODE_STEREO)
1062 /* apply scaling to coefficients (headroom, dynrng) */
1063 for(ch=1; ch<=s->channels; ch++) {
1064 float gain = 2.0f * s->mul_bias;
1065 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1066 gain *= s->dynamic_range[ch-1];
1068 gain *= s->dynamic_range[0];
1070 for(i=0; i<s->end_freq[ch]; i++) {
1071 s->transform_coeffs[ch][i] *= gain;
1077 /* downmix output if needed */
1078 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1079 s->fbw_channels == s->out_channels)) {
1083 /* convert float to 16-bit integer */
1084 for(ch=0; ch<s->out_channels; ch++) {
1085 for(i=0; i<256; i++) {
1086 s->output[ch][i] += s->add_bias;
1088 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1095 * Decode a single AC-3 frame.
1097 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
1099 AC3DecodeContext *s = avctx->priv_data;
1100 int16_t *out_samples = (int16_t *)data;
1101 int i, blk, ch, err;
1103 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1104 init_get_bits(&s->gbc, buf, buf_size * 8);
1106 /* parse the syncinfo */
1107 err = ac3_parse_header(s);
1110 case AC3_PARSE_ERROR_SYNC:
1111 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1113 case AC3_PARSE_ERROR_BSID:
1114 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1116 case AC3_PARSE_ERROR_SAMPLE_RATE:
1117 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1119 case AC3_PARSE_ERROR_FRAME_SIZE:
1120 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1123 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1129 /* check that reported frame size fits in input buffer */
1130 if(s->frame_size > buf_size) {
1131 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1135 /* check for crc mismatch */
1136 if(avctx->error_resilience >= FF_ER_CAREFUL) {
1137 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1138 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1141 /* TODO: error concealment */
1144 avctx->sample_rate = s->sample_rate;
1145 avctx->bit_rate = s->bit_rate;
1147 /* channel config */
1148 s->out_channels = s->channels;
1149 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1150 avctx->request_channels < s->channels) {
1151 s->out_channels = avctx->request_channels;
1152 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1154 avctx->channels = s->out_channels;
1156 /* set downmixing coefficients if needed */
1157 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1158 s->fbw_channels == s->out_channels)) {
1159 set_downmix_coeffs(s);
1162 /* parse the audio blocks */
1163 for (blk = 0; blk < NB_BLOCKS; blk++) {
1164 if (ac3_parse_audio_block(s, blk)) {
1165 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1167 return s->frame_size;
1169 for (i = 0; i < 256; i++)
1170 for (ch = 0; ch < s->out_channels; ch++)
1171 *(out_samples++) = s->int_output[ch][i];
1173 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1174 return s->frame_size;
1178 * Uninitialize the AC-3 decoder.
1180 static int ac3_decode_end(AVCodecContext *avctx)
1182 AC3DecodeContext *s = avctx->priv_data;
1183 ff_mdct_end(&s->imdct_512);
1184 ff_mdct_end(&s->imdct_256);
1189 AVCodec ac3_decoder = {
1191 .type = CODEC_TYPE_AUDIO,
1193 .priv_data_size = sizeof (AC3DecodeContext),
1194 .init = ac3_decode_init,
1195 .close = ac3_decode_end,
1196 .decode = ac3_decode_frame,