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"
42 * Table of bin locations for rematrixing bands
43 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
45 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
48 * table for exponent to scale_factor mapping
49 * scale_factors[i] = 2 ^ -i
51 static float scale_factors[25];
53 /** table for grouping exponents */
54 static uint8_t exp_ungroup_tab[128][3];
57 /** tables for ungrouping mantissas */
58 static float b1_mantissas[32][3];
59 static float b2_mantissas[128][3];
60 static float b3_mantissas[8];
61 static float b4_mantissas[128][2];
62 static float b5_mantissas[16];
65 * Quantization table: levels for symmetric. bits for asymmetric.
66 * reference: Table 7.18 Mapping of bap to Quantizer
68 static const uint8_t quantization_tab[16] = {
70 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
73 /** dynamic range table. converts codes to scale factors. */
74 static float dynamic_range_tab[256];
76 /** Adjustments in dB gain */
77 #define LEVEL_MINUS_3DB 0.7071067811865476
78 #define LEVEL_MINUS_4POINT5DB 0.5946035575013605
79 #define LEVEL_MINUS_6DB 0.5000000000000000
80 #define LEVEL_MINUS_9DB 0.3535533905932738
81 #define LEVEL_ZERO 0.0000000000000000
82 #define LEVEL_ONE 1.0000000000000000
84 static const float gain_levels[6] = {
88 LEVEL_MINUS_4POINT5DB,
94 * Table for center mix levels
95 * reference: Section 5.4.2.4 cmixlev
97 static const uint8_t center_levels[4] = { 2, 3, 4, 3 };
100 * Table for surround mix levels
101 * reference: Section 5.4.2.5 surmixlev
103 static const uint8_t surround_levels[4] = { 2, 4, 0, 4 };
106 * Table for default stereo downmixing coefficients
107 * reference: Section 7.8.2 Downmixing Into Two Channels
109 static const uint8_t ac3_default_coeffs[8][5][2] = {
110 { { 1, 0 }, { 0, 1 }, },
112 { { 1, 0 }, { 0, 1 }, },
113 { { 1, 0 }, { 3, 3 }, { 0, 1 }, },
114 { { 1, 0 }, { 0, 1 }, { 4, 4 }, },
115 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 5, 5 }, },
116 { { 1, 0 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
117 { { 1, 0 }, { 3, 3 }, { 0, 1 }, { 4, 0 }, { 0, 4 }, },
120 /* override ac3.h to include coupling channel */
121 #undef AC3_MAX_CHANNELS
122 #define AC3_MAX_CHANNELS 7
125 #define AC3_OUTPUT_LFEON 8
128 int channel_mode; ///< channel mode (acmod)
129 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
130 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
131 int dither_all; ///< true if all channels are dithered
132 int cpl_in_use; ///< coupling in use
133 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
134 int phase_flags_in_use; ///< phase flags in use
135 int cpl_band_struct[18]; ///< coupling band structure
136 int rematrixing_strategy; ///< rematrixing strategy
137 int num_rematrixing_bands; ///< number of rematrixing bands
138 int rematrixing_flags[4]; ///< rematrixing flags
139 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
140 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
141 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
142 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
143 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
144 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
145 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
146 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
148 int sampling_rate; ///< sample frequency, in Hz
149 int bit_rate; ///< stream bit rate, in bits-per-second
150 int frame_size; ///< current frame size, in bytes
152 int channels; ///< number of total channels
153 int fbw_channels; ///< number of full-bandwidth channels
154 int lfe_on; ///< lfe channel in use
155 int lfe_ch; ///< index of LFE channel
156 int output_mode; ///< output channel configuration
157 int out_channels; ///< number of output channels
159 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
160 float dynamic_range[2]; ///< dynamic range
161 float cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
162 int num_cpl_bands; ///< number of coupling bands
163 int num_cpl_subbands; ///< number of coupling sub bands
164 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
165 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
166 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
168 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
169 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
170 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
171 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
172 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
174 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
177 MDCTContext imdct_512; ///< for 512 sample IMDCT
178 MDCTContext imdct_256; ///< for 256 sample IMDCT
179 DSPContext dsp; ///< for optimization
180 float add_bias; ///< offset for float_to_int16 conversion
181 float mul_bias; ///< scaling for float_to_int16 conversion
183 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS-1][256]); ///< output after imdct transform and windowing
184 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
185 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS-1][256]); ///< delay - added to the next block
186 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
187 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
188 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
191 GetBitContext gbc; ///< bitstream reader
192 AVRandomState dith_state; ///< for dither generation
193 AVCodecContext *avctx; ///< parent context
197 * Generate a Kaiser-Bessel Derived Window.
199 static void ac3_window_init(float *window)
202 double sum = 0.0, bessel, tmp;
203 double local_window[256];
204 double alpha2 = (5.0 * M_PI / 256.0) * (5.0 * M_PI / 256.0);
206 for (i = 0; i < 256; i++) {
207 tmp = i * (256 - i) * alpha2;
209 for (j = 100; j > 0; j--) /* default to 100 iterations */
210 bessel = bessel * tmp / (j * j) + 1;
212 local_window[i] = sum;
216 for (i = 0; i < 256; i++)
217 window[i] = sqrt(local_window[i] / sum);
221 * Symmetrical Dequantization
222 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
223 * Tables 7.19 to 7.23
226 symmetric_dequant(int code, int levels)
228 return (code - (levels >> 1)) * (2.0f / levels);
232 * Initialize tables at runtime.
234 static void ac3_tables_init(void)
238 /* generate grouped mantissa tables
239 reference: Section 7.3.5 Ungrouping of Mantissas */
240 for(i=0; i<32; i++) {
241 /* bap=1 mantissas */
242 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
243 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
244 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
246 for(i=0; i<128; i++) {
247 /* bap=2 mantissas */
248 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
249 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
250 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
252 /* bap=4 mantissas */
253 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
254 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
256 /* generate ungrouped mantissa tables
257 reference: Tables 7.21 and 7.23 */
259 /* bap=3 mantissas */
260 b3_mantissas[i] = symmetric_dequant(i, 7);
262 for(i=0; i<15; i++) {
263 /* bap=5 mantissas */
264 b5_mantissas[i] = symmetric_dequant(i, 15);
267 /* generate dynamic range table
268 reference: Section 7.7.1 Dynamic Range Control */
269 for(i=0; i<256; i++) {
270 int v = (i >> 5) - ((i >> 7) << 3) - 5;
271 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
274 /* generate scale factors for exponents and asymmetrical dequantization
275 reference: Section 7.3.2 Expansion of Mantissas for Asymmetric Quantization */
276 for (i = 0; i < 25; i++)
277 scale_factors[i] = pow(2.0, -i);
279 /* generate exponent tables
280 reference: Section 7.1.3 Exponent Decoding */
281 for(i=0; i<128; i++) {
282 exp_ungroup_tab[i][0] = i / 25;
283 exp_ungroup_tab[i][1] = (i % 25) / 5;
284 exp_ungroup_tab[i][2] = (i % 25) % 5;
290 * AVCodec initialization
292 static int ac3_decode_init(AVCodecContext *avctx)
294 AC3DecodeContext *ctx = avctx->priv_data;
299 ff_mdct_init(&ctx->imdct_256, 8, 1);
300 ff_mdct_init(&ctx->imdct_512, 9, 1);
301 ac3_window_init(ctx->window);
302 dsputil_init(&ctx->dsp, avctx);
303 av_init_random(0, &ctx->dith_state);
305 /* set bias values for float to int16 conversion */
306 if(ctx->dsp.float_to_int16 == ff_float_to_int16_c) {
307 ctx->add_bias = 385.0f;
308 ctx->mul_bias = 1.0f;
310 ctx->add_bias = 0.0f;
311 ctx->mul_bias = 32767.0f;
318 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
319 * GetBitContext within AC3DecodeContext must point to
320 * start of the synchronized ac3 bitstream.
322 static int ac3_parse_header(AC3DecodeContext *ctx)
325 GetBitContext *gbc = &ctx->gbc;
326 float center_mix_level, surround_mix_level;
329 err = ff_ac3_parse_header(gbc->buffer, &hdr);
333 /* get decoding parameters from header info */
334 ctx->bit_alloc_params.sr_code = hdr.sr_code;
335 ctx->channel_mode = hdr.channel_mode;
336 center_mix_level = gain_levels[center_levels[hdr.center_mix_level]];
337 surround_mix_level = gain_levels[surround_levels[hdr.surround_mix_level]];
338 ctx->lfe_on = hdr.lfe_on;
339 ctx->bit_alloc_params.sr_shift = hdr.sr_shift;
340 ctx->sampling_rate = hdr.sample_rate;
341 ctx->bit_rate = hdr.bit_rate;
342 ctx->channels = hdr.channels;
343 ctx->fbw_channels = ctx->channels - ctx->lfe_on;
344 ctx->lfe_ch = ctx->fbw_channels + 1;
345 ctx->frame_size = hdr.frame_size;
347 /* set default output to all source channels */
348 ctx->out_channels = ctx->channels;
349 ctx->output_mode = ctx->channel_mode;
351 ctx->output_mode |= AC3_OUTPUT_LFEON;
353 /* skip over portion of header which has already been read */
354 skip_bits(gbc, 16); // skip the sync_word
355 skip_bits(gbc, 16); // skip crc1
356 skip_bits(gbc, 8); // skip fscod and frmsizecod
357 skip_bits(gbc, 11); // skip bsid, bsmod, and acmod
358 if(ctx->channel_mode == AC3_CHMODE_STEREO) {
359 skip_bits(gbc, 2); // skip dsurmod
361 if((ctx->channel_mode & 1) && ctx->channel_mode != AC3_CHMODE_MONO)
362 skip_bits(gbc, 2); // skip cmixlev
363 if(ctx->channel_mode & 4)
364 skip_bits(gbc, 2); // skip surmixlev
366 skip_bits1(gbc); // skip lfeon
368 /* read the rest of the bsi. read twice for dual mono mode. */
369 i = !(ctx->channel_mode);
371 skip_bits(gbc, 5); // skip dialog normalization
373 skip_bits(gbc, 8); //skip compression
375 skip_bits(gbc, 8); //skip language code
377 skip_bits(gbc, 7); //skip audio production information
380 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
382 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
383 TODO: read & use the xbsi1 downmix levels */
385 skip_bits(gbc, 14); //skip timecode1 / xbsi1
387 skip_bits(gbc, 14); //skip timecode2 / xbsi2
389 /* skip additional bitstream info */
390 if (get_bits1(gbc)) {
391 i = get_bits(gbc, 6);
397 /* set stereo downmixing coefficients
398 reference: Section 7.8.2 Downmixing Into Two Channels */
399 for(i=0; i<ctx->fbw_channels; i++) {
400 ctx->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[ctx->channel_mode][i][0]];
401 ctx->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[ctx->channel_mode][i][1]];
403 if(ctx->channel_mode > 1 && ctx->channel_mode & 1) {
404 ctx->downmix_coeffs[1][0] = ctx->downmix_coeffs[1][1] = center_mix_level;
406 if(ctx->channel_mode == AC3_CHMODE_2F1R || ctx->channel_mode == AC3_CHMODE_3F1R) {
407 int nf = ctx->channel_mode - 2;
408 ctx->downmix_coeffs[nf][0] = ctx->downmix_coeffs[nf][1] = surround_mix_level * LEVEL_MINUS_3DB;
410 if(ctx->channel_mode == AC3_CHMODE_2F2R || ctx->channel_mode == AC3_CHMODE_3F2R) {
411 int nf = ctx->channel_mode - 4;
412 ctx->downmix_coeffs[nf][0] = ctx->downmix_coeffs[nf+1][1] = surround_mix_level;
419 * Decode the grouped exponents according to exponent strategy.
420 * reference: Section 7.1.3 Exponent Decoding
422 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
423 uint8_t absexp, int8_t *dexps)
425 int i, j, grp, group_size;
430 group_size = exp_strategy + (exp_strategy == EXP_D45);
431 for(grp=0,i=0; grp<ngrps; grp++) {
432 expacc = get_bits(gbc, 7);
433 dexp[i++] = exp_ungroup_tab[expacc][0];
434 dexp[i++] = exp_ungroup_tab[expacc][1];
435 dexp[i++] = exp_ungroup_tab[expacc][2];
438 /* convert to absolute exps and expand groups */
440 for(i=0; i<ngrps*3; i++) {
441 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
442 for(j=0; j<group_size; j++) {
443 dexps[(i*group_size)+j] = prevexp;
449 * Generate transform coefficients for each coupled channel in the coupling
450 * range using the coupling coefficients and coupling coordinates.
451 * reference: Section 7.4.3 Coupling Coordinate Format
453 static void uncouple_channels(AC3DecodeContext *ctx)
455 int i, j, ch, bnd, subbnd;
458 i = ctx->start_freq[CPL_CH];
459 for(bnd=0; bnd<ctx->num_cpl_bands; bnd++) {
462 for(j=0; j<12; j++) {
463 for(ch=1; ch<=ctx->fbw_channels; ch++) {
464 if(ctx->channel_in_cpl[ch])
465 ctx->transform_coeffs[ch][i] = ctx->transform_coeffs[CPL_CH][i] * ctx->cpl_coords[ch][bnd] * 8.0f;
469 } while(ctx->cpl_band_struct[subbnd]);
474 * Grouped mantissas for 3-level 5-level and 11-level quantization
486 * Get the transform coefficients for a particular channel
487 * reference: Section 7.3 Quantization and Decoding of Mantissas
489 static int get_transform_coeffs_ch(AC3DecodeContext *ctx, int ch_index, mant_groups *m)
491 GetBitContext *gbc = &ctx->gbc;
492 int i, gcode, tbap, start, end;
497 exps = ctx->dexps[ch_index];
498 bap = ctx->bap[ch_index];
499 coeffs = ctx->transform_coeffs[ch_index];
500 start = ctx->start_freq[ch_index];
501 end = ctx->end_freq[ch_index];
503 for (i = start; i < end; i++) {
507 coeffs[i] = ((av_random(&ctx->dith_state) & 0xFFFF) / 65535.0f) - 0.5f;
512 gcode = get_bits(gbc, 5);
513 m->b1_mant[0] = b1_mantissas[gcode][0];
514 m->b1_mant[1] = b1_mantissas[gcode][1];
515 m->b1_mant[2] = b1_mantissas[gcode][2];
518 coeffs[i] = m->b1_mant[m->b1ptr++];
523 gcode = get_bits(gbc, 7);
524 m->b2_mant[0] = b2_mantissas[gcode][0];
525 m->b2_mant[1] = b2_mantissas[gcode][1];
526 m->b2_mant[2] = b2_mantissas[gcode][2];
529 coeffs[i] = m->b2_mant[m->b2ptr++];
533 coeffs[i] = b3_mantissas[get_bits(gbc, 3)];
538 gcode = get_bits(gbc, 7);
539 m->b4_mant[0] = b4_mantissas[gcode][0];
540 m->b4_mant[1] = b4_mantissas[gcode][1];
543 coeffs[i] = m->b4_mant[m->b4ptr++];
547 coeffs[i] = b5_mantissas[get_bits(gbc, 4)];
551 /* asymmetric dequantization */
552 coeffs[i] = get_sbits(gbc, quantization_tab[tbap]) * scale_factors[quantization_tab[tbap]-1];
555 coeffs[i] *= scale_factors[exps[i]];
562 * Remove random dithering from coefficients with zero-bit mantissas
563 * reference: Section 7.3.4 Dither for Zero Bit Mantissas (bap=0)
565 static void remove_dithering(AC3DecodeContext *ctx) {
571 for(ch=1; ch<=ctx->fbw_channels; ch++) {
572 if(!ctx->dither_flag[ch]) {
573 coeffs = ctx->transform_coeffs[ch];
575 if(ctx->channel_in_cpl[ch])
576 end = ctx->start_freq[CPL_CH];
578 end = ctx->end_freq[ch];
579 for(i=0; i<end; i++) {
583 if(ctx->channel_in_cpl[ch]) {
584 bap = ctx->bap[CPL_CH];
585 for(; i<ctx->end_freq[CPL_CH]; i++) {
595 * Get the transform coefficients.
597 static int get_transform_coeffs(AC3DecodeContext * ctx)
603 m.b1ptr = m.b2ptr = m.b4ptr = 3;
605 for (ch = 1; ch <= ctx->channels; ch++) {
606 /* transform coefficients for full-bandwidth channel */
607 if (get_transform_coeffs_ch(ctx, ch, &m))
609 /* tranform coefficients for coupling channel come right after the
610 coefficients for the first coupled channel*/
611 if (ctx->channel_in_cpl[ch]) {
613 if (get_transform_coeffs_ch(ctx, CPL_CH, &m)) {
614 av_log(ctx->avctx, AV_LOG_ERROR, "error in decoupling channels\n");
617 uncouple_channels(ctx);
620 end = ctx->end_freq[CPL_CH];
622 end = ctx->end_freq[ch];
625 ctx->transform_coeffs[ch][end] = 0;
629 /* if any channel doesn't use dithering, zero appropriate coefficients */
631 remove_dithering(ctx);
637 * Stereo rematrixing.
638 * reference: Section 7.5.4 Rematrixing : Decoding Technique
640 static void do_rematrixing(AC3DecodeContext *ctx)
646 end = FFMIN(ctx->end_freq[1], ctx->end_freq[2]);
648 for(bnd=0; bnd<ctx->num_rematrixing_bands; bnd++) {
649 if(ctx->rematrixing_flags[bnd]) {
650 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
651 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
652 tmp0 = ctx->transform_coeffs[1][i];
653 tmp1 = ctx->transform_coeffs[2][i];
654 ctx->transform_coeffs[1][i] = tmp0 + tmp1;
655 ctx->transform_coeffs[2][i] = tmp0 - tmp1;
662 * Perform the 256-point IMDCT
664 static void do_imdct_256(AC3DecodeContext *ctx, int chindex)
667 DECLARE_ALIGNED_16(float, x[128]);
669 float *o_ptr = ctx->tmp_output;
672 /* de-interleave coefficients */
673 for(k=0; k<128; k++) {
674 x[k] = ctx->transform_coeffs[chindex][2*k+i];
677 /* run standard IMDCT */
678 ctx->imdct_256.fft.imdct_calc(&ctx->imdct_256, o_ptr, x, ctx->tmp_imdct);
680 /* reverse the post-rotation & reordering from standard IMDCT */
681 for(k=0; k<32; k++) {
682 z[i][32+k].re = -o_ptr[128+2*k];
683 z[i][32+k].im = -o_ptr[2*k];
684 z[i][31-k].re = o_ptr[2*k+1];
685 z[i][31-k].im = o_ptr[128+2*k+1];
689 /* apply AC-3 post-rotation & reordering */
690 for(k=0; k<64; k++) {
691 o_ptr[ 2*k ] = -z[0][ k].im;
692 o_ptr[ 2*k+1] = z[0][63-k].re;
693 o_ptr[128+2*k ] = -z[0][ k].re;
694 o_ptr[128+2*k+1] = z[0][63-k].im;
695 o_ptr[256+2*k ] = -z[1][ k].re;
696 o_ptr[256+2*k+1] = z[1][63-k].im;
697 o_ptr[384+2*k ] = z[1][ k].im;
698 o_ptr[384+2*k+1] = -z[1][63-k].re;
703 * Inverse MDCT Transform.
704 * Convert frequency domain coefficients to time-domain audio samples.
705 * reference: Section 7.9.4 Transformation Equations
707 static inline void do_imdct(AC3DecodeContext *ctx)
712 /* Don't perform the IMDCT on the LFE channel unless it's used in the output */
713 channels = ctx->fbw_channels;
714 if(ctx->output_mode & AC3_OUTPUT_LFEON)
717 for (ch=1; ch<=channels; ch++) {
718 if (ctx->block_switch[ch]) {
719 do_imdct_256(ctx, ch);
721 ctx->imdct_512.fft.imdct_calc(&ctx->imdct_512, ctx->tmp_output,
722 ctx->transform_coeffs[ch],
725 /* For the first half of the block, apply the window, add the delay
726 from the previous block, and send to output */
727 ctx->dsp.vector_fmul_add_add(ctx->output[ch-1], ctx->tmp_output,
728 ctx->window, ctx->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 ctx->dsp.vector_fmul_reverse(ctx->delay[ch-1], ctx->tmp_output+256,
737 * Downmix the output to mono or stereo.
739 static void ac3_downmix(float samples[AC3_MAX_CHANNELS][256], int fbw_channels,
740 int output_mode, float coef[AC3_MAX_CHANNELS][2])
743 float v0, v1, s0, s1;
745 for(i=0; i<256; i++) {
746 v0 = v1 = s0 = s1 = 0.0f;
747 for(j=0; j<fbw_channels; j++) {
748 v0 += samples[j][i] * coef[j][0];
749 v1 += samples[j][i] * coef[j][1];
755 if(output_mode == AC3_CHMODE_MONO) {
756 samples[0][i] = (v0 + v1) * LEVEL_MINUS_3DB;
757 } else if(output_mode == AC3_CHMODE_STEREO) {
765 * Parse an audio block from AC-3 bitstream.
767 static int ac3_parse_audio_block(AC3DecodeContext *ctx, int blk)
769 int fbw_channels = ctx->fbw_channels;
770 int channel_mode = ctx->channel_mode;
772 GetBitContext *gbc = &ctx->gbc;
773 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
775 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
777 /* block switch flags */
778 for (ch = 1; ch <= fbw_channels; ch++)
779 ctx->block_switch[ch] = get_bits1(gbc);
781 /* dithering flags */
783 for (ch = 1; ch <= fbw_channels; ch++) {
784 ctx->dither_flag[ch] = get_bits1(gbc);
785 if(!ctx->dither_flag[ch])
790 i = !(ctx->channel_mode);
793 ctx->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
794 ctx->avctx->drc_scale)+1.0;
795 } else if(blk == 0) {
796 ctx->dynamic_range[i] = 1.0f;
800 /* coupling strategy */
801 if (get_bits1(gbc)) {
802 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
803 ctx->cpl_in_use = get_bits1(gbc);
804 if (ctx->cpl_in_use) {
805 /* coupling in use */
806 int cpl_begin_freq, cpl_end_freq;
808 /* determine which channels are coupled */
809 for (ch = 1; ch <= fbw_channels; ch++)
810 ctx->channel_in_cpl[ch] = get_bits1(gbc);
812 /* phase flags in use */
813 if (channel_mode == AC3_CHMODE_STEREO)
814 ctx->phase_flags_in_use = get_bits1(gbc);
816 /* coupling frequency range and band structure */
817 cpl_begin_freq = get_bits(gbc, 4);
818 cpl_end_freq = get_bits(gbc, 4);
819 if (3 + cpl_end_freq - cpl_begin_freq < 0) {
820 av_log(ctx->avctx, AV_LOG_ERROR, "3+cplendf = %d < cplbegf = %d\n", 3+cpl_end_freq, cpl_begin_freq);
823 ctx->num_cpl_bands = ctx->num_cpl_subbands = 3 + cpl_end_freq - cpl_begin_freq;
824 ctx->start_freq[CPL_CH] = cpl_begin_freq * 12 + 37;
825 ctx->end_freq[CPL_CH] = cpl_end_freq * 12 + 73;
826 for (bnd = 0; bnd < ctx->num_cpl_subbands - 1; bnd++) {
827 if (get_bits1(gbc)) {
828 ctx->cpl_band_struct[bnd] = 1;
829 ctx->num_cpl_bands--;
833 /* coupling not in use */
834 for (ch = 1; ch <= fbw_channels; ch++)
835 ctx->channel_in_cpl[ch] = 0;
839 /* coupling coordinates */
840 if (ctx->cpl_in_use) {
841 int cpl_coords_exist = 0;
843 for (ch = 1; ch <= fbw_channels; ch++) {
844 if (ctx->channel_in_cpl[ch]) {
845 if (get_bits1(gbc)) {
846 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
847 cpl_coords_exist = 1;
848 master_cpl_coord = 3 * get_bits(gbc, 2);
849 for (bnd = 0; bnd < ctx->num_cpl_bands; bnd++) {
850 cpl_coord_exp = get_bits(gbc, 4);
851 cpl_coord_mant = get_bits(gbc, 4);
852 if (cpl_coord_exp == 15)
853 ctx->cpl_coords[ch][bnd] = cpl_coord_mant / 16.0f;
855 ctx->cpl_coords[ch][bnd] = (cpl_coord_mant + 16.0f) / 32.0f;
856 ctx->cpl_coords[ch][bnd] *= scale_factors[cpl_coord_exp + master_cpl_coord];
862 if (channel_mode == AC3_CHMODE_STEREO && ctx->phase_flags_in_use && cpl_coords_exist) {
863 for (bnd = 0; bnd < ctx->num_cpl_bands; bnd++) {
865 ctx->cpl_coords[2][bnd] = -ctx->cpl_coords[2][bnd];
870 /* stereo rematrixing strategy and band structure */
871 if (channel_mode == AC3_CHMODE_STEREO) {
872 ctx->rematrixing_strategy = get_bits1(gbc);
873 if (ctx->rematrixing_strategy) {
874 ctx->num_rematrixing_bands = 4;
875 if(ctx->cpl_in_use && ctx->start_freq[CPL_CH] <= 61)
876 ctx->num_rematrixing_bands -= 1 + (ctx->start_freq[CPL_CH] == 37);
877 for(bnd=0; bnd<ctx->num_rematrixing_bands; bnd++)
878 ctx->rematrixing_flags[bnd] = get_bits1(gbc);
882 /* exponent strategies for each channel */
883 ctx->exp_strategy[CPL_CH] = EXP_REUSE;
884 ctx->exp_strategy[ctx->lfe_ch] = EXP_REUSE;
885 for (ch = !ctx->cpl_in_use; ch <= ctx->channels; ch++) {
886 if(ch == ctx->lfe_ch)
887 ctx->exp_strategy[ch] = get_bits(gbc, 1);
889 ctx->exp_strategy[ch] = get_bits(gbc, 2);
890 if(ctx->exp_strategy[ch] != EXP_REUSE)
891 bit_alloc_stages[ch] = 3;
894 /* channel bandwidth */
895 for (ch = 1; ch <= fbw_channels; ch++) {
896 ctx->start_freq[ch] = 0;
897 if (ctx->exp_strategy[ch] != EXP_REUSE) {
898 int prev = ctx->end_freq[ch];
899 if (ctx->channel_in_cpl[ch])
900 ctx->end_freq[ch] = ctx->start_freq[CPL_CH];
902 int bandwidth_code = get_bits(gbc, 6);
903 if (bandwidth_code > 60) {
904 av_log(ctx->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
907 ctx->end_freq[ch] = bandwidth_code * 3 + 73;
909 if(blk > 0 && ctx->end_freq[ch] != prev)
910 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
913 ctx->start_freq[ctx->lfe_ch] = 0;
914 ctx->end_freq[ctx->lfe_ch] = 7;
916 /* decode exponents for each channel */
917 for (ch = !ctx->cpl_in_use; ch <= ctx->channels; ch++) {
918 if (ctx->exp_strategy[ch] != EXP_REUSE) {
919 int group_size, num_groups;
920 group_size = 3 << (ctx->exp_strategy[ch] - 1);
922 num_groups = (ctx->end_freq[ch] - ctx->start_freq[ch]) / group_size;
923 else if(ch == ctx->lfe_ch)
926 num_groups = (ctx->end_freq[ch] + group_size - 4) / group_size;
927 ctx->dexps[ch][0] = get_bits(gbc, 4) << !ch;
928 decode_exponents(gbc, ctx->exp_strategy[ch], num_groups, ctx->dexps[ch][0],
929 &ctx->dexps[ch][ctx->start_freq[ch]+!!ch]);
930 if(ch != CPL_CH && ch != ctx->lfe_ch)
931 skip_bits(gbc, 2); /* skip gainrng */
935 /* bit allocation information */
936 if (get_bits1(gbc)) {
937 ctx->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> ctx->bit_alloc_params.sr_shift;
938 ctx->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> ctx->bit_alloc_params.sr_shift;
939 ctx->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
940 ctx->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
941 ctx->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
942 for(ch=!ctx->cpl_in_use; ch<=ctx->channels; ch++) {
943 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
947 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
948 if (get_bits1(gbc)) {
950 csnr = (get_bits(gbc, 6) - 15) << 4;
951 for (ch = !ctx->cpl_in_use; ch <= ctx->channels; ch++) { /* snr offset and fast gain */
952 ctx->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
953 ctx->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
955 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
958 /* coupling leak information */
959 if (ctx->cpl_in_use && get_bits1(gbc)) {
960 ctx->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
961 ctx->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
962 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
965 /* delta bit allocation information */
966 if (get_bits1(gbc)) {
967 /* delta bit allocation exists (strategy) */
968 for (ch = !ctx->cpl_in_use; ch <= fbw_channels; ch++) {
969 ctx->dba_mode[ch] = get_bits(gbc, 2);
970 if (ctx->dba_mode[ch] == DBA_RESERVED) {
971 av_log(ctx->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
974 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
976 /* channel delta offset, len and bit allocation */
977 for (ch = !ctx->cpl_in_use; ch <= fbw_channels; ch++) {
978 if (ctx->dba_mode[ch] == DBA_NEW) {
979 ctx->dba_nsegs[ch] = get_bits(gbc, 3);
980 for (seg = 0; seg <= ctx->dba_nsegs[ch]; seg++) {
981 ctx->dba_offsets[ch][seg] = get_bits(gbc, 5);
982 ctx->dba_lengths[ch][seg] = get_bits(gbc, 4);
983 ctx->dba_values[ch][seg] = get_bits(gbc, 3);
987 } else if(blk == 0) {
988 for(ch=0; ch<=ctx->channels; ch++) {
989 ctx->dba_mode[ch] = DBA_NONE;
994 for(ch=!ctx->cpl_in_use; ch<=ctx->channels; ch++) {
995 if(bit_alloc_stages[ch] > 2) {
996 /* Exponent mapping into PSD and PSD integration */
997 ff_ac3_bit_alloc_calc_psd(ctx->dexps[ch],
998 ctx->start_freq[ch], ctx->end_freq[ch],
999 ctx->psd[ch], ctx->band_psd[ch]);
1001 if(bit_alloc_stages[ch] > 1) {
1002 /* Compute excitation function, Compute masking curve, and
1003 Apply delta bit allocation */
1004 ff_ac3_bit_alloc_calc_mask(&ctx->bit_alloc_params, ctx->band_psd[ch],
1005 ctx->start_freq[ch], ctx->end_freq[ch],
1006 ctx->fast_gain[ch], (ch == ctx->lfe_ch),
1007 ctx->dba_mode[ch], ctx->dba_nsegs[ch],
1008 ctx->dba_offsets[ch], ctx->dba_lengths[ch],
1009 ctx->dba_values[ch], ctx->mask[ch]);
1011 if(bit_alloc_stages[ch] > 0) {
1012 /* Compute bit allocation */
1013 ff_ac3_bit_alloc_calc_bap(ctx->mask[ch], ctx->psd[ch],
1014 ctx->start_freq[ch], ctx->end_freq[ch],
1015 ctx->snr_offset[ch],
1016 ctx->bit_alloc_params.floor,
1021 /* unused dummy data */
1022 if (get_bits1(gbc)) {
1023 int skipl = get_bits(gbc, 9);
1028 /* unpack the transform coefficients
1029 this also uncouples channels if coupling is in use. */
1030 if (get_transform_coeffs(ctx)) {
1031 av_log(ctx->avctx, AV_LOG_ERROR, "Error in routine get_transform_coeffs\n");
1035 /* recover coefficients if rematrixing is in use */
1036 if(ctx->channel_mode == AC3_CHMODE_STEREO)
1037 do_rematrixing(ctx);
1039 /* apply scaling to coefficients (headroom, dynrng) */
1040 for(ch=1; ch<=ctx->channels; ch++) {
1041 float gain = 2.0f * ctx->mul_bias;
1042 if(ctx->channel_mode == AC3_CHMODE_DUALMONO) {
1043 gain *= ctx->dynamic_range[ch-1];
1045 gain *= ctx->dynamic_range[0];
1047 for(i=0; i<ctx->end_freq[ch]; i++) {
1048 ctx->transform_coeffs[ch][i] *= gain;
1054 /* downmix output if needed */
1055 if(ctx->channels != ctx->out_channels && !((ctx->output_mode & AC3_OUTPUT_LFEON) &&
1056 ctx->fbw_channels == ctx->out_channels)) {
1057 ac3_downmix(ctx->output, ctx->fbw_channels, ctx->output_mode,
1058 ctx->downmix_coeffs);
1061 /* convert float to 16-bit integer */
1062 for(ch=0; ch<ctx->out_channels; ch++) {
1063 for(i=0; i<256; i++) {
1064 ctx->output[ch][i] += ctx->add_bias;
1066 ctx->dsp.float_to_int16(ctx->int_output[ch], ctx->output[ch], 256);
1073 * Decode a single AC-3 frame.
1075 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size, uint8_t *buf, int buf_size)
1077 AC3DecodeContext *ctx = (AC3DecodeContext *)avctx->priv_data;
1078 int16_t *out_samples = (int16_t *)data;
1079 int i, blk, ch, err;
1081 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1082 init_get_bits(&ctx->gbc, buf, buf_size * 8);
1084 /* parse the syncinfo */
1085 err = ac3_parse_header(ctx);
1088 case AC3_PARSE_ERROR_SYNC:
1089 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1091 case AC3_PARSE_ERROR_BSID:
1092 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1094 case AC3_PARSE_ERROR_SAMPLE_RATE:
1095 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1097 case AC3_PARSE_ERROR_FRAME_SIZE:
1098 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1101 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1107 avctx->sample_rate = ctx->sampling_rate;
1108 avctx->bit_rate = ctx->bit_rate;
1110 /* check that reported frame size fits in input buffer */
1111 if(ctx->frame_size > buf_size) {
1112 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1116 /* channel config */
1117 ctx->out_channels = ctx->channels;
1118 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1119 avctx->request_channels < ctx->channels) {
1120 ctx->out_channels = avctx->request_channels;
1121 ctx->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1123 avctx->channels = ctx->out_channels;
1125 /* parse the audio blocks */
1126 for (blk = 0; blk < NB_BLOCKS; blk++) {
1127 if (ac3_parse_audio_block(ctx, blk)) {
1128 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1130 return ctx->frame_size;
1132 for (i = 0; i < 256; i++)
1133 for (ch = 0; ch < ctx->out_channels; ch++)
1134 *(out_samples++) = ctx->int_output[ch][i];
1136 *data_size = NB_BLOCKS * 256 * avctx->channels * sizeof (int16_t);
1137 return ctx->frame_size;
1141 * Uninitialize the AC-3 decoder.
1143 static int ac3_decode_end(AVCodecContext *avctx)
1145 AC3DecodeContext *ctx = (AC3DecodeContext *)avctx->priv_data;
1146 ff_mdct_end(&ctx->imdct_512);
1147 ff_mdct_end(&ctx->imdct_256);
1152 AVCodec ac3_decoder = {
1154 .type = CODEC_TYPE_AUDIO,
1156 .priv_data_size = sizeof (AC3DecodeContext),
1157 .init = ac3_decode_init,
1158 .close = ac3_decode_end,
1159 .decode = ac3_decode_frame,