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
35 #include "libavutil/crc.h"
36 #include "libavutil/random.h"
38 #include "ac3_parser.h"
39 #include "bitstream.h"
42 /** Maximum possible frame size when the specification limit is ignored */
43 #define AC3_MAX_FRAME_SIZE 21695
46 * Table of bin locations for rematrixing bands
47 * reference: Section 7.5.2 Rematrixing : Frequency Band Definitions
49 static const uint8_t rematrix_band_tab[5] = { 13, 25, 37, 61, 253 };
51 /** table for grouping exponents */
52 static uint8_t exp_ungroup_tab[128][3];
55 /** tables for ungrouping mantissas */
56 static int b1_mantissas[32][3];
57 static int b2_mantissas[128][3];
58 static int b3_mantissas[8];
59 static int b4_mantissas[128][2];
60 static int b5_mantissas[16];
63 * Quantization table: levels for symmetric. bits for asymmetric.
64 * reference: Table 7.18 Mapping of bap to Quantizer
66 static const uint8_t quantization_tab[16] = {
68 5, 6, 7, 8, 9, 10, 11, 12, 14, 16
71 /** dynamic range table. converts codes to scale factors. */
72 static float dynamic_range_tab[256];
74 /** Adjustments in dB gain */
75 #define LEVEL_PLUS_3DB 1.4142135623730950
76 #define LEVEL_PLUS_1POINT5DB 1.1892071150027209
77 #define LEVEL_MINUS_1POINT5DB 0.8408964152537145
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[9] = {
89 LEVEL_MINUS_1POINT5DB,
91 LEVEL_MINUS_4POINT5DB,
98 * Table for default stereo downmixing coefficients
99 * reference: Section 7.8.2 Downmixing Into Two Channels
101 static const uint8_t ac3_default_coeffs[8][5][2] = {
102 { { 2, 7 }, { 7, 2 }, },
104 { { 2, 7 }, { 7, 2 }, },
105 { { 2, 7 }, { 5, 5 }, { 7, 2 }, },
106 { { 2, 7 }, { 7, 2 }, { 6, 6 }, },
107 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 8, 8 }, },
108 { { 2, 7 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
109 { { 2, 7 }, { 5, 5 }, { 7, 2 }, { 6, 7 }, { 7, 6 }, },
112 /* override ac3.h to include coupling channel */
113 #undef AC3_MAX_CHANNELS
114 #define AC3_MAX_CHANNELS 7
117 #define AC3_OUTPUT_LFEON 8
120 int num_blocks; ///< number of audio blocks
121 int channel_mode; ///< channel mode (acmod)
122 int block_switch[AC3_MAX_CHANNELS]; ///< block switch flags
123 int dither_flag[AC3_MAX_CHANNELS]; ///< dither flags
124 int dither_all; ///< true if all channels are dithered
125 int cpl_in_use; ///< coupling in use
126 int channel_in_cpl[AC3_MAX_CHANNELS]; ///< channel in coupling
127 int phase_flags_in_use; ///< phase flags in use
128 int phase_flags[18]; ///< phase flags
129 int cpl_band_struct[18]; ///< coupling band structure
130 int num_rematrixing_bands; ///< number of rematrixing bands
131 int rematrixing_flags[4]; ///< rematrixing flags
132 int exp_strategy[AC3_MAX_CHANNELS]; ///< exponent strategies
133 int snr_offset[AC3_MAX_CHANNELS]; ///< signal-to-noise ratio offsets
134 int fast_gain[AC3_MAX_CHANNELS]; ///< fast gain values (signal-to-mask ratio)
135 int dba_mode[AC3_MAX_CHANNELS]; ///< delta bit allocation mode
136 int dba_nsegs[AC3_MAX_CHANNELS]; ///< number of delta segments
137 uint8_t dba_offsets[AC3_MAX_CHANNELS][8]; ///< delta segment offsets
138 uint8_t dba_lengths[AC3_MAX_CHANNELS][8]; ///< delta segment lengths
139 uint8_t dba_values[AC3_MAX_CHANNELS][8]; ///< delta values for each segment
141 int sample_rate; ///< sample frequency, in Hz
142 int bit_rate; ///< stream bit rate, in bits-per-second
143 int frame_type; ///< frame type (strmtyp)
144 int substreamid; ///< substream identification
145 int frame_size; ///< current frame size, in bytes
147 int channels; ///< number of total channels
148 int fbw_channels; ///< number of full-bandwidth channels
149 int lfe_on; ///< lfe channel in use
150 int lfe_ch; ///< index of LFE channel
151 int output_mode; ///< output channel configuration
152 int out_channels; ///< number of output channels
154 int center_mix_level; ///< Center mix level index
155 int surround_mix_level; ///< Surround mix level index
156 float downmix_coeffs[AC3_MAX_CHANNELS][2]; ///< stereo downmix coefficients
157 float downmix_coeff_adjust[2]; ///< adjustment needed for each output channel when downmixing
158 float dynamic_range[2]; ///< dynamic range
159 int cpl_coords[AC3_MAX_CHANNELS][18]; ///< coupling coordinates
160 int num_cpl_bands; ///< number of coupling bands
161 int num_cpl_subbands; ///< number of coupling sub bands
162 int start_freq[AC3_MAX_CHANNELS]; ///< start frequency bin
163 int end_freq[AC3_MAX_CHANNELS]; ///< end frequency bin
164 AC3BitAllocParameters bit_alloc_params; ///< bit allocation parameters
166 int num_exp_groups[AC3_MAX_CHANNELS]; ///< Number of exponent groups
167 int8_t dexps[AC3_MAX_CHANNELS][256]; ///< decoded exponents
168 uint8_t bap[AC3_MAX_CHANNELS][256]; ///< bit allocation pointers
169 int16_t psd[AC3_MAX_CHANNELS][256]; ///< scaled exponents
170 int16_t band_psd[AC3_MAX_CHANNELS][50]; ///< interpolated exponents
171 int16_t mask[AC3_MAX_CHANNELS][50]; ///< masking curve values
173 int fixed_coeffs[AC3_MAX_CHANNELS][256]; ///> fixed-point transform coefficients
174 DECLARE_ALIGNED_16(float, transform_coeffs[AC3_MAX_CHANNELS][256]); ///< transform coefficients
175 int downmixed; ///< indicates if coeffs are currently downmixed
178 MDCTContext imdct_512; ///< for 512 sample IMDCT
179 MDCTContext imdct_256; ///< for 256 sample IMDCT
180 DSPContext dsp; ///< for optimization
181 float add_bias; ///< offset for float_to_int16 conversion
182 float mul_bias; ///< scaling for float_to_int16 conversion
184 DECLARE_ALIGNED_16(float, output[AC3_MAX_CHANNELS][256]); ///< output after imdct transform and windowing
185 DECLARE_ALIGNED_16(short, int_output[AC3_MAX_CHANNELS-1][256]); ///< final 16-bit integer output
186 DECLARE_ALIGNED_16(float, delay[AC3_MAX_CHANNELS][256]); ///< delay - added to the next block
187 DECLARE_ALIGNED_16(float, tmp_imdct[256]); ///< temporary storage for imdct transform
188 DECLARE_ALIGNED_16(float, tmp_output[512]); ///< temporary storage for output before windowing
189 DECLARE_ALIGNED_16(float, window[256]); ///< window coefficients
192 GetBitContext gbc; ///< bitstream reader
193 AVRandomState dith_state; ///< for dither generation
194 AVCodecContext *avctx; ///< parent context
195 uint8_t *input_buffer; ///< temp buffer to prevent overread
199 * Symmetrical Dequantization
200 * reference: Section 7.3.3 Expansion of Mantissas for Symmetrical Quantization
201 * Tables 7.19 to 7.23
204 symmetric_dequant(int code, int levels)
206 return ((code - (levels >> 1)) << 24) / levels;
210 * Initialize tables at runtime.
212 static av_cold void ac3_tables_init(void)
216 /* generate grouped mantissa tables
217 reference: Section 7.3.5 Ungrouping of Mantissas */
218 for(i=0; i<32; i++) {
219 /* bap=1 mantissas */
220 b1_mantissas[i][0] = symmetric_dequant( i / 9 , 3);
221 b1_mantissas[i][1] = symmetric_dequant((i % 9) / 3, 3);
222 b1_mantissas[i][2] = symmetric_dequant((i % 9) % 3, 3);
224 for(i=0; i<128; i++) {
225 /* bap=2 mantissas */
226 b2_mantissas[i][0] = symmetric_dequant( i / 25 , 5);
227 b2_mantissas[i][1] = symmetric_dequant((i % 25) / 5, 5);
228 b2_mantissas[i][2] = symmetric_dequant((i % 25) % 5, 5);
230 /* bap=4 mantissas */
231 b4_mantissas[i][0] = symmetric_dequant(i / 11, 11);
232 b4_mantissas[i][1] = symmetric_dequant(i % 11, 11);
234 /* generate ungrouped mantissa tables
235 reference: Tables 7.21 and 7.23 */
237 /* bap=3 mantissas */
238 b3_mantissas[i] = symmetric_dequant(i, 7);
240 for(i=0; i<15; i++) {
241 /* bap=5 mantissas */
242 b5_mantissas[i] = symmetric_dequant(i, 15);
245 /* generate dynamic range table
246 reference: Section 7.7.1 Dynamic Range Control */
247 for(i=0; i<256; i++) {
248 int v = (i >> 5) - ((i >> 7) << 3) - 5;
249 dynamic_range_tab[i] = powf(2.0f, v) * ((i & 0x1F) | 0x20);
252 /* generate exponent tables
253 reference: Section 7.1.3 Exponent Decoding */
254 for(i=0; i<128; i++) {
255 exp_ungroup_tab[i][0] = i / 25;
256 exp_ungroup_tab[i][1] = (i % 25) / 5;
257 exp_ungroup_tab[i][2] = (i % 25) % 5;
263 * AVCodec initialization
265 static av_cold int ac3_decode_init(AVCodecContext *avctx)
267 AC3DecodeContext *s = avctx->priv_data;
272 ff_mdct_init(&s->imdct_256, 8, 1);
273 ff_mdct_init(&s->imdct_512, 9, 1);
274 ff_kbd_window_init(s->window, 5.0, 256);
275 dsputil_init(&s->dsp, avctx);
276 av_init_random(0, &s->dith_state);
278 /* set bias values for float to int16 conversion */
279 if(s->dsp.float_to_int16 == ff_float_to_int16_c) {
280 s->add_bias = 385.0f;
284 s->mul_bias = 32767.0f;
287 /* allow downmixing to stereo or mono */
288 if (avctx->channels > 0 && avctx->request_channels > 0 &&
289 avctx->request_channels < avctx->channels &&
290 avctx->request_channels <= 2) {
291 avctx->channels = avctx->request_channels;
295 /* allocate context input buffer */
296 if (avctx->error_resilience >= FF_ER_CAREFUL) {
297 s->input_buffer = av_mallocz(AC3_MAX_FRAME_SIZE + FF_INPUT_BUFFER_PADDING_SIZE);
298 if (!s->input_buffer)
299 return AVERROR_NOMEM;
306 * Parse the 'sync info' and 'bit stream info' from the AC-3 bitstream.
307 * GetBitContext within AC3DecodeContext must point to
308 * start of the synchronized ac3 bitstream.
310 static int ac3_parse_header(AC3DecodeContext *s)
313 GetBitContext *gbc = &s->gbc;
316 err = ff_ac3_parse_header(gbc, &hdr);
320 if(hdr.bitstream_id > 10)
321 return AC3_PARSE_ERROR_BSID;
323 /* get decoding parameters from header info */
324 s->bit_alloc_params.sr_code = hdr.sr_code;
325 s->channel_mode = hdr.channel_mode;
326 s->lfe_on = hdr.lfe_on;
327 s->bit_alloc_params.sr_shift = hdr.sr_shift;
328 s->sample_rate = hdr.sample_rate;
329 s->bit_rate = hdr.bit_rate;
330 s->channels = hdr.channels;
331 s->fbw_channels = s->channels - s->lfe_on;
332 s->lfe_ch = s->fbw_channels + 1;
333 s->frame_size = hdr.frame_size;
334 s->center_mix_level = hdr.center_mix_level;
335 s->surround_mix_level = hdr.surround_mix_level;
336 s->num_blocks = hdr.num_blocks;
337 s->frame_type = hdr.frame_type;
338 s->substreamid = hdr.substreamid;
341 s->start_freq[s->lfe_ch] = 0;
342 s->end_freq[s->lfe_ch] = 7;
343 s->num_exp_groups[s->lfe_ch] = 2;
344 s->channel_in_cpl[s->lfe_ch] = 0;
347 /* read the rest of the bsi. read twice for dual mono mode. */
348 i = !(s->channel_mode);
350 skip_bits(gbc, 5); // skip dialog normalization
352 skip_bits(gbc, 8); //skip compression
354 skip_bits(gbc, 8); //skip language code
356 skip_bits(gbc, 7); //skip audio production information
359 skip_bits(gbc, 2); //skip copyright bit and original bitstream bit
361 /* skip the timecodes (or extra bitstream information for Alternate Syntax)
362 TODO: read & use the xbsi1 downmix levels */
364 skip_bits(gbc, 14); //skip timecode1 / xbsi1
366 skip_bits(gbc, 14); //skip timecode2 / xbsi2
368 /* skip additional bitstream info */
369 if (get_bits1(gbc)) {
370 i = get_bits(gbc, 6);
380 * Set stereo downmixing coefficients based on frame header info.
381 * reference: Section 7.8.2 Downmixing Into Two Channels
383 static void set_downmix_coeffs(AC3DecodeContext *s)
386 float cmix = gain_levels[s->center_mix_level];
387 float smix = gain_levels[s->surround_mix_level];
389 for(i=0; i<s->fbw_channels; i++) {
390 s->downmix_coeffs[i][0] = gain_levels[ac3_default_coeffs[s->channel_mode][i][0]];
391 s->downmix_coeffs[i][1] = gain_levels[ac3_default_coeffs[s->channel_mode][i][1]];
393 if(s->channel_mode > 1 && s->channel_mode & 1) {
394 s->downmix_coeffs[1][0] = s->downmix_coeffs[1][1] = cmix;
396 if(s->channel_mode == AC3_CHMODE_2F1R || s->channel_mode == AC3_CHMODE_3F1R) {
397 int nf = s->channel_mode - 2;
398 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf][1] = smix * LEVEL_MINUS_3DB;
400 if(s->channel_mode == AC3_CHMODE_2F2R || s->channel_mode == AC3_CHMODE_3F2R) {
401 int nf = s->channel_mode - 4;
402 s->downmix_coeffs[nf][0] = s->downmix_coeffs[nf+1][1] = smix;
405 /* calculate adjustment needed for each channel to avoid clipping */
406 s->downmix_coeff_adjust[0] = s->downmix_coeff_adjust[1] = 0.0f;
407 for(i=0; i<s->fbw_channels; i++) {
408 s->downmix_coeff_adjust[0] += s->downmix_coeffs[i][0];
409 s->downmix_coeff_adjust[1] += s->downmix_coeffs[i][1];
411 s->downmix_coeff_adjust[0] = 1.0f / s->downmix_coeff_adjust[0];
412 s->downmix_coeff_adjust[1] = 1.0f / s->downmix_coeff_adjust[1];
416 * Decode the grouped exponents according to exponent strategy.
417 * reference: Section 7.1.3 Exponent Decoding
419 static void decode_exponents(GetBitContext *gbc, int exp_strategy, int ngrps,
420 uint8_t absexp, int8_t *dexps)
422 int i, j, grp, group_size;
427 group_size = exp_strategy + (exp_strategy == EXP_D45);
428 for(grp=0,i=0; grp<ngrps; grp++) {
429 expacc = get_bits(gbc, 7);
430 dexp[i++] = exp_ungroup_tab[expacc][0];
431 dexp[i++] = exp_ungroup_tab[expacc][1];
432 dexp[i++] = exp_ungroup_tab[expacc][2];
435 /* convert to absolute exps and expand groups */
437 for(i=0; i<ngrps*3; i++) {
438 prevexp = av_clip(prevexp + dexp[i]-2, 0, 24);
439 for(j=0; j<group_size; j++) {
440 dexps[(i*group_size)+j] = prevexp;
446 * Generate transform coefficients for each coupled channel in the coupling
447 * range using the coupling coefficients and coupling coordinates.
448 * reference: Section 7.4.3 Coupling Coordinate Format
450 static void uncouple_channels(AC3DecodeContext *s)
452 int i, j, ch, bnd, subbnd;
455 i = s->start_freq[CPL_CH];
456 for(bnd=0; bnd<s->num_cpl_bands; bnd++) {
459 for(j=0; j<12; j++) {
460 for(ch=1; ch<=s->fbw_channels; ch++) {
461 if(s->channel_in_cpl[ch]) {
462 s->fixed_coeffs[ch][i] = ((int64_t)s->fixed_coeffs[CPL_CH][i] * (int64_t)s->cpl_coords[ch][bnd]) >> 23;
463 if (ch == 2 && s->phase_flags[bnd])
464 s->fixed_coeffs[ch][i] = -s->fixed_coeffs[ch][i];
469 } while(s->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 void get_transform_coeffs_ch(AC3DecodeContext *s, int ch_index, mant_groups *m)
491 GetBitContext *gbc = &s->gbc;
492 int i, gcode, tbap, start, end;
497 exps = s->dexps[ch_index];
498 bap = s->bap[ch_index];
499 coeffs = s->fixed_coeffs[ch_index];
500 start = s->start_freq[ch_index];
501 end = s->end_freq[ch_index];
503 for (i = start; i < end; i++) {
507 coeffs[i] = (av_random(&s->dith_state) & 0x7FFFFF) - 4194304;
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 int qlevel = quantization_tab[tbap];
553 coeffs[i] = get_sbits(gbc, qlevel) << (24 - qlevel);
557 coeffs[i] >>= 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 *s) {
571 for(ch=1; ch<=s->fbw_channels; ch++) {
572 if(!s->dither_flag[ch]) {
573 coeffs = s->fixed_coeffs[ch];
575 if(s->channel_in_cpl[ch])
576 end = s->start_freq[CPL_CH];
578 end = s->end_freq[ch];
579 for(i=0; i<end; i++) {
583 if(s->channel_in_cpl[ch]) {
584 bap = s->bap[CPL_CH];
585 for(; i<s->end_freq[CPL_CH]; i++) {
595 * Get the transform coefficients.
597 static void get_transform_coeffs(AC3DecodeContext *s)
603 m.b1ptr = m.b2ptr = m.b4ptr = 3;
605 for (ch = 1; ch <= s->channels; ch++) {
606 /* transform coefficients for full-bandwidth channel */
607 get_transform_coeffs_ch(s, ch, &m);
608 /* tranform coefficients for coupling channel come right after the
609 coefficients for the first coupled channel*/
610 if (s->channel_in_cpl[ch]) {
612 get_transform_coeffs_ch(s, CPL_CH, &m);
613 uncouple_channels(s);
616 end = s->end_freq[CPL_CH];
618 end = s->end_freq[ch];
621 s->fixed_coeffs[ch][end] = 0;
625 /* if any channel doesn't use dithering, zero appropriate coefficients */
631 * Stereo rematrixing.
632 * reference: Section 7.5.4 Rematrixing : Decoding Technique
634 static void do_rematrixing(AC3DecodeContext *s)
640 end = FFMIN(s->end_freq[1], s->end_freq[2]);
642 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++) {
643 if(s->rematrixing_flags[bnd]) {
644 bndend = FFMIN(end, rematrix_band_tab[bnd+1]);
645 for(i=rematrix_band_tab[bnd]; i<bndend; i++) {
646 tmp0 = s->fixed_coeffs[1][i];
647 tmp1 = s->fixed_coeffs[2][i];
648 s->fixed_coeffs[1][i] = tmp0 + tmp1;
649 s->fixed_coeffs[2][i] = tmp0 - tmp1;
656 * Perform the 256-point IMDCT
658 static void do_imdct_256(AC3DecodeContext *s, int chindex)
661 DECLARE_ALIGNED_16(float, x[128]);
663 float *o_ptr = s->tmp_output;
666 /* de-interleave coefficients */
667 for(k=0; k<128; k++) {
668 x[k] = s->transform_coeffs[chindex][2*k+i];
671 /* run standard IMDCT */
672 s->imdct_256.fft.imdct_calc(&s->imdct_256, o_ptr, x, s->tmp_imdct);
674 /* reverse the post-rotation & reordering from standard IMDCT */
675 for(k=0; k<32; k++) {
676 z[i][32+k].re = -o_ptr[128+2*k];
677 z[i][32+k].im = -o_ptr[2*k];
678 z[i][31-k].re = o_ptr[2*k+1];
679 z[i][31-k].im = o_ptr[128+2*k+1];
683 /* apply AC-3 post-rotation & reordering */
684 for(k=0; k<64; k++) {
685 o_ptr[ 2*k ] = -z[0][ k].im;
686 o_ptr[ 2*k+1] = z[0][63-k].re;
687 o_ptr[128+2*k ] = -z[0][ k].re;
688 o_ptr[128+2*k+1] = z[0][63-k].im;
689 o_ptr[256+2*k ] = -z[1][ k].re;
690 o_ptr[256+2*k+1] = z[1][63-k].im;
691 o_ptr[384+2*k ] = z[1][ k].im;
692 o_ptr[384+2*k+1] = -z[1][63-k].re;
697 * Inverse MDCT Transform.
698 * Convert frequency domain coefficients to time-domain audio samples.
699 * reference: Section 7.9.4 Transformation Equations
701 static inline void do_imdct(AC3DecodeContext *s, int channels)
705 for (ch=1; ch<=channels; ch++) {
706 if (s->block_switch[ch]) {
709 s->imdct_512.fft.imdct_calc(&s->imdct_512, s->tmp_output,
710 s->transform_coeffs[ch], s->tmp_imdct);
712 /* For the first half of the block, apply the window, add the delay
713 from the previous block, and send to output */
714 s->dsp.vector_fmul_add_add(s->output[ch-1], s->tmp_output,
715 s->window, s->delay[ch-1], 0, 256, 1);
716 /* For the second half of the block, apply the window and store the
717 samples to delay, to be combined with the next block */
718 s->dsp.vector_fmul_reverse(s->delay[ch-1], s->tmp_output+256,
724 * Downmix the output to mono or stereo.
726 static void ac3_downmix(AC3DecodeContext *s,
727 float samples[AC3_MAX_CHANNELS][256], int ch_offset)
732 for(i=0; i<256; i++) {
734 for(j=0; j<s->fbw_channels; j++) {
735 v0 += samples[j+ch_offset][i] * s->downmix_coeffs[j][0];
736 v1 += samples[j+ch_offset][i] * s->downmix_coeffs[j][1];
738 v0 *= s->downmix_coeff_adjust[0];
739 v1 *= s->downmix_coeff_adjust[1];
740 if(s->output_mode == AC3_CHMODE_MONO) {
741 samples[ch_offset][i] = (v0 + v1) * LEVEL_MINUS_3DB;
742 } else if(s->output_mode == AC3_CHMODE_STEREO) {
743 samples[ ch_offset][i] = v0;
744 samples[1+ch_offset][i] = v1;
750 * Upmix delay samples from stereo to original channel layout.
752 static void ac3_upmix_delay(AC3DecodeContext *s)
754 int channel_data_size = sizeof(s->delay[0]);
755 switch(s->channel_mode) {
756 case AC3_CHMODE_DUALMONO:
757 case AC3_CHMODE_STEREO:
758 /* upmix mono to stereo */
759 memcpy(s->delay[1], s->delay[0], channel_data_size);
761 case AC3_CHMODE_2F2R:
762 memset(s->delay[3], 0, channel_data_size);
763 case AC3_CHMODE_2F1R:
764 memset(s->delay[2], 0, channel_data_size);
766 case AC3_CHMODE_3F2R:
767 memset(s->delay[4], 0, channel_data_size);
768 case AC3_CHMODE_3F1R:
769 memset(s->delay[3], 0, channel_data_size);
771 memcpy(s->delay[2], s->delay[1], channel_data_size);
772 memset(s->delay[1], 0, channel_data_size);
778 * Parse an audio block from AC-3 bitstream.
780 static int ac3_parse_audio_block(AC3DecodeContext *s, int blk)
782 int fbw_channels = s->fbw_channels;
783 int channel_mode = s->channel_mode;
785 int different_transforms;
787 GetBitContext *gbc = &s->gbc;
788 uint8_t bit_alloc_stages[AC3_MAX_CHANNELS];
790 memset(bit_alloc_stages, 0, AC3_MAX_CHANNELS);
792 /* block switch flags */
793 different_transforms = 0;
794 for (ch = 1; ch <= fbw_channels; ch++) {
795 s->block_switch[ch] = get_bits1(gbc);
796 if(ch > 1 && s->block_switch[ch] != s->block_switch[1])
797 different_transforms = 1;
800 /* dithering flags */
802 for (ch = 1; ch <= fbw_channels; ch++) {
803 s->dither_flag[ch] = get_bits1(gbc);
804 if(!s->dither_flag[ch])
809 i = !(s->channel_mode);
812 s->dynamic_range[i] = ((dynamic_range_tab[get_bits(gbc, 8)]-1.0) *
813 s->avctx->drc_scale)+1.0;
814 } else if(blk == 0) {
815 s->dynamic_range[i] = 1.0f;
819 /* coupling strategy */
820 if (get_bits1(gbc)) {
821 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
822 s->cpl_in_use = get_bits1(gbc);
824 /* coupling in use */
825 int cpl_begin_freq, cpl_end_freq;
827 if (channel_mode < AC3_CHMODE_STEREO) {
828 av_log(s->avctx, AV_LOG_ERROR, "coupling not allowed in mono or dual-mono\n");
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;
863 av_log(s->avctx, AV_LOG_ERROR, "new coupling strategy must be present in block 0\n");
867 /* coupling coordinates */
869 int cpl_coords_exist = 0;
871 for (ch = 1; ch <= fbw_channels; ch++) {
872 if (s->channel_in_cpl[ch]) {
873 if (get_bits1(gbc)) {
874 int master_cpl_coord, cpl_coord_exp, cpl_coord_mant;
875 cpl_coords_exist = 1;
876 master_cpl_coord = 3 * get_bits(gbc, 2);
877 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
878 cpl_coord_exp = get_bits(gbc, 4);
879 cpl_coord_mant = get_bits(gbc, 4);
880 if (cpl_coord_exp == 15)
881 s->cpl_coords[ch][bnd] = cpl_coord_mant << 22;
883 s->cpl_coords[ch][bnd] = (cpl_coord_mant + 16) << 21;
884 s->cpl_coords[ch][bnd] >>= (cpl_coord_exp + master_cpl_coord);
887 av_log(s->avctx, AV_LOG_ERROR, "new coupling coordinates must be present in block 0\n");
893 if (channel_mode == AC3_CHMODE_STEREO && cpl_coords_exist) {
894 for (bnd = 0; bnd < s->num_cpl_bands; bnd++) {
895 s->phase_flags[bnd] = s->phase_flags_in_use? get_bits1(gbc) : 0;
900 /* stereo rematrixing strategy and band structure */
901 if (channel_mode == AC3_CHMODE_STEREO) {
902 if (get_bits1(gbc)) {
903 s->num_rematrixing_bands = 4;
904 if(s->cpl_in_use && s->start_freq[CPL_CH] <= 61)
905 s->num_rematrixing_bands -= 1 + (s->start_freq[CPL_CH] == 37);
906 for(bnd=0; bnd<s->num_rematrixing_bands; bnd++)
907 s->rematrixing_flags[bnd] = get_bits1(gbc);
909 av_log(s->avctx, AV_LOG_ERROR, "new rematrixing strategy must be present in block 0\n");
914 /* exponent strategies for each channel */
915 s->exp_strategy[CPL_CH] = EXP_REUSE;
916 s->exp_strategy[s->lfe_ch] = EXP_REUSE;
917 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
918 s->exp_strategy[ch] = get_bits(gbc, 2 - (ch == s->lfe_ch));
919 if(s->exp_strategy[ch] != EXP_REUSE)
920 bit_alloc_stages[ch] = 3;
923 /* channel bandwidth */
924 for (ch = 1; ch <= fbw_channels; ch++) {
925 s->start_freq[ch] = 0;
926 if (s->exp_strategy[ch] != EXP_REUSE) {
928 int prev = s->end_freq[ch];
929 if (s->channel_in_cpl[ch])
930 s->end_freq[ch] = s->start_freq[CPL_CH];
932 int bandwidth_code = get_bits(gbc, 6);
933 if (bandwidth_code > 60) {
934 av_log(s->avctx, AV_LOG_ERROR, "bandwidth code = %d > 60", bandwidth_code);
937 s->end_freq[ch] = bandwidth_code * 3 + 73;
939 group_size = 3 << (s->exp_strategy[ch] - 1);
940 s->num_exp_groups[ch] = (s->end_freq[ch]+group_size-4) / group_size;
941 if(blk > 0 && s->end_freq[ch] != prev)
942 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
945 if (s->cpl_in_use && s->exp_strategy[CPL_CH] != EXP_REUSE) {
946 s->num_exp_groups[CPL_CH] = (s->end_freq[CPL_CH] - s->start_freq[CPL_CH]) /
947 (3 << (s->exp_strategy[CPL_CH] - 1));
950 /* decode exponents for each channel */
951 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) {
952 if (s->exp_strategy[ch] != EXP_REUSE) {
953 s->dexps[ch][0] = get_bits(gbc, 4) << !ch;
954 decode_exponents(gbc, s->exp_strategy[ch],
955 s->num_exp_groups[ch], s->dexps[ch][0],
956 &s->dexps[ch][s->start_freq[ch]+!!ch]);
957 if(ch != CPL_CH && ch != s->lfe_ch)
958 skip_bits(gbc, 2); /* skip gainrng */
962 /* bit allocation information */
963 if (get_bits1(gbc)) {
964 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
965 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[get_bits(gbc, 2)] >> s->bit_alloc_params.sr_shift;
966 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab[get_bits(gbc, 2)];
967 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[get_bits(gbc, 2)];
968 s->bit_alloc_params.floor = ff_ac3_floor_tab[get_bits(gbc, 3)];
969 for(ch=!s->cpl_in_use; ch<=s->channels; ch++)
970 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
972 av_log(s->avctx, AV_LOG_ERROR, "new bit allocation info must be present in block 0\n");
976 /* signal-to-noise ratio offsets and fast gains (signal-to-mask ratios) */
977 if (get_bits1(gbc)) {
979 csnr = (get_bits(gbc, 6) - 15) << 4;
980 for (ch = !s->cpl_in_use; ch <= s->channels; ch++) { /* snr offset and fast gain */
981 s->snr_offset[ch] = (csnr + get_bits(gbc, 4)) << 2;
982 s->fast_gain[ch] = ff_ac3_fast_gain_tab[get_bits(gbc, 3)];
984 memset(bit_alloc_stages, 3, AC3_MAX_CHANNELS);
986 av_log(s->avctx, AV_LOG_ERROR, "new snr offsets must be present in block 0\n");
990 /* coupling leak information */
992 if (get_bits1(gbc)) {
993 s->bit_alloc_params.cpl_fast_leak = get_bits(gbc, 3);
994 s->bit_alloc_params.cpl_slow_leak = get_bits(gbc, 3);
995 bit_alloc_stages[CPL_CH] = FFMAX(bit_alloc_stages[CPL_CH], 2);
997 av_log(s->avctx, AV_LOG_ERROR, "new coupling leak info must be present in block 0\n");
1002 /* delta bit allocation information */
1003 if (get_bits1(gbc)) {
1004 /* delta bit allocation exists (strategy) */
1005 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1006 s->dba_mode[ch] = get_bits(gbc, 2);
1007 if (s->dba_mode[ch] == DBA_RESERVED) {
1008 av_log(s->avctx, AV_LOG_ERROR, "delta bit allocation strategy reserved\n");
1011 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1013 /* channel delta offset, len and bit allocation */
1014 for (ch = !s->cpl_in_use; ch <= fbw_channels; ch++) {
1015 if (s->dba_mode[ch] == DBA_NEW) {
1016 s->dba_nsegs[ch] = get_bits(gbc, 3);
1017 for (seg = 0; seg <= s->dba_nsegs[ch]; seg++) {
1018 s->dba_offsets[ch][seg] = get_bits(gbc, 5);
1019 s->dba_lengths[ch][seg] = get_bits(gbc, 4);
1020 s->dba_values[ch][seg] = get_bits(gbc, 3);
1022 /* run last 2 bit allocation stages if new dba values */
1023 bit_alloc_stages[ch] = FFMAX(bit_alloc_stages[ch], 2);
1026 } else if(blk == 0) {
1027 for(ch=0; ch<=s->channels; ch++) {
1028 s->dba_mode[ch] = DBA_NONE;
1032 /* Bit allocation */
1033 for(ch=!s->cpl_in_use; ch<=s->channels; ch++) {
1034 if(bit_alloc_stages[ch] > 2) {
1035 /* Exponent mapping into PSD and PSD integration */
1036 ff_ac3_bit_alloc_calc_psd(s->dexps[ch],
1037 s->start_freq[ch], s->end_freq[ch],
1038 s->psd[ch], s->band_psd[ch]);
1040 if(bit_alloc_stages[ch] > 1) {
1041 /* Compute excitation function, Compute masking curve, and
1042 Apply delta bit allocation */
1043 ff_ac3_bit_alloc_calc_mask(&s->bit_alloc_params, s->band_psd[ch],
1044 s->start_freq[ch], s->end_freq[ch],
1045 s->fast_gain[ch], (ch == s->lfe_ch),
1046 s->dba_mode[ch], s->dba_nsegs[ch],
1047 s->dba_offsets[ch], s->dba_lengths[ch],
1048 s->dba_values[ch], s->mask[ch]);
1050 if(bit_alloc_stages[ch] > 0) {
1051 /* Compute bit allocation */
1052 ff_ac3_bit_alloc_calc_bap(s->mask[ch], s->psd[ch],
1053 s->start_freq[ch], s->end_freq[ch],
1055 s->bit_alloc_params.floor,
1056 ff_ac3_bap_tab, s->bap[ch]);
1060 /* unused dummy data */
1061 if (get_bits1(gbc)) {
1062 int skipl = get_bits(gbc, 9);
1067 /* unpack the transform coefficients
1068 this also uncouples channels if coupling is in use. */
1069 get_transform_coeffs(s);
1071 /* recover coefficients if rematrixing is in use */
1072 if(s->channel_mode == AC3_CHMODE_STEREO)
1075 /* apply scaling to coefficients (headroom, dynrng) */
1076 for(ch=1; ch<=s->channels; ch++) {
1077 float gain = s->mul_bias / 4194304.0f;
1078 if(s->channel_mode == AC3_CHMODE_DUALMONO) {
1079 gain *= s->dynamic_range[ch-1];
1081 gain *= s->dynamic_range[0];
1083 for(i=0; i<256; i++) {
1084 s->transform_coeffs[ch][i] = s->fixed_coeffs[ch][i] * gain;
1088 /* downmix and MDCT. order depends on whether block switching is used for
1089 any channel in this block. this is because coefficients for the long
1090 and short transforms cannot be mixed. */
1091 downmix_output = s->channels != s->out_channels &&
1092 !((s->output_mode & AC3_OUTPUT_LFEON) &&
1093 s->fbw_channels == s->out_channels);
1094 if(different_transforms) {
1095 /* the delay samples have already been downmixed, so we upmix the delay
1096 samples in order to reconstruct all channels before downmixing. */
1102 do_imdct(s, s->channels);
1104 if(downmix_output) {
1105 ac3_downmix(s, s->output, 0);
1108 if(downmix_output) {
1109 ac3_downmix(s, s->transform_coeffs, 1);
1114 ac3_downmix(s, s->delay, 0);
1117 do_imdct(s, s->out_channels);
1120 /* convert float to 16-bit integer */
1121 for(ch=0; ch<s->out_channels; ch++) {
1122 for(i=0; i<256; i++) {
1123 s->output[ch][i] += s->add_bias;
1125 s->dsp.float_to_int16(s->int_output[ch], s->output[ch], 256);
1132 * Decode a single AC-3 frame.
1134 static int ac3_decode_frame(AVCodecContext * avctx, void *data, int *data_size,
1135 const uint8_t *buf, int buf_size)
1137 AC3DecodeContext *s = avctx->priv_data;
1138 int16_t *out_samples = (int16_t *)data;
1139 int i, blk, ch, err;
1141 /* initialize the GetBitContext with the start of valid AC-3 Frame */
1142 if (s->input_buffer) {
1143 /* copy input buffer to decoder context to avoid reading past the end
1144 of the buffer, which can be caused by a damaged input stream. */
1145 memcpy(s->input_buffer, buf, FFMIN(buf_size, AC3_MAX_FRAME_SIZE));
1146 init_get_bits(&s->gbc, s->input_buffer, buf_size * 8);
1148 init_get_bits(&s->gbc, buf, buf_size * 8);
1151 /* parse the syncinfo */
1153 err = ac3_parse_header(s);
1155 /* check that reported frame size fits in input buffer */
1156 if(s->frame_size > buf_size) {
1157 av_log(avctx, AV_LOG_ERROR, "incomplete frame\n");
1158 err = AC3_PARSE_ERROR_FRAME_SIZE;
1161 /* check for crc mismatch */
1162 if(err != AC3_PARSE_ERROR_FRAME_SIZE && avctx->error_resilience >= FF_ER_CAREFUL) {
1163 if(av_crc(av_crc_get_table(AV_CRC_16_ANSI), 0, &buf[2], s->frame_size-2)) {
1164 av_log(avctx, AV_LOG_ERROR, "frame CRC mismatch\n");
1165 err = AC3_PARSE_ERROR_CRC;
1169 if(err && err != AC3_PARSE_ERROR_CRC) {
1171 case AC3_PARSE_ERROR_SYNC:
1172 av_log(avctx, AV_LOG_ERROR, "frame sync error\n");
1174 case AC3_PARSE_ERROR_BSID:
1175 av_log(avctx, AV_LOG_ERROR, "invalid bitstream id\n");
1177 case AC3_PARSE_ERROR_SAMPLE_RATE:
1178 av_log(avctx, AV_LOG_ERROR, "invalid sample rate\n");
1180 case AC3_PARSE_ERROR_FRAME_SIZE:
1181 av_log(avctx, AV_LOG_ERROR, "invalid frame size\n");
1183 case AC3_PARSE_ERROR_FRAME_TYPE:
1184 /* skip frame if CRC is ok. otherwise use error concealment. */
1185 /* TODO: add support for substreams and dependent frames */
1186 if(s->frame_type == EAC3_FRAME_TYPE_DEPENDENT || s->substreamid) {
1187 av_log(avctx, AV_LOG_ERROR, "unsupported frame type : skipping frame\n");
1188 return s->frame_size;
1190 av_log(avctx, AV_LOG_ERROR, "invalid frame type\n");
1194 av_log(avctx, AV_LOG_ERROR, "invalid header\n");
1199 /* if frame is ok, set audio parameters */
1201 avctx->sample_rate = s->sample_rate;
1202 avctx->bit_rate = s->bit_rate;
1204 /* channel config */
1205 s->out_channels = s->channels;
1206 s->output_mode = s->channel_mode;
1208 s->output_mode |= AC3_OUTPUT_LFEON;
1209 if (avctx->request_channels > 0 && avctx->request_channels <= 2 &&
1210 avctx->request_channels < s->channels) {
1211 s->out_channels = avctx->request_channels;
1212 s->output_mode = avctx->request_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1214 avctx->channels = s->out_channels;
1216 /* set downmixing coefficients if needed */
1217 if(s->channels != s->out_channels && !((s->output_mode & AC3_OUTPUT_LFEON) &&
1218 s->fbw_channels == s->out_channels)) {
1219 set_downmix_coeffs(s);
1221 } else if (!s->out_channels) {
1222 s->out_channels = avctx->channels;
1223 if(s->out_channels < s->channels)
1224 s->output_mode = s->out_channels == 1 ? AC3_CHMODE_MONO : AC3_CHMODE_STEREO;
1227 /* parse the audio blocks */
1228 for (blk = 0; blk < s->num_blocks; blk++) {
1229 if (!err && ac3_parse_audio_block(s, blk)) {
1230 av_log(avctx, AV_LOG_ERROR, "error parsing the audio block\n");
1233 /* interleave output samples */
1234 for (i = 0; i < 256; i++)
1235 for (ch = 0; ch < s->out_channels; ch++)
1236 *(out_samples++) = s->int_output[ch][i];
1238 *data_size = s->num_blocks * 256 * avctx->channels * sizeof (int16_t);
1239 return s->frame_size;
1243 * Uninitialize the AC-3 decoder.
1245 static av_cold int ac3_decode_end(AVCodecContext *avctx)
1247 AC3DecodeContext *s = avctx->priv_data;
1248 ff_mdct_end(&s->imdct_512);
1249 ff_mdct_end(&s->imdct_256);
1251 av_freep(&s->input_buffer);
1256 AVCodec ac3_decoder = {
1258 .type = CODEC_TYPE_AUDIO,
1260 .priv_data_size = sizeof (AC3DecodeContext),
1261 .init = ac3_decode_init,
1262 .close = ac3_decode_end,
1263 .decode = ac3_decode_frame,
1264 .long_name = "ATSC A/52 / AC-3",