3 * Copyright (c) 2007 Bartlomiej Wolowiec <bartek.wolowiec@gmail.com>
4 * Copyright (c) 2008 Justin Ruggles
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 * There are several features of E-AC-3 that this decoder does not yet support.
27 * No known samples exist. If any ever surface, this feature should not be
28 * too difficult to implement.
30 * Reduced Sample Rates
31 * No known samples exist. The spec also does not give clear information
32 * on how this is to be implemented.
35 * Only the independent stream is currently decoded. Any dependent
36 * streams are skipped. We have only come across two examples of this, and
37 * they are both just test streams, one for HD-DVD and the other for
40 * Transient Pre-noise Processing
41 * This is side information which a decoder should use to reduce artifacts
42 * caused by transients. There are samples which are known to have this
43 * information, but this decoder currently ignores it.
49 #include "aac_ac3_parser.h"
51 #include "ac3_parser.h"
53 #include "ac3dec_data.h"
54 #include "eac3_data.h"
56 /** gain adaptive quantization mode */
64 #define EAC3_SR_CODE_REDUCED 3
66 void ff_eac3_apply_spectral_extension(AC3DecodeContext *s)
69 uint8_t wrapflag[SPX_MAX_BANDS]={1,0,}, num_copy_sections, copy_sizes[SPX_MAX_BANDS];
70 float rms_energy[SPX_MAX_BANDS];
72 /* Set copy index mapping table. Set wrap flags to apply a notch filter at
73 wrap points later on. */
74 bin = s->spx_dst_start_freq;
75 num_copy_sections = 0;
76 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
78 int bandsize = s->spx_band_sizes[bnd];
79 if (bin + bandsize > s->spx_src_start_freq) {
80 copy_sizes[num_copy_sections++] = bin - s->spx_dst_start_freq;
81 bin = s->spx_dst_start_freq;
84 for (i = 0; i < bandsize; i += copysize) {
85 if (bin == s->spx_src_start_freq) {
86 copy_sizes[num_copy_sections++] = bin - s->spx_dst_start_freq;
87 bin = s->spx_dst_start_freq;
89 copysize = FFMIN(bandsize - i, s->spx_src_start_freq - bin);
93 copy_sizes[num_copy_sections++] = bin - s->spx_dst_start_freq;
95 for (ch = 1; ch <= s->fbw_channels; ch++) {
96 if (!s->channel_uses_spx[ch])
99 /* Copy coeffs from normal bands to extension bands */
100 bin = s->spx_src_start_freq;
101 for (i = 0; i < num_copy_sections; i++) {
102 memcpy(&s->transform_coeffs[ch][bin],
103 &s->transform_coeffs[ch][s->spx_dst_start_freq],
104 copy_sizes[i]*sizeof(float));
105 bin += copy_sizes[i];
108 /* Calculate RMS energy for each SPX band. */
109 bin = s->spx_src_start_freq;
110 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
111 int bandsize = s->spx_band_sizes[bnd];
113 for (i = 0; i < bandsize; i++) {
114 float coeff = s->transform_coeffs[ch][bin++];
115 accum += coeff * coeff;
117 rms_energy[bnd] = sqrtf(accum / bandsize);
120 /* Apply a notch filter at transitions between normal and extension
121 bands and at all wrap points. */
122 if (s->spx_atten_code[ch] >= 0) {
123 const float *atten_tab = ff_eac3_spx_atten_tab[s->spx_atten_code[ch]];
124 bin = s->spx_src_start_freq - 2;
125 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
127 float *coeffs = &s->transform_coeffs[ch][bin];
128 coeffs[0] *= atten_tab[0];
129 coeffs[1] *= atten_tab[1];
130 coeffs[2] *= atten_tab[2];
131 coeffs[3] *= atten_tab[1];
132 coeffs[4] *= atten_tab[0];
134 bin += s->spx_band_sizes[bnd];
138 /* Apply noise-blended coefficient scaling based on previously
139 calculated RMS energy, blending factors, and SPX coordinates for
141 bin = s->spx_src_start_freq;
142 for (bnd = 0; bnd < s->num_spx_bands; bnd++) {
143 float nscale = s->spx_noise_blend[ch][bnd] * rms_energy[bnd] * (1.0f / INT32_MIN);
144 float sscale = s->spx_signal_blend[ch][bnd];
145 for (i = 0; i < s->spx_band_sizes[bnd]; i++) {
146 float noise = nscale * (int32_t)av_lfg_get(&s->dith_state);
147 s->transform_coeffs[ch][bin] *= sscale;
148 s->transform_coeffs[ch][bin++] += noise;
155 /** lrint(M_SQRT2*cos(2*M_PI/12)*(1<<23)) */
156 #define COEFF_0 10273905LL
158 /** lrint(M_SQRT2*cos(0*M_PI/12)*(1<<23)) = lrint(M_SQRT2*(1<<23)) */
159 #define COEFF_1 11863283LL
161 /** lrint(M_SQRT2*cos(5*M_PI/12)*(1<<23)) */
162 #define COEFF_2 3070444LL
165 * Calculate 6-point IDCT of the pre-mantissas.
166 * All calculations are 24-bit fixed-point.
168 static void idct6(int pre_mant[6])
171 int even0, even1, even2, odd0, odd1, odd2;
173 odd1 = pre_mant[1] - pre_mant[3] - pre_mant[5];
175 even2 = ( pre_mant[2] * COEFF_0) >> 23;
176 tmp = ( pre_mant[4] * COEFF_1) >> 23;
177 odd0 = ((pre_mant[1] + pre_mant[5]) * COEFF_2) >> 23;
179 even0 = pre_mant[0] + (tmp >> 1);
180 even1 = pre_mant[0] - tmp;
187 odd0 = tmp + pre_mant[1] + pre_mant[3];
188 odd2 = tmp + pre_mant[5] - pre_mant[3];
190 pre_mant[0] = even0 + odd0;
191 pre_mant[1] = even1 + odd1;
192 pre_mant[2] = even2 + odd2;
193 pre_mant[3] = even2 - odd2;
194 pre_mant[4] = even1 - odd1;
195 pre_mant[5] = even0 - odd0;
198 void ff_eac3_decode_transform_coeffs_aht_ch(AC3DecodeContext *s, int ch)
201 int end_bap, gaq_mode;
202 GetBitContext *gbc = &s->gbc;
203 int gaq_gain[AC3_MAX_COEFS];
205 gaq_mode = get_bits(gbc, 2);
206 end_bap = (gaq_mode < 2) ? 12 : 17;
208 /* if GAQ gain is used, decode gain codes for bins with hebap between
211 if (gaq_mode == EAC3_GAQ_12 || gaq_mode == EAC3_GAQ_14) {
212 /* read 1-bit GAQ gain codes */
213 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
214 if (s->bap[ch][bin] > 7 && s->bap[ch][bin] < end_bap)
215 gaq_gain[gs++] = get_bits1(gbc) << (gaq_mode-1);
217 } else if (gaq_mode == EAC3_GAQ_124) {
218 /* read 1.67-bit GAQ gain codes (3 codes in 5 bits) */
220 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
221 if (s->bap[ch][bin] > 7 && s->bap[ch][bin] < 17) {
223 int group_code = get_bits(gbc, 5);
224 if (group_code > 26) {
225 av_log(s->avctx, AV_LOG_WARNING, "GAQ gain group code out-of-range\n");
228 gaq_gain[gs++] = ff_ac3_ungroup_3_in_5_bits_tab[group_code][0];
229 gaq_gain[gs++] = ff_ac3_ungroup_3_in_5_bits_tab[group_code][1];
230 gaq_gain[gs++] = ff_ac3_ungroup_3_in_5_bits_tab[group_code][2];
238 for (bin = s->start_freq[ch]; bin < s->end_freq[ch]; bin++) {
239 int hebap = s->bap[ch][bin];
240 int bits = ff_eac3_bits_vs_hebap[hebap];
242 /* zero-mantissa dithering */
243 for (blk = 0; blk < 6; blk++) {
244 s->pre_mantissa[ch][bin][blk] = (av_lfg_get(&s->dith_state) & 0x7FFFFF) - 0x400000;
246 } else if (hebap < 8) {
247 /* Vector Quantization */
248 int v = get_bits(gbc, bits);
249 for (blk = 0; blk < 6; blk++) {
250 s->pre_mantissa[ch][bin][blk] = ff_eac3_mantissa_vq[hebap][v][blk] << 8;
253 /* Gain Adaptive Quantization */
255 if (gaq_mode != EAC3_GAQ_NO && hebap < end_bap) {
256 log_gain = gaq_gain[gs++];
260 gbits = bits - log_gain;
262 for (blk = 0; blk < 6; blk++) {
263 int mant = get_sbits(gbc, gbits);
264 if (log_gain && mant == -(1 << (gbits-1))) {
267 int mbits = bits - (2 - log_gain);
268 mant = get_sbits(gbc, mbits);
269 mant <<= (23 - (mbits - 1));
270 /* remap mantissa value to correct for asymmetric quantization */
272 b = 1 << (23 - log_gain);
274 b = ff_eac3_gaq_remap_2_4_b[hebap-8][log_gain-1] << 8;
275 mant += ((ff_eac3_gaq_remap_2_4_a[hebap-8][log_gain-1] * (int64_t)mant) >> 15) + b;
277 /* small mantissa, no GAQ, or Gk=1 */
280 /* remap mantissa value for no GAQ or Gk=1 */
281 mant += (ff_eac3_gaq_remap_1[hebap-8] * (int64_t)mant) >> 15;
284 s->pre_mantissa[ch][bin][blk] = mant;
287 idct6(s->pre_mantissa[ch][bin]);
291 int ff_eac3_parse_header(AC3DecodeContext *s)
294 int ac3_exponent_strategy, parse_aht_info, parse_spx_atten_data;
295 int parse_transient_proc_info;
297 GetBitContext *gbc = &s->gbc;
299 /* An E-AC-3 stream can have multiple independent streams which the
300 application can select from. each independent stream can also contain
301 dependent streams which are used to add or replace channels. */
302 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
303 av_log_missing_feature(s->avctx, "Dependent substream decoding", 1);
304 return AAC_AC3_PARSE_ERROR_FRAME_TYPE;
305 } else if (s->frame_type == EAC3_FRAME_TYPE_RESERVED) {
306 av_log(s->avctx, AV_LOG_ERROR, "Reserved frame type\n");
307 return AAC_AC3_PARSE_ERROR_FRAME_TYPE;
310 /* The substream id indicates which substream this frame belongs to. each
311 independent stream has its own substream id, and the dependent streams
312 associated to an independent stream have matching substream id's. */
313 if (s->substreamid) {
314 /* only decode substream with id=0. skip any additional substreams. */
315 av_log_missing_feature(s->avctx, "Additional substreams", 1);
316 return AAC_AC3_PARSE_ERROR_FRAME_TYPE;
319 if (s->bit_alloc_params.sr_code == EAC3_SR_CODE_REDUCED) {
320 /* The E-AC-3 specification does not tell how to handle reduced sample
321 rates in bit allocation. The best assumption would be that it is
322 handled like AC-3 DolbyNet, but we cannot be sure until we have a
323 sample which utilizes this feature. */
324 av_log_missing_feature(s->avctx, "Reduced sampling rates", 1);
327 skip_bits(gbc, 5); // skip bitstream id
329 /* volume control params */
330 for (i = 0; i < (s->channel_mode ? 1 : 2); i++) {
331 skip_bits(gbc, 5); // skip dialog normalization
332 if (get_bits1(gbc)) {
333 skip_bits(gbc, 8); // skip compression gain word
337 /* dependent stream channel map */
338 if (s->frame_type == EAC3_FRAME_TYPE_DEPENDENT) {
339 if (get_bits1(gbc)) {
340 skip_bits(gbc, 16); // skip custom channel map
344 /* mixing metadata */
345 if (get_bits1(gbc)) {
346 /* center and surround mix levels */
347 if (s->channel_mode > AC3_CHMODE_STEREO) {
348 skip_bits(gbc, 2); // skip preferred stereo downmix mode
349 if (s->channel_mode & 1) {
350 /* if three front channels exist */
351 skip_bits(gbc, 3); //skip Lt/Rt center mix level
352 s->center_mix_level = get_bits(gbc, 3);
354 if (s->channel_mode & 4) {
355 /* if a surround channel exists */
356 skip_bits(gbc, 3); //skip Lt/Rt surround mix level
357 s->surround_mix_level = get_bits(gbc, 3);
362 if (s->lfe_on && get_bits1(gbc)) {
363 // TODO: use LFE mix level
364 skip_bits(gbc, 5); // skip LFE mix level code
367 /* info for mixing with other streams and substreams */
368 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT) {
369 for (i = 0; i < (s->channel_mode ? 1 : 2); i++) {
370 // TODO: apply program scale factor
371 if (get_bits1(gbc)) {
372 skip_bits(gbc, 6); // skip program scale factor
375 if (get_bits1(gbc)) {
376 skip_bits(gbc, 6); // skip external program scale factor
378 /* skip mixing parameter data */
379 switch(get_bits(gbc, 2)) {
380 case 1: skip_bits(gbc, 5); break;
381 case 2: skip_bits(gbc, 12); break;
383 int mix_data_size = (get_bits(gbc, 5) + 2) << 3;
384 skip_bits_long(gbc, mix_data_size);
388 /* skip pan information for mono or dual mono source */
389 if (s->channel_mode < AC3_CHMODE_STEREO) {
390 for (i = 0; i < (s->channel_mode ? 1 : 2); i++) {
391 if (get_bits1(gbc)) {
392 /* note: this is not in the ATSC A/52B specification
393 reference: ETSI TS 102 366 V1.1.1
394 section: E.1.3.1.25 */
395 skip_bits(gbc, 8); // skip pan mean direction index
396 skip_bits(gbc, 6); // skip reserved paninfo bits
400 /* skip mixing configuration information */
401 if (get_bits1(gbc)) {
402 for (blk = 0; blk < s->num_blocks; blk++) {
403 if (s->num_blocks == 1 || get_bits1(gbc)) {
411 /* informational metadata */
412 if (get_bits1(gbc)) {
413 s->bitstream_mode = get_bits(gbc, 3);
414 skip_bits(gbc, 2); // skip copyright bit and original bitstream bit
415 if (s->channel_mode == AC3_CHMODE_STEREO) {
416 skip_bits(gbc, 4); // skip Dolby surround and headphone mode
418 if (s->channel_mode >= AC3_CHMODE_2F2R) {
419 skip_bits(gbc, 2); // skip Dolby surround EX mode
421 for (i = 0; i < (s->channel_mode ? 1 : 2); i++) {
422 if (get_bits1(gbc)) {
423 skip_bits(gbc, 8); // skip mix level, room type, and A/D converter type
426 if (s->bit_alloc_params.sr_code != EAC3_SR_CODE_REDUCED) {
427 skip_bits1(gbc); // skip source sample rate code
431 /* converter synchronization flag
432 If frames are less than six blocks, this bit should be turned on
433 once every 6 blocks to indicate the start of a frame set.
434 reference: RFC 4598, Section 2.1.3 Frame Sets */
435 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT && s->num_blocks != 6) {
436 skip_bits1(gbc); // skip converter synchronization flag
439 /* original frame size code if this stream was converted from AC-3 */
440 if (s->frame_type == EAC3_FRAME_TYPE_AC3_CONVERT &&
441 (s->num_blocks == 6 || get_bits1(gbc))) {
442 skip_bits(gbc, 6); // skip frame size code
445 /* additional bitstream info */
446 if (get_bits1(gbc)) {
447 int addbsil = get_bits(gbc, 6);
448 for (i = 0; i < addbsil + 1; i++) {
449 skip_bits(gbc, 8); // skip additional bit stream info
453 /* audio frame syntax flags, strategy data, and per-frame data */
455 if (s->num_blocks == 6) {
456 ac3_exponent_strategy = get_bits1(gbc);
457 parse_aht_info = get_bits1(gbc);
459 /* less than 6 blocks, so use AC-3-style exponent strategy syntax, and
461 ac3_exponent_strategy = 1;
465 s->snr_offset_strategy = get_bits(gbc, 2);
466 parse_transient_proc_info = get_bits1(gbc);
468 s->block_switch_syntax = get_bits1(gbc);
469 if (!s->block_switch_syntax)
470 memset(s->block_switch, 0, sizeof(s->block_switch));
472 s->dither_flag_syntax = get_bits1(gbc);
473 if (!s->dither_flag_syntax) {
474 for (ch = 1; ch <= s->fbw_channels; ch++)
475 s->dither_flag[ch] = 1;
477 s->dither_flag[CPL_CH] = s->dither_flag[s->lfe_ch] = 0;
479 s->bit_allocation_syntax = get_bits1(gbc);
480 if (!s->bit_allocation_syntax) {
481 /* set default bit allocation parameters */
482 s->bit_alloc_params.slow_decay = ff_ac3_slow_decay_tab[2];
483 s->bit_alloc_params.fast_decay = ff_ac3_fast_decay_tab[1];
484 s->bit_alloc_params.slow_gain = ff_ac3_slow_gain_tab [1];
485 s->bit_alloc_params.db_per_bit = ff_ac3_db_per_bit_tab[2];
486 s->bit_alloc_params.floor = ff_ac3_floor_tab [7];
489 s->fast_gain_syntax = get_bits1(gbc);
490 s->dba_syntax = get_bits1(gbc);
491 s->skip_syntax = get_bits1(gbc);
492 parse_spx_atten_data = get_bits1(gbc);
494 /* coupling strategy occurance and coupling use per block */
496 if (s->channel_mode > 1) {
497 for (blk = 0; blk < s->num_blocks; blk++) {
498 s->cpl_strategy_exists[blk] = (!blk || get_bits1(gbc));
499 if (s->cpl_strategy_exists[blk]) {
500 s->cpl_in_use[blk] = get_bits1(gbc);
502 s->cpl_in_use[blk] = s->cpl_in_use[blk-1];
504 num_cpl_blocks += s->cpl_in_use[blk];
507 memset(s->cpl_in_use, 0, sizeof(s->cpl_in_use));
510 /* exponent strategy data */
511 if (ac3_exponent_strategy) {
512 /* AC-3-style exponent strategy syntax */
513 for (blk = 0; blk < s->num_blocks; blk++) {
514 for (ch = !s->cpl_in_use[blk]; ch <= s->fbw_channels; ch++) {
515 s->exp_strategy[blk][ch] = get_bits(gbc, 2);
519 /* LUT-based exponent strategy syntax */
520 for (ch = !((s->channel_mode > 1) && num_cpl_blocks); ch <= s->fbw_channels; ch++) {
521 int frmchexpstr = get_bits(gbc, 5);
522 for (blk = 0; blk < 6; blk++) {
523 s->exp_strategy[blk][ch] = ff_eac3_frm_expstr[frmchexpstr][blk];
527 /* LFE exponent strategy */
529 for (blk = 0; blk < s->num_blocks; blk++) {
530 s->exp_strategy[blk][s->lfe_ch] = get_bits1(gbc);
533 /* original exponent strategies if this stream was converted from AC-3 */
534 if (s->frame_type == EAC3_FRAME_TYPE_INDEPENDENT &&
535 (s->num_blocks == 6 || get_bits1(gbc))) {
536 skip_bits(gbc, 5 * s->fbw_channels); // skip converter channel exponent strategy
539 /* determine which channels use AHT */
540 if (parse_aht_info) {
541 /* For AHT to be used, all non-zero blocks must reuse exponents from
542 the first block. Furthermore, for AHT to be used in the coupling
543 channel, all blocks must use coupling and use the same coupling
545 s->channel_uses_aht[CPL_CH]=0;
546 for (ch = (num_cpl_blocks != 6); ch <= s->channels; ch++) {
548 for (blk = 1; blk < 6; blk++) {
549 if ((s->exp_strategy[blk][ch] != EXP_REUSE) ||
550 (!ch && s->cpl_strategy_exists[blk])) {
555 s->channel_uses_aht[ch] = use_aht && get_bits1(gbc);
558 memset(s->channel_uses_aht, 0, sizeof(s->channel_uses_aht));
561 /* per-frame SNR offset */
562 if (!s->snr_offset_strategy) {
563 int csnroffst = (get_bits(gbc, 6) - 15) << 4;
564 int snroffst = (csnroffst + get_bits(gbc, 4)) << 2;
565 for (ch = 0; ch <= s->channels; ch++)
566 s->snr_offset[ch] = snroffst;
569 /* transient pre-noise processing data */
570 if (parse_transient_proc_info) {
571 for (ch = 1; ch <= s->fbw_channels; ch++) {
572 if (get_bits1(gbc)) { // channel in transient processing
573 skip_bits(gbc, 10); // skip transient processing location
574 skip_bits(gbc, 8); // skip transient processing length
579 /* spectral extension attenuation data */
580 for (ch = 1; ch <= s->fbw_channels; ch++) {
581 if (parse_spx_atten_data && get_bits1(gbc)) {
582 s->spx_atten_code[ch] = get_bits(gbc, 5);
584 s->spx_atten_code[ch] = -1;
588 /* block start information */
589 if (s->num_blocks > 1 && get_bits1(gbc)) {
590 /* reference: Section E2.3.2.27
591 nblkstrtbits = (numblks - 1) * (4 + ceiling(log2(words_per_frame)))
592 The spec does not say what this data is or what it's used for.
593 It is likely the offset of each block within the frame. */
594 int block_start_bits = (s->num_blocks-1) * (4 + av_log2(s->frame_size-2));
595 skip_bits_long(gbc, block_start_bits);
596 av_log_missing_feature(s->avctx, "Block start info", 1);
599 /* syntax state initialization */
600 for (ch = 1; ch <= s->fbw_channels; ch++) {
601 s->first_spx_coords[ch] = 1;
602 s->first_cpl_coords[ch] = 1;
604 s->first_cpl_leak = 1;