3 * Copyright (C) 2008-2012 Alexander E. Patrakov
4 * 2010 Benjamin Larsson
7 * This file is part of FFmpeg.
9 * FFmpeg is free software; you can redistribute it and/or
10 * modify it under the terms of the GNU Lesser General Public
11 * License as published by the Free Software Foundation; either
12 * version 2.1 of the License, or (at your option) any later version.
14 * FFmpeg is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
17 * Lesser General Public License for more details.
19 * You should have received a copy of the GNU Lesser General Public
20 * License along with FFmpeg; if not, write to the Free Software
21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
24 #include "libavutil/avassert.h"
25 #include "libavutil/channel_layout.h"
26 #include "libavutil/common.h"
27 #include "libavutil/internal.h"
36 #define MAX_CHANNELS 6
37 #define DCA_MAX_FRAME_SIZE 16384
38 #define DCA_HEADER_SIZE 13
39 #define DCA_LFE_SAMPLES 8
41 #define DCAENC_SUBBANDS 32
43 #define SUBSUBFRAMES 2
44 #define SUBBAND_SAMPLES (SUBFRAMES * SUBSUBFRAMES * 8)
47 typedef struct DCAEncContext {
51 int fullband_channels;
57 const int32_t *band_interpolation;
58 const int32_t *band_spectrum;
62 const int8_t *channel_order_tab; ///< channel reordering table, lfe and non lfe
64 int32_t history[512][MAX_CHANNELS]; /* This is a circular buffer */
65 int32_t subband[SUBBAND_SAMPLES][DCAENC_SUBBANDS][MAX_CHANNELS];
66 int32_t quantized[SUBBAND_SAMPLES][DCAENC_SUBBANDS][MAX_CHANNELS];
67 int32_t peak_cb[DCAENC_SUBBANDS][MAX_CHANNELS];
68 int32_t downsampled_lfe[DCA_LFE_SAMPLES];
69 int32_t masking_curve_cb[SUBSUBFRAMES][256];
70 int abits[DCAENC_SUBBANDS][MAX_CHANNELS];
71 int scale_factor[DCAENC_SUBBANDS][MAX_CHANNELS];
72 softfloat quant[DCAENC_SUBBANDS][MAX_CHANNELS];
73 int32_t eff_masking_curve_cb[256];
74 int32_t band_masking_cb[32];
75 int32_t worst_quantization_noise;
76 int32_t worst_noise_ever;
80 static int32_t cos_table[2048];
81 static int32_t band_interpolation[2][512];
82 static int32_t band_spectrum[2][8];
83 static int32_t auf[9][AUBANDS][256];
84 static int32_t cb_to_add[256];
85 static int32_t cb_to_level[2048];
86 static int32_t lfe_fir_64i[512];
88 /* Transfer function of outer and middle ear, Hz -> dB */
89 static double hom(double f)
93 return -3.64 * pow(f1, -0.8)
94 + 6.8 * exp(-0.6 * (f1 - 3.4) * (f1 - 3.4))
95 - 6.0 * exp(-0.15 * (f1 - 8.7) * (f1 - 8.7))
96 - 0.0006 * (f1 * f1) * (f1 * f1);
99 static double gammafilter(int i, double f)
101 double h = (f - fc[i]) / erb[i];
105 return 20 * log10(h);
108 static int encode_init(AVCodecContext *avctx)
110 DCAEncContext *c = avctx->priv_data;
111 uint64_t layout = avctx->channel_layout;
112 int i, min_frame_bits;
114 c->fullband_channels = c->channels = avctx->channels;
115 c->lfe_channel = (avctx->channels == 3 || avctx->channels == 6);
116 c->band_interpolation = band_interpolation[1];
117 c->band_spectrum = band_spectrum[1];
118 c->worst_quantization_noise = -2047;
119 c->worst_noise_ever = -2047;
122 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
123 "encoder will guess the layout, but it "
124 "might be incorrect.\n");
125 layout = av_get_default_channel_layout(avctx->channels);
128 case AV_CH_LAYOUT_MONO: c->channel_config = 0; break;
129 case AV_CH_LAYOUT_STEREO: c->channel_config = 2; break;
130 case AV_CH_LAYOUT_2_2: c->channel_config = 8; break;
131 case AV_CH_LAYOUT_5POINT0: c->channel_config = 9; break;
132 case AV_CH_LAYOUT_5POINT1: c->channel_config = 9; break;
134 av_log(avctx, AV_LOG_ERROR, "Unsupported channel layout!\n");
135 return AVERROR_PATCHWELCOME;
138 if (c->lfe_channel) {
139 c->fullband_channels--;
140 c->channel_order_tab = ff_dca_channel_reorder_lfe[c->channel_config];
142 c->channel_order_tab = ff_dca_channel_reorder_nolfe[c->channel_config];
145 for (i = 0; i < 9; i++) {
146 if (sample_rates[i] == avctx->sample_rate)
150 return AVERROR(EINVAL);
151 c->samplerate_index = i;
153 if (avctx->bit_rate < 32000 || avctx->bit_rate > 3840000) {
154 av_log(avctx, AV_LOG_ERROR, "Bit rate %"PRId64" not supported.", (int64_t)avctx->bit_rate);
155 return AVERROR(EINVAL);
157 for (i = 0; ff_dca_bit_rates[i] < avctx->bit_rate; i++)
159 c->bitrate_index = i;
160 avctx->bit_rate = ff_dca_bit_rates[i];
161 c->frame_bits = FFALIGN((avctx->bit_rate * 512 + avctx->sample_rate - 1) / avctx->sample_rate, 32);
162 min_frame_bits = 132 + (493 + 28 * 32) * c->fullband_channels + c->lfe_channel * 72;
163 if (c->frame_bits < min_frame_bits || c->frame_bits > (DCA_MAX_FRAME_SIZE << 3))
164 return AVERROR(EINVAL);
166 c->frame_size = (c->frame_bits + 7) / 8;
168 avctx->frame_size = 32 * SUBBAND_SAMPLES;
173 cos_table[0] = 0x7fffffff;
175 cos_table[1024] = -cos_table[0];
176 for (i = 1; i < 512; i++) {
177 cos_table[i] = (int32_t)(0x7fffffff * cos(M_PI * i / 1024));
178 cos_table[1024-i] = -cos_table[i];
179 cos_table[1024+i] = -cos_table[i];
180 cos_table[2048-i] = cos_table[i];
182 for (i = 0; i < 2048; i++) {
183 cb_to_level[i] = (int32_t)(0x7fffffff * ff_exp10(-0.005 * i));
186 for (k = 0; k < 32; k++) {
187 for (j = 0; j < 8; j++) {
188 lfe_fir_64i[64 * j + k] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);
189 lfe_fir_64i[64 * (7-j) + (63 - k)] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);
193 for (i = 0; i < 512; i++) {
194 band_interpolation[0][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_perfect[i]);
195 band_interpolation[1][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_nonperfect[i]);
198 for (i = 0; i < 9; i++) {
199 for (j = 0; j < AUBANDS; j++) {
200 for (k = 0; k < 256; k++) {
201 double freq = sample_rates[i] * (k + 0.5) / 512;
203 auf[i][j][k] = (int32_t)(10 * (hom(freq) + gammafilter(j, freq)));
208 for (i = 0; i < 256; i++) {
209 double add = 1 + ff_exp10(-0.01 * i);
210 cb_to_add[i] = (int32_t)(100 * log10(add));
212 for (j = 0; j < 8; j++) {
214 for (i = 0; i < 512; i++) {
215 double reconst = ff_dca_fir_32bands_perfect[i] * ((i & 64) ? (-1) : 1);
216 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);
218 band_spectrum[0][j] = (int32_t)(200 * log10(accum));
220 for (j = 0; j < 8; j++) {
222 for (i = 0; i < 512; i++) {
223 double reconst = ff_dca_fir_32bands_nonperfect[i] * ((i & 64) ? (-1) : 1);
224 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);
226 band_spectrum[1][j] = (int32_t)(200 * log10(accum));
232 static inline int32_t cos_t(int x)
234 return cos_table[x & 2047];
237 static inline int32_t sin_t(int x)
239 return cos_t(x - 512);
242 static inline int32_t half32(int32_t a)
247 static inline int32_t mul32(int32_t a, int32_t b)
249 int64_t r = (int64_t)a * b + 0x80000000ULL;
253 static void subband_transform(DCAEncContext *c, const int32_t *input)
255 int ch, subs, i, k, j;
257 for (ch = 0; ch < c->fullband_channels; ch++) {
258 /* History is copied because it is also needed for PSY */
261 const int chi = c->channel_order_tab[ch];
263 for (i = 0; i < 512; i++)
264 hist[i] = c->history[i][ch];
266 for (subs = 0; subs < SUBBAND_SAMPLES; subs++) {
271 /* Calculate the convolutions at once */
272 for (i = 0; i < 64; i++)
275 for (k = 0, i = hist_start, j = 0;
276 i < 512; k = (k + 1) & 63, i++, j++)
277 accum[k] += mul32(hist[i], c->band_interpolation[j]);
278 for (i = 0; i < hist_start; k = (k + 1) & 63, i++, j++)
279 accum[k] += mul32(hist[i], c->band_interpolation[j]);
281 for (k = 16; k < 32; k++)
282 accum[k] = accum[k] - accum[31 - k];
283 for (k = 32; k < 48; k++)
284 accum[k] = accum[k] + accum[95 - k];
286 for (band = 0; band < 32; band++) {
288 for (i = 16; i < 48; i++) {
289 int s = (2 * band + 1) * (2 * (i + 16) + 1);
290 resp += mul32(accum[i], cos_t(s << 3)) >> 3;
293 c->subband[subs][band][ch] = ((band + 1) & 2) ? -resp : resp;
296 /* Copy in 32 new samples from input */
297 for (i = 0; i < 32; i++)
298 hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi];
299 hist_start = (hist_start + 32) & 511;
304 static void lfe_downsample(DCAEncContext *c, const int32_t *input)
306 /* FIXME: make 128x LFE downsampling possible */
307 const int lfech = ff_dca_lfe_index[c->channel_config];
313 for (i = 0; i < 512; i++)
314 hist[i] = c->history[i][c->channels - 1];
316 for (lfes = 0; lfes < DCA_LFE_SAMPLES; lfes++) {
317 /* Calculate the convolution */
320 for (i = hist_start, j = 0; i < 512; i++, j++)
321 accum += mul32(hist[i], lfe_fir_64i[j]);
322 for (i = 0; i < hist_start; i++, j++)
323 accum += mul32(hist[i], lfe_fir_64i[j]);
325 c->downsampled_lfe[lfes] = accum;
327 /* Copy in 64 new samples from input */
328 for (i = 0; i < 64; i++)
329 hist[i + hist_start] = input[(lfes * 64 + i) * c->channels + lfech];
331 hist_start = (hist_start + 64) & 511;
340 static void fft(const int32_t in[2 * 256], cplx32 out[256])
342 cplx32 buf[256], rin[256], rout[256];
345 /* do two transforms in parallel */
346 for (i = 0; i < 256; i++) {
347 /* Apply the Hann window */
348 rin[i].re = mul32(in[2 * i], 0x3fffffff - (cos_t(8 * i + 2) >> 1));
349 rin[i].im = mul32(in[2 * i + 1], 0x3fffffff - (cos_t(8 * i + 6) >> 1));
352 for (i = 0; i < 256; i++) {
353 buf[i].re = mul32(cos_t(4 * i + 2), rin[i].re)
354 - mul32(sin_t(4 * i + 2), rin[i].im);
355 buf[i].im = mul32(cos_t(4 * i + 2), rin[i].im)
356 + mul32(sin_t(4 * i + 2), rin[i].re);
359 for (j = 256, l = 1; j != 1; j >>= 1, l <<= 1) {
360 for (k = 0; k < 256; k += j) {
361 for (i = k; i < k + j / 2; i++) {
365 sum.re = buf[i].re + buf[i + j / 2].re;
366 sum.im = buf[i].im + buf[i + j / 2].im;
368 diff.re = buf[i].re - buf[i + j / 2].re;
369 diff.im = buf[i].im - buf[i + j / 2].im;
371 buf[i].re = half32(sum.re);
372 buf[i].im = half32(sum.im);
374 buf[i + j / 2].re = mul32(diff.re, cos_t(t))
375 - mul32(diff.im, sin_t(t));
376 buf[i + j / 2].im = mul32(diff.im, cos_t(t))
377 + mul32(diff.re, sin_t(t));
382 for (i = 0; i < 256; i++) {
383 int b = ff_reverse[i];
384 rout[i].re = mul32(buf[b].re, cos_t(4 * i))
385 - mul32(buf[b].im, sin_t(4 * i));
386 rout[i].im = mul32(buf[b].im, cos_t(4 * i))
387 + mul32(buf[b].re, sin_t(4 * i));
389 for (i = 0; i < 256; i++) {
390 /* separate the results of the two transforms */
393 o1.re = rout[i].re - rout[255 - i].re;
394 o1.im = rout[i].im + rout[255 - i].im;
396 o2.re = rout[i].im - rout[255 - i].im;
397 o2.im = -rout[i].re - rout[255 - i].re;
399 /* combine them into one long transform */
400 out[i].re = mul32( o1.re + o2.re, cos_t(2 * i + 1))
401 + mul32( o1.im - o2.im, sin_t(2 * i + 1));
402 out[i].im = mul32( o1.im + o2.im, cos_t(2 * i + 1))
403 + mul32(-o1.re + o2.re, sin_t(2 * i + 1));
407 static int32_t get_cb(int32_t in)
414 for (i = 1024; i > 0; i >>= 1) {
415 if (cb_to_level[i + res] >= in)
421 static int32_t add_cb(int32_t a, int32_t b)
424 FFSWAP(int32_t, a, b);
428 return a + cb_to_add[a - b];
431 static void adjust_jnd(int samplerate_index,
432 const int32_t in[512], int32_t out_cb[256])
436 int32_t out_cb_unnorm[256];
438 const int32_t ca_cb = -1114;
439 const int32_t cs_cb = 928;
444 for (j = 0; j < 256; j++) {
445 power[j] = add_cb(get_cb(out[j].re), get_cb(out[j].im));
446 out_cb_unnorm[j] = -2047; /* and can only grow */
449 for (i = 0; i < AUBANDS; i++) {
450 denom = ca_cb; /* and can only grow */
451 for (j = 0; j < 256; j++)
452 denom = add_cb(denom, power[j] + auf[samplerate_index][i][j]);
453 for (j = 0; j < 256; j++)
454 out_cb_unnorm[j] = add_cb(out_cb_unnorm[j],
455 -denom + auf[samplerate_index][i][j]);
458 for (j = 0; j < 256; j++)
459 out_cb[j] = add_cb(out_cb[j], -out_cb_unnorm[j] - ca_cb - cs_cb);
462 typedef void (*walk_band_t)(DCAEncContext *c, int band1, int band2, int f,
463 int32_t spectrum1, int32_t spectrum2, int channel,
466 static void walk_band_low(DCAEncContext *c, int band, int channel,
467 walk_band_t walk, int32_t *arg)
472 for (f = 0; f < 4; f++)
473 walk(c, 0, 0, f, 0, -2047, channel, arg);
475 for (f = 0; f < 8; f++)
476 walk(c, band, band - 1, 8 * band - 4 + f,
477 c->band_spectrum[7 - f], c->band_spectrum[f], channel, arg);
481 static void walk_band_high(DCAEncContext *c, int band, int channel,
482 walk_band_t walk, int32_t *arg)
487 for (f = 0; f < 4; f++)
488 walk(c, 31, 31, 256 - 4 + f, 0, -2047, channel, arg);
490 for (f = 0; f < 8; f++)
491 walk(c, band, band + 1, 8 * band + 4 + f,
492 c->band_spectrum[f], c->band_spectrum[7 - f], channel, arg);
496 static void update_band_masking(DCAEncContext *c, int band1, int band2,
497 int f, int32_t spectrum1, int32_t spectrum2,
498 int channel, int32_t * arg)
500 int32_t value = c->eff_masking_curve_cb[f] - spectrum1;
502 if (value < c->band_masking_cb[band1])
503 c->band_masking_cb[band1] = value;
506 static void calc_masking(DCAEncContext *c, const int32_t *input)
508 int i, k, band, ch, ssf;
511 for (i = 0; i < 256; i++)
512 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
513 c->masking_curve_cb[ssf][i] = -2047;
515 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
516 for (ch = 0; ch < c->fullband_channels; ch++) {
517 const int chi = c->channel_order_tab[ch];
519 for (i = 0, k = 128 + 256 * ssf; k < 512; i++, k++)
520 data[i] = c->history[k][ch];
521 for (k -= 512; i < 512; i++, k++)
522 data[i] = input[k * c->channels + chi];
523 adjust_jnd(c->samplerate_index, data, c->masking_curve_cb[ssf]);
525 for (i = 0; i < 256; i++) {
528 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
529 if (c->masking_curve_cb[ssf][i] < m)
530 m = c->masking_curve_cb[ssf][i];
531 c->eff_masking_curve_cb[i] = m;
534 for (band = 0; band < 32; band++) {
535 c->band_masking_cb[band] = 2048;
536 walk_band_low(c, band, 0, update_band_masking, NULL);
537 walk_band_high(c, band, 0, update_band_masking, NULL);
541 static void find_peaks(DCAEncContext *c)
545 for (band = 0; band < 32; band++)
546 for (ch = 0; ch < c->fullband_channels; ch++) {
550 for (sample = 0; sample < SUBBAND_SAMPLES; sample++) {
551 int32_t s = abs(c->subband[sample][band][ch]);
555 c->peak_cb[band][ch] = get_cb(m);
558 if (c->lfe_channel) {
562 for (sample = 0; sample < DCA_LFE_SAMPLES; sample++)
563 if (m < abs(c->downsampled_lfe[sample]))
564 m = abs(c->downsampled_lfe[sample]);
565 c->lfe_peak_cb = get_cb(m);
569 static const int snr_fudge = 128;
570 #define USED_1ABITS 1
571 #define USED_NABITS 2
572 #define USED_26ABITS 4
574 static int init_quantization_noise(DCAEncContext *c, int noise)
576 int ch, band, ret = 0;
578 c->consumed_bits = 132 + 493 * c->fullband_channels;
580 c->consumed_bits += 72;
582 /* attempt to guess the bit distribution based on the prevoius frame */
583 for (ch = 0; ch < c->fullband_channels; ch++) {
584 for (band = 0; band < 32; band++) {
585 int snr_cb = c->peak_cb[band][ch] - c->band_masking_cb[band] - noise;
587 if (snr_cb >= 1312) {
588 c->abits[band][ch] = 26;
590 } else if (snr_cb >= 222) {
591 c->abits[band][ch] = 8 + mul32(snr_cb - 222, 69000000);
593 } else if (snr_cb >= 0) {
594 c->abits[band][ch] = 2 + mul32(snr_cb, 106000000);
597 c->abits[band][ch] = 1;
603 for (band = 0; band < 32; band++)
604 for (ch = 0; ch < c->fullband_channels; ch++) {
605 c->consumed_bits += bit_consumption[c->abits[band][ch]];
611 static void assign_bits(DCAEncContext *c)
613 /* Find the bounds where the binary search should work */
617 init_quantization_noise(c, c->worst_quantization_noise);
618 low = high = c->worst_quantization_noise;
619 if (c->consumed_bits > c->frame_bits) {
620 while (c->consumed_bits > c->frame_bits) {
621 av_assert0(used_abits != USED_1ABITS);
624 used_abits = init_quantization_noise(c, high);
627 while (c->consumed_bits <= c->frame_bits) {
629 if (used_abits == USED_26ABITS)
630 goto out; /* The requested bitrate is too high, pad with zeros */
632 used_abits = init_quantization_noise(c, low);
636 /* Now do a binary search between low and high to see what fits */
637 for (down = snr_fudge >> 1; down; down >>= 1) {
638 init_quantization_noise(c, high - down);
639 if (c->consumed_bits <= c->frame_bits)
642 init_quantization_noise(c, high);
644 c->worst_quantization_noise = high;
645 if (high > c->worst_noise_ever)
646 c->worst_noise_ever = high;
649 static void shift_history(DCAEncContext *c, const int32_t *input)
653 for (k = 0; k < 512; k++)
654 for (ch = 0; ch < c->channels; ch++) {
655 const int chi = c->channel_order_tab[ch];
657 c->history[k][ch] = input[k * c->channels + chi];
661 static int32_t quantize_value(int32_t value, softfloat quant)
663 int32_t offset = 1 << (quant.e - 1);
665 value = mul32(value, quant.m) + offset;
666 value = value >> quant.e;
670 static int calc_one_scale(int32_t peak_cb, int abits, softfloat *quant)
673 int our_nscale, try_remove;
676 av_assert0(peak_cb <= 0);
677 av_assert0(peak_cb >= -2047);
680 peak = cb_to_level[-peak_cb];
682 for (try_remove = 64; try_remove > 0; try_remove >>= 1) {
683 if (scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e <= 17)
685 our_quant.m = mul32(scalefactor_inv[our_nscale - try_remove].m, stepsize_inv[abits].m);
686 our_quant.e = scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e - 17;
687 if ((quant_levels[abits] - 1) / 2 < quantize_value(peak, our_quant))
689 our_nscale -= try_remove;
692 if (our_nscale >= 125)
695 quant->m = mul32(scalefactor_inv[our_nscale].m, stepsize_inv[abits].m);
696 quant->e = scalefactor_inv[our_nscale].e + stepsize_inv[abits].e - 17;
697 av_assert0((quant_levels[abits] - 1) / 2 >= quantize_value(peak, *quant));
702 static void calc_scales(DCAEncContext *c)
706 for (band = 0; band < 32; band++)
707 for (ch = 0; ch < c->fullband_channels; ch++)
708 c->scale_factor[band][ch] = calc_one_scale(c->peak_cb[band][ch],
710 &c->quant[band][ch]);
713 c->lfe_scale_factor = calc_one_scale(c->lfe_peak_cb, 11, &c->lfe_quant);
716 static void quantize_all(DCAEncContext *c)
718 int sample, band, ch;
720 for (sample = 0; sample < SUBBAND_SAMPLES; sample++)
721 for (band = 0; band < 32; band++)
722 for (ch = 0; ch < c->fullband_channels; ch++)
723 c->quantized[sample][band][ch] = quantize_value(c->subband[sample][band][ch], c->quant[band][ch]);
726 static void put_frame_header(DCAEncContext *c)
729 put_bits(&c->pb, 16, 0x7ffe);
730 put_bits(&c->pb, 16, 0x8001);
732 /* Frame type: normal */
733 put_bits(&c->pb, 1, 1);
735 /* Deficit sample count: none */
736 put_bits(&c->pb, 5, 31);
738 /* CRC is not present */
739 put_bits(&c->pb, 1, 0);
741 /* Number of PCM sample blocks */
742 put_bits(&c->pb, 7, SUBBAND_SAMPLES - 1);
744 /* Primary frame byte size */
745 put_bits(&c->pb, 14, c->frame_size - 1);
747 /* Audio channel arrangement */
748 put_bits(&c->pb, 6, c->channel_config);
750 /* Core audio sampling frequency */
751 put_bits(&c->pb, 4, bitstream_sfreq[c->samplerate_index]);
753 /* Transmission bit rate */
754 put_bits(&c->pb, 5, c->bitrate_index);
756 /* Embedded down mix: disabled */
757 put_bits(&c->pb, 1, 0);
759 /* Embedded dynamic range flag: not present */
760 put_bits(&c->pb, 1, 0);
762 /* Embedded time stamp flag: not present */
763 put_bits(&c->pb, 1, 0);
765 /* Auxiliary data flag: not present */
766 put_bits(&c->pb, 1, 0);
768 /* HDCD source: no */
769 put_bits(&c->pb, 1, 0);
771 /* Extension audio ID: N/A */
772 put_bits(&c->pb, 3, 0);
774 /* Extended audio data: not present */
775 put_bits(&c->pb, 1, 0);
777 /* Audio sync word insertion flag: after each sub-frame */
778 put_bits(&c->pb, 1, 0);
780 /* Low frequency effects flag: not present or 64x subsampling */
781 put_bits(&c->pb, 2, c->lfe_channel ? 2 : 0);
783 /* Predictor history switch flag: on */
784 put_bits(&c->pb, 1, 1);
787 /* Multirate interpolator switch: non-perfect reconstruction */
788 put_bits(&c->pb, 1, 0);
790 /* Encoder software revision: 7 */
791 put_bits(&c->pb, 4, 7);
793 /* Copy history: 0 */
794 put_bits(&c->pb, 2, 0);
796 /* Source PCM resolution: 16 bits, not DTS ES */
797 put_bits(&c->pb, 3, 0);
799 /* Front sum/difference coding: no */
800 put_bits(&c->pb, 1, 0);
802 /* Surrounds sum/difference coding: no */
803 put_bits(&c->pb, 1, 0);
805 /* Dialog normalization: 0 dB */
806 put_bits(&c->pb, 4, 0);
809 static void put_primary_audio_header(DCAEncContext *c)
811 static const int bitlen[11] = { 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3 };
812 static const int thr[11] = { 0, 1, 3, 3, 3, 3, 7, 7, 7, 7, 7 };
815 /* Number of subframes */
816 put_bits(&c->pb, 4, SUBFRAMES - 1);
818 /* Number of primary audio channels */
819 put_bits(&c->pb, 3, c->fullband_channels - 1);
821 /* Subband activity count */
822 for (ch = 0; ch < c->fullband_channels; ch++)
823 put_bits(&c->pb, 5, DCAENC_SUBBANDS - 2);
825 /* High frequency VQ start subband */
826 for (ch = 0; ch < c->fullband_channels; ch++)
827 put_bits(&c->pb, 5, DCAENC_SUBBANDS - 1);
829 /* Joint intensity coding index: 0, 0 */
830 for (ch = 0; ch < c->fullband_channels; ch++)
831 put_bits(&c->pb, 3, 0);
833 /* Transient mode codebook: A4, A4 (arbitrary) */
834 for (ch = 0; ch < c->fullband_channels; ch++)
835 put_bits(&c->pb, 2, 0);
837 /* Scale factor code book: 7 bit linear, 7-bit sqrt table (for each channel) */
838 for (ch = 0; ch < c->fullband_channels; ch++)
839 put_bits(&c->pb, 3, 6);
841 /* Bit allocation quantizer select: linear 5-bit */
842 for (ch = 0; ch < c->fullband_channels; ch++)
843 put_bits(&c->pb, 3, 6);
845 /* Quantization index codebook select: dummy data
846 to avoid transmission of scale factor adjustment */
847 for (i = 1; i < 11; i++)
848 for (ch = 0; ch < c->fullband_channels; ch++)
849 put_bits(&c->pb, bitlen[i], thr[i]);
851 /* Scale factor adjustment index: not transmitted */
852 /* Audio header CRC check word: not transmitted */
855 static void put_subframe_samples(DCAEncContext *c, int ss, int band, int ch)
857 if (c->abits[band][ch] <= 7) {
859 for (i = 0; i < 8; i += 4) {
861 for (j = 3; j >= 0; j--) {
862 sum *= quant_levels[c->abits[band][ch]];
863 sum += c->quantized[ss * 8 + i + j][band][ch];
864 sum += (quant_levels[c->abits[band][ch]] - 1) / 2;
866 put_bits(&c->pb, bit_consumption[c->abits[band][ch]] / 4, sum);
870 for (i = 0; i < 8; i++) {
871 int bits = bit_consumption[c->abits[band][ch]] / 16;
872 put_sbits(&c->pb, bits, c->quantized[ss * 8 + i][band][ch]);
877 static void put_subframe(DCAEncContext *c, int subframe)
881 /* Subsubframes count */
882 put_bits(&c->pb, 2, SUBSUBFRAMES -1);
884 /* Partial subsubframe sample count: dummy */
885 put_bits(&c->pb, 3, 0);
887 /* Prediction mode: no ADPCM, in each channel and subband */
888 for (ch = 0; ch < c->fullband_channels; ch++)
889 for (band = 0; band < DCAENC_SUBBANDS; band++)
890 put_bits(&c->pb, 1, 0);
892 /* Prediction VQ address: not transmitted */
893 /* Bit allocation index */
894 for (ch = 0; ch < c->fullband_channels; ch++)
895 for (band = 0; band < DCAENC_SUBBANDS; band++)
896 put_bits(&c->pb, 5, c->abits[band][ch]);
898 if (SUBSUBFRAMES > 1) {
899 /* Transition mode: none for each channel and subband */
900 for (ch = 0; ch < c->fullband_channels; ch++)
901 for (band = 0; band < DCAENC_SUBBANDS; band++)
902 put_bits(&c->pb, 1, 0); /* codebook A4 */
906 for (ch = 0; ch < c->fullband_channels; ch++)
907 for (band = 0; band < DCAENC_SUBBANDS; band++)
908 put_bits(&c->pb, 7, c->scale_factor[band][ch]);
910 /* Joint subband scale factor codebook select: not transmitted */
911 /* Scale factors for joint subband coding: not transmitted */
912 /* Stereo down-mix coefficients: not transmitted */
913 /* Dynamic range coefficient: not transmitted */
914 /* Stde information CRC check word: not transmitted */
915 /* VQ encoded high frequency subbands: not transmitted */
917 /* LFE data: 8 samples and scalefactor */
918 if (c->lfe_channel) {
919 for (i = 0; i < DCA_LFE_SAMPLES; i++)
920 put_bits(&c->pb, 8, quantize_value(c->downsampled_lfe[i], c->lfe_quant) & 0xff);
921 put_bits(&c->pb, 8, c->lfe_scale_factor);
924 /* Audio data (subsubframes) */
925 for (ss = 0; ss < SUBSUBFRAMES ; ss++)
926 for (ch = 0; ch < c->fullband_channels; ch++)
927 for (band = 0; band < DCAENC_SUBBANDS; band++)
928 put_subframe_samples(c, ss, band, ch);
931 put_bits(&c->pb, 16, 0xffff);
934 static int encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
935 const AVFrame *frame, int *got_packet_ptr)
937 DCAEncContext *c = avctx->priv_data;
938 const int32_t *samples;
941 if ((ret = ff_alloc_packet2(avctx, avpkt, c->frame_size , 0)) < 0)
944 samples = (const int32_t *)frame->data[0];
946 subband_transform(c, samples);
948 lfe_downsample(c, samples);
950 calc_masking(c, samples);
955 shift_history(c, samples);
957 init_put_bits(&c->pb, avpkt->data, avpkt->size);
959 put_primary_audio_header(c);
960 for (i = 0; i < SUBFRAMES; i++)
964 for (i = put_bits_count(&c->pb); i < 8*c->frame_size; i++)
965 put_bits(&c->pb, 1, 0);
967 flush_put_bits(&c->pb);
969 avpkt->pts = frame->pts;
970 avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
971 avpkt->size = c->frame_size + 1;
976 static const AVCodecDefault defaults[] = {
981 AVCodec ff_dca_encoder = {
983 .long_name = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
984 .type = AVMEDIA_TYPE_AUDIO,
985 .id = AV_CODEC_ID_DTS,
986 .priv_data_size = sizeof(DCAEncContext),
988 .encode2 = encode_frame,
989 .capabilities = AV_CODEC_CAP_EXPERIMENTAL,
990 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32,
991 AV_SAMPLE_FMT_NONE },
992 .supported_samplerates = sample_rates,
993 .channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_MONO,
996 AV_CH_LAYOUT_5POINT0,
997 AV_CH_LAYOUT_5POINT1,
999 .defaults = defaults,