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"
35 #define MAX_CHANNELS 6
36 #define DCA_MAX_FRAME_SIZE 16384
37 #define DCA_HEADER_SIZE 13
38 #define DCA_LFE_SAMPLES 8
40 #define DCAENC_SUBBANDS 32
42 #define SUBSUBFRAMES 2
43 #define SUBBAND_SAMPLES (SUBFRAMES * SUBSUBFRAMES * 8)
46 typedef struct DCAEncContext {
50 int fullband_channels;
56 const int32_t *band_interpolation;
57 const int32_t *band_spectrum;
61 const int8_t *channel_order_tab; ///< channel reordering table, lfe and non lfe
63 int32_t history[512][MAX_CHANNELS]; /* This is a circular buffer */
64 int32_t subband[SUBBAND_SAMPLES][DCAENC_SUBBANDS][MAX_CHANNELS];
65 int32_t quantized[SUBBAND_SAMPLES][DCAENC_SUBBANDS][MAX_CHANNELS];
66 int32_t peak_cb[DCAENC_SUBBANDS][MAX_CHANNELS];
67 int32_t downsampled_lfe[DCA_LFE_SAMPLES];
68 int32_t masking_curve_cb[SUBSUBFRAMES][256];
69 int abits[DCAENC_SUBBANDS][MAX_CHANNELS];
70 int scale_factor[DCAENC_SUBBANDS][MAX_CHANNELS];
71 softfloat quant[DCAENC_SUBBANDS][MAX_CHANNELS];
72 int32_t eff_masking_curve_cb[256];
73 int32_t band_masking_cb[32];
74 int32_t worst_quantization_noise;
75 int32_t worst_noise_ever;
79 static int32_t cos_table[2048];
80 static int32_t band_interpolation[2][512];
81 static int32_t band_spectrum[2][8];
82 static int32_t auf[9][AUBANDS][256];
83 static int32_t cb_to_add[256];
84 static int32_t cb_to_level[2048];
85 static int32_t lfe_fir_64i[512];
87 /* Transfer function of outer and middle ear, Hz -> dB */
88 static double hom(double f)
92 return -3.64 * pow(f1, -0.8)
93 + 6.8 * exp(-0.6 * (f1 - 3.4) * (f1 - 3.4))
94 - 6.0 * exp(-0.15 * (f1 - 8.7) * (f1 - 8.7))
95 - 0.0006 * (f1 * f1) * (f1 * f1);
98 static double gammafilter(int i, double f)
100 double h = (f - fc[i]) / erb[i];
104 return 20 * log10(h);
107 static int encode_init(AVCodecContext *avctx)
109 DCAEncContext *c = avctx->priv_data;
110 uint64_t layout = avctx->channel_layout;
111 int i, min_frame_bits;
113 c->fullband_channels = c->channels = avctx->channels;
114 c->lfe_channel = (avctx->channels == 3 || avctx->channels == 6);
115 c->band_interpolation = band_interpolation[1];
116 c->band_spectrum = band_spectrum[1];
117 c->worst_quantization_noise = -2047;
118 c->worst_noise_ever = -2047;
121 av_log(avctx, AV_LOG_WARNING, "No channel layout specified. The "
122 "encoder will guess the layout, but it "
123 "might be incorrect.\n");
124 layout = av_get_default_channel_layout(avctx->channels);
127 case AV_CH_LAYOUT_MONO: c->channel_config = 0; break;
128 case AV_CH_LAYOUT_STEREO: c->channel_config = 2; break;
129 case AV_CH_LAYOUT_2_2: c->channel_config = 8; break;
130 case AV_CH_LAYOUT_5POINT0: c->channel_config = 9; break;
131 case AV_CH_LAYOUT_5POINT1: c->channel_config = 9; break;
133 av_log(avctx, AV_LOG_ERROR, "Unsupported channel layout!\n");
134 return AVERROR_PATCHWELCOME;
137 if (c->lfe_channel) {
138 c->fullband_channels--;
139 c->channel_order_tab = ff_dca_channel_reorder_lfe[c->channel_config];
141 c->channel_order_tab = ff_dca_channel_reorder_nolfe[c->channel_config];
144 for (i = 0; i < 9; i++) {
145 if (sample_rates[i] == avctx->sample_rate)
149 return AVERROR(EINVAL);
150 c->samplerate_index = i;
152 if (avctx->bit_rate < 32000 || avctx->bit_rate > 3840000) {
153 av_log(avctx, AV_LOG_ERROR, "Bit rate %"PRId64" not supported.", (int64_t)avctx->bit_rate);
154 return AVERROR(EINVAL);
156 for (i = 0; ff_dca_bit_rates[i] < avctx->bit_rate; i++)
158 c->bitrate_index = i;
159 avctx->bit_rate = ff_dca_bit_rates[i];
160 c->frame_bits = FFALIGN((avctx->bit_rate * 512 + avctx->sample_rate - 1) / avctx->sample_rate, 32);
161 min_frame_bits = 132 + (493 + 28 * 32) * c->fullband_channels + c->lfe_channel * 72;
162 if (c->frame_bits < min_frame_bits || c->frame_bits > (DCA_MAX_FRAME_SIZE << 3))
163 return AVERROR(EINVAL);
165 c->frame_size = (c->frame_bits + 7) / 8;
167 avctx->frame_size = 32 * SUBBAND_SAMPLES;
172 for (i = 0; i < 2048; i++) {
173 cos_table[i] = (int32_t)(0x7fffffff * cos(M_PI * i / 1024));
174 cb_to_level[i] = (int32_t)(0x7fffffff * pow(10, -0.005 * i));
177 for (k = 0; k < 32; k++) {
178 for (j = 0; j < 8; j++) {
179 lfe_fir_64i[64 * j + k] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);
180 lfe_fir_64i[64 * (7-j) + (63 - k)] = (int32_t)(0xffffff800000ULL * ff_dca_lfe_fir_64[8 * k + j]);
184 for (i = 0; i < 512; i++) {
185 band_interpolation[0][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_perfect[i]);
186 band_interpolation[1][i] = (int32_t)(0x1000000000ULL * ff_dca_fir_32bands_nonperfect[i]);
189 for (i = 0; i < 9; i++) {
190 for (j = 0; j < AUBANDS; j++) {
191 for (k = 0; k < 256; k++) {
192 double freq = sample_rates[i] * (k + 0.5) / 512;
194 auf[i][j][k] = (int32_t)(10 * (hom(freq) + gammafilter(j, freq)));
199 for (i = 0; i < 256; i++) {
200 double add = 1 + pow(10, -0.01 * i);
201 cb_to_add[i] = (int32_t)(100 * log10(add));
203 for (j = 0; j < 8; j++) {
205 for (i = 0; i < 512; i++) {
206 double reconst = ff_dca_fir_32bands_perfect[i] * ((i & 64) ? (-1) : 1);
207 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);
209 band_spectrum[0][j] = (int32_t)(200 * log10(accum));
211 for (j = 0; j < 8; j++) {
213 for (i = 0; i < 512; i++) {
214 double reconst = ff_dca_fir_32bands_nonperfect[i] * ((i & 64) ? (-1) : 1);
215 accum += reconst * cos(2 * M_PI * (i + 0.5 - 256) * (j + 0.5) / 512);
217 band_spectrum[1][j] = (int32_t)(200 * log10(accum));
223 static inline int32_t cos_t(int x)
225 return cos_table[x & 2047];
228 static inline int32_t sin_t(int x)
230 return cos_t(x - 512);
233 static inline int32_t half32(int32_t a)
238 static inline int32_t mul32(int32_t a, int32_t b)
240 int64_t r = (int64_t)a * b + 0x80000000ULL;
244 static void subband_transform(DCAEncContext *c, const int32_t *input)
246 int ch, subs, i, k, j;
248 for (ch = 0; ch < c->fullband_channels; ch++) {
249 /* History is copied because it is also needed for PSY */
252 const int chi = c->channel_order_tab[ch];
254 for (i = 0; i < 512; i++)
255 hist[i] = c->history[i][ch];
257 for (subs = 0; subs < SUBBAND_SAMPLES; subs++) {
262 /* Calculate the convolutions at once */
263 for (i = 0; i < 64; i++)
266 for (k = 0, i = hist_start, j = 0;
267 i < 512; k = (k + 1) & 63, i++, j++)
268 accum[k] += mul32(hist[i], c->band_interpolation[j]);
269 for (i = 0; i < hist_start; k = (k + 1) & 63, i++, j++)
270 accum[k] += mul32(hist[i], c->band_interpolation[j]);
272 for (k = 16; k < 32; k++)
273 accum[k] = accum[k] - accum[31 - k];
274 for (k = 32; k < 48; k++)
275 accum[k] = accum[k] + accum[95 - k];
277 for (band = 0; band < 32; band++) {
279 for (i = 16; i < 48; i++) {
280 int s = (2 * band + 1) * (2 * (i + 16) + 1);
281 resp += mul32(accum[i], cos_t(s << 3)) >> 3;
284 c->subband[subs][band][ch] = ((band + 1) & 2) ? -resp : resp;
287 /* Copy in 32 new samples from input */
288 for (i = 0; i < 32; i++)
289 hist[i + hist_start] = input[(subs * 32 + i) * c->channels + chi];
290 hist_start = (hist_start + 32) & 511;
295 static void lfe_downsample(DCAEncContext *c, const int32_t *input)
297 /* FIXME: make 128x LFE downsampling possible */
298 const int lfech = ff_dca_lfe_index[c->channel_config];
304 for (i = 0; i < 512; i++)
305 hist[i] = c->history[i][c->channels - 1];
307 for (lfes = 0; lfes < DCA_LFE_SAMPLES; lfes++) {
308 /* Calculate the convolution */
311 for (i = hist_start, j = 0; i < 512; i++, j++)
312 accum += mul32(hist[i], lfe_fir_64i[j]);
313 for (i = 0; i < hist_start; i++, j++)
314 accum += mul32(hist[i], lfe_fir_64i[j]);
316 c->downsampled_lfe[lfes] = accum;
318 /* Copy in 64 new samples from input */
319 for (i = 0; i < 64; i++)
320 hist[i + hist_start] = input[(lfes * 64 + i) * c->channels + lfech];
322 hist_start = (hist_start + 64) & 511;
331 static void fft(const int32_t in[2 * 256], cplx32 out[256])
333 cplx32 buf[256], rin[256], rout[256];
336 /* do two transforms in parallel */
337 for (i = 0; i < 256; i++) {
338 /* Apply the Hann window */
339 rin[i].re = mul32(in[2 * i], 0x3fffffff - (cos_t(8 * i + 2) >> 1));
340 rin[i].im = mul32(in[2 * i + 1], 0x3fffffff - (cos_t(8 * i + 6) >> 1));
343 for (i = 0; i < 256; i++) {
344 buf[i].re = mul32(cos_t(4 * i + 2), rin[i].re)
345 - mul32(sin_t(4 * i + 2), rin[i].im);
346 buf[i].im = mul32(cos_t(4 * i + 2), rin[i].im)
347 + mul32(sin_t(4 * i + 2), rin[i].re);
350 for (j = 256, l = 1; j != 1; j >>= 1, l <<= 1) {
351 for (k = 0; k < 256; k += j) {
352 for (i = k; i < k + j / 2; i++) {
356 sum.re = buf[i].re + buf[i + j / 2].re;
357 sum.im = buf[i].im + buf[i + j / 2].im;
359 diff.re = buf[i].re - buf[i + j / 2].re;
360 diff.im = buf[i].im - buf[i + j / 2].im;
362 buf[i].re = half32(sum.re);
363 buf[i].im = half32(sum.im);
365 buf[i + j / 2].re = mul32(diff.re, cos_t(t))
366 - mul32(diff.im, sin_t(t));
367 buf[i + j / 2].im = mul32(diff.im, cos_t(t))
368 + mul32(diff.re, sin_t(t));
373 for (i = 0; i < 256; i++) {
374 int b = ff_reverse[i];
375 rout[i].re = mul32(buf[b].re, cos_t(4 * i))
376 - mul32(buf[b].im, sin_t(4 * i));
377 rout[i].im = mul32(buf[b].im, cos_t(4 * i))
378 + mul32(buf[b].re, sin_t(4 * i));
380 for (i = 0; i < 256; i++) {
381 /* separate the results of the two transforms */
384 o1.re = rout[i].re - rout[255 - i].re;
385 o1.im = rout[i].im + rout[255 - i].im;
387 o2.re = rout[i].im - rout[255 - i].im;
388 o2.im = -rout[i].re - rout[255 - i].re;
390 /* combine them into one long transform */
391 out[i].re = mul32( o1.re + o2.re, cos_t(2 * i + 1))
392 + mul32( o1.im - o2.im, sin_t(2 * i + 1));
393 out[i].im = mul32( o1.im + o2.im, cos_t(2 * i + 1))
394 + mul32(-o1.re + o2.re, sin_t(2 * i + 1));
398 static int32_t get_cb(int32_t in)
405 for (i = 1024; i > 0; i >>= 1) {
406 if (cb_to_level[i + res] >= in)
412 static int32_t add_cb(int32_t a, int32_t b)
415 FFSWAP(int32_t, a, b);
419 return a + cb_to_add[a - b];
422 static void adjust_jnd(int samplerate_index,
423 const int32_t in[512], int32_t out_cb[256])
427 int32_t out_cb_unnorm[256];
429 const int32_t ca_cb = -1114;
430 const int32_t cs_cb = 928;
435 for (j = 0; j < 256; j++) {
436 power[j] = add_cb(get_cb(out[j].re), get_cb(out[j].im));
437 out_cb_unnorm[j] = -2047; /* and can only grow */
440 for (i = 0; i < AUBANDS; i++) {
441 denom = ca_cb; /* and can only grow */
442 for (j = 0; j < 256; j++)
443 denom = add_cb(denom, power[j] + auf[samplerate_index][i][j]);
444 for (j = 0; j < 256; j++)
445 out_cb_unnorm[j] = add_cb(out_cb_unnorm[j],
446 -denom + auf[samplerate_index][i][j]);
449 for (j = 0; j < 256; j++)
450 out_cb[j] = add_cb(out_cb[j], -out_cb_unnorm[j] - ca_cb - cs_cb);
453 typedef void (*walk_band_t)(DCAEncContext *c, int band1, int band2, int f,
454 int32_t spectrum1, int32_t spectrum2, int channel,
457 static void walk_band_low(DCAEncContext *c, int band, int channel,
458 walk_band_t walk, int32_t *arg)
463 for (f = 0; f < 4; f++)
464 walk(c, 0, 0, f, 0, -2047, channel, arg);
466 for (f = 0; f < 8; f++)
467 walk(c, band, band - 1, 8 * band - 4 + f,
468 c->band_spectrum[7 - f], c->band_spectrum[f], channel, arg);
472 static void walk_band_high(DCAEncContext *c, int band, int channel,
473 walk_band_t walk, int32_t *arg)
478 for (f = 0; f < 4; f++)
479 walk(c, 31, 31, 256 - 4 + f, 0, -2047, channel, arg);
481 for (f = 0; f < 8; f++)
482 walk(c, band, band + 1, 8 * band + 4 + f,
483 c->band_spectrum[f], c->band_spectrum[7 - f], channel, arg);
487 static void update_band_masking(DCAEncContext *c, int band1, int band2,
488 int f, int32_t spectrum1, int32_t spectrum2,
489 int channel, int32_t * arg)
491 int32_t value = c->eff_masking_curve_cb[f] - spectrum1;
493 if (value < c->band_masking_cb[band1])
494 c->band_masking_cb[band1] = value;
497 static void calc_masking(DCAEncContext *c, const int32_t *input)
499 int i, k, band, ch, ssf;
502 for (i = 0; i < 256; i++)
503 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
504 c->masking_curve_cb[ssf][i] = -2047;
506 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
507 for (ch = 0; ch < c->fullband_channels; ch++) {
508 const int chi = c->channel_order_tab[ch];
510 for (i = 0, k = 128 + 256 * ssf; k < 512; i++, k++)
511 data[i] = c->history[k][ch];
512 for (k -= 512; i < 512; i++, k++)
513 data[i] = input[k * c->channels + chi];
514 adjust_jnd(c->samplerate_index, data, c->masking_curve_cb[ssf]);
516 for (i = 0; i < 256; i++) {
519 for (ssf = 0; ssf < SUBSUBFRAMES; ssf++)
520 if (c->masking_curve_cb[ssf][i] < m)
521 m = c->masking_curve_cb[ssf][i];
522 c->eff_masking_curve_cb[i] = m;
525 for (band = 0; band < 32; band++) {
526 c->band_masking_cb[band] = 2048;
527 walk_band_low(c, band, 0, update_band_masking, NULL);
528 walk_band_high(c, band, 0, update_band_masking, NULL);
532 static void find_peaks(DCAEncContext *c)
536 for (band = 0; band < 32; band++)
537 for (ch = 0; ch < c->fullband_channels; ch++) {
541 for (sample = 0; sample < SUBBAND_SAMPLES; sample++) {
542 int32_t s = abs(c->subband[sample][band][ch]);
546 c->peak_cb[band][ch] = get_cb(m);
549 if (c->lfe_channel) {
553 for (sample = 0; sample < DCA_LFE_SAMPLES; sample++)
554 if (m < abs(c->downsampled_lfe[sample]))
555 m = abs(c->downsampled_lfe[sample]);
556 c->lfe_peak_cb = get_cb(m);
560 static const int snr_fudge = 128;
561 #define USED_1ABITS 1
562 #define USED_NABITS 2
563 #define USED_26ABITS 4
565 static int init_quantization_noise(DCAEncContext *c, int noise)
567 int ch, band, ret = 0;
569 c->consumed_bits = 132 + 493 * c->fullband_channels;
571 c->consumed_bits += 72;
573 /* attempt to guess the bit distribution based on the prevoius frame */
574 for (ch = 0; ch < c->fullband_channels; ch++) {
575 for (band = 0; band < 32; band++) {
576 int snr_cb = c->peak_cb[band][ch] - c->band_masking_cb[band] - noise;
578 if (snr_cb >= 1312) {
579 c->abits[band][ch] = 26;
581 } else if (snr_cb >= 222) {
582 c->abits[band][ch] = 8 + mul32(snr_cb - 222, 69000000);
584 } else if (snr_cb >= 0) {
585 c->abits[band][ch] = 2 + mul32(snr_cb, 106000000);
588 c->abits[band][ch] = 1;
594 for (band = 0; band < 32; band++)
595 for (ch = 0; ch < c->fullband_channels; ch++) {
596 c->consumed_bits += bit_consumption[c->abits[band][ch]];
602 static void assign_bits(DCAEncContext *c)
604 /* Find the bounds where the binary search should work */
608 init_quantization_noise(c, c->worst_quantization_noise);
609 low = high = c->worst_quantization_noise;
610 if (c->consumed_bits > c->frame_bits) {
611 while (c->consumed_bits > c->frame_bits) {
612 av_assert0(used_abits != USED_1ABITS);
615 used_abits = init_quantization_noise(c, high);
618 while (c->consumed_bits <= c->frame_bits) {
620 if (used_abits == USED_26ABITS)
621 goto out; /* The requested bitrate is too high, pad with zeros */
623 used_abits = init_quantization_noise(c, low);
627 /* Now do a binary search between low and high to see what fits */
628 for (down = snr_fudge >> 1; down; down >>= 1) {
629 init_quantization_noise(c, high - down);
630 if (c->consumed_bits <= c->frame_bits)
633 init_quantization_noise(c, high);
635 c->worst_quantization_noise = high;
636 if (high > c->worst_noise_ever)
637 c->worst_noise_ever = high;
640 static void shift_history(DCAEncContext *c, const int32_t *input)
644 for (k = 0; k < 512; k++)
645 for (ch = 0; ch < c->channels; ch++) {
646 const int chi = c->channel_order_tab[ch];
648 c->history[k][ch] = input[k * c->channels + chi];
652 static int32_t quantize_value(int32_t value, softfloat quant)
654 int32_t offset = 1 << (quant.e - 1);
656 value = mul32(value, quant.m) + offset;
657 value = value >> quant.e;
661 static int calc_one_scale(int32_t peak_cb, int abits, softfloat *quant)
664 int our_nscale, try_remove;
667 av_assert0(peak_cb <= 0);
668 av_assert0(peak_cb >= -2047);
671 peak = cb_to_level[-peak_cb];
673 for (try_remove = 64; try_remove > 0; try_remove >>= 1) {
674 if (scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e <= 17)
676 our_quant.m = mul32(scalefactor_inv[our_nscale - try_remove].m, stepsize_inv[abits].m);
677 our_quant.e = scalefactor_inv[our_nscale - try_remove].e + stepsize_inv[abits].e - 17;
678 if ((quant_levels[abits] - 1) / 2 < quantize_value(peak, our_quant))
680 our_nscale -= try_remove;
683 if (our_nscale >= 125)
686 quant->m = mul32(scalefactor_inv[our_nscale].m, stepsize_inv[abits].m);
687 quant->e = scalefactor_inv[our_nscale].e + stepsize_inv[abits].e - 17;
688 av_assert0((quant_levels[abits] - 1) / 2 >= quantize_value(peak, *quant));
693 static void calc_scales(DCAEncContext *c)
697 for (band = 0; band < 32; band++)
698 for (ch = 0; ch < c->fullband_channels; ch++)
699 c->scale_factor[band][ch] = calc_one_scale(c->peak_cb[band][ch],
701 &c->quant[band][ch]);
704 c->lfe_scale_factor = calc_one_scale(c->lfe_peak_cb, 11, &c->lfe_quant);
707 static void quantize_all(DCAEncContext *c)
709 int sample, band, ch;
711 for (sample = 0; sample < SUBBAND_SAMPLES; sample++)
712 for (band = 0; band < 32; band++)
713 for (ch = 0; ch < c->fullband_channels; ch++)
714 c->quantized[sample][band][ch] = quantize_value(c->subband[sample][band][ch], c->quant[band][ch]);
717 static void put_frame_header(DCAEncContext *c)
720 put_bits(&c->pb, 16, 0x7ffe);
721 put_bits(&c->pb, 16, 0x8001);
723 /* Frame type: normal */
724 put_bits(&c->pb, 1, 1);
726 /* Deficit sample count: none */
727 put_bits(&c->pb, 5, 31);
729 /* CRC is not present */
730 put_bits(&c->pb, 1, 0);
732 /* Number of PCM sample blocks */
733 put_bits(&c->pb, 7, SUBBAND_SAMPLES - 1);
735 /* Primary frame byte size */
736 put_bits(&c->pb, 14, c->frame_size - 1);
738 /* Audio channel arrangement */
739 put_bits(&c->pb, 6, c->channel_config);
741 /* Core audio sampling frequency */
742 put_bits(&c->pb, 4, bitstream_sfreq[c->samplerate_index]);
744 /* Transmission bit rate */
745 put_bits(&c->pb, 5, c->bitrate_index);
747 /* Embedded down mix: disabled */
748 put_bits(&c->pb, 1, 0);
750 /* Embedded dynamic range flag: not present */
751 put_bits(&c->pb, 1, 0);
753 /* Embedded time stamp flag: not present */
754 put_bits(&c->pb, 1, 0);
756 /* Auxiliary data flag: not present */
757 put_bits(&c->pb, 1, 0);
759 /* HDCD source: no */
760 put_bits(&c->pb, 1, 0);
762 /* Extension audio ID: N/A */
763 put_bits(&c->pb, 3, 0);
765 /* Extended audio data: not present */
766 put_bits(&c->pb, 1, 0);
768 /* Audio sync word insertion flag: after each sub-frame */
769 put_bits(&c->pb, 1, 0);
771 /* Low frequency effects flag: not present or 64x subsampling */
772 put_bits(&c->pb, 2, c->lfe_channel ? 2 : 0);
774 /* Predictor history switch flag: on */
775 put_bits(&c->pb, 1, 1);
778 /* Multirate interpolator switch: non-perfect reconstruction */
779 put_bits(&c->pb, 1, 0);
781 /* Encoder software revision: 7 */
782 put_bits(&c->pb, 4, 7);
784 /* Copy history: 0 */
785 put_bits(&c->pb, 2, 0);
787 /* Source PCM resolution: 16 bits, not DTS ES */
788 put_bits(&c->pb, 3, 0);
790 /* Front sum/difference coding: no */
791 put_bits(&c->pb, 1, 0);
793 /* Surrounds sum/difference coding: no */
794 put_bits(&c->pb, 1, 0);
796 /* Dialog normalization: 0 dB */
797 put_bits(&c->pb, 4, 0);
800 static void put_primary_audio_header(DCAEncContext *c)
802 static const int bitlen[11] = { 0, 1, 2, 2, 2, 2, 3, 3, 3, 3, 3 };
803 static const int thr[11] = { 0, 1, 3, 3, 3, 3, 7, 7, 7, 7, 7 };
806 /* Number of subframes */
807 put_bits(&c->pb, 4, SUBFRAMES - 1);
809 /* Number of primary audio channels */
810 put_bits(&c->pb, 3, c->fullband_channels - 1);
812 /* Subband activity count */
813 for (ch = 0; ch < c->fullband_channels; ch++)
814 put_bits(&c->pb, 5, DCAENC_SUBBANDS - 2);
816 /* High frequency VQ start subband */
817 for (ch = 0; ch < c->fullband_channels; ch++)
818 put_bits(&c->pb, 5, DCAENC_SUBBANDS - 1);
820 /* Joint intensity coding index: 0, 0 */
821 for (ch = 0; ch < c->fullband_channels; ch++)
822 put_bits(&c->pb, 3, 0);
824 /* Transient mode codebook: A4, A4 (arbitrary) */
825 for (ch = 0; ch < c->fullband_channels; ch++)
826 put_bits(&c->pb, 2, 0);
828 /* Scale factor code book: 7 bit linear, 7-bit sqrt table (for each channel) */
829 for (ch = 0; ch < c->fullband_channels; ch++)
830 put_bits(&c->pb, 3, 6);
832 /* Bit allocation quantizer select: linear 5-bit */
833 for (ch = 0; ch < c->fullband_channels; ch++)
834 put_bits(&c->pb, 3, 6);
836 /* Quantization index codebook select: dummy data
837 to avoid transmission of scale factor adjustment */
838 for (i = 1; i < 11; i++)
839 for (ch = 0; ch < c->fullband_channels; ch++)
840 put_bits(&c->pb, bitlen[i], thr[i]);
842 /* Scale factor adjustment index: not transmitted */
843 /* Audio header CRC check word: not transmitted */
846 static void put_subframe_samples(DCAEncContext *c, int ss, int band, int ch)
848 if (c->abits[band][ch] <= 7) {
850 for (i = 0; i < 8; i += 4) {
852 for (j = 3; j >= 0; j--) {
853 sum *= quant_levels[c->abits[band][ch]];
854 sum += c->quantized[ss * 8 + i + j][band][ch];
855 sum += (quant_levels[c->abits[band][ch]] - 1) / 2;
857 put_bits(&c->pb, bit_consumption[c->abits[band][ch]] / 4, sum);
861 for (i = 0; i < 8; i++) {
862 int bits = bit_consumption[c->abits[band][ch]] / 16;
863 put_sbits(&c->pb, bits, c->quantized[ss * 8 + i][band][ch]);
868 static void put_subframe(DCAEncContext *c, int subframe)
872 /* Subsubframes count */
873 put_bits(&c->pb, 2, SUBSUBFRAMES -1);
875 /* Partial subsubframe sample count: dummy */
876 put_bits(&c->pb, 3, 0);
878 /* Prediction mode: no ADPCM, in each channel and subband */
879 for (ch = 0; ch < c->fullband_channels; ch++)
880 for (band = 0; band < DCAENC_SUBBANDS; band++)
881 put_bits(&c->pb, 1, 0);
883 /* Prediction VQ address: not transmitted */
884 /* Bit allocation index */
885 for (ch = 0; ch < c->fullband_channels; ch++)
886 for (band = 0; band < DCAENC_SUBBANDS; band++)
887 put_bits(&c->pb, 5, c->abits[band][ch]);
889 if (SUBSUBFRAMES > 1) {
890 /* Transition mode: none for each channel and subband */
891 for (ch = 0; ch < c->fullband_channels; ch++)
892 for (band = 0; band < DCAENC_SUBBANDS; band++)
893 put_bits(&c->pb, 1, 0); /* codebook A4 */
897 for (ch = 0; ch < c->fullband_channels; ch++)
898 for (band = 0; band < DCAENC_SUBBANDS; band++)
899 put_bits(&c->pb, 7, c->scale_factor[band][ch]);
901 /* Joint subband scale factor codebook select: not transmitted */
902 /* Scale factors for joint subband coding: not transmitted */
903 /* Stereo down-mix coefficients: not transmitted */
904 /* Dynamic range coefficient: not transmitted */
905 /* Stde information CRC check word: not transmitted */
906 /* VQ encoded high frequency subbands: not transmitted */
908 /* LFE data: 8 samples and scalefactor */
909 if (c->lfe_channel) {
910 for (i = 0; i < DCA_LFE_SAMPLES; i++)
911 put_bits(&c->pb, 8, quantize_value(c->downsampled_lfe[i], c->lfe_quant) & 0xff);
912 put_bits(&c->pb, 8, c->lfe_scale_factor);
915 /* Audio data (subsubframes) */
916 for (ss = 0; ss < SUBSUBFRAMES ; ss++)
917 for (ch = 0; ch < c->fullband_channels; ch++)
918 for (band = 0; band < DCAENC_SUBBANDS; band++)
919 put_subframe_samples(c, ss, band, ch);
922 put_bits(&c->pb, 16, 0xffff);
925 static int encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
926 const AVFrame *frame, int *got_packet_ptr)
928 DCAEncContext *c = avctx->priv_data;
929 const int32_t *samples;
932 if ((ret = ff_alloc_packet2(avctx, avpkt, c->frame_size , 0)) < 0)
935 samples = (const int32_t *)frame->data[0];
937 subband_transform(c, samples);
939 lfe_downsample(c, samples);
941 calc_masking(c, samples);
946 shift_history(c, samples);
948 init_put_bits(&c->pb, avpkt->data, avpkt->size);
950 put_primary_audio_header(c);
951 for (i = 0; i < SUBFRAMES; i++)
955 for (i = put_bits_count(&c->pb); i < 8*c->frame_size; i++)
956 put_bits(&c->pb, 1, 0);
958 flush_put_bits(&c->pb);
960 avpkt->pts = frame->pts;
961 avpkt->duration = ff_samples_to_time_base(avctx, frame->nb_samples);
962 avpkt->size = c->frame_size + 1;
967 static const AVCodecDefault defaults[] = {
972 AVCodec ff_dca_encoder = {
974 .long_name = NULL_IF_CONFIG_SMALL("DCA (DTS Coherent Acoustics)"),
975 .type = AVMEDIA_TYPE_AUDIO,
976 .id = AV_CODEC_ID_DTS,
977 .priv_data_size = sizeof(DCAEncContext),
979 .encode2 = encode_frame,
980 .capabilities = AV_CODEC_CAP_EXPERIMENTAL,
981 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S32,
982 AV_SAMPLE_FMT_NONE },
983 .supported_samplerates = sample_rates,
984 .channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_MONO,
987 AV_CH_LAYOUT_5POINT0,
988 AV_CH_LAYOUT_5POINT1,
990 .defaults = defaults,