FFmpeg
aptxenc.c
Go to the documentation of this file.
1 /*
2  * Audio Processing Technology codec for Bluetooth (aptX)
3  *
4  * Copyright (C) 2017 Aurelien Jacobs <aurel@gnuage.org>
5  *
6  * This file is part of FFmpeg.
7  *
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.
12  *
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.
17  *
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
21  */
22 
23 #include "config_components.h"
24 
26 #include "aptx.h"
27 #include "codec_internal.h"
28 #include "encode.h"
29 
30 /*
31  * Half-band QMF analysis filter realized with a polyphase FIR filter.
32  * Split into 2 subbands and downsample by 2.
33  * So for each pair of samples that goes in, one sample goes out,
34  * split into 2 separate subbands.
35  */
38  const int32_t coeffs[NB_FILTERS][FILTER_TAPS],
39  int shift,
41  int32_t *low_subband_output,
42  int32_t *high_subband_output)
43 {
45  int i;
46 
47  for (i = 0; i < NB_FILTERS; i++) {
49  subbands[i] = aptx_qmf_convolution(&signal[i], coeffs[i], shift);
50  }
51 
52  *low_subband_output = av_clip_intp2(subbands[0] + subbands[1], 23);
53  *high_subband_output = av_clip_intp2(subbands[0] - subbands[1], 23);
54 }
55 
56 /*
57  * Two stage QMF analysis tree.
58  * Split 4 input samples into 4 subbands and downsample by 4.
59  * So for each group of 4 samples that goes in, one sample goes out,
60  * split into 4 separate subbands.
61  */
63  int32_t samples[4],
64  int32_t subband_samples[4])
65 {
66  int32_t intermediate_samples[4];
67  int i;
68 
69  /* Split 4 input samples into 2 intermediate subbands downsampled to 2 samples */
70  for (i = 0; i < 2; i++)
73  &samples[2*i],
74  &intermediate_samples[0+i],
75  &intermediate_samples[2+i]);
76 
77  /* Split 2 intermediate subband samples into 4 final subbands downsampled to 1 sample */
78  for (i = 0; i < 2; i++)
81  &intermediate_samples[2*i],
82  &subband_samples[2*i+0],
83  &subband_samples[2*i+1]);
84 }
85 
88  const int32_t *intervals, int32_t nb_intervals)
89 {
90  int32_t idx = 0;
91  int i;
92 
93  for (i = nb_intervals >> 1; i > 0; i >>= 1)
94  if (MUL64(factor, intervals[idx + i]) <= ((int64_t)value << 24))
95  idx += i;
96 
97  return idx;
98 }
99 
101  int32_t sample_difference,
102  int32_t dither,
103  int32_t quantization_factor,
105 {
106  const int32_t *intervals = tables->quantize_intervals;
107  int32_t quantized_sample, dithered_sample, parity_change;
108  int32_t d, mean, interval, inv, sample_difference_abs;
109  int64_t error;
110 
111  sample_difference_abs = FFABS(sample_difference);
112  sample_difference_abs = FFMIN(sample_difference_abs, (1 << 23) - 1);
113 
114  quantized_sample = aptx_bin_search(sample_difference_abs >> 4,
115  quantization_factor,
116  intervals, tables->tables_size);
117 
118  d = rshift32_clip24(MULH(dither, dither), 7) - (1 << 23);
119  d = rshift64(MUL64(d, tables->quantize_dither_factors[quantized_sample]), 23);
120 
121  intervals += quantized_sample;
122  mean = (intervals[1] + intervals[0]) / 2;
123  interval = (intervals[1] - intervals[0]) * (-(sample_difference < 0) | 1);
124 
125  dithered_sample = rshift64_clip24(MUL64(dither, interval) + ((int64_t)av_clip_intp2(mean + d, 23) << 32), 32);
126  error = ((int64_t)sample_difference_abs << 20) - MUL64(dithered_sample, quantization_factor);
127  quantize->error = FFABS(rshift64(error, 23));
128 
129  parity_change = quantized_sample;
130  if (error < 0)
131  quantized_sample--;
132  else
133  parity_change--;
134 
135  inv = -(sample_difference < 0);
136  quantize->quantized_sample = quantized_sample ^ inv;
137  quantize->quantized_sample_parity_change = parity_change ^ inv;
138 }
139 
141 {
142  int32_t subband_samples[4];
143  int subband;
144  aptx_qmf_tree_analysis(&channel->qmf, samples, subband_samples);
146  for (subband = 0; subband < NB_SUBBANDS; subband++) {
147  int32_t diff = av_clip_intp2(subband_samples[subband] - channel->prediction[subband].predicted_sample, 23);
148  aptx_quantize_difference(&channel->quantize[subband], diff,
149  channel->dither[subband],
150  channel->invert_quantize[subband].quantization_factor,
151  &ff_aptx_quant_tables[hd][subband]);
152  }
153 }
154 
156 {
157  if (aptx_check_parity(channels, idx)) {
158  int i;
159  Channel *c;
160  static const int map[] = { 1, 2, 0, 3 };
161  Quantize *min = &channels[NB_CHANNELS-1].quantize[map[0]];
162  for (c = &channels[NB_CHANNELS-1]; c >= channels; c--)
163  for (i = 0; i < NB_SUBBANDS; i++)
164  if (c->quantize[map[i]].error < min->error)
165  min = &c->quantize[map[i]];
166 
167  /* Forcing the desired parity is done by offsetting by 1 the quantized
168  * sample from the subband featuring the smallest quantization error. */
169  min->quantized_sample = min->quantized_sample_parity_change;
170  }
171 }
172 
174 {
176  return (((channel->quantize[3].quantized_sample & 0x06) | parity) << 13)
177  | (((channel->quantize[2].quantized_sample & 0x03) ) << 11)
178  | (((channel->quantize[1].quantized_sample & 0x0F) ) << 7)
179  | (((channel->quantize[0].quantized_sample & 0x7F) ) << 0);
180 }
181 
183 {
185  return (((channel->quantize[3].quantized_sample & 0x01E) | parity) << 19)
186  | (((channel->quantize[2].quantized_sample & 0x00F) ) << 15)
187  | (((channel->quantize[1].quantized_sample & 0x03F) ) << 9)
188  | (((channel->quantize[0].quantized_sample & 0x1FF) ) << 0);
189 }
190 
193  uint8_t *output)
194 {
195  int channel;
196  for (channel = 0; channel < NB_CHANNELS; channel++)
197  aptx_encode_channel(&ctx->channels[channel], samples[channel], ctx->hd);
198 
199  aptx_insert_sync(ctx->channels, &ctx->sync_idx);
200 
201  for (channel = 0; channel < NB_CHANNELS; channel++) {
203  if (ctx->hd)
204  AV_WB24(output + 3*channel,
205  aptxhd_pack_codeword(&ctx->channels[channel]));
206  else
207  AV_WB16(output + 2*channel,
208  aptx_pack_codeword(&ctx->channels[channel]));
209  }
210 }
211 
212 static int aptx_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
213  const AVFrame *frame, int *got_packet_ptr)
214 {
215  AptXContext *s = avctx->priv_data;
216  int pos, ipos, channel, sample, output_size, ret;
217 
218  if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
219  return ret;
220 
221  output_size = s->block_size * frame->nb_samples/4;
222  if ((ret = ff_get_encode_buffer(avctx, avpkt, output_size, 0)) < 0)
223  return ret;
224 
225  for (pos = 0, ipos = 0; pos < output_size; pos += s->block_size, ipos += 4) {
227 
228  for (channel = 0; channel < NB_CHANNELS; channel++)
229  for (sample = 0; sample < 4; sample++)
230  samples[channel][sample] = (int32_t)AV_RN32A(&frame->data[channel][4*(ipos+sample)]) >> 8;
231 
232  aptx_encode_samples(s, samples, avpkt->data + pos);
233  }
234 
235  ff_af_queue_remove(&s->afq, frame->nb_samples, &avpkt->pts, &avpkt->duration);
236  *got_packet_ptr = 1;
237  return 0;
238 }
239 
241 {
242  AptXContext *s = avctx->priv_data;
243  ff_af_queue_close(&s->afq);
244  return 0;
245 }
246 
247 #if CONFIG_APTX_ENCODER
248 const FFCodec ff_aptx_encoder = {
249  .p.name = "aptx",
250  .p.long_name = NULL_IF_CONFIG_SMALL("aptX (Audio Processing Technology for Bluetooth)"),
251  .p.type = AVMEDIA_TYPE_AUDIO,
252  .p.id = AV_CODEC_ID_APTX,
254  .priv_data_size = sizeof(AptXContext),
255  .init = ff_aptx_init,
257  .close = aptx_close,
258  .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
259 #if FF_API_OLD_CHANNEL_LAYOUT
260  .p.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
261 #endif
262  .p.ch_layouts = (const AVChannelLayout[]) { AV_CHANNEL_LAYOUT_STEREO, { 0 } },
263  .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
265  .p.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
266 };
267 #endif
268 
269 #if CONFIG_APTX_HD_ENCODER
270 const FFCodec ff_aptx_hd_encoder = {
271  .p.name = "aptx_hd",
272  .p.long_name = NULL_IF_CONFIG_SMALL("aptX HD (Audio Processing Technology for Bluetooth)"),
273  .p.type = AVMEDIA_TYPE_AUDIO,
274  .p.id = AV_CODEC_ID_APTX_HD,
276  .priv_data_size = sizeof(AptXContext),
277  .init = ff_aptx_init,
279  .close = aptx_close,
280  .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE,
281 #if FF_API_OLD_CHANNEL_LAYOUT
282  .p.channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_STEREO, 0},
283 #endif
284  .p.ch_layouts = (const AVChannelLayout[]) { AV_CHANNEL_LAYOUT_STEREO, { 0 } },
285  .p.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_S32P,
287  .p.supported_samplerates = (const int[]) {8000, 16000, 24000, 32000, 44100, 48000, 0},
288 };
289 #endif
ff_aptx_quant_tables
ConstTables ff_aptx_quant_tables[2][NB_SUBBANDS]
Definition: aptx.c:313
error
static void error(const char *err)
Definition: target_bsf_fuzzer.c:31
Channel
Definition: aptx.h:82
FILTER_TAPS
#define FILTER_TAPS
Definition: aptx.h:47
ff_af_queue_remove
void ff_af_queue_remove(AudioFrameQueue *afq, int nb_samples, int64_t *pts, int64_t *duration)
Remove frame(s) from the queue.
Definition: audio_frame_queue.c:75
aptx_quantized_parity
static int32_t aptx_quantized_parity(Channel *channel)
Definition: aptx.h:190
ff_af_queue_close
void ff_af_queue_close(AudioFrameQueue *afq)
Close AudioFrameQueue.
Definition: audio_frame_queue.c:36
aptx_encode_frame
static int aptx_encode_frame(AVCodecContext *avctx, AVPacket *avpkt, const AVFrame *frame, int *got_packet_ptr)
Definition: aptxenc.c:212
output
filter_frame For filters that do not use the this method is called when a frame is pushed to the filter s input It can be called at any time except in a reentrant way If the input frame is enough to produce output
Definition: filter_design.txt:225
QMFAnalysis
Definition: aptx.h:54
AVFrame
This structure describes decoded (raw) audio or video data.
Definition: frame.h:325
AVPacket::data
uint8_t * data
Definition: packet.h:374
ff_aptx_hd_encoder
const FFCodec ff_aptx_hd_encoder
encode.h
ff_aptx_generate_dither
void ff_aptx_generate_dither(Channel *channel)
Definition: aptx.c:385
AV_SAMPLE_FMT_S32P
@ AV_SAMPLE_FMT_S32P
signed 32 bits, planar
Definition: samplefmt.h:65
FFCodec
Definition: codec_internal.h:112
AVPacket::duration
int64_t duration
Duration of this packet in AVStream->time_base units, 0 if unknown.
Definition: packet.h:392
AV_CHANNEL_LAYOUT_STEREO
#define AV_CHANNEL_LAYOUT_STEREO
Definition: channel_layout.h:354
subbands
subbands
Definition: aptx.h:38
AptXContext
Definition: aptx.h:93
init
static int init
Definition: av_tx.c:47
QMFAnalysis::inner_filter_signal
FilterSignal inner_filter_signal[NB_FILTERS][NB_FILTERS]
Definition: aptx.h:56
FFCodec::p
AVCodec p
The public AVCodec.
Definition: codec_internal.h:116
aptx_qmf_tree_analysis
static void aptx_qmf_tree_analysis(QMFAnalysis *qmf, int32_t samples[4], int32_t subband_samples[4])
Definition: aptxenc.c:62
NB_FILTERS
@ NB_FILTERS
Definition: vf_waveform.c:54
quantize
static int quantize(CinepakEncContext *s, int h, uint8_t *data[4], int linesize[4], int v1mode, strip_info *info, mb_encoding encoding)
Definition: cinepakenc.c:699
MULH
#define MULH
Definition: mathops.h:42
tables
Writing a table generator This documentation is preliminary Parts of the API are not good and should be changed Basic concepts A table generator consists of two *_tablegen c and *_tablegen h The h file will provide the variable declarations and initialization code for the tables
Definition: tablegen.txt:10
FF_CODEC_ENCODE_CB
#define FF_CODEC_ENCODE_CB(func)
Definition: codec_internal.h:263
ff_af_queue_add
int ff_af_queue_add(AudioFrameQueue *afq, const AVFrame *f)
Add a frame to the queue.
Definition: audio_frame_queue.c:44
AV_CH_LAYOUT_STEREO
#define AV_CH_LAYOUT_STEREO
Definition: channel_layout.h:205
aptx_pack_codeword
static uint16_t aptx_pack_codeword(Channel *channel)
Definition: aptxenc.c:173
av_cold
#define av_cold
Definition: attributes.h:90
ff_aptx_encoder
const FFCodec ff_aptx_encoder
s
#define s(width, name)
Definition: cbs_vp9.c:256
ConstTables
Definition: aptx.h:101
AVMEDIA_TYPE_AUDIO
@ AVMEDIA_TYPE_AUDIO
Definition: avutil.h:202
NB_CHANNELS
@ NB_CHANNELS
Definition: aptx.h:35
ctx
AVFormatContext * ctx
Definition: movenc.c:48
channels
channels
Definition: aptx.h:32
aptxhd_pack_codeword
static uint32_t aptxhd_pack_codeword(Channel *channel)
Definition: aptxenc.c:182
FFABS
#define FFABS(a)
Absolute value, Note, INT_MIN / INT64_MIN result in undefined behavior as they are not representable ...
Definition: common.h:64
QMFAnalysis::outer_filter_signal
FilterSignal outer_filter_signal[NB_FILTERS]
Definition: aptx.h:55
aptx_qmf_convolution
static av_always_inline int32_t aptx_qmf_convolution(FilterSignal *signal, const int32_t coeffs[FILTER_TAPS], int shift)
Definition: aptx.h:176
aptx_qmf_polyphase_analysis
static av_always_inline void aptx_qmf_polyphase_analysis(FilterSignal signal[NB_FILTERS], const int32_t coeffs[NB_FILTERS][FILTER_TAPS], int shift, int32_t samples[NB_FILTERS], int32_t *low_subband_output, int32_t *high_subband_output)
Definition: aptxenc.c:37
av_clip_intp2
#define av_clip_intp2
Definition: common.h:116
AV_WB16
#define AV_WB16(p, v)
Definition: intreadwrite.h:405
aptx_qmf_outer_coeffs
static const int32_t aptx_qmf_outer_coeffs[NB_FILTERS][FILTER_TAPS]
Definition: aptx.h:134
FilterSignal
Definition: aptx.h:49
aptx_encode_samples
static void aptx_encode_samples(AptXContext *ctx, int32_t samples[NB_CHANNELS][4], uint8_t *output)
Definition: aptxenc.c:191
aptx.h
c
Undefined Behavior In the C some operations are like signed integer dereferencing freed accessing outside allocated Undefined Behavior must not occur in a C it is not safe even if the output of undefined operations is unused The unsafety may seem nit picking but Optimizing compilers have in fact optimized code on the assumption that no undefined Behavior occurs Optimizing code based on wrong assumptions can and has in some cases lead to effects beyond the output of computations The signed integer overflow problem in speed critical code Code which is highly optimized and works with signed integers sometimes has the problem that often the output of the computation does not c
Definition: undefined.txt:32
AV_CODEC_CAP_DR1
#define AV_CODEC_CAP_DR1
Codec uses get_buffer() or get_encode_buffer() for allocating buffers and supports custom allocators.
Definition: codec.h:52
NULL_IF_CONFIG_SMALL
#define NULL_IF_CONFIG_SMALL(x)
Return NULL if CONFIG_SMALL is true, otherwise the argument without modification.
Definition: internal.h:117
AVChannelLayout
An AVChannelLayout holds information about the channel layout of audio data.
Definition: channel_layout.h:290
codec_internal.h
aptx_qmf_inner_coeffs
static const int32_t aptx_qmf_inner_coeffs[NB_FILTERS][FILTER_TAPS]
Definition: aptx.h:149
AV_SAMPLE_FMT_NONE
@ AV_SAMPLE_FMT_NONE
Definition: samplefmt.h:56
sample
#define sample
Definition: flacdsp_template.c:44
parity
mcdeint parity
Definition: vf_mcdeint.c:266
AV_WB24
#define AV_WB24(p, d)
Definition: intreadwrite.h:450
aptx_quantize_difference
static void aptx_quantize_difference(Quantize *quantize, int32_t sample_difference, int32_t dither, int32_t quantization_factor, ConstTables *tables)
Definition: aptxenc.c:100
i
#define i(width, name, range_min, range_max)
Definition: cbs_h2645.c:271
AVPacket::pts
int64_t pts
Presentation timestamp in AVStream->time_base units; the time at which the decompressed packet will b...
Definition: packet.h:367
AVSampleFormat
AVSampleFormat
Audio sample formats.
Definition: samplefmt.h:55
av_always_inline
#define av_always_inline
Definition: attributes.h:49
value
it s the only field you need to keep assuming you have a context There is some magic you don t need to care about around this just let it vf default value
Definition: writing_filters.txt:86
FFMIN
#define FFMIN(a, b)
Definition: macros.h:49
FF_CODEC_CAP_INIT_THREADSAFE
#define FF_CODEC_CAP_INIT_THREADSAFE
The codec does not modify any global variables in the init function, allowing to call the init functi...
Definition: codec_internal.h:31
aptx_qmf_filter_signal_push
static av_always_inline void aptx_qmf_filter_signal_push(FilterSignal *signal, int32_t sample)
Definition: aptx.h:164
AVCodec::name
const char * name
Name of the codec implementation.
Definition: codec.h:203
AV_RN32A
#define AV_RN32A(p)
Definition: intreadwrite.h:526
ret
ret
Definition: filter_design.txt:187
frame
these buffered frames must be flushed immediately if a new input produces new the filter must not call request_frame to get more It must just process the frame or queue it The task of requesting more frames is left to the filter s request_frame method or the application If a filter has several the filter must be ready for frames arriving randomly on any input any filter with several inputs will most likely require some kind of queuing mechanism It is perfectly acceptable to have a limited queue and to drop frames when the inputs are too unbalanced request_frame For filters that do not use the this method is called when a frame is wanted on an output For a it should directly call filter_frame on the corresponding output For a if there are queued frames already one of these frames should be pushed If the filter should request a frame on one of its repeatedly until at least one frame has been pushed Return or at least make progress towards producing a frame
Definition: filter_design.txt:264
pos
unsigned int pos
Definition: spdifenc.c:412
ff_aptx_init
av_cold int ff_aptx_init(AVCodecContext *avctx)
Definition: aptx.c:508
AVCodecContext
main external API structure.
Definition: avcodec.h:389
channel_layout.h
ff_aptx_invert_quantize_and_prediction
void ff_aptx_invert_quantize_and_prediction(Channel *channel, int hd)
Definition: aptx.c:497
aptx_bin_search
static av_always_inline int32_t aptx_bin_search(int32_t value, int32_t factor, const int32_t *intervals, int32_t nb_intervals)
Definition: aptxenc.c:87
ff_get_encode_buffer
int ff_get_encode_buffer(AVCodecContext *avctx, AVPacket *avpkt, int64_t size, int flags)
Get a buffer for a packet.
Definition: encode.c:79
MUL64
#define MUL64(a, b)
Definition: mathops.h:54
samples
Filter the word “frame” indicates either a video frame or a group of audio samples
Definition: filter_design.txt:8
mean
static float mean(const float *input, int size)
Definition: vf_nnedi.c:857
factor
static const int factor[16]
Definition: vf_pp7.c:76
shift
static int shift(int a, int b)
Definition: sonic.c:88
map
const VDPAUPixFmtMap * map
Definition: hwcontext_vdpau.c:71
diff
static av_always_inline int diff(const uint32_t a, const uint32_t b)
Definition: vf_palettegen.c:139
AVCodecContext::priv_data
void * priv_data
Definition: avcodec.h:416
AVPacket
This structure stores compressed data.
Definition: packet.h:351
d
d
Definition: ffmpeg_filter.c:153
int32_t
int32_t
Definition: audioconvert.c:56
aptx_encode_channel
static void aptx_encode_channel(Channel *channel, int32_t samples[4], int hd)
Definition: aptxenc.c:140
aptx_close
static av_cold int aptx_close(AVCodecContext *avctx)
Definition: aptxenc.c:240
AV_CODEC_ID_APTX
@ AV_CODEC_ID_APTX
Definition: codec_id.h:511
NB_SUBBANDS
@ NB_SUBBANDS
Definition: aptx.h:43
AV_CODEC_CAP_SMALL_LAST_FRAME
#define AV_CODEC_CAP_SMALL_LAST_FRAME
Codec can be fed a final frame with a smaller size.
Definition: codec.h:87
AV_CODEC_ID_APTX_HD
@ AV_CODEC_ID_APTX_HD
Definition: codec_id.h:512
channel
channel
Definition: ebur128.h:39
Quantize
Definition: aptx.h:59
aptx_insert_sync
static void aptx_insert_sync(Channel channels[NB_CHANNELS], int32_t *idx)
Definition: aptxenc.c:155
aptx_check_parity
static int aptx_check_parity(Channel channels[NB_CHANNELS], int32_t *idx)
Definition: aptx.h:203
min
float min
Definition: vorbis_enc_data.h:429
dither
static const uint8_t dither[8][8]
Definition: vf_fspp.c:58