1 /*
2 ** Copyright 2011, The Android Open-Source Project
3 **
4 ** Licensed under the Apache License, Version 2.0 (the "License");
5 ** you may not use this file except in compliance with the License.
6 ** You may obtain a copy of the License at
7 **
8 ** http://www.apache.org/licenses/LICENSE-2.0
9 **
10 ** Unless required by applicable law or agreed to in writing, software
11 ** distributed under the License is distributed on an "AS IS" BASIS,
12 ** WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 ** See the License for the specific language governing permissions and
14 ** limitations under the License.
15 */
16
17 //#define LOG_NDEBUG 0
18 #define LOG_TAG "echo_reference"
19
20 #include <errno.h>
21 #include <stdlib.h>
22 #include <pthread.h>
23 #include <cutils/log.h>
24 #include <system/audio.h>
25 #include <audio_utils/resampler.h>
26 #include <audio_utils/echo_reference.h>
27
28 // echo reference state: bit field indicating if read, write or both are active.
29 enum state {
30 ECHOREF_IDLE = 0x00, // idle
31 ECHOREF_READING = 0x01, // reading is active
32 ECHOREF_WRITING = 0x02 // writing is active
33 };
34
35 struct echo_reference {
36 struct echo_reference_itfe itfe;
37 int status; // init status
38 uint32_t state; // active state: reading, writing or both
39 audio_format_t rd_format; // read sample format
40 uint32_t rd_channel_count; // read number of channels
41 uint32_t rd_sampling_rate; // read sampling rate in Hz
42 size_t rd_frame_size; // read frame size (bytes per sample)
43 audio_format_t wr_format; // write sample format
44 uint32_t wr_channel_count; // write number of channels
45 uint32_t wr_sampling_rate; // write sampling rate in Hz
46 size_t wr_frame_size; // write frame size (bytes per sample)
47 void *buffer; // main buffer
48 size_t buf_size; // main buffer size in frames
49 size_t frames_in; // number of frames in main buffer
50 void *wr_buf; // buffer for input conversions
51 size_t wr_buf_size; // size of conversion buffer in frames
52 size_t wr_frames_in; // number of frames in conversion buffer
53 size_t wr_curr_frame_size; // number of frames given to current write() function
54 void *wr_src_buf; // resampler input buf (either wr_buf or buffer used by write())
55 struct timespec wr_render_time; // latest render time indicated by write()
56 // default ALSA gettimeofday() format
57 int32_t playback_delay; // playback buffer delay indicated by last write()
58 int16_t prev_delta_sign; // sign of previous delay difference:
59 // 1: positive, -1: negative, 0: unknown
60 uint16_t delta_count; // number of consecutive delay differences with same sign
61 pthread_mutex_t lock; // mutex protecting read/write concurrency
62 pthread_cond_t cond; // condition signaled when data is ready to read
63 struct resampler_itfe *resampler; // input resampler
64 struct resampler_buffer_provider provider; // resampler buffer provider
65 };
66
67
echo_reference_get_next_buffer(struct resampler_buffer_provider * buffer_provider,struct resampler_buffer * buffer)68 int echo_reference_get_next_buffer(struct resampler_buffer_provider *buffer_provider,
69 struct resampler_buffer* buffer)
70 {
71 struct echo_reference *er;
72
73 if (buffer_provider == NULL) {
74 return -EINVAL;
75 }
76
77 er = (struct echo_reference *)((char *)buffer_provider -
78 offsetof(struct echo_reference, provider));
79
80 if (er->wr_src_buf == NULL || er->wr_frames_in == 0) {
81 buffer->raw = NULL;
82 buffer->frame_count = 0;
83 return -ENODATA;
84 }
85
86 buffer->frame_count = (buffer->frame_count > er->wr_frames_in) ? er->wr_frames_in : buffer->frame_count;
87 // this is er->rd_channel_count here as we resample after stereo to mono conversion if any
88 buffer->i16 = (int16_t *)er->wr_src_buf + (er->wr_curr_frame_size - er->wr_frames_in) * er->rd_channel_count;
89
90 return 0;
91 }
92
echo_reference_release_buffer(struct resampler_buffer_provider * buffer_provider,struct resampler_buffer * buffer)93 void echo_reference_release_buffer(struct resampler_buffer_provider *buffer_provider,
94 struct resampler_buffer* buffer)
95 {
96 struct echo_reference *er;
97
98 if (buffer_provider == NULL) {
99 return;
100 }
101
102 er = (struct echo_reference *)((char *)buffer_provider -
103 offsetof(struct echo_reference, provider));
104
105 er->wr_frames_in -= buffer->frame_count;
106 }
107
echo_reference_reset_l(struct echo_reference * er)108 static void echo_reference_reset_l(struct echo_reference *er)
109 {
110 ALOGV("echo_reference_reset_l()");
111 free(er->buffer);
112 er->buffer = NULL;
113 er->buf_size = 0;
114 er->frames_in = 0;
115 free(er->wr_buf);
116 er->wr_buf = NULL;
117 er->wr_buf_size = 0;
118 er->wr_render_time.tv_sec = 0;
119 er->wr_render_time.tv_nsec = 0;
120 er->delta_count = 0;
121 er->prev_delta_sign = 0;
122 }
123
124 /* additional space in resampler buffer allowing for extra samples to be returned
125 * by speex resampler when sample rates ratio is not an integer.
126 */
127 #define RESAMPLER_HEADROOM_SAMPLES 10
128
echo_reference_write(struct echo_reference_itfe * echo_reference,struct echo_reference_buffer * buffer)129 static int echo_reference_write(struct echo_reference_itfe *echo_reference,
130 struct echo_reference_buffer *buffer)
131 {
132 struct echo_reference *er = (struct echo_reference *)echo_reference;
133 int status = 0;
134
135 if (er == NULL) {
136 return -EINVAL;
137 }
138
139 pthread_mutex_lock(&er->lock);
140
141 if (buffer == NULL) {
142 ALOGV("echo_reference_write() stop write");
143 er->state &= ~ECHOREF_WRITING;
144 echo_reference_reset_l(er);
145 goto exit;
146 }
147
148 ALOGV("echo_reference_write() START trying to write %d frames", buffer->frame_count);
149 ALOGV("echo_reference_write() playbackTimestamp:[%d].[%d], er->playback_delay:[%d]",
150 (int)buffer->time_stamp.tv_sec,
151 (int)buffer->time_stamp.tv_nsec, er->playback_delay);
152
153 //ALOGV("echo_reference_write() %d frames", buffer->frame_count);
154 // discard writes until a valid time stamp is provided.
155
156 if ((buffer->time_stamp.tv_sec == 0) && (buffer->time_stamp.tv_nsec == 0) &&
157 (er->wr_render_time.tv_sec == 0) && (er->wr_render_time.tv_nsec == 0)) {
158 goto exit;
159 }
160
161 if ((er->state & ECHOREF_WRITING) == 0) {
162 ALOGV("echo_reference_write() start write");
163 if (er->resampler != NULL) {
164 er->resampler->reset(er->resampler);
165 }
166 er->state |= ECHOREF_WRITING;
167 }
168
169 if ((er->state & ECHOREF_READING) == 0) {
170 goto exit;
171 }
172
173 er->wr_render_time.tv_sec = buffer->time_stamp.tv_sec;
174 er->wr_render_time.tv_nsec = buffer->time_stamp.tv_nsec;
175
176 er->playback_delay = buffer->delay_ns;
177
178 // this will be used in the get_next_buffer, to support variable input buffer sizes
179 er->wr_curr_frame_size = buffer->frame_count;
180
181 void *srcBuf;
182 size_t inFrames;
183 // do stereo to mono and down sampling if necessary
184 if (er->rd_channel_count != er->wr_channel_count ||
185 er->rd_sampling_rate != er->wr_sampling_rate) {
186 size_t wrBufSize = buffer->frame_count;
187
188 inFrames = buffer->frame_count;
189
190 if (er->rd_sampling_rate != er->wr_sampling_rate) {
191 inFrames = (buffer->frame_count * er->rd_sampling_rate) / er->wr_sampling_rate +
192 RESAMPLER_HEADROOM_SAMPLES;
193 // wr_buf is not only used as resampler output but also for stereo to mono conversion
194 // output so buffer size is driven by both write and read sample rates
195 if (inFrames > wrBufSize) {
196 wrBufSize = inFrames;
197 }
198 }
199
200 if (er->wr_buf_size < wrBufSize) {
201 ALOGV("echo_reference_write() increasing write buffer size from %d to %d",
202 er->wr_buf_size, wrBufSize);
203 er->wr_buf_size = wrBufSize;
204 er->wr_buf = realloc(er->wr_buf, er->wr_buf_size * er->rd_frame_size);
205 }
206
207 if (er->rd_channel_count != er->wr_channel_count) {
208 // must be stereo to mono
209 int16_t *src16 = (int16_t *)buffer->raw;
210 int16_t *dst16 = (int16_t *)er->wr_buf;
211 size_t frames = buffer->frame_count;
212 while (frames--) {
213 *dst16++ = (int16_t)(((int32_t)*src16 + (int32_t)*(src16 + 1)) >> 1);
214 src16 += 2;
215 }
216 }
217 if (er->wr_sampling_rate != er->rd_sampling_rate) {
218 if (er->resampler == NULL) {
219 int rc;
220 ALOGV("echo_reference_write() new ReSampler(%d, %d)",
221 er->wr_sampling_rate, er->rd_sampling_rate);
222 er->provider.get_next_buffer = echo_reference_get_next_buffer;
223 er->provider.release_buffer = echo_reference_release_buffer;
224 rc = create_resampler(er->wr_sampling_rate,
225 er->rd_sampling_rate,
226 er->rd_channel_count,
227 RESAMPLER_QUALITY_DEFAULT,
228 &er->provider,
229 &er->resampler);
230 if (rc != 0) {
231 er->resampler = NULL;
232 ALOGV("echo_reference_write() failure to create resampler %d", rc);
233 status = -ENODEV;
234 goto exit;
235 }
236 }
237 // er->wr_src_buf and er->wr_frames_in are used by getNexBuffer() called by the resampler
238 // to get new frames
239 if (er->rd_channel_count != er->wr_channel_count) {
240 er->wr_src_buf = er->wr_buf;
241 } else {
242 er->wr_src_buf = buffer->raw;
243 }
244 er->wr_frames_in = buffer->frame_count;
245 // inFrames is always more than we need here to get frames remaining from previous runs
246 // inFrames is updated by resample() with the number of frames produced
247 ALOGV("echo_reference_write() ReSampling(%d, %d)",
248 er->wr_sampling_rate, er->rd_sampling_rate);
249 er->resampler->resample_from_provider(er->resampler,
250 (int16_t *)er->wr_buf, &inFrames);
251 ALOGV_IF(er->wr_frames_in != 0,
252 "echo_reference_write() er->wr_frames_in not 0 (%d) after resampler",
253 er->wr_frames_in);
254 }
255 srcBuf = er->wr_buf;
256 } else {
257 inFrames = buffer->frame_count;
258 srcBuf = buffer->raw;
259 }
260
261 if (er->frames_in + inFrames > er->buf_size) {
262 ALOGV("echo_reference_write() increasing buffer size from %d to %d",
263 er->buf_size, er->frames_in + inFrames);
264 er->buf_size = er->frames_in + inFrames;
265 er->buffer = realloc(er->buffer, er->buf_size * er->rd_frame_size);
266 }
267 memcpy((char *)er->buffer + er->frames_in * er->rd_frame_size,
268 srcBuf,
269 inFrames * er->rd_frame_size);
270 er->frames_in += inFrames;
271
272 ALOGV("echo_reference_write() frames written:[%d], frames total:[%d] buffer size:[%d]\n"
273 " er->wr_render_time:[%d].[%d], er->playback_delay:[%d]",
274 inFrames, er->frames_in, er->buf_size,
275 (int)er->wr_render_time.tv_sec, (int)er->wr_render_time.tv_nsec, er->playback_delay);
276
277 pthread_cond_signal(&er->cond);
278 exit:
279 pthread_mutex_unlock(&er->lock);
280 ALOGV("echo_reference_write() END");
281 return status;
282 }
283
284 // delay jump threshold to update ref buffer: 6 samples at 8kHz in nsecs
285 #define MIN_DELAY_DELTA_NS (375000*2)
286 // number of consecutive delta with same sign between expected and actual delay before adjusting
287 // the buffer
288 #define MIN_DELTA_NUM 4
289
290
echo_reference_read(struct echo_reference_itfe * echo_reference,struct echo_reference_buffer * buffer)291 static int echo_reference_read(struct echo_reference_itfe *echo_reference,
292 struct echo_reference_buffer *buffer)
293 {
294 struct echo_reference *er = (struct echo_reference *)echo_reference;
295
296 if (er == NULL) {
297 return -EINVAL;
298 }
299
300 pthread_mutex_lock(&er->lock);
301
302 if (buffer == NULL) {
303 ALOGV("echo_reference_read() stop read");
304 er->state &= ~ECHOREF_READING;
305 goto exit;
306 }
307
308 ALOGV("echo_reference_read() START, delayCapture:[%d], "
309 "er->frames_in:[%d],buffer->frame_count:[%d]",
310 buffer->delay_ns, er->frames_in, buffer->frame_count);
311
312 if ((er->state & ECHOREF_READING) == 0) {
313 ALOGV("echo_reference_read() start read");
314 echo_reference_reset_l(er);
315 er->state |= ECHOREF_READING;
316 }
317
318 if ((er->state & ECHOREF_WRITING) == 0) {
319 memset(buffer->raw, 0, er->rd_frame_size * buffer->frame_count);
320 buffer->delay_ns = 0;
321 goto exit;
322 }
323
324 // ALOGV("echo_reference_read() %d frames", buffer->frame_count);
325
326 // allow some time for new frames to arrive if not enough frames are ready for read
327 if (er->frames_in < buffer->frame_count) {
328 uint32_t timeoutMs = (uint32_t)((1000 * buffer->frame_count) / er->rd_sampling_rate / 2);
329 struct timespec ts;
330
331 ts.tv_sec = timeoutMs/1000;
332 ts.tv_nsec = timeoutMs%1000;
333 pthread_cond_timedwait_relative_np(&er->cond, &er->lock, &ts);
334
335 ALOGV_IF((er->frames_in < buffer->frame_count),
336 "echo_reference_read() waited %d ms but still not enough frames"\
337 " er->frames_in: %d, buffer->frame_count = %d",
338 timeoutMs, er->frames_in, buffer->frame_count);
339 }
340
341 int64_t timeDiff;
342 struct timespec tmp;
343
344 if ((er->wr_render_time.tv_sec == 0 && er->wr_render_time.tv_nsec == 0) ||
345 (buffer->time_stamp.tv_sec == 0 && buffer->time_stamp.tv_nsec == 0)) {
346 ALOGV("echo_reference_read(): NEW:timestamp is zero---------setting timeDiff = 0, "\
347 "not updating delay this time");
348 timeDiff = 0;
349 } else {
350 if (buffer->time_stamp.tv_nsec < er->wr_render_time.tv_nsec) {
351 tmp.tv_sec = buffer->time_stamp.tv_sec - er->wr_render_time.tv_sec - 1;
352 tmp.tv_nsec = 1000000000 + buffer->time_stamp.tv_nsec - er->wr_render_time.tv_nsec;
353 } else {
354 tmp.tv_sec = buffer->time_stamp.tv_sec - er->wr_render_time.tv_sec;
355 tmp.tv_nsec = buffer->time_stamp.tv_nsec - er->wr_render_time.tv_nsec;
356 }
357 timeDiff = (((int64_t)tmp.tv_sec * 1000000000 + tmp.tv_nsec));
358
359 int64_t expectedDelayNs = er->playback_delay + buffer->delay_ns - timeDiff;
360
361 if (er->resampler != NULL) {
362 // Resampler already compensates part of the delay
363 int32_t rsmp_delay = er->resampler->delay_ns(er->resampler);
364 expectedDelayNs -= rsmp_delay;
365 }
366
367 ALOGV("echo_reference_read(): expectedDelayNs[%lld] = "
368 "er->playback_delay[%d] + delayCapture[%d] - timeDiff[%lld]",
369 expectedDelayNs, er->playback_delay, buffer->delay_ns, timeDiff);
370
371 if (expectedDelayNs > 0) {
372 int64_t delayNs = ((int64_t)er->frames_in * 1000000000) / er->rd_sampling_rate;
373
374 int64_t deltaNs = delayNs - expectedDelayNs;
375
376 ALOGV("echo_reference_read(): EchoPathDelayDeviation between reference and DMA [%lld]", deltaNs);
377 if (abs(deltaNs) >= MIN_DELAY_DELTA_NS) {
378 // smooth the variation and update the reference buffer only
379 // if a deviation in the same direction is observed for more than MIN_DELTA_NUM
380 // consecutive reads.
381 int16_t delay_sign = (deltaNs >= 0) ? 1 : -1;
382 if (delay_sign == er->prev_delta_sign) {
383 er->delta_count++;
384 } else {
385 er->delta_count = 1;
386 }
387 er->prev_delta_sign = delay_sign;
388
389 if (er->delta_count > MIN_DELTA_NUM) {
390 size_t previousFrameIn = er->frames_in;
391 er->frames_in = (size_t)((expectedDelayNs * er->rd_sampling_rate)/1000000000);
392 int offset = er->frames_in - previousFrameIn;
393
394 ALOGV("echo_reference_read(): deltaNs ENOUGH and %s: "
395 "er->frames_in: %d, previousFrameIn = %d",
396 delay_sign ? "positive" : "negative", er->frames_in, previousFrameIn);
397
398 if (deltaNs < 0) {
399 // Less data available in the reference buffer than expected
400 if (er->frames_in > er->buf_size) {
401 er->buf_size = er->frames_in;
402 er->buffer = realloc(er->buffer, er->buf_size * er->rd_frame_size);
403 ALOGV("echo_reference_read(): increasing buffer size to %d",
404 er->buf_size);
405 }
406
407 if (offset > 0) {
408 memset((char *)er->buffer + previousFrameIn * er->rd_frame_size,
409 0, offset * er->rd_frame_size);
410 ALOGV("echo_reference_read(): pushing ref buffer by [%d]", offset);
411 }
412 } else {
413 // More data available in the reference buffer than expected
414 offset = -offset;
415 if (offset > 0) {
416 memcpy(er->buffer, (char *)er->buffer + (offset * er->rd_frame_size),
417 er->frames_in * er->rd_frame_size);
418 ALOGV("echo_reference_read(): shifting ref buffer by [%d]",
419 er->frames_in);
420 }
421 }
422 }
423 } else {
424 er->delta_count = 0;
425 er->prev_delta_sign = 0;
426 ALOGV("echo_reference_read(): Constant EchoPathDelay - difference "
427 "between reference and DMA %lld", deltaNs);
428 }
429 } else {
430 ALOGV("echo_reference_read(): NEGATIVE expectedDelayNs[%lld] = "\
431 "er->playback_delay[%d] + delayCapture[%d] - timeDiff[%lld]",
432 expectedDelayNs, er->playback_delay, buffer->delay_ns, timeDiff);
433 }
434 }
435
436 if (er->frames_in < buffer->frame_count) {
437 if (buffer->frame_count > er->buf_size) {
438 er->buf_size = buffer->frame_count;
439 er->buffer = realloc(er->buffer, er->buf_size * er->rd_frame_size);
440 ALOGV("echo_reference_read(): increasing buffer size to %d", er->buf_size);
441 }
442 // filling up the reference buffer with 0s to match the expected delay.
443 memset((char *)er->buffer + er->frames_in * er->rd_frame_size,
444 0, (buffer->frame_count - er->frames_in) * er->rd_frame_size);
445 er->frames_in = buffer->frame_count;
446 }
447
448 memcpy(buffer->raw,
449 (char *)er->buffer,
450 buffer->frame_count * er->rd_frame_size);
451
452 er->frames_in -= buffer->frame_count;
453 memcpy(er->buffer,
454 (char *)er->buffer + buffer->frame_count * er->rd_frame_size,
455 er->frames_in * er->rd_frame_size);
456
457 // As the reference buffer is now time aligned to the microphone signal there is a zero delay
458 buffer->delay_ns = 0;
459
460 ALOGV("echo_reference_read() END %d frames, total frames in %d",
461 buffer->frame_count, er->frames_in);
462
463 pthread_cond_signal(&er->cond);
464
465 exit:
466 pthread_mutex_unlock(&er->lock);
467 return 0;
468 }
469
470
create_echo_reference(audio_format_t rdFormat,uint32_t rdChannelCount,uint32_t rdSamplingRate,audio_format_t wrFormat,uint32_t wrChannelCount,uint32_t wrSamplingRate,struct echo_reference_itfe ** echo_reference)471 int create_echo_reference(audio_format_t rdFormat,
472 uint32_t rdChannelCount,
473 uint32_t rdSamplingRate,
474 audio_format_t wrFormat,
475 uint32_t wrChannelCount,
476 uint32_t wrSamplingRate,
477 struct echo_reference_itfe **echo_reference)
478 {
479 struct echo_reference *er;
480
481 ALOGV("create_echo_reference()");
482
483 if (echo_reference == NULL) {
484 return -EINVAL;
485 }
486
487 *echo_reference = NULL;
488
489 if (rdFormat != AUDIO_FORMAT_PCM_16_BIT ||
490 rdFormat != wrFormat) {
491 ALOGW("create_echo_reference bad format rd %d, wr %d", rdFormat, wrFormat);
492 return -EINVAL;
493 }
494 if ((rdChannelCount != 1 && rdChannelCount != 2) ||
495 wrChannelCount != 2) {
496 ALOGW("create_echo_reference bad channel count rd %d, wr %d", rdChannelCount, wrChannelCount);
497 return -EINVAL;
498 }
499
500 er = (struct echo_reference *)calloc(1, sizeof(struct echo_reference));
501
502 er->itfe.read = echo_reference_read;
503 er->itfe.write = echo_reference_write;
504
505 er->state = ECHOREF_IDLE;
506 er->rd_format = rdFormat;
507 er->rd_channel_count = rdChannelCount;
508 er->rd_sampling_rate = rdSamplingRate;
509 er->wr_format = wrFormat;
510 er->wr_channel_count = wrChannelCount;
511 er->wr_sampling_rate = wrSamplingRate;
512 er->rd_frame_size = audio_bytes_per_sample(rdFormat) * rdChannelCount;
513 er->wr_frame_size = audio_bytes_per_sample(wrFormat) * wrChannelCount;
514 *echo_reference = &er->itfe;
515 return 0;
516 }
517
release_echo_reference(struct echo_reference_itfe * echo_reference)518 void release_echo_reference(struct echo_reference_itfe *echo_reference) {
519 struct echo_reference *er = (struct echo_reference *)echo_reference;
520
521 if (er == NULL) {
522 return;
523 }
524
525 ALOGV("EchoReference dstor");
526 echo_reference_reset_l(er);
527 if (er->resampler != NULL) {
528 release_resampler(er->resampler);
529 }
530 free(er);
531 }
532
533