stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 1 | /* |
| 2 | * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. |
| 3 | * |
| 4 | * Use of this source code is governed by a BSD-style license |
| 5 | * that can be found in the LICENSE file in the root of the source |
| 6 | * tree. An additional intellectual property rights grant can be found |
| 7 | * in the file PATENTS. All contributing project authors may |
| 8 | * be found in the AUTHORS file in the root of the source tree. |
| 9 | */ |
| 10 | |
| 11 | #include "video_engine/stream_synchronization.h" |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 12 | |
| 13 | #include <assert.h> |
| 14 | #include <algorithm> |
| 15 | #include <cmath> |
| 16 | |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 17 | #include "system_wrappers/interface/trace.h" |
| 18 | |
| 19 | namespace webrtc { |
| 20 | |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 21 | const int kMaxVideoDiffMs = 80; |
| 22 | const int kMaxAudioDiffMs = 80; |
| 23 | const int kMaxDelay = 1500; |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 24 | |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 25 | const double kNtpFracPerMs = 4.294967296E6; |
| 26 | |
| 27 | namespace synchronization { |
| 28 | |
| 29 | RtcpMeasurement::RtcpMeasurement() |
| 30 | : ntp_secs(0), ntp_frac(0), rtp_timestamp(0) {} |
| 31 | |
| 32 | RtcpMeasurement::RtcpMeasurement(uint32_t ntp_secs, uint32_t ntp_frac, |
| 33 | uint32_t timestamp) |
| 34 | : ntp_secs(ntp_secs), ntp_frac(ntp_frac), rtp_timestamp(timestamp) {} |
| 35 | |
| 36 | // Calculates the RTP timestamp frequency from two pairs of NTP and RTP |
| 37 | // timestamps. |
| 38 | bool CalculateFrequency( |
| 39 | int64_t rtcp_ntp_ms1, |
| 40 | uint32_t rtp_timestamp1, |
| 41 | int64_t rtcp_ntp_ms2, |
| 42 | uint32_t rtp_timestamp2, |
| 43 | double* frequency_khz) { |
| 44 | if (rtcp_ntp_ms1 == rtcp_ntp_ms2) { |
| 45 | return false; |
| 46 | } |
| 47 | assert(rtcp_ntp_ms1 > rtcp_ntp_ms2); |
| 48 | *frequency_khz = static_cast<double>(rtp_timestamp1 - rtp_timestamp2) / |
| 49 | static_cast<double>(rtcp_ntp_ms1 - rtcp_ntp_ms2); |
| 50 | return true; |
| 51 | } |
| 52 | |
| 53 | // Detects if there has been a wraparound between |old_timestamp| and |
| 54 | // |new_timestamp|, and compensates by adding 2^32 if that is the case. |
| 55 | bool CompensateForWrapAround(uint32_t new_timestamp, |
| 56 | uint32_t old_timestamp, |
| 57 | int64_t* compensated_timestamp) { |
| 58 | assert(compensated_timestamp); |
| 59 | int64_t wraps = synchronization::CheckForWrapArounds(new_timestamp, |
| 60 | old_timestamp); |
| 61 | if (wraps < 0) { |
| 62 | // Reordering, don't use this packet. |
| 63 | return false; |
| 64 | } |
| 65 | *compensated_timestamp = new_timestamp + (wraps << 32); |
| 66 | return true; |
| 67 | } |
| 68 | |
| 69 | // Converts an NTP timestamp to a millisecond timestamp. |
| 70 | int64_t NtpToMs(uint32_t ntp_secs, uint32_t ntp_frac) { |
| 71 | const double ntp_frac_ms = static_cast<double>(ntp_frac) / kNtpFracPerMs; |
| 72 | return ntp_secs * 1000 + ntp_frac_ms + 0.5; |
| 73 | } |
| 74 | |
| 75 | // Converts |rtp_timestamp| to the NTP time base using the NTP and RTP timestamp |
| 76 | // pairs in |rtcp|. The converted timestamp is returned in |
| 77 | // |rtp_timestamp_in_ms|. This function compensates for wrap arounds in RTP |
| 78 | // timestamps and returns false if it can't do the conversion due to reordering. |
| 79 | bool RtpToNtpMs(int64_t rtp_timestamp, |
| 80 | const synchronization::RtcpList& rtcp, |
| 81 | int64_t* rtp_timestamp_in_ms) { |
| 82 | assert(rtcp.size() == 2); |
| 83 | int64_t rtcp_ntp_ms_new = synchronization::NtpToMs(rtcp.front().ntp_secs, |
| 84 | rtcp.front().ntp_frac); |
| 85 | int64_t rtcp_ntp_ms_old = synchronization::NtpToMs(rtcp.back().ntp_secs, |
| 86 | rtcp.back().ntp_frac); |
| 87 | int64_t rtcp_timestamp_new = rtcp.front().rtp_timestamp; |
| 88 | int64_t rtcp_timestamp_old = rtcp.back().rtp_timestamp; |
| 89 | if (!CompensateForWrapAround(rtcp_timestamp_new, |
| 90 | rtcp_timestamp_old, |
| 91 | &rtcp_timestamp_new)) { |
| 92 | return false; |
| 93 | } |
| 94 | double freq_khz; |
| 95 | if (!CalculateFrequency(rtcp_ntp_ms_new, |
| 96 | rtcp_timestamp_new, |
| 97 | rtcp_ntp_ms_old, |
| 98 | rtcp_timestamp_old, |
| 99 | &freq_khz)) { |
| 100 | return false; |
| 101 | } |
| 102 | double offset = rtcp_timestamp_new - freq_khz * rtcp_ntp_ms_new; |
| 103 | int64_t rtp_timestamp_unwrapped; |
| 104 | if (!CompensateForWrapAround(rtp_timestamp, rtcp_timestamp_old, |
| 105 | &rtp_timestamp_unwrapped)) { |
| 106 | return false; |
| 107 | } |
| 108 | double rtp_timestamp_ntp_ms = (static_cast<double>(rtp_timestamp_unwrapped) - |
| 109 | offset) / freq_khz + 0.5f; |
| 110 | assert(rtp_timestamp_ntp_ms >= 0); |
| 111 | *rtp_timestamp_in_ms = rtp_timestamp_ntp_ms; |
| 112 | return true; |
| 113 | } |
| 114 | |
| 115 | int CheckForWrapArounds(uint32_t new_timestamp, uint32_t old_timestamp) { |
| 116 | if (new_timestamp < old_timestamp) { |
| 117 | // This difference should be less than -2^31 if we have had a wrap around |
| 118 | // (e.g. |new_timestamp| = 1, |rtcp_rtp_timestamp| = 2^32 - 1). Since it is |
| 119 | // cast to a int32_t, it should be positive. |
| 120 | if (static_cast<int32_t>(new_timestamp - old_timestamp) > 0) { |
| 121 | // Forward wrap around. |
| 122 | return 1; |
| 123 | } |
| 124 | } else if (static_cast<int32_t>(old_timestamp - new_timestamp) > 0) { |
| 125 | // This difference should be less than -2^31 if we have had a backward wrap |
| 126 | // around. Since it is cast to a int32_t, it should be positive. |
| 127 | return -1; |
| 128 | } |
| 129 | return 0; |
| 130 | } |
| 131 | } // namespace synchronization |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 132 | |
| 133 | struct ViESyncDelay { |
| 134 | ViESyncDelay() { |
| 135 | extra_video_delay_ms = 0; |
| 136 | last_video_delay_ms = 0; |
| 137 | extra_audio_delay_ms = 0; |
| 138 | last_sync_delay = 0; |
| 139 | network_delay = 120; |
| 140 | } |
| 141 | |
| 142 | int extra_video_delay_ms; |
| 143 | int last_video_delay_ms; |
| 144 | int extra_audio_delay_ms; |
| 145 | int last_sync_delay; |
| 146 | int network_delay; |
| 147 | }; |
| 148 | |
| 149 | StreamSynchronization::StreamSynchronization(int audio_channel_id, |
| 150 | int video_channel_id) |
| 151 | : channel_delay_(new ViESyncDelay), |
| 152 | audio_channel_id_(audio_channel_id), |
| 153 | video_channel_id_(video_channel_id) {} |
| 154 | |
| 155 | StreamSynchronization::~StreamSynchronization() { |
| 156 | delete channel_delay_; |
| 157 | } |
| 158 | |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 159 | bool StreamSynchronization::ComputeRelativeDelay( |
| 160 | const Measurements& audio_measurement, |
| 161 | const Measurements& video_measurement, |
| 162 | int* relative_delay_ms) { |
| 163 | assert(relative_delay_ms); |
| 164 | if (audio_measurement.rtcp.size() < 2 || video_measurement.rtcp.size() < 2) { |
| 165 | // We need two RTCP SR reports per stream to do synchronization. |
| 166 | return false; |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 167 | } |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 168 | int64_t audio_last_capture_time_ms; |
| 169 | if (!synchronization::RtpToNtpMs(audio_measurement.latest_timestamp, |
| 170 | audio_measurement.rtcp, |
| 171 | &audio_last_capture_time_ms)) { |
| 172 | return false; |
| 173 | } |
| 174 | int64_t video_last_capture_time_ms; |
| 175 | if (!synchronization::RtpToNtpMs(video_measurement.latest_timestamp, |
| 176 | video_measurement.rtcp, |
| 177 | &video_last_capture_time_ms)) { |
| 178 | return false; |
| 179 | } |
| 180 | if (video_last_capture_time_ms < 0) { |
| 181 | return false; |
| 182 | } |
| 183 | // Positive diff means that video_measurement is behind audio_measurement. |
| 184 | *relative_delay_ms = video_measurement.latest_receive_time_ms - |
| 185 | audio_measurement.latest_receive_time_ms - |
| 186 | (video_last_capture_time_ms - audio_last_capture_time_ms); |
| 187 | if (*relative_delay_ms > 1000 || *relative_delay_ms < -1000) { |
| 188 | return false; |
| 189 | } |
| 190 | return true; |
| 191 | } |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 192 | |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 193 | bool StreamSynchronization::ComputeDelays(int relative_delay_ms, |
| 194 | int current_audio_delay_ms, |
| 195 | int* extra_audio_delay_ms, |
| 196 | int* total_video_delay_target_ms) { |
| 197 | assert(extra_audio_delay_ms && total_video_delay_target_ms); |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 198 | WEBRTC_TRACE(webrtc::kTraceInfo, webrtc::kTraceVideo, video_channel_id_, |
| 199 | "Audio delay is: %d for voice channel: %d", |
| 200 | current_audio_delay_ms, audio_channel_id_); |
| 201 | WEBRTC_TRACE(webrtc::kTraceInfo, webrtc::kTraceVideo, video_channel_id_, |
| 202 | "Network delay diff is: %d for voice channel: %d", |
| 203 | channel_delay_->network_delay, audio_channel_id_); |
| 204 | // Calculate the difference between the lowest possible video delay and |
| 205 | // the current audio delay. |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 206 | WEBRTC_TRACE(webrtc::kTraceInfo, webrtc::kTraceVideo, video_channel_id_, |
| 207 | "Current diff is: %d for audio channel: %d", |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 208 | relative_delay_ms, audio_channel_id_); |
| 209 | |
| 210 | int current_diff_ms = *total_video_delay_target_ms - current_audio_delay_ms + |
| 211 | relative_delay_ms; |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 212 | |
| 213 | int video_delay_ms = 0; |
| 214 | if (current_diff_ms > 0) { |
| 215 | // The minimum video delay is longer than the current audio delay. |
| 216 | // We need to decrease extra video delay, if we have added extra delay |
| 217 | // earlier, or add extra audio delay. |
| 218 | if (channel_delay_->extra_video_delay_ms > 0) { |
| 219 | // We have extra delay added to ViE. Reduce this delay before adding |
| 220 | // extra delay to VoE. |
| 221 | |
| 222 | // This is the desired delay, we can't reduce more than this. |
| 223 | video_delay_ms = *total_video_delay_target_ms; |
| 224 | |
| 225 | // Check that we don't reduce the delay more than what is allowed. |
| 226 | if (video_delay_ms < |
| 227 | channel_delay_->last_video_delay_ms - kMaxVideoDiffMs) { |
| 228 | video_delay_ms = |
| 229 | channel_delay_->last_video_delay_ms - kMaxVideoDiffMs; |
| 230 | channel_delay_->extra_video_delay_ms = |
| 231 | video_delay_ms - *total_video_delay_target_ms; |
| 232 | } else { |
| 233 | channel_delay_->extra_video_delay_ms = 0; |
| 234 | } |
| 235 | channel_delay_->last_video_delay_ms = video_delay_ms; |
| 236 | channel_delay_->last_sync_delay = -1; |
| 237 | channel_delay_->extra_audio_delay_ms = 0; |
| 238 | } else { // channel_delay_->extra_video_delay_ms > 0 |
| 239 | // We have no extra video delay to remove, increase the audio delay. |
| 240 | if (channel_delay_->last_sync_delay >= 0) { |
| 241 | // We have increased the audio delay earlier, increase it even more. |
| 242 | int audio_diff_ms = current_diff_ms / 2; |
| 243 | if (audio_diff_ms > kMaxAudioDiffMs) { |
| 244 | // We only allow a maximum change of KMaxAudioDiffMS for audio |
| 245 | // due to NetEQ maximum changes. |
| 246 | audio_diff_ms = kMaxAudioDiffMs; |
| 247 | } |
| 248 | // Increase the audio delay |
| 249 | channel_delay_->extra_audio_delay_ms += audio_diff_ms; |
| 250 | |
| 251 | // Don't set a too high delay. |
| 252 | if (channel_delay_->extra_audio_delay_ms > kMaxDelay) { |
| 253 | channel_delay_->extra_audio_delay_ms = kMaxDelay; |
| 254 | } |
| 255 | |
| 256 | // Don't add any extra video delay. |
| 257 | video_delay_ms = *total_video_delay_target_ms; |
| 258 | channel_delay_->extra_video_delay_ms = 0; |
| 259 | channel_delay_->last_video_delay_ms = video_delay_ms; |
| 260 | channel_delay_->last_sync_delay = 1; |
| 261 | } else { // channel_delay_->last_sync_delay >= 0 |
| 262 | // First time after a delay change, don't add any extra delay. |
| 263 | // This is to not toggle back and forth too much. |
| 264 | channel_delay_->extra_audio_delay_ms = 0; |
| 265 | // Set minimum video delay |
| 266 | video_delay_ms = *total_video_delay_target_ms; |
| 267 | channel_delay_->extra_video_delay_ms = 0; |
| 268 | channel_delay_->last_video_delay_ms = video_delay_ms; |
| 269 | channel_delay_->last_sync_delay = 0; |
| 270 | } |
| 271 | } |
| 272 | } else { // if (current_diffMS > 0) |
| 273 | // The minimum video delay is lower than the current audio delay. |
| 274 | // We need to decrease possible extra audio delay, or |
| 275 | // add extra video delay. |
| 276 | |
| 277 | if (channel_delay_->extra_audio_delay_ms > 0) { |
| 278 | // We have extra delay in VoiceEngine |
| 279 | // Start with decreasing the voice delay |
| 280 | int audio_diff_ms = current_diff_ms / 2; |
| 281 | if (audio_diff_ms < -1 * kMaxAudioDiffMs) { |
| 282 | // Don't change the delay too much at once. |
| 283 | audio_diff_ms = -1 * kMaxAudioDiffMs; |
| 284 | } |
| 285 | // Add the negative difference. |
| 286 | channel_delay_->extra_audio_delay_ms += audio_diff_ms; |
| 287 | |
| 288 | if (channel_delay_->extra_audio_delay_ms < 0) { |
| 289 | // Negative values not allowed. |
| 290 | channel_delay_->extra_audio_delay_ms = 0; |
| 291 | channel_delay_->last_sync_delay = 0; |
| 292 | } else { |
| 293 | // There is more audio delay to use for the next round. |
| 294 | channel_delay_->last_sync_delay = 1; |
| 295 | } |
| 296 | |
| 297 | // Keep the video delay at the minimum values. |
| 298 | video_delay_ms = *total_video_delay_target_ms; |
| 299 | channel_delay_->extra_video_delay_ms = 0; |
| 300 | channel_delay_->last_video_delay_ms = video_delay_ms; |
| 301 | } else { // channel_delay_->extra_audio_delay_ms > 0 |
| 302 | // We have no extra delay in VoiceEngine, increase the video delay. |
| 303 | channel_delay_->extra_audio_delay_ms = 0; |
| 304 | |
| 305 | // Make the difference positive. |
| 306 | int video_diff_ms = -1 * current_diff_ms; |
| 307 | |
| 308 | // This is the desired delay. |
| 309 | video_delay_ms = *total_video_delay_target_ms + video_diff_ms; |
| 310 | if (video_delay_ms > channel_delay_->last_video_delay_ms) { |
| 311 | if (video_delay_ms > |
| 312 | channel_delay_->last_video_delay_ms + kMaxVideoDiffMs) { |
| 313 | // Don't increase the delay too much at once |
| 314 | video_delay_ms = |
| 315 | channel_delay_->last_video_delay_ms + kMaxVideoDiffMs; |
| 316 | } |
| 317 | // Verify we don't go above the maximum allowed delay |
| 318 | if (video_delay_ms > kMaxDelay) { |
| 319 | video_delay_ms = kMaxDelay; |
| 320 | } |
| 321 | } else { |
| 322 | if (video_delay_ms < |
| 323 | channel_delay_->last_video_delay_ms - kMaxVideoDiffMs) { |
| 324 | // Don't decrease the delay too much at once |
| 325 | video_delay_ms = |
| 326 | channel_delay_->last_video_delay_ms - kMaxVideoDiffMs; |
| 327 | } |
| 328 | // Verify we don't go below the minimum delay |
| 329 | if (video_delay_ms < *total_video_delay_target_ms) { |
| 330 | video_delay_ms = *total_video_delay_target_ms; |
| 331 | } |
| 332 | } |
| 333 | // Store the values |
| 334 | channel_delay_->extra_video_delay_ms = |
| 335 | video_delay_ms - *total_video_delay_target_ms; |
| 336 | channel_delay_->last_video_delay_ms = video_delay_ms; |
| 337 | channel_delay_->last_sync_delay = -1; |
| 338 | } |
| 339 | } |
| 340 | |
| 341 | WEBRTC_TRACE(webrtc::kTraceInfo, webrtc::kTraceVideo, video_channel_id_, |
| 342 | "Sync video delay %d ms for video channel and audio delay %d for audio " |
| 343 | "channel %d", |
| 344 | video_delay_ms, channel_delay_->extra_audio_delay_ms, audio_channel_id_); |
| 345 | |
| 346 | *extra_audio_delay_ms = channel_delay_->extra_audio_delay_ms; |
| 347 | |
| 348 | if (video_delay_ms < 0) { |
| 349 | video_delay_ms = 0; |
| 350 | } |
| 351 | *total_video_delay_target_ms = |
| 352 | (*total_video_delay_target_ms > video_delay_ms) ? |
| 353 | *total_video_delay_target_ms : video_delay_ms; |
stefan@webrtc.org | 7c3523c | 2012-09-11 07:00:42 +0000 | [diff] [blame^] | 354 | return true; |
stefan@webrtc.org | 5f28498 | 2012-06-28 07:51:16 +0000 | [diff] [blame] | 355 | } |
| 356 | } // namespace webrtc |