WebRTC研究：丢包判断

使用RTP协议封装数据时，我们可以通过RTP报头的序列号连续性判断是否丢包。但由于RTP报头序列号只有两字节表示，值范围[0,65535]，存在回绕问题（参考之前文章：WebRTC研究：RTP中的序列号以及时间戳比较，建议先阅读一遍此文章）。所以判断序列号连续性时得考虑回绕问题。下面我们就结合WebRTC这相关源码，讲下如何有效地根据序列号进行丢包判断。

首先看下这块代码，代码位于jitter_buffer.cc中：

bool VCMJitterBuffer::UpdateNackList(uint16_t sequence_number) {
  // Make sure we don't add packets which are already too old to be decoded.
  if (!last_decoded_state_.in_initial_state()) {
    latest_received_sequence_number_ = LatestSequenceNumber(
        latest_received_sequence_number_, last_decoded_state_.sequence_num());
  }
  if (IsNewerSequenceNumber(sequence_number,
                            latest_received_sequence_number_)) {
    // Push any missing sequence numbers to the NACK list.
    for (uint16_t i = latest_received_sequence_number_ + 1;
         IsNewerSequenceNumber(sequence_number, i); ++i) {
      missing_sequence_numbers_.insert(missing_sequence_numbers_.end(), i);
    }
    if (TooLargeNackList() && !HandleTooLargeNackList()) {
      RTC_LOG(LS_WARNING) << "Requesting key frame due to too large NACK list.";
      return false;
    }
    if (MissingTooOldPacket(sequence_number) &&
        !HandleTooOldPackets(sequence_number)) {
      RTC_LOG(LS_WARNING)
          << "Requesting key frame due to missing too old packets";
      return false;
    }
  } else {
    missing_sequence_numbers_.erase(sequence_number);
  }
  return true;
}

bool VCMJitterBuffer::UpdateNackList(uint16_t sequence_number) {

// Make sure we don't add packets which are already too old to be decoded.

if (!last_decoded_state_.in_initial_state()) {

latest_received_sequence_number_ = LatestSequenceNumber(

latest_received_sequence_number_, last_decoded_state_.sequence_num());

}

if (IsNewerSequenceNumber(sequence_number,

latest_received_sequence_number_)) {

// Push any missing sequence numbers to the NACK list.

for (uint16_t i = latest_received_sequence_number_ + 1;

IsNewerSequenceNumber(sequence_number, i); ++i) {

missing_sequence_numbers_.insert(missing_sequence_numbers_.end(), i);

}

if (TooLargeNackList() && !HandleTooLargeNackList()) {

RTC_LOG(LS_WARNING) << "Requesting key frame due to too large NACK list.";

return false;

}

if (MissingTooOldPacket(sequence_number) &&

!HandleTooOldPackets(sequence_number)) {

RTC_LOG(LS_WARNING)

<< "Requesting key frame due to missing too old packets";

return false;

}

} else {

missing_sequence_numbers_.erase(sequence_number);

}

return true;

}

IsNewerSequenceNumber函数以及latest_received_sequence_number_介绍在之前文章说过，读者可参考前面给的文字链接。上述代码丢包判断用到的相关定义如下：

IsNewerSequenceNumber：判断是否是最新的序列号
sequence_number：当前收到的包序列号
latest_received_sequence_number_：上一次收到的最新RTP包序列号
missing_sequence_numbers_：存储丢包序列号

丢包判断流程如下：

首先IsNewerSequenceNumber传入当前收到的包序列号sequence_number以及之前收到的最新RTP包序列号latest_received_sequence_number_，判断当前RTP包是否是新的数据包，而不是重传包或者乱序包
在for循环中，从latest_received_sequence_number_ + 1开始判断（因为如果丢包了，两个包序列号间隔大于1），满足IsNewerSequenceNumber(sequence_number, i)则说明i为丢失的RTP包序列号，插入missing_sequence_numbers_中，接着++i继续判断，否则退出

通过如上方法，我们就可以获取那些丢失的包序列号，是不是很简单。