节点文献

基于TCP的实时流媒体自适应传输策略及其应用研究

Study and Application of Rate Adaptive Real-Time Multimedia Streaming Transmission Scheme over TCP

【作者】 熊永华

【导师】 吴敏; 贾维嘉;

【作者基本信息】 中南大学 , 计算机应用技术, 2009, 博士

【摘要】 实时流媒体是下一代网络的主要传输对象,将极大地影响人们日常生活。使用TCP传输实时流媒体具有对网络状况反应迅速、TCP友好、易于为防火墙所接受、便于实现与维护等优点,但同时也存在着发送速率无法自适应网络可用带宽的波动、TCP重传易增加媒体包的延时抖动和乱序等问题,这些都大幅降低了实时流媒体的服务质量。为解决上述问题,本文从TCP实时流媒体传输性能分析、延时预测模型、自适应传输策略及其应用等方面开展研究,取得的创新型研究成果如下。(1) TCP实时流媒体传输具有可接受播放性能的条件通过深入分析TCP实时流媒体传输过程,指出使用TCP在Internet上传输实时音频时,具有可接受播放性能的条件:TCP最大报文段的大小应设置为单个语音帧的大小,但这将使得TCP的传输效率非常低下,不适用于Internet应用。指出实时视频传输具有可接受播放性能的条件:视频帧的发送延时应小于播放缓冲延时与网络往返延时一半的差值。模拟结果表明了所指出条件的合理性。(2)影响TCP实时视频端到端延时的关键因素通过推导和模拟实验剖析了实时视频的端到端延时,指出等待延时是实时视频端到端延时的主要构成,发送延时则是影响端到端延时的关键因素,提出可以通过减少等待延时和减少发送延时的方式,来减少视频帧端到端延时抖动。由于发送延时由TCP协议栈管理,应用程序无法对其进行控制,提出了一种对发送延时进行预测判断的方法:预测视频帧的发送延时,若发送延时不满足实时视频传输的条件,则选择性丢弃该视频帧,从而确保了所有被传输的视频帧都具有较小的发送延时。模拟结果表明使用该方法可以有效减少实时视频的端到端延时。(3)两种实时视频发送延时预测模型针对Internet上两种使用最为广泛的路由器丢包机制,即尾部丢包(DropTail)和随机早期丢包(RED)机制,分别建立了两种实时视频发送延时预测模型。1)分阶随机预测模型。从微观角度分别考虑视频帧在传输过程中处于TCP各个阶段的概率,特别是DropTail机制下易出现的慢开始和RTO阶段,以帧大小、丢包率、拥塞窗口大小、网络往返时间和TCP初始重传计时器时间作为输入参数,以视频帧发送延时的期望值作为预测值。2)递阶式马尔科夫预测模型。从宏观角度将RED机制下整个传输过程的拥塞窗口变化作为一个离散时间离散状态的马尔科夫链;设计了一个递阶式马尔科夫预测算法,算法以窗口大小、网络往返时延和丢包率作为状态转移矩阵函数的输入参数,以拥塞窗口大小作为增益矩阵函数的输入参数,以各个状态拥塞窗口大小之和大于等于视频帧大小为递阶结束条件,对视频帧发送延时进行预测。模拟结果表明两种模型的预测精度符合要求,预测值可以被用于视频帧的预测判断方法中。(4)两种实时视频自适应传输策略从减少等待延时和发送延时入手,建立了两种实时视频自适应传输策略:1)基于多缓冲区调度的自适应策略。设计了一个多缓冲区调度模型,通过建立一个具有延时级别的应用层发送缓冲区,对实时视频帧在该缓冲区、TCP发送缓冲区和播放缓冲区之间进行调度,使用基于缓冲区延时级别的选择性丢弃视频帧的方法,将所有视频帧的等待延时减少为最大仅为编码延时的两倍,同时根据丢弃率的变化对视频帧率和播放缓冲大小进行自动调整。2)基于发送延时预测模型的自适应策略。在视频帧被发送以前,首先利用发送延时预测模型对发送延时进行预测,然后使用预测判断的方法,对达不到可接受播放性能的视频帧进行选择性丢弃,确保了所有被传输的视频帧都具有较小的发送延时,并将所有视频帧的等待延时减少为最大仅为编码延时的一倍,同时以丢弃率为约束条件对视频帧率和播放缓冲大小进行自动调整。模拟结果表明,与直接使用TCP或UDP传输相比,两类自适应策略都能够大幅度减少端到端延时抖动,并使得发送速率自适应网络拥塞状况,有效避免关键帧的丢失。基于发送延时预测模型的策略在减少端到端延时上具有更好的性能,但要求预知瓶颈路由器的丢包机制。(5)自适应传输策略在基于3G的实时通信系统中的应用与实现采用自适应传输策略,利用面向对象设计方法、IO完成端口、原始套接字和TCP防火墙穿透等技术,设计并实现了一种新型的基于3G的实时通信系统3G-VIM,系统实现了IP网内的PC和3G网内的3G手机之间的双向实时通信。系统运行结果表明,所提出的实时视频自适应传输策略能够明显提高实时视频的视觉质量。

【Abstract】 Real-time multimedia streaming will be the main transmission objective on the next generation network, which could have large effect on our lives. There exist many advantages such as reliable transmission, rapid reaction to network congestion, TCP friendliness, acceptableness to fire walls, convenient for implementation and maintenance and so on while using TCP to transmitting real-time multimedia streaming. However, on the other hand, the disadvantages are obvious as follows. The sending rate is unable to be adaptive to the available network bandwidth. The retransmission arisen from TCP would likely increase the delay jitter and confusion of multimedia packet. These disadvantages decrease notablely the quality of service of real-time multimedia streaming. Amied at the above problems, performance analysis of real-time multimedia streaming transmission, sending-delays prediction model, rate adaptive transmitting scheme and its application are researched in this thesis and the main creative achievements included are the following five aspects.(1) Requirement for transmitting real-time multimedia streaming via TCP with acceptable playing performanceThe process of transmitting real-time multimedia via TCP is deeply analyzed and then the requirement of transmitting real-time audio with acceptable playing performance via TCP is pointed as that the TCP MSS should be equal with the size of a audio frame. However, the transmitting efficiency of TCP will be very low under this condition so that it is infeasible in the Internet. The requirement of transmitting real-time video is also pointed as that the video frame sending-delays should be less than the results that play out buffer delays subtract half of the round trip time (RTT). The simulation results show that the requirement presended is reasonable.(2) The critical factor impacting on the end-to-end delays of real-time video transmitted via TCPThe end-to-end delays of real-time video are dissected by deriving and simulating, and it is discoveried that the waiting-delays are the main part of end-to-end delays of real-time video and the sending-delays are the critical factor impacting on the end-to-end delays. Two ways to decrease the end-to-end delays and jitters are presented, which are reducing the waiting-delays and cutting donw the sending-delays. Since the sending-delays are straightly and rigidly managed by the TCP stack, it is impossible for applications to control it. Therefore, a method to decrease the end-to-end delays by predicting and judging the sending-delays is proposed as follows. Predict the sending-delays of a video frame and if it can not achieve the requirement mentioned above, then the video frame will be selectively discarded. Utilizing the method of predicting and judging, the sending-delays of all the video frames deliveried are little. The simulation results show that the end-to-end delays of real-time video can be cut down notablely by using the method.(3) two real-time video sending-delays prediction modelsAimed to the two widely-used router loss strategies, DropTail and RED, two real-time video sending-delays prediction model are established in the thesis.1) The well-phased stochastic prediction model. The model microscopically considers the probability of video frame placed in diverse phases, especially the slow starting and RTO phases, which happen more easily while using DropTail strategy. The model’s input parameters include frame size, loss ratio, congestion windows size, RTT, RTO time and so on. The mathematic expectation of sending-delays is looked as the prediction value of sending-delays.2) The recursive Markov prediction model. The model macroscopically treats the variation of TCP congestion windows during the whole TCP transmitting process as a Markov chain with discrete time and state. A recursive Markov prediction algrithm is presented, in which current congestion window size, RTT and loss ratio are input as the parameters of the state jumping matrix function, and congestion window size is input as the parameters of the reward matrix function. The recursive process of algrithm ends while the summations of prediction values are larger than the video frame sizes, and then outputs the prediction value of sending-delays.The simulation results show that the accuracy of both the prediction model are acceptable and the prediction values can be used in the method of predicting and judging mentioned above.(4) two real-time video rate adaptive transmission schemeConsidering the two way of decreasing the end-to-end delays pointed above, two real-time video rate adaptive transmission scheme are established.1) Rate adaptive transmission scheme based on multi-buffer scheduling. A sending buffer on application layer with delays level is set up to schedule real-time video frames between the sending buffer, TCP sender buffer, and play out buffer. Using the method of selectively discard video frames based on buffer delays level, the waiting-delays of each video frame is decreased to be only less than twice of encoding delays. Simultaneously, the video frame rate and sizes of play out buffer can be adjusted automaticly according the fluctuation of discarding ratio.2) Rate adaptive transmission scheme based on the sending-delays prediction model. Firstly, the sending-delays of video frame are predicted before it is sended using the sending-delays prediction model. Afterwards, the method of predicting and judging is ultilized to selectively discard the video frames without the acceptable playing performance. Therefore, the sending-delays of each video frame transmitted can be little and the waiting-delays of each video frame is cut down to be only less than encoding delays. At the same time, the video frame rate and sizes of play out buffer can be regulated automaticly according the fluctuation of discarding ratio.The simulation results show that compared with UDP and TCP, using either the adaptive schemes established in this thesis, the end-to-end delays and jitters can be reduced to a larger extent, the sending rate can be adaptive to the network congestion condition, and the key video frames can be avoided to be discarded. Moreover, the rate adaptive transmission scheme based on the sending-delays prediction model is better to decrease the end-to-end delays, howerver, it requires to get the loss strategy of bottleneck router, which brings forth some limitations.(5) Application and implementation of the rate adaptive transmission scheme in the real-time communication system based on 3G networkUsing the rate adaptive transmitting scheme based on TCP and utilizing the methods and techniques of object-oriented design, database optimization access, IO completion port, raw socket, and TCP NAT penetrating, a novel real-time communication system based on 3G network is designed and implemented. The system realizes the two-way real-time audio and video communication between PC in IP network and 3G phone in 3G network with the requirement of quality of service. The running results of system show that the visual quality of video can be improved obviously after using the rate adaptive transmission schemes proposed in the thesis.

  • 【网络出版投稿人】 中南大学
  • 【网络出版年期】2010年 02期
节点文献中: 

本文链接的文献网络图示:

本文的引文网络