网络多媒体服务系统的码率控制与性能优化

发布时间：2018-01-19 19:06

本文关键词： 网络多媒体码率控制软件定义网络大偏差原理 Markov决策过程混合高斯模型下溢概率　出处：《中国科学技术大学》2016年博士论文　论文类型：学位论文

【摘要】：互联网时代,网络多媒体技术正在深刻地改变着人们的生活方式。随着与不同行业的融合,网络多媒体服务已经深入生活的方方面面,例如网络电视(Internet Protocol Television,IPTV),视频会议,远程教育,远程医疗,新闻出版,娱乐,电子商务等等。近年来,无论从应用的数量还是从用户的规模上都呈现爆发式增长,凸显出了巨大的市场价值和发展前景。但是,用户需求的激增随之而来地带来各种挑战。高清、流畅的多媒体服务需要稳定的带宽,然而用户行为和网络状态的随机性、时变性、和复杂性通常难以保证服务的质量。研究如何克服这些问题具有现实意义,也是当前研究的热点。此外,互联网技术的变革催生了各种新的思想,如数据面与控制面分离、名字和地址分离、可编程路由等,也为网络多媒体的发展注入新的活力。研究如何将多媒体技术与未来网络技术融合,从而更好地保障用户的服务质量、提高网络资源利用率。对于满足用户日益增长的需求,具有深远意义。本文以网络多媒体服务系统为切入点,以控制与优化理论为技术手段,对所涉及的自适应资源配置和调度等问题进行深入研究,具体工作总结如下:1)无线网络中,为克服用户移动和信道变化的随机性,研究如何设计有效的媒体自适应策略,从而保证多媒体业务的服务质量。为实现媒体自适应性,本文采用可伸缩的编码方式,并设计实现了客户端驱动的可伸缩视频传输系统。通过对播放过程进行分析和建模,提出和定义下溢概率。其综合考虑了信道的状态和接收端的缓冲区状态。随后,该问题被建模为一个带有下溢概率约束的优化模型。为了获得更好的用户体验,用户通常希望下溢的发生是小概率事件。因此,采用大偏差原理构建下溢概率的估计模型,从而得到自适应的码率控制算法。进一步地,本文考虑“闪烁效应”对视频播放的影响,并提出基于摄动理论的改进码率控制算法。为验证算法部署的可行性,设计实现原型系统。全面的仿真系统实验表明,算法能够适应不同的应用场景,并且很好地实现播放流畅性、平滑性以及视频质量的折中。2)软件定义网络(Software DefinedNetwork,SDN)中,控制器可以观测到网络中间节点的状态(瞬时吞吐量、时延、丢包率等)。因此,通过合适的控制器算法可以提前感知网络拥塞的变化。本文依托流行的基于HTTP的动态自适应流媒体(Dynamic Adaptive Streaming over HTTP,DASH)技术,研究如何实现网络感知的自适应多媒体服务。问题的核心是要研究建立合适的数学模型预测未来长时间内网络带宽的变化。为克服网络瞬时流量短时随机变化的影响,引入链路拥塞概率。该问题被建模为链路拥塞概率约束的优化问题。为了尽量避免因链路拥塞引起的丢包,用户通常期望链路拥塞的发生是小概率事件,因此,可以采用大偏差原理进行估计。此外,由于此模型依赖于链路流量分布等先验知识。本文采用基于混合高斯模型的极大似然估计方法对该分布进行在线估计,从而得到在线的码率自适应算法。原型系统实验结果表明,相比于传统的DASH服务平台,算法能够在有效避免链路拥塞的同时提供流畅、高质量的多媒体服务。3)利用SDN网络路由可编程的特点,可以实现动态路由和多路径传输。本文将传统的码率控制与动态路由结合,考虑如何设计联合优化策略,从而实现更流畅、高质量的视频服务。具体来说,在数据传输过程中,如果某一网络中间位置发生拥塞,控制器可以探知到该拥塞,并能够动态寻找和路由到另一条通畅的路径,如果这样的路径不存在则降低码率。本文采用Markov决策过程对该联合优化问题进行建模。求解时,为避免对转移概率等先验知识的要求,提出基于Q学习的在线最优策略求解算法。Mininet平台的仿真结果表明,该系统能够充分利用网络带宽,进一步提高视频质量。
[Abstract]:The age of the Internet, network multimedia technology is profoundly changing the way of people's life. With the integration of different industries and the network multimedia service has in-depth every aspect of life, such as network television (Internet Protocol, Television, IPTV), video conference, distance education, telemedicine, news publishing, entertainment, electronic commerce and so on. In recent years, in terms of the number of applications or from the user scale are showing explosive growth, highlights the huge market value and development prospects. However, the subsequent surge in user demand has brought many challenges. HD, smooth multimedia services require stable bandwidth, however, randomness, user behavior and network status when degeneration and complexity are often difficult to guarantee the quality of service. Research on how to overcome these problems is of practical significance, but also a research focus. In addition, the Internet. Operation changes spawned a variety of new ideas, such as data plane and control plane separation, separation of name and address, programmable routing, has injected new vitality into the development of multimedia network. Research on how to apply multimedia technology and the integration of network technology in the future, so as to better protect the user's quality of service, improve the utilization rate of network resources. To meet the growing demand of users, has far-reaching significance. This paper takes the network multimedia service system as the starting point, to control and optimization theory of technology, in-depth study of the adaptive resource allocation and scheduling problem, the specific work are summarized as follows: 1) in a wireless network, to overcome randomness mobile users and channel change, study how to design adaptive strategy of media effectively, so as to ensure the quality of multimedia services. In order to realize the media using adaptive, scalable Encoding method, and the design and implementation of client driven scalable video transmission system. Through the analysis and modeling of the playing process, and put forward the definition of underflow probability. Considering the channel state and the receiver buffer status. Then, the problem is modeled as an optimization model with underflow probability constraints. In order to obtain a better user experience, users often want to underflow occurs is a small probability event. Therefore, the construction of the underflow probability estimation model of large deviation principle, so as to obtain the adaptive control algorithm of rate. Further, considering the "flicker effect" on the impact of video playback, and the perturbation theory of the improved rate control algorithm in order to verify the feasibility of the algorithm. Based on the deployment of the prototype system. Experiments show that the simulation system comprehensively, the algorithm can adapt to different application scenarios, and well To achieve playback fluency, smoothness and video quality tradeoff.2 software defined network (Software) DefinedNetwork, SDN), the controller can be observed at intermediate nodes in the network state (instantaneous throughput, delay, packet loss rate). Therefore, by changing the appropriate controller algorithm can advance perception network congestion. This paper relies on the popular dynamic adaptive streaming media based on HTTP (Dynamic Adaptive Streaming over HTTP DASH) technology, studies how to realize the adaptive multimedia service network. The core of the problem is to establish an appropriate mathematical model to predict the future changes in network bandwidth in a long time. In order to overcome the influence of random network instantaneous flow short-time change, the introduction of link congestion probability the problem is modeled as optimization of link congestion probability constraints. In order to avoid congestion caused by the lost packets, users usually period At the occurrence of link congestion is a small probability event, therefore, can be used to estimate the large deviation principle. In addition, because this model depends on the link flow distribution of prior knowledge. This paper uses maximum likelihood estimation method of mixed Gauss model to estimate the distribution of the resulting bit rate based on adaptive Online algorithms. The experimental results of prototype system show that compared to the traditional DASH service platform, the algorithm can effectively avoid congestion and provide smooth, high quality multimedia service.3) using SDN network routing programming, can realize dynamic routing and multi-path transmission. This paper will combine traditional rate control and dynamic routing, consider how to design the joint optimization strategy and in order to achieve a more smooth, high quality video service. Specifically, in the data transmission process, if a middle position of network congestion, The controller can detect the congestion, and can dynamically search and routing path to another unobstructed, if such a path does not exist on the lower bit rate. This paper uses the combined optimization model of Markov decision process. When solving the problem, in order to avoid the transfer probability of the prior knowledge of the requirements, puts forward the simulation Q online learning the optimal strategy algorithm.Mininet platform based on the results, the system can make full use of network bandwidth, further improve the quality of the video.

【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP393.0

【相似文献】