【作者】 曾涛；

【导师】 戴琼海；

【作者基本信息】 清华大学 ， 控制科学与工程， 2005， 硕士

【摘要】 流媒体技术是应用于因特网上的一种重要的音视频传输技术,这项日趋成熟的技术保证了多媒体数据能在网络传输的同时进行解码播放。由于音视频数据量非常庞大,有限的因特网带宽如何能最大限度的担负如此巨大的传输任务是流媒体应用中最需要解决的问题。从对媒体采集,编码压缩,到网络传输,客户端的接收,解码,播放的研究,最终目的都是提供一定码率下最优的终端播放质量。作为一个具有巨大市场潜力的运营平台,流媒体业务吸引了众多的研究单位和开发商的参与,得到迅速的发展,一系列标准协议也逐步形成。清华大学宽带网数字媒体实验室开发了具有自主知识产权的基于 Linux 的流媒体业务平台(LSMP)系统,目标是提供基于 IP 网络的流媒体增值业务整体解决方案,涵盖了流媒体直播点播服务器、客户端播放器、负载均衡集群系统,用户认证计费系统,媒体制作工具等多个模块。本论文是在整个 LSMP 项目的背景中,研究、设计、搭建并完善了其中的客户端部分。论文内容主要包括:一是设计开发流媒体客户端播放器系统。通过大量的实际开发工作,实现了客户端对音视频数据的传输与解码功能,用户 VCR 控制,音视频变速播放技术,以及其他许多相关的应用。作为 LSMP 项目的一个重要组成部分,客户端涵盖了相当多成熟的技术,不断的完善和升级使它已经可以稳定高效的运行于Windows 平台上,实现网络与本地媒体的高质量播放。二是基于神经网络控制的自适应媒体播放(AMP)的算法研究。这种新的方法在原有的可提高客户端缓冲稳定性的自适应播放算法的基础上,引入了人工神经网络的控制算法,实现输出的可变播放速度能随当前缓冲状态自适应动态变化,减少由于 AMP 的变速播放带来的对质量的损害,同时保证缓冲仍然具有较高的鲁棒性。在统一的控制系统结构中分别讨论了单层和多层神经网络及其相应的学习算法,通过离线训练保证了控制器能获得理想的速度输出,实现了客户端的动态自适应播放,为终端用户提供了更高质量的服务。论文还涉及了运营平台,差错控制,媒体制作和无线信道传输等相关研究。更多还原

【Abstract】 Streaming media is a significant and developing technique transmitting videoand audio data on Internet, which ensures that the processes of transmission and playcan proceed concurrently. Because media’s quantities are usually greatly huge, howto afford such a massive carrying task furthest within a limited network bandwidthbecomes a problem to be resolved in streaming application. From media source’scollection, encode to data’s send, receive, decode and play, the final aim is to achievethe optimization of the quality of service. As a business platform with great marketpotential, streaming media service absorbed plentiful researchers and developers andgrowing up very soon during less than a decade. A series of relevant standardprotocols are established gradually.The Broadband Network and Digital Media Lab in Tsinghua University builds aLinux-based Streaming Media Platform (LSMP), aiming to provide a holistic solutionfor streaming value-added service based IP network, which includes streaming vodand live servers, client player, load-balance cluster, authentication and accountingsystem, meida producer, etc. This paper studies, designes and achieves the client ofstreaming system. The main work consists of two parts:The first is the streaming client player system. With lots of actual work, itrealizes the receiving and decoding functionality, VCR control, variable speedplayout and some other correlative applications. As a basilic module, the clientinvolves quite a few mature techniques. It could run in Windows steadily, performingmedia’s network and local play with high quality.The second is the research of adaptive media playout (AMP) based on neuralnetwork control. This new scheme presents neural network control arithmetic,realizing speed’s dynamic variance according to current buffer’s status. It reduces thehurt of quality from speed’s variance and ensures buffer’s high robustness. In thecontrol architecture with single-layer or multi-layer controller, trained by relevantlearning algorithm, it exports speed in expectation, which achieves client buffer’sdynamic adaptive playout and provides service of high quality for users.The paper has also included some relative research on business platform, errorcontrol, media producer and wireless transmission.更多还原