Research Of Adaptive Media Playout Based On Neural Network Control And Realization Of Streaming Player

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.