Research On The Problem Of Adapative Streaming Media Transmission Based FLV File

With the increasing requirements of audio and video playback, streaming media has become atrend. Streaming media means that the continuous time-based medias transfer using steamingtechnology. The streaming media accord with the follow features: the server firstly spilt the audioand video files which are compressed through a special way into continuous compression packages,then the server sends theses real-time packages to customer via the networks continuously; thecustomers do not need to download the entire video file, but just wait a short start-up and the firstfew compressed packest transmission time; as the same time subsequent compression packages aredownloaded until the transfer is completed. But the traditional streaming media system cannot adaptto changing network, and makes the client cannot get good video playback or uses networkbandwidth unefficiently. To solve this problem, we need to design a straming media transmissionsystem with flow control. This thesis designs a transmission system based on Darwin StreamingServer using RTP/RTCP and RTSP, whick can monitor the real-time network bandwidth and providvideo slices to of different quality.The thesis firstly analyzes technology related to streaming media transmission, mainly study theRTP/RTCP transport protocol, summarizy and compare two adaptive streaming systems based onRTSP and HTTP, make the decision to design a system using RTSP. Secondly, the thesis analyse thefile structure of FLV. With utilizing the analysis, the thesis gives the real-time algorithm of FLV fileand logic slicing algorithm. Thirdly, the thesis designs a straming media model taking advantage ofthe Darwin system. By expanding its module, we add the fuction to support for sending FLV slices,and uses a simple fow control strategy, namely using TFRC to determine the quality of slice byestimating the bandwith of the network. Lastly, in order to make up for the deficiencies of TFRCbandwidth estimation, the thesis introduce Markov model into flow control strategy. The systemsimulate client side buffer model on the transmit side, and use the extended RTSP packet content topredict the condition of client buffer. According to Markov decision set, the server can determinethe quality level of the next slice.As the result from the NS2simulation, compared to the original“best effort” strategy, Markov decision model is more suitable for changing network environment.