Transmission Control Of Streaming Video Based On Neural Network And Particle Swarm Algorithm

With the rapid spread in scales and wide permeation in applications, the Internet is more and more indispensable both in our works and daily lives. It has brought so much convenience to us especially with the emergence of network multimedia, and there are increasing demands for the applications of video conference, distance education, VOD (Video on Demand), IPTV (Internet Protocol Television), etc., but obviously, the traditional method which is to browse the content after downloading the entire media file from the Internet is no longer meet their requirements, thereby streaming media came into being. A brief account of streaming media is that it is a transmission technique for multimedia supporting simultaneous downloading and audiovisual.But there exist lots of challenges in streaming media transmission currently. Firstly, the bandwidth is insufficient for streaming media transmission, because most bandwidths of end users are very limited. Secondly, the current best-effort Internet cannot satisfy any QoS (Quality of Service) guarantees, considering the delay, jitter, package error ratio, package loss ratio, smoothing, etc. All of the pending issues make streaming media transmission control a research hotspot nowadays.In this dissertation, some relative background knowledge is introduced in the beginning chapters, and then on the basis of synthetical analysis of the state of arts of streaming media transmission control, a universal streaming video transmission control method is proposed. The traffic model and a suit of differential equations presenting the status of the system are given first, from which an objective function is derived, and then the transmission is optimally controlled by the neural network which is characterized by nonlinear map and the particle swarm optimization algorithm which is characterized by stochastic optimization, namely the neural network is employed to generate variable rate of token generation, and the particle swarm optimization algorithm with inertia weight is employed to optimally train neural network in the form of finding a sub-optimal resolution in acceptable computation time. In order to prove its feasibility and superiority, a segment of MPEG-4 encoded streaming video is used for simulation. The experimental results and analyses show that this method outperforms the traditional ones that are based on the fixed token generation rate in the same condition of network throughput.