Research And Optimization Of MPEG-DASH Based Dynamic Adaptive Video Streaming Algorithm

With the rapid development of wireless communication and Internet technologies,mobile devices represented by mobile phones and tablet PCs have rapidly spread,and the demand for multimedia is rapidly increasing.Among various multimedia information content,video has become the main choice of network users due to the richness of its content.Therefore,research related to network video has been a hot content.Video Streaming is the core technology of the network video industry.In video streaming,dynamic adaptive streaming media over HTTP(DASH)is the most widely adopted streaming technology.The key advantage of DASH(HAS)is that the video bitrate adaptive switching is supported,so that the client can dynamically adjust the bitrate of video according to its own environment and network conditions,thereby ensuring the quality of experience(QoE)in different network environments.Previously,many researchers have proposed their own solutions,but these technologies depend on different platforms.In order to solve this problem,MPEG organization introduced the cross-platform standard streaming video transmission protocol MPEG-DASH.In April 2012,MPEG-DASH officially became an international standard.At present,MPEGDASH is widely supported by the industry and is being adopted by more and more multimedia providers.This paper deeply studies the MPEG-DASH protocol and its related dynamic adaptive streaming media technology.Since the existing adaptive bitrate(ABR)algorithms have their own limitations and shortcomings,as a result,it is difficult for them to achieve good performance under the condition of network bandwidth fluctuation.After studying the deep reinforcement learning which has been successfully implemented in many research such as image,speech and natural language processing,this paper then proposes and designs an adaptive bitrate(ABR)algorithms based on deep reinforcement learning.By using the real network dataset to train the reinforcement learning model,it then generates adaptive bitrate algorithm based on reinforcement learning.Finally,this paper also implements a complete environment according to the MPEG-DASH.It describes in detail the implementation and the technology adopted by each component of the framework.By using this system as the test environment,this paper compares the designed algorithm with other algorithms and draws conclusions.The simulation results show that the proposed algorithm based on deep reinforcement learning can select appropriate bitrates for future video segments according to all the information collected by users during video playback,thus significantly improving the quality of video service.