Research On Real Time Media Streaming Distribution Strategy For Ultra Low Latency Scenarios

With the rapid development of the Internet,video conferencing,online games,live video and other emerging applications with streaming media technology as the core emerge in large numbers.They all have the common characteristics of low delay.The pursuit of delay ranges from seconds(Live),to hundreds of milliseconds(video conference),and then to milliseconds(cloud games,VR/AR,Internet of vehicles).Driven by the needs of these new application scenarios,the direction of the Internet must be faster and faster,more and more real-time.These new ultra low latency applications are richer than traditional network intensive applications(simultaneous interpreting Web,FTP,multimedia streaming,etc.),including voice,video,voice,text,and operation information.Among them,the video stream can be divided into different definition and different quality video slices(such as dash).The research of choosing to download video with different bit rate under different network conditions has become a hot research direction in the academic and industrial circles,that is,the field of bit rate adaptation.In addition,these concurrent media streams usually have different priorities and specific deadline,which makes it possible to schedule selectively in the transport layer according to their characteristics.Aiming at the ultra-low delay requirements of emerging applications,this paper designs a delay sensitive real-time media stream distribution strategy to reduce the delay from two aspects:one is the delay sensitive rate adaptive algorithm integrated with deep reinforcement learning,and the development of new applications poses many new technical challenges to the adaptive bit rate(ABR)algorithm,which not only requires stable high-quality transmission,but also has the advantages of low delay And low end-to-end delay is needed.Reinforcement learning(RL)can automatically learn ABR algorithm without using any pre programmed control rules,and has achieved good results.However,the existing methods only consider the bit rate control and ignore the delay control.Therefore,in order to effectively reduce the end-to-end delay,this paper proposes an independent delay limiting model to control frame skipping.In addition,a model integration algorithm is proposed to reduce the performance variance and improve the quality of user experience.The second is delay sensitive transport layer scheduling algorithm,which includes two scheduling algorithms:congestion control algorithm based on heuristic rules and DRL,packet selection scheduling algorithm,which determines the sending rate and which packets to send respectively.On the basis of detecting network bandwidth,bandwidth can be used as effectively as possible to reduce bandwidth waste or network congestion and meet the deadline of packets as much as possible.To sum up,for the emerging ultra-low latency applications,the design of a suitable streaming media distribution strategy can effectively adapt to the rapid development of network transmission and provide higher quality services.This paper will try to solve this problem from two aspects:the terminal side distribution strategy and the network side distribution strategy.The experimental results show that the two algorithms are significantly better than the baseline method,which proves the effectiveness and practicability of the strategy.