Design And Implementation Of Adaptive Bit-Rate Algorithm For Low-Latency Live Streaming

With the development of Internet and the popularity of network entertainment,the application of live video streaming becomes more and more concerned.People use mobile phones,computers and other electronic devices to watch various live applications,such as live games and live sports competitions,at any time and any place.This makes the problem of video streaming transmission in the live scenes extremely difficult.How to transmit stable video streams with high quality and low latency in the complex network environment and diverse video types has become a major challenge in the field of real-time streaming.In response to the above challenges,this paper proposes an adaptive bitrate algorithm for low-delay live streaming based on reinforcement learning theory.The main work includes:(1)Aiming at the problem that the traditional algorithm with video block as the basic scheduling unit cannot effectively reduce the latency,this paper will take video frame as the basic scheduling unit to construct the algorithm and implement the corresponding video emulator.(2)For the low latency demand of live streaming video,we established a new user experience quality model.In this paper,latency is taken as a significant factor to evaluate the quality of user experience,and a variety of auxiliary mechanisms such as buffer threshold mechanism,benefit limitation mechanism and skip-frame penalty mechanism are added.(3)Aiming at the problems of "inaccurate bitrate decision" and"error learning" caused by different video frame rates,this paper proposes the concept of training mode and constructs a fixed-point memory strategy.By recording the corresponding relationship between decision points and learning points,the training effect of the algorithm model is improved successfully.Experimental results show that the fixed-point memory strategy improves the algorithm performance by an average of 16%compared with other training modes.(4)Aiming at the problem of convergence accuracy in algorithm training,we proposed a dynamic reward method in this paper.According to the value of reward generated during each round of training,the calculation method of reward function is changed dynamically to make the algorithm model converge more quickly and accurately.The experimental results show that the dynamic reward method improves the effect of fixed point memory strategy by 5%to 10%.Finally,based on the above work,this paper designs an adaptive bitrate algorithm for low-latency live streaming.Compared with other advanced bitrate adaptive algorithms under multiple evaluation models,the algorithm improves the effect by 15%-58%...