Research And Implementation Of Deep Reinforcement Learning Based Transmission Control Algorithm And Transmission Evaluation Platform

With the rapid development of the Internet and the arrival of the 5G era,more and more novel network services and applications have emerged.Some of these applications,such as Metaverse,cloud VR gaming,and selfdriving,are delay-sensitive,requiring the data to arrive within the users’acceptable end-to-end latency(i.e.,meeting deadline).These novel applications have brought new challenges to guarantee users’ QoE(Quality of Experience).DRL-TC(Deep Reinforcement Learning based Transmission Control Algorithm)is proposed in this thesis to meet the delay requirements of these delay-sensitive multimedia applications and improve users’ QoE of these novel applications.DRL-TC is composed of a block scheduling component and a congestion control component.The block scheduling component predicts the blocks that may miss the deadline and takes the blocks’ deadline,priority,and network conditions into full consideration to decide the sending order of blocks.A deep reinforcement learning framework is combined with the congestion control component to solve the problem of insufficient performance of heuristic congestion control algorithms in delay-sensitive multimedia applications.This congestion control component adapts to complex network statuses and improve users’QoE of novel applications.The simulation results show that DRL-TC performs well in various scenarios and outperforms other benchmark algorithms.In addition,an open network platform that provides differentiated network scenarios is conducive to exploring and evaluating various learning-based transmission control algorithms,thus promoting the further research of learning-based transmission algorithms.Therefore,a transmission control evaluation platform is designed and developed in this thesis for algorithm researchers to evaluate and compare the QoE performance of different transmission algorithms.The evaluation platform automates the deployment and multi-scene evaluation of transmission control algorithms.It monitors and manages deployment tasks.