Resarch On Deterministic Latency Routing Algorithm Of LEO Satellite Constellation Network

With the advent of the 5G era,advanced applications such as digital twins and smart cities have emerged,presenting an urgent demand for deterministic latency in terms of Quality of Service(Qo S).Unlike traditional low-latency,deterministic latency must not only ensure an upper bound on delay but also achieve bounded delay jitter.To meet this demand,both the industry and academia are actively conducting research.Among them,low Earth orbit satellite constellation networks,with their low latency,high bandwidth,wide coverage,and low cost,have become an important solution.However,due to the time-varying nature of satellite networks,the design of routing algorithms faces severe challenges,such as constructing stable end-to-end transmission paths,ensuring the Qo S requirements of multiple services,and improving network resource utilization.In response to these issues,this thesis investigates a deterministic latency routing algorithm for multi-service applications,suitable for low Earth orbit satellite constellation networks with time-varying topology and resources.The research work in this thesis is mainly divided into three aspects.Firstly,a comprehensive modeling of LEO satellite constellation networks is conducted,including constellation models,network models,communication models,delay models,and service models,providing theoretical support for the design of routing algorithms and the construction of LEO satellite constellation network simulation test platforms.Secondly,with the optimization goal of minimizing average end-to-end delay and delay jitter,the data packet routing problem in LEO satellite constellation network scenarios is modeled as a pure integer linear programming problem.Considering the NP-hardness of the problem and the complex characteristics of the satellite network environment,a solution approach is proposed that transforms the problem into a distributed partially observable Markov decision process.Based on this,a deterministic latency routing algorithm based on multi-agent deep reinforcement learning(MADRL-based Deterministic Latency Routing Algorithm,MADRA)is proposed,which allows each satellite to make adaptive routing decisions and achieve optimal routing strategies.Finally,a LEO satellite constellation network simulation test platform is built,and the performance of the MADRA algorithm is analyzed through simulations.To evaluate the application performance of the MADRA algorithm in low Earth orbit satellite constellation networks,this thesis uses an Iridium design target simulation scenario and compares the end-to-end delay,throughput,packet loss rate,and delay jitter network performance indicators with the shortest path first algorithm,explicit load balancing algorithm,and traffic classification-based routing algorithm.Simulation results demonstrate that,in terms of convergence performance,the MADRA algorithm enables agents to achieve effective convergence in the satellite network environment,and as the training rounds increase,the algorithm gradually learns more optimal strategies.Regarding network performance indicators,as data transmission rates increase,the MADRA algorithm outperforms the three comparison algorithms while meeting the Qo S requirements of various services.Under high network load conditions,compared to the traffic classification-based routing algorithm,the end-to-end delay of the MADRA algorithm is reduced by 11%,delay jitter is decreased by 17%,packet loss rate is lowered by 14%,and throughput is increased by 5%.Simulation results indicate that the MADRA algorithm can effectively optimize the routing performance of low Earth orbit satellite constellation networks.