Research On HARQ Of B5G Satellite-Terrestial Link Based On Reinforcement Learning

The integration of B5G network and satellite communication network can meet the ubiquitous business needs of users.It is an important direction for the development of mobile communication network in the future.Satellite-terrestial communication system has the characteristics of large propagation delay and high Doppler frequency shift.Therefore,the application of HARQ mechanism in satellite-terrestial fusion network is facing new challenges.HARQ needs multiple retransmissions to ensure the reliable delivery of data packets.How to reduce the impact of retransmission on satellite-terrestial communication system,reduce transmission delay and increase system throughput has become an urgent problem to be solved.Therefore,combined with reinforcement learning,this paper studies the HARQ strategy in the scenario of B5G satellite-terrestial fusion scenario.Firstly,in order to solve the problem of excessive total packet transmission delay caused by the long propagation distance of satellite-terrestial channel,a fast retransmission HARQ strategy based on reinforcement learning is proposed.The feedback process of some NACK signals is omitted through the continuous transmission of the first transmission,so as to reduce the packet reception delay.Because the selection of continuous transmission threshold is closely related to CSI,this paper utilizes reinforcement learning to obtain the mapping relationship between transmission threshold and CSI,so as to improve the accuracy of transmission threshold selection.The experimental results show that the average transmission delay of HARQ packet can be reduced by about 30%compared with the traditional feedback strategy.Secondly,in the satellite communication scenario,the traditional HARQ feedback granularity is single and can not adapt to the change of channel state.Therefore,an adaptive feedback HARQ strategy based on reinforcement learning is proposed.This paper adopts reinforcement learning method,selects the best feedback mode under the current channel state in multi-level block group feedback and equally divided block group feedback,and adaptively adjusts the feedback granularity.The simulation results show that the average throughput of the proposed adaptive feedback strategy is improved by about 25%compared with the traditional feedback strategy.