Research On Satellite Adaptive Transmission Technology Based On Channel Prediction And Reinforcement Learning

With the continuous development of satellite communication technology,the requirements for transmission rate and spectrum utilization have become higher and higher,but the characteristics of satellite channels such as complex channel environment,rapid changes,and weather conditions have greatly limited satellite communication Transmission performance of the system.In this case,an adaptive transmission technology capable of improving system throughput and ensuring communication quality has emerged.This technology can dynamically change transmission parameters of the transmitting end through real-time feedback of Channel State Information(CSI).Find the right balance between throughput rate and transmission quality to improve spectrum utilization.Adaptive Modulation and Coding(AMC)is one of many adaptive transmission technologies.It mainly adapts to the channel environment by changing the modulation method and coding rate.The traditional application of AMC technology mainly selects the corresponding modulation and coding combination by uniformly determining a good signal-to-noise ratio switching threshold,which conforms to a fixed mathematical model.However,due to the long feedback delay under the satellite channel,and the estimation of the signal-to-noise ratio is not ideal,there is an estimation error.Therefore,the traditional AMC method often fails to exert the performance of adaptive transmission,and even occasionally causes communication interruption.It is difficult to guarantee the quality of satellite communication services.In view of the above problems,this thesis builds a dynamic communication link simulation platform based on the DVB-S2 protocol,applies channel prediction technology,and proposes an Online-LSTM algorithm based on online iterative model training to predict the signal-to-noise ratio time series value.At the same time,it is proposed to apply the Q-learning and Dyna-Q algorithms in reinforcement learning to adaptive coding and modulation.It doesn't need rely on a fixed mathematical model and perform autonomous learning in the process of continuous interaction with the channel environment.According to the actual performance of the system determine the selection relationship between the signal-to-noise ratio information and the combination of modulation and coding.Finally,based on the spectral efficiency as the performance evaluation index,the transmission performance of several algorithms is compared through simulation experiments in the presence of channel information feedback delays and SNR estimation errors.It was found that the AMC algorithm based on Online-LSTM channel prediction and reinforcement learning has higher spectral efficiency than the traditional AMC algorithm under a fixed threshold,and the AMC algorithm based on reinforcement learning has better performance,is more robust and more general abilities.