Research On LEO Satellites Adaptive Coding And Modulation Technology Based On Machine Learning

One of the visions of the 6th generation of mobile communications(6G)is to converge terrestrial communications and satellite communications to build an integrated air-space-ground network to achieve true global full-coverage communications.Compared with high orbit satellites,low orbit satellites(LEO)have higher energy efficiency and are more suitable for 6G communications.This thesis is dedicated to effectively increase the communication capacity of LEO to cope with the increasing business needs under the premise of limited on-board resources.Aiming at the problems of complex LEO channel environment,long transmission distance,and limited transmission resources,adaptive modulation and coding(AMC)technology can be adopted to improve system performance under the premise of satisfying communication quality.By analyzing the characteristics of the LEO communication scene,the application of AMC in the LEO communication system is studied in depth.The research contents of this thesis are as follows:1.The LEO satellite communication system and channel state information(CSI)prediction technology are studied.The long delay of satellite link leads to the outdated CSI received by the transmitter,and the nonlinear distortion of the signal will be introduced when the high power amplifier(HPA)works near the saturation point.To solve the above problems,a CSI prediction algorithm based on machine learning(ML)for LEO satellites is proposed.The algorithm determines the order of Autoregressive Integrated Moving Average(ARIMA)through the nonlinear residuals of the time series of predicted SNR obtained by the Long Short Term Memory network(LSTM).The ARIMA can effectively predict the SNR at the current moment.The LSTM can rely on a shorter time series to make up for the data in the nonlinear distortion.Simulation results show that compared with ARIMA model,the prediction accuracy of the proposed algorithm for CSI is improved by 50%,and the accuracy of CSI is effectively improved.2.On the basis of obtaining real-time and accurate CSI at the transmitter,the LEO adaptive system is studied.The variability of the satellite channel leads to the failure of the mapping table between SNR and modulation and coding scheme(MCS),and the model-based Dyna-q algorithm has a slow convergence rate when the channel state changes.To solve the above problems,a LEO adaptive strategy based on reinforcement learning(RL)is proposed.Based on the model-based Dyna-q architecture,a priority function of state-action pairs is added,namely the priority function of SNR-MCS.This strategy learns virtual experience through the environment model and effectively improves the learning efficiency.The agent focuses on more valuable MCSs,accelerates the convergence of the algorithm,and can quickly capture the changes in the LEO communication environment and correct the errors of the model.The simulation results show that in the LEO communication scenario,compared with the Q-learning and the Dyna-q algorithm,the proposed MCS selection algorithm has faster convergence speed and effectively improves the spectrum efficiency of the system.3.A LEO link level simulation platform based on the 5G protocol is built,and the LEO channel state prediction algorithm based on ML and the LEO adaptive coding modulation strategy based on RL are jointly designed to realize the AMC system for LEO satellite based on machine learning.The simulation results show that the LEO link level simulation platform can support the verification of key technologies,and the performance of the adaptive communication system based on machine learning is significantly improved compared with the existing algorithms.The channel state prediction algorithm proposed in this thesis improves the spectrum efficiency of the LEO adaptive communication system by 50%.The MCS selection algorithm proposed in this thesis not only increases the system capacity,but also doubles the convergence speed.