Research On Adaptive Scheme Of LEO Multi-beam Satellite Based On Artificial Intelligence

Due to the rapid development of satellite systems,low earth orbit(LEO)satellite systems can improve the service coverage of overall communication and increase the capacity of communication system,which have attracted extensive attention of scholars.In the LEO Satellite Internet scenario,the traditional single-beam satellite can not meet the broadband access needs of a large number of users,while the multi-beam satellite can provide higher system throughput and improve the coverage of the satellite.Therefore,multi-beam satellite has become an ideal way to realize LEO broadband satellite,and low earth orbit multi-beam satellite systems have become an attractive research area in recent years.In order to improve the throughput and transmission performance of LEO multi-beam satellite communication system,scholars have proposed an adaptive coding and modulation(ACM)technology to ensure the link transmission quality by adjusting the modulation mode and coding rate of wireless link transmission.However,effective channel state information conforming to the current channel quality is the basis of using adaptive coding and modulation technology.From the proposal of least square(LS)channel estimation method to the optimization and improvement of minimum mean square error(MMSE)channel estimation method,how to improve the estimation accuracy of channel state has become one of the hot area of the research.For LEO multi-beam satellite communication systems,the characteristics of wide coverage make the environmental conditions of the receiving terminal different.In addition,different channel types will change the mapping criteria of modulation coding strategy(MCS)of adaptive coding modulation.The wrong mapping criterion of modulation coding strategy will lead to a waste of resources and seriously affect the throughput performance of the system.According to the differences of channel scenarios corresponding to different receiving terminal environments,how to quickly identify and obtain the current channel types from the receiving terminal to select the most appropriate modulation coding strategy mapping criteria,so as to optimize the existing adaptive coding and modulation schemes has certain research significance.In addition,the existing ACM systems have the problems that frequent feedback leads to high resource overhead,channel transmission delay leads to outdated and inaccurate channel state information,which still need to be solved.With the development of artificial intelligence technology,machine learning,reinforcement learning and other technologies provide a new idea for the optimization of traditional communication systems.However,at present,the artificial intelligence algorithms for improving the accuracy of channel state information are generally aimed at determining the complex parameters in the communication scene,so as to estimate and predict the channel state;For the artificial intelligence algorithm combined with modulation and coding strategy mapping criteria,the workload of establishing database for different satellites and climates is huge and difficult.For LEO multi-beam satellite system,more attention can be paid to the characteristics of channel and the statistical information of model to optimize the communication system.In this thesis,the adaptive scheme of LEO multi-beam satellite based on artificial intelligence is studied by means of deep learning.In this thesis,first,the LEO satellite channel model and traditional ACM system are introduced in detail.By analyzing the channel characteristics and combining the key parameter information of satellite channel characteristics with deep neural network,a channel state estimator based on deep learning is proposed.In order to quickly identify and obtain the current channel types from the receiving terminal,this thesis also proposed a channel state classifier based on bidirectional long and shortterm memory network.Finally,combined with the above two neural networks,this thesis proposed a semi-static adaptive transmission scheme,which is verified by experiments in the actual satellite communication scene.The main contributions and innovations of this thesis are as follows:1.In order to obtain channel state information with higher estimation accuracy,a channel state estimator based on deep learning is proposed in this paper.The key parameter information in the satellite channel model can be obtained through the trained deep neural network.As an auxiliary priori statistical information,it is applied to the minimum mean square error channel estimation algorithm to optimize the estimation results,so as to obtain the channel state information of LEO multi-beam satellite system with higher estimation accuracy.2.The semi-static adaptive transmission scheme is proposed to optimize the LEO multi-beam satellite communication ACM system.A channel state classifier based on bidirectional long short-term memory(BILSTM)network is proposed,which is used to quickly identify different communication scenarios of mobile terminals in order to select the most appropriate modulation and coding strategy mapping criteria.Furthermore,we optimized the online LEO satellite ACM communication system with two offline DL network and improved the transmission performance.3.In this thesis,we carried out the trial experiment for LEO multibeam satellite system by the real satellite to prove the effectiveness of the proposed scheme.This paper compared the performance of different schemes and the proposed architecture of deep learning combined with traditional communication system has better performance.