Research On Spectrum Occupancy State Fitting And Prediction For Low Earth Orbit Satellites

With the development trend of the integrated information network,the low-orbit satellite network has gradually become a research hotspot.In the last 20 years,the proportion of spacecraft launched annually in Low Earth Orbit(LEO)has increased from 57% to 97%,and the average number of satellites deployed annually has increased by two orders of magnitude.The frequency data of low-orbit satellites has attracted much attention.These data are very important for the study of low-orbit satellite communication and navigation technology,satellite orbit dynamics and so on.In addition,it is also of great significance for the design and optimization of satellite communication systems and radio spectrum management.However,due to the difficulty in obtaining the measured data of low-orbit satellites,it is necessary to build a simulation system to simulate and generate satellite spectrum data,so as to provide an evaluation and verification environment for spectrum occupancy fitting and prediction.Therefore,this paper focuses on the following three aspects of research: 1)For satellite spectrum data,explore a method based on low-orbit satellite spectrum occupancy state fitting;2)Explore the prediction method of satellite spectrum occupation time series in dynamic scenarios.3)Conduct research on the design of the simulation system for generating spectrum data of low-orbit satellites,and use the simulation system to simulate and generate satellite received data to provide data support for the analysis of low-orbit satellite spectrum characteristics and spectrum situation prediction;At present,with the rapid development of artificial intelligence technology,its application in various fields is becoming more and more mature.Compared with traditional methods,artificial intelligence technology has the advantage of realizing rapid deployment based on a unified platform.Therefore,this paper studies the fitting method of low-orbit satellite spectrum occupation state,the prediction method of satellite spectrum occupation time series and the simulation system of low-orbit satellite spectrum data generation respectively.The main research work and innovations are as follows:Firstly,to solve the problem of low accuracy of satellite spectral occupation time series fitting by traditional methods,the spectral occupation time series fitting is designed based on Poisson distribution and kernel density estimation respectively.Specifically,the Poisson distribution method was used to fit the time series of satellite spectrum occupation,the feature modeling of LEO satellite signals was carried out,the spectrum occupation situation information was extracted,the kernel density estimation method in machine learning was used to build a unified model of heterogeneous user spectrum occupation situation,and the probability density function distribution model of this set of data was determined.Finally,the correctness of kernel density estimation modeling method is verified by quantization and comparison of KL divergence error.The experimental results show that the proposed kernel density estimation method has better fitting performance than the non-parametric estimation method.The KL divergence value of kernel density estimation is only 0.1099.Compared with the KL divergence value of non-parametric estimation,the KL divergence value of exponential distribution is 0.4286,and the KL divergence value of GP distribution is0.7184.It is concluded that the fitting method based on KDE has better fitting performance than other distributions,and the designed method improves the fitting accuracy.Secondly,to solve the problem of large error in the prediction of satellite spectrum occupation time series in dynamic scenarios,this paper proposes a method of satellite spectrum occupation time series prediction based on backpropagation neural network.Constructing a backpropagation neural network model to mine the historical spectrum occupancy characteristics of LEO satellites can not only obtain the future spectrum occupancy state,but also predict the future spectrum occupancy degree.The experimental evaluation results show that the proposed model has obvious advantages in forecasting accuracy compared with the traditional long short-term memory neural networks.The standard error(RMSE)of the backpropagation network model is 0.1728,while the standard error of the short-duration memory neural network is 2.2081.BP neural network has lower loss than LSTM model,and the proposed BP neural network model can obtain higher prediction accuracy and lower RMSE.Finally,in order to solve the problem of the lack of spectrum sensing data of low-orbit satellites,this paper studies the design of the simulation system of low-orbit satellite spectrum data generation to generate the spectrum occupancy data of low-orbit satellites.Specifically,this paper first conducts constellation modeling based on Starlink constellation;Secondly,considering the factors of channel influence,rain failure,antenna pattern and service arrival,the link information between the satellite and the user is calculated to simulate the satellite simulation received signal.Finally,a graphical user interface is used to visualize the received signals,which provides data support for the analysis of the spectral characteristics and the prediction of the spectral situation of the low-orbit satellite.In this paper,the data accuracy is verified in three communication scenarios:single user-single star,single user-multi-star and multi-user-single star.The verification results show that the signal in time domain and frequency domain is consistent with the theoretical values.