Research On Interference Recognition Technology In Satellite Communication System

With the development of communication technology,satellite communication system has been widely used in various fields.However,due to the long-term exposure to open environments,satellite communication system will be easily affected by complex electromagnetic environments.As an important part of anti-interference in the system,accurate and efficient recognition of interference types can support for the selection of interference suppression and elimination methods for communication receivers,thus making the anti-interference methods more intelligent and improving the performance of the system.At present,the study of interference recognition technology in satellite communication system only extracts feature by adding interference and Gaussian noise together,and the simulation and validation are only for JNR.However,the superposition of communication signals will reduce the accuracy of interference feature extraction,which results in the blurred meaning of the extracted interference features,and the correct probability of interference recognition will also be affected.The interference recognition algorithms studied in this thesis are based on the superposition of communication signals,interference and Gaussian noise.The simulation analysis is also for JSR.In addition,because of the cumbersome and inefficient methods for extracting interference features,this thesis applies convolutional neural network(CNN)in deep learning to interference recognition to avoid the explicit extraction of interference feature parameters,which can implicitly extract the local characteristics of interference and utilize the correlation characteristics for interference recognition.The communication signals modulated by BPSK are superimposed with seven typical interferences respectively in this thesis.Eight characteristic parameters of the interferences to be identified are extracted and normalized from the time domain,frequency domain and transform domain respectively.Eight types of samples are composed of non-interference and seven interferences to realize the integration of the existence detection and recognition of the interferences.Based on the feature extraction of interference recognition technology,this thesis designs a support vector machine(SVM)based decision tree classifier firstly.Because the decision tree is sensitive to error samples,a BP neural network model is also designed in this thesis.Samples are generated to train and test the two models under the same communication signal and noise background.The simulation results show that when the JSR is 5d B,the correct recognition probability of interference of SVM decision tree classifier and BP neural network model can reach 95% and 99% respectively,but when the JSR is reduced to 0d B,the two results are reduced to 85% and 95% respectively.When all samples were mixed,the number and complexity of samples increased,resulting in a serious decrease in the correct recognition probability of both samples,which decreased to 73% and 77% respectively.Because of the complex feature extraction of interference signal,a CNN model for interference recognition is also designed in this thesis.The time-domain sampling sequence of the signal is normalized directly as the input of this network.The simulation results show that when the JSR is 0d B,the correct recognition probability of the model is 98%.When all samples are mixed,the correct recognition probability is 95%,which is higher than that of the feature extraction method,and by contrast,the convergence speed of this model is the fastest.