Research On Specific Radar Emitter Recongnition Algorithm

Radar emitter individual recognition is developed to adapt to the new radar technology and the increasingly complex electromagnetic environment.It is the key development direction in the field of electronic countermeasures.Therefore,it is of great significance for radar individual recognition.At present,the method of radar individual recognition is more complex,and the recognition rate is low under the condition of low signal-to-noise ratio(SNR)and small sample,so the research on the algorithm which can effectively identify the emitter under the condition of low SNR has great practical value.In this paper,five signal sources and two kinds of modulation signals(LFM,BPSK)are studied.By taking the signal source as the research object,the "fingerprint" information of the signal generating device is studied as the reference method of the radar transmitter.Based on this,the wavelet method,the variational mode decomposition method and the depth learning method of LSTM and CNN are proposed,the results show that it has a good recognition effect in low SNR.The main research work is as follows:Firstly,the signal unintentional modulation is modeled.This paper introduces two kinds of unintentional modulation modeling of LFM signal and BPSK signal.Then,three radar radiation sources with phase noise are simulated.The difference between different radar transmitters is well reflected by spectrum and power spectrum.Secondly,aiming at the problem of radar individual recognition,a method based on wavelet method and variational mode decomposition is proposed.The energy feature of signal is extracted by wavelet method as unintentional modulation feature,and is recognized by classifier.For the method of variational mode decomposition,firstly,the parameters are optimized by leapfrog algorithm,and then the second features are obtained by singular value decomposition of the eigenmatrix composed of the decomposed modal components.In this paper,PSO?PF-SVM is selected as the classifier,and particle filter algorithm(PF)is optimized by PSO to realize support vector machine(SVM)classification optimization,the optimized classifier has better classification effect and good generalization ability.Two methods are used to realize the purpose of using few samples to identify radar individuals,and the performance is good.The simulation results show that when the signal-to-noise ratio is 16 d B,the recognition rate of each group of five radar signal sources is 100%;when the training sample is 5,the recognition rate is 100%,and the stability is high.Finally,in order to study more intelligent methods,the long-term memory network(LSTM)and convolutional neural network(CNN)models are studied.The two models are trained and recognized by envelope front feature and phase feature respectively.In CNN network,the classifier still uses PSO?PF-SVM,and the recognition effect is good.For rising edge features,the recognition rate of CNN model proposed in this paper is better than that of LSTM model at low SNR.With the improvement of SNR,the recognition rate of LSTM is higher.At 15 d B,the recognition rate based on LSTM model reaches 100%.For phase features,the recognition effect of LSTM model and CNN model is very good.Finally,the advantages of each method under different conditions are illustrated by comparing them.