Research On Methods Of Low Probability Intercept Radar Waveform Recognition

Low probability of intercept(LPI)radar waveform recognition is an important branch in the field of electronic reconnaissance,which plays a significant role in non-cooperating radar cognition.To recognize the hostile LPI radar and extract the key information on the battlefield in time helps us to present appropriate countermeasures and decisions,which not only plays a key role in improving the success rate of tracking and attacking specific targets but also indicates important theoretical significance and practical application.This dissertation focuses on multi-signal separation of LPI radar and recognition algorithms after separation.Waveform recognition algorithms for LPI radar waveforms are completed based on its feature and structure.The former algorithm uses feature extraction while the latter utilizes deep learning.Both of them improve the ratio of successful recognition(RSR)of waveforms at low signal-to-noise ratio(SNR)condition.The main works are as follows:Firstly,in the mixed signal separation with one sensor,we propose the discrete prolate spheroidal sequence(DPSS)and Hough transform(HT)joint separation methods.Taking DPSS as a window function of short time Fourier transform(STFT)can reduce noise interference of multiple linear frequency modulation(MLFM)after STFT.Therefore,MLFM can be easily distinguished by HT in time-frequency images,which increases the separation effect for MLFM at low SNR condition.Simulation results show that the algorithem can separate MLFM when SNR<-20 dB.In the mixed signal separation with multiple sensors,we propose the joint separation methods of single source points detection(SSPD)and subspace projection(SP).The method is no longer restricted to signal type.Single source point detection method estimates the mixing matrix of the signal relied on the sparseness of different waveforms in the time-frequency domain.While the subspace projection algorithm recovers the mixing signals on the basis of the estimated mixing matrix.Simulation results show that the algorithm has good performance in mixing matrix estimation and waveform reduction when different waveforms mix together.Secondly,the recognition method based on waveform feature is proposed in the research of frequency modulated waveforms,binary phase shift keying(BPSK)and polyphase codes.The method recognizes LPI radar waveform relied on feature extraction and Elman neural network classifier.Elman neural network is robuster than traditional ones because of its layer feedback.The classifier uses master-slave structure.The complete classifier is composed of two sub-classification networks.Each subclassification network is responsible for classifing different waveforms.The advanced structure can not only reduces the training time and training complexity of each subclassification network but also improves the reliability and anti-interference of the classifier.Meanwhile,new features are employed to opitimize extraction process.Compared with traditional menthods,less prior information is needed while extracted features are robuster.Simulation results show that the method can be used in linear frequency modulation(LFM),BPSK,Costas,Frank code and P1-P4 codes.The total RSR of waveform recognition is over 97% at SNR of 0dB.Finally,the recognition method based on waveform structure is proposed in the research of frequency modulated waveforms,BPSK and polytime codes.A new classifier –CNN-EntropyRBM classifier is presented,which no longer needs artificial extraction features.The process of feature extraction can be finished automatically during the training of classifier.Neuron optimization algorithm based on the information entropy is utilized in the process.The algorithm determines the neurons number in each hidden layer according to the principle of maximum transmission information,which improves the accuracy of the classifier.Simulation results show that the classifier can recognize 7 kinds of waveforms including LFM,BPSK,Costas and T1-T4 codes.The total RSR of waveform recognition is above 96% at SNR of 0dB.To solve the recognition of frequency modulated waveforms,polyphase codes and polytime cods,the joint recognition menthod based on waveform feature and structure is presented.The method combines the advantage of feature recognition and structure recognition,which can recognize 12 kinds of waveforms including LFM?BPSK?Costas?Frank code?P1-P4 codes and T1-T4 codes.Simulation results show that the RSR of joint recognition method for the above 12 waveforms is over 93% at SNR of 0dB.To sum up,this dissertation introduces some new methods for LPI radar waveform recognition,which not only improves the RSR of waveform recognition at low SNR condition,but also increases the number of recognizable LPI radar waveforms.It also has strong practical in application.