Research On LPI Radar Signal Modulation Recognition Based On Dictionary Learning

The modulation recognition of low probability of intercept(LPI)radar signals is an important link in electronic reconnaissance and a significant means to obtain non-cooperative radar information.Only when the key information of non-cooperative radar is fully mastered,can the enemy radar be targeted for interference,suppression,deception and even accurate strike,so as to establish an advantage in the war.However,due to the complex waveform design of the LPI radar,the existing reconnaissance technology is difficult to identify the modulation type of the signals,and it is difficult to build a recognition framework with wide applicability.In this paper,aiming at the blind recognition system of radar signals in non-cooperative passive reconnaissance scenarios,we deeply study the LPI radar recognition technologies under the background of Gaussian white noise and strong pulse interference.A novel framework of LPI radar signal scheme is proposed based on dictionary learning(DL)and sparse representation(SR)theory.The main research contents are summarized as follows:Firstly,the fuzzy performance and low interception characteristics of typical LPI signals are discussed.The time-frequency(T-F)analysis method of non-stationary signals is studied,and the T-F graphs of fourteen types of LPI radar signals at low signal to noise ratio(SNR)are given.In the case that the sparseness prior information of radar signal is unknown,the blind compression process based on Cauchy random matrix is studied,which not only guarantees lossless compression,removes information redundancy,but also simplifies signal processing.It has excellent suppression effect on strong pulse interference.Secondly,aiming at the problems of existing recognition methods such as difficulty in effective feature extraction and poor recognition effect under low SNR,a radar single signal recognition method based on blind compressive label consistant K-SVD(BCLC-KSVD)algorithm is proposed.This method starts from SR theory,analyzes the basic principle and method of SR for signal recognition,and proposes a LPI radar recognition framework based on SR.This is a brand new recognition framework,using the dictionary learning method to obtain the best sparse representation results and classification,which does not need to extract any features,but uses the location of the sparse coefficients for signal type discrimination.The effectiveness of the proposed method under low SNR and strong pulse interference is verified by simulation.The robustness and convergence under different simulation conditions are analyzed.Finally,aiming at the difficulty of signal separation in the multi-signal modulation recognition method,this paper proposes an improved radar multi-signal recognition method based on D-S evidence theory and kernel DL algorithm under blind compression(DS-BCK-KSVD).This method can realize multi-signal recognition without separating multiple signals.D-S evidence theory combines the classification judgement based on sparse coefficient aggregation with the classification judgement based on sub-dictionary reconstruction contribution,to get a decision-level fusion recognition method,so as to promote the overall recognition effect.The simulation results show that the fusion of the two discriminant methods can complement the information,and the quality of the fusion can help to correct various misjudgements caused by single factor discrimination,which can improve the decision-making ability.