A Recognition Method for Radar Emitter Signals Based on Deep Belief Network and Ambiguity Function Matrix Singular Value Vectors

摘要

Aiming at the weak representation ability under low signal noise ratio (SNR) in radar emitter signals recognition based on ambiguity function main ridge or contour lines, etc. A recognition method based on Deep Belief Network (DBN) and ambiguity function matrix singular value vectors (SVV) is proposed. First, the appropriate window parameter of one-dimensional median filtering is selected to denoise the time domain signals and ambiguity function matrix singular value vectors datasets are created in different SNR. Then, a deep network based on DBN model is established to complete the training of labeled data. Finally, the network is used to achieve radar emitter signals recognition. The simulated experiments show that the average recognition accuracy rate of six kinds of complex modulated signals, i.e., BPSK, BFSK, FMCW, QPSK, MSEQ, and LFM-BC, by proposed method keeps above 99.17% in fixed SNR environment above -10dB with good generalization ability and strong robustness.