Non-stationary Signal Analysis Based On Synchrosqueezing S-transform

Radar emitter identification is the key research issue in the field of modern electronic warfare.As a powerful tool for non-stationary signals processing,the time-frequency analysis can get the signal's time-varying energy and frequency information simultaneously,it has been widely used in radar emitter signal analysis.In this thesis,we aimed at non-stationary signal analysis,combined with synchro-squeezing transform and Viterbi algorithm,the high-performance time-frequency analysis method is studied which is based on the S-transform.A time-frequency analysis method with high concentration and reversibility is proposed as well as a high-precision method for instantaneous frequency estimation.On this basis,the time-frequency feature extraction of radar emitter signal is preliminary studied combined with the measured radar data.The main innovations of this thesis include: 1.Aiming at the problem of phase loss in standard synchro-squeezing S transform,a modified S transform is proposed and its synchro-squeezing transform form is derived.The algorithm solves the problem of phase information loss after synchro-squeezing while retaining the linear characteristics and multi-resolution of the original S transform by adding a phase factor.An algorithm for fast inverse transformation is proposed.Compared with the standard algorithm,the computational complexity is greatly reduced.The synchronous inverse compression can be realized quickly in the engineering practice.The simulation experiments show that it has higher signal reconstruction accuracy within the error tolerance range,and has a certain engineering application value in signal separation,time-frequency filtering and signal synthesis.2.For the problem that the S-transform has low accuracy of instantaneous frequency estimation under complex noise,the reason of high-frequency distortion is analyzed and an improved Viterbi algorithm is proposed.The algorithm improves the suppression performance of high frequency noise by adjusting the original Viterbi algorithm penalty factor.Simulation experiments show that the improved algorithm exhibits better accuracy and robustness under mixed noise interference for many types of FM signals.3.Based on high-performance time-frequency analysis,a time-frequency feature extraction method for radar emitter signals is proposed.The method draws on the theory of compressed sensing and uses high-concentration and high-precision time-frequency analysis methods to extract signal energy and instantaneous frequency information to form sparse time-frequency features.This feature contains most of the information of the signal,and has a clearer physical meaning while reducing the feature dimension,and has a wider application.The characteristic performance was verified by the actual data of civil flight meteorological radar and laboratory radar,which showed a good classification and recognition rate.