Research On Multicomponent Signal Separation Based On Synchrosqueezing Transform

In the radar electronic reconnaissance,due to the complexity of intercepted signal's pulse density,environment and the waveform is changeable,lead to the probability of time-frequency overlapping increase greatly,the traditional time-frequency analysis(TFA)method has been unable to effectively separate the components of multicomponent signal.In order to improve algorithm performance and extract more effective feature information,the paper mainly focus on the improvement of temporal and frequency resolution,the instantaneous frequency(IF)estimation and separation of multicomponent non-stationary signal based on short-time Fourier transform-based synchrosqueezing transform(SST).Firstly,we introduce the principle of SST,in view of traditional time-frequency distribution(TFD)can't achieve good results among time-frequency concentration,suppression of the cross-term interference and reversibility.Then the STFT-based SST with a time-varying parameter is proposed to enhance the time-frequency concentration and resolution.And the relevant theoretical formula is derived.Given the unknown prior knowledge of the signal,a local rényi entropy algorithm is also present to estimate time-varying parameters.The simulation results show that the proposed method matches the theoretical analysis,and has a certain reliability.Secondly,on account of the limitations of the commonly used IF estimation method for the multicomponent signal.A new IF estimation method is proposed based on Viterbi algorithm and FSST,which combines the high concentration of FSST and the optimal route searching of Viterbi algorithm.According to the different types of signals,the corresponding algorithm flow chart is given.Especially for the signal of there are intersection points of the IF,it can be effectively dealt with by classification.Lots of simulation and contrast experiments support the superiority of the proposed algorithm.Finally,for the preset of the traditional Fourier transform's basis function,the lack of theoretical basis and guarantee that the convergence properties of EMD,the paper brings in the adaptive harmonic model analysis method.Because of the imperfect of separation condition,we present an updation by using the linear frequency modulation to approximate a non-stationary signal during any local time,which make more accurate component separation for multicomponent signals.At the same time,a detailed theoretical derivation analysis is provided.Due to the unknown prior knowledge of the signal,a localized algorithm based on separation condition is presented to estimate the time-varying parameter adaptively and automatically.The final simulation results on synthetic and the bat echolocation signal demonstrate the effectiveness and robustness of the proposed method.