Abstract:

S transform (ST), which is developed from the short time Fourier transform (STFT), overcomes the disadvantages of STFT, i.e., the window width is fixed and it cannot show the information at high frequency and low frequency simultaneously. However, in the strong impulsive based α stable distribution noise environment, the ST algorithm usually experiences severe degradation or even failure. Based on the generalized Cauchy distribution (GCD), a class of loss function, which can be applied effectively to the impulsive noise condition, is established and a novel ST called the robust ST (RST) is proposed under the framework of the maximum likelihood estimation (MLE) theory. Afterwards, the robustness of the influence function (IF) is analyzed. The proposed RST can be obtained by the iteration in timefrequency domain whose initial value comes from the conventional ST. Consequently, it reserves the flexibility in window width selection of the ST. Simulation results show that, in dealing with the chirp signal in impulse noise environments, the RST can perform better in impulse noise compared with the conventional nonlinear filter based methods such as the Myriad, and the Meridian filters. Moreover, the proposed algorithm shows good robustness and excellent linear frequency modulation (LFM) signal parameter estimation performance especially in the strong impulsive based α stable distribution noise environment.