Non-Stationary Signal Analysis Based On HHT And Machine Learning

With the development of science and technology,non-stationary signal analysis has become an important part of many engineering fields.The time-frequency analysis method is a common method for non-stationary signal analysis.The Hilbert-Huang Transform(HHT)is a new adaptive non-stationary and nonlinear signal analysis method with sharper time-frequency resolution than traditional time-frequency analysis methods.The HHT consists of two key parts: empirical mode decomposition(EMD)and Hilbert spectral analysis(HSA).The EMD can adaptively decompose the non-stationary signal into a finite number of intrinsic mode functions(IMFs),and the Hilbert transform(HT)of the IMFs can obtain the time-frequency distribution of the non-stationary signal.For lack of a firm mathematical foundation,the implementation of EMD is still empirical,which leads to some problems in the HHT method,and seriously affects its analysis results.In this paper,the end effects and mode mixing problems in HHT are studied.A class of non-stationary communication signals is analyzed using HHT and its improved method.The following points are the main research contents of this paper:Firstly,We use HHT and several traditional time-frequency analysis methods to perform time-frequency analysis on common non-stationary signals.The results show that the HHT method has better time-frequency analysis performance for non-stationary signal analysisSecondly,aiming at the end effects existing in the HHT.We clarify the reasons for the end effect and summarize the existing solutions.We mainly introduce the extreme value mirror extension method,the waveform extension method based on support vector regression,and the waveform extension method based on extreme learning machine.In addition,we propose an extension method based on the combination of extreme learning machine and extrema mirror.According to the evaluation indexes such as orthogonality coefficient,similarity coefficient and model training time,this method can effectively suppress the end effects.Thirdly,we study the mode mixing problem in the HHT.The causes of mode mixing and the current solutions are introduced.We highlight three methods for solving mode mixing,including complete ensemble empirical mode decomposition with adaptive noise,frequency shifting method,and mask signal method.In addition,we propose an implementation principle of the mask signal method,which can guide researchers to make better use of the mask signal method.Finally,we use the HHT and its improved method to analyze a class of nonstationary communication signals.The results show that under the condition of no Gaussian white noise,the HHT method can accurately analyze the time-frequency characteristics of the signal.Under the condition of Gaussian white noise,the analysis performance of the HHT method is weakened,and using CEEMDAN and HSA as an improved method of the HHT can effectively extract the frequency information of nonstationary communication signals.