Research And Application Of Time-frequency Analysis Of Non-stationary Signals Based On Wavelet Transform

Time-frequency analysis is hot in signal processing.It uses a united function about time and frequency to represent time and frequency domain information at the same time.Time-frequency analysis solves the problem that Fourier transform is a global analysis.Now,it has been widely used in many fields,such as national defense,industry,medicine,and so on.Wavelet transform is a kind of time-scale analysis method of signal.It has "zoom" feature of multi-resolution,which makes it be suitable to analyze time-varying non-stationary signals.This paper makes a deep theoretical research on commonly used methods of time-frequency analysis,and then makes simulations to study the relationship between them and find their advantages and disadvantages,on the basic of studying characteristics and representation methods of non-stationary signals.In order to solving the problem of the common methods,we propose two improved methods of time-frequency analysis and apply them to analyze the actual frequency hopping signal.The main contents of this paper are as follows.Firstly,this paper studies the linear and quadratic time-frequency analysis methods of non-stationary signal.Besides,it makes a theoretical research and contrast between short-time Fourier transform(STFT),Gabor extension,wavelet transform,Wigner-Ville distribution(WVD)and Cohen class distribution.And it also studies their advantages and disadvantages through simulation.Secondly,this paper discusses and improves some problems that exist in commonly used methods of time-frequency analysis.First,it gives specific parameter selection and optimization criteria to solve the problem that Wavelet transform is sensitive to wavelet parameters and has low resolution of frequency in high frequency,then studies its anti-noise performance.Second,because of the fixed window function,the resolution of short-time Fourier transform is single and low.A method that based on the wavelet transform and short-time Fourier transform which introduces the theory of combining window and the idea of wavelet's multi-scale analysis is proposed to solve the problem.Third,Wigner-Ville distribution suffers from severe interference terms for multi-component signal's analysis.A method of time-frequency analysis that combines the idea of wavelet filtering with Wigner-Ville distribution is proposed to solve the problem.The simulation shows that these two improved methods can effectively solve the problems that exist in the original single algorithm.They eliminate interference terms of time-frequency feature representation and keep good time-frequency aggregation at the same time.Finally,this paper applies the improved time-frequency analysis methods to the blind parameter estimation of hopping signal,and makes performance evaluation.The simulation results show that the improved time-frequency analysis methods can better describe the local characteristics of the signal.It has high potential value for practical application.