Research On Hybrid Signal Separation Based On Empirical Mode Decomposition And Independent Component Analysis

Rapid development of electronic information technology makes electronic warfare a dominant part of the modern battlefield.The unmanned aerial vehicle cluster operation also plays a very important role in electronic reconnaissance.Because the electromagnetic environment of the battlefield is complicated,there is a great possibility that the intercepted signal contains multiple signals which are overlapping both in time and frequency domain.The effective separation of hybrid radar signals and the judgement of the types and parameters are prerequisites for making strategy correctly.This thesis focuses on the separation of hybrid radar signals based on the basic theory of empirical mode decomposition and independent component analysis.Firstly,the characteristics of common radar signals and hybrid radar signals are studied.Hybrid radar signals refer to the intercepted signals which is overlapping in time or frequency domain.The classical timefrequency analysis method can not deal with the hybrid signals effectively,Therefore,the method of empirical mode decomposition is introduced to analyze the time-frequency distribution of the signal.The signal type studied in this thesis is the linear combination of LFM signal and single frequency signal,and the type of channel noise is Gaussian white noise.Secondly,the basic principles and main problems of EMD algorithm are studied.The algorithm can decompose the signal into several components with physical meaning adaptively,which have better timefrequency aggregation characteristics in time-frequency distribution.In this thesis,the problems of noise sensitivity,mode mixing and false components are analyzed.The source and influence of the main problems are studied,and the signal is processed by noise reduction and mode demixing.In part of the noise reduction,the algorithm of EMD iterative interval thresholding denoising under the wavelet packet frame is proposed to suppress the noise contained in the hybrid signal.In the part of the mode de-mixing,the phase space reconstruction method is used to expand the signal to multidimension,and the mode de-mixing of the component is realized by the Fast ICA method.Finally,based on the time-frequency distribution of hybrid signal,the time-frequency characteristics of different signals are extracted by Hough transform,and the basic theory of processing multidimensional signals by independent component analysis is introduced in the solution of the coefficient matrix.The weight of each component is determined by the correlation between the component and the estimated values.The source signals in the hybrid signal are recovered effectively.Experiments show that when the SNR is higher than 10 d B,the similarity coefficient of the time-frequency characteristic of the separation results is higher than 0.9.In this thesis,the empirical mode decomposition algorithm is introduced to obtain multiple signal component to deal with the insufficient number of observed signals in the signal separation.Furthermore,the thesis combines the basic theory of independent component analysis deeply and constructs a mixed coefficient matrix to evaluate the weight of each component,and the separation of the hybrid radar signals is carried out from four parts,which are signal interception,signal denoising,signal decomposition and signal recovery.The channel hybrid radar signal is effectively separated and reconstructed.