Parameter Estimation Of Backscattered Echoes From Underwater Objects Using Sparse Representation

The detection and identification of underwater proud or buried objects is one of the most significant research areas with wide application in underwater acoustic field.Due to the time-varying and space-varying signatures of the backscattering echoes,and the interference from the environment,the problems such as signal detection,parameter estimation and feature extraction are most important that must be solved.In this dissertation,based on the properties of the backscattering echoes,the sparse signal representation is introduced into the parameter estimation of the backscattering echoes.Moreover,the methods can be extended into the estimation problem in the interference of reverberation.The algorithms proposed in this dissertation may provide new approached tin underwater objects detection and identification.The dissertation begins with the acoustic scattering theory,which may provide theoretical basis to the following research.The backscattering properties of objects with simple shapes,such as spherical shell and finite cylindrical shell,are introduced.The main properties used here are the time structure and spectral characteristics.Then,experiments carried on both tank and lake are introduced.These contents may provide theoretical basis and data support of the following proposed signal processing algorithms.The dissertation considers the parameter estimation of the backscattering echoes into two aspects,i.e.,the frequency estimation and the time-delay estimation.Firstly,a model with weighted combination of multiple linear frequency modulated signals is used as the approximation of the backscattering echoes,under the assumption that the incident pulse is linearly frequency modulated.Moreover,the parameters are sparse considering the whole parameter space.Thus the sparse signal representation theory can be extended with the features of the backscattering echoes to estimate the parameters.Considering the frequency estimation problem,a method that can estimate the frequency of multiple linear frequency modulated signals with constant amplitudes is proposed,which is based on the sparse signal representation.The method model the signal as sparse combination of the dictionary,and the estimator operates on the penalties of the weight.On the other hand,the backscattering echoes has time-varying properties on power.In order to location of the energy of the components,a parametric model of the time-varying amplitude is exploited.And the estimation of multiple linear frequency modulated signals with time-varying amplitudes is proposed.This method can estimate the amplitude and frequency simultaneously,which can be used to analyze the time-frequency properties of backscattering echoes.This algorithm is a convex problem,which can be solved with iterative equations with low computational complexity.The research is proceeded with the application of sparse signal representation into the time-delay estimation of backscattering echoes.In order to solve the estimation of the weak elastic scattering waves,based on the time structure of the echoes,a block signal model is exploited to describe it and a estimation method based on block sparsity is proposed.The method can treat the periodic elastic waves as a pattern,and penalty the sparsity of the patterns,which can be helpful to find the weak components.Considering the diffuse of the time-frequency properties of the backscattering echoes,and in order to reduce the computational complexity of the estimation,an integrated dictionary method is propose,which can cover the continuous parameter space.Moreover,a signal reconstruction method based on dictionary learning is proposed to solve the different signatures between the scattering waves and the template in the dictionary.This method can recover the backscattering components.Under the practical background,in order to avoid the selection problem of the tuning parameters,sparse iterative covariance-based estimators are used to estimate the parameters of the backscattering echoes.This method is extended from the covariance fitting criterion in array signal processing,but with only one snapshot.The match between the signal and dictionary is built in statistical fashion,and signal part and interference part are divided,which can be helpful to reduce the interact influence of the interference on the signal,and can avoid the difficult selection of the tuning parameters.The validity of the proposed algorithms are illustrated by simulation and experimental signal processing,which can act as new approaches to the parameter estimation of backscattering echoes and detection in practical environment.