Investigation On Near-field Source Localization And The Corresponding Applications

Source localization is a key technology in array signal processing,which is widely applied in radar,geologic prospecting,sonar,electronic surveillance,medical electronics,and other fields.Source localization can be classified into far-field source localization and near-field source localization according to the distance between the sources and the array.The wavefronts of incoming signals are different in these two situations,thus leading to different signal models.In the far-field source localization,the wavefront of the incoming signal is assumed to be a plane wave when it impinges the array,and each source is parameterized by only the Direction-Of-Arrival(DOA).But in the near-field,the signal wavefront is spherical,and both the DOAs and ranges are required to localize near-field sources.This dissertation concentrates on the computational complexity or the estimation accuracy of the near-field source localization,and also studies the application on the Ground Penetrating Radar(GPR).The main contributions of this dissertation are summarized as follows:1)Based on the existing high-order modified 2D MUltiple SIgnal Classification(MUSIC),this dissertation has developed three new methods,either to reduce the computational complexity or to improve the estimation accuracy.a)For the existing high-order modified 2D MUSIC methods,two Hermitian matrices are constructed,and two eigenvalue decomposition(EVD)are used to estimate parameters.This dissertation proves that the modified 2D MUSIC can be carried out with a non-Hermitian matrix and proposes the Low-Complexity MUSIC(LCM).The DOA estimation can be achieved with the eigenvectors associated with the zero eigenvalues,and to further improve the efficiency,the remained eigenvectors associated with the non-zero eigenvalues are orthogonalized to estimate the range.Only1 matrix and 1 EVD are needed,and the results of simulation show that the proposed method maintains the same high accuracy with other high-order MUSIC,but with a lower computational complexity and higher processing speed.b)The second proposal is an improvement to LCM.A propagator-based method is proposed based on the fact that the high-order cumulant is free from the Gaussian noise.The relationship between the constructed cumulant matrix and the two different steering matrices is studied,and two propagators can be built with the columns or rows of the cumulant matrix.The columns and rows can be used to build two propagators,which can be directly used to construct two subspaces orthogonal with the two steering matrices.Then the DOA and range estimation can be achieved respectively.Based on LCM,this propagator-based method needs to construct only one cumulant matrix,but without the application of EVD,reducing the computational complexity of LCM.c)The apertures of the existing high-order MUSIC methods are limited by the array.In this dissertation,an Aperture-Expanded MUSIC(AEM)is proposed to enlarge the array aperture,which is a very important factor for the resolution and accuracy.By making full use of the high degrees of freedom of high-order cumulant,the aperture for range estimation is expanded significantly without enlarging the distance between two adjacent array elements or increasing the number of the elements.Compared with other high-order MUSIC,the achieved aperture expansion allows to improve the accuracy for the range estimation.The simulation results demonstrate its effectiveness.2)Inspiring by the principle of the modified 2D MUSIC,a method for near-field source localization is proposed combining the clustering technology and Compressive Sensing(CS)theory.But different from the modified 2D MUSIC,which requires several 1D spectrum search,the proposed method only needs to build two 1D overcomplete dictionaries and reconstruct two signals.By making full use of the reconstructed signal,a paring method is introduced to decide the correct combination of the two parameters,leading to the final estimation of DOAs and ranges.3)Some decorrelation algorithms are necessary to deal with the coherence among the echoes before traditional methods can be applied to estimate the time-delay.We propose in this dissertation to apply CS to directly carry out the Time-Delay Estimation(TDE).Besides,the signal enhancement,which was originally used in the communication system,is also introduced to ensure that the TDE can still be accurate enough even when the SNR is low.