Research On Spatial Spectrum Estimation Of Far-field Sources And Its Application

Spatial spectrum estimation has been extensively studied in the field of array signal processing.The main objective of spatial spectrum estimation is to utilize the received signals of the array to estimate the spatial parameter,i.e.,direction of arrival(DOA),and other characteristic parameters,e.g.,Doppler frequency.Electromagnetic waves(or sound waves)travel through space in the form of spherical wave.However,since the wavefront curvature of the source signals is negligible when the distance between the source and the sensor array is significantly larger than the array aperture,the spherical waves can be approximated as plane ones and the source is then referred to as the far-field source.In many applications,e.g.,radar,sonar and mobile communications,the far-field model is frequently satisfied.Hence,spatial spectrum estimation of far-field sources has always been a research hotspot for decades.The far-field source is always characterized into two different types: the point source and the distributed source.This dissertation focus on the spatial spectrum estimation of point sources and distributed sources and the main contributions are listed as follows:(1)The angular parameters estimation of incoherently distributed source is studied and a novel algorithm for nominal DOA and angular spread estimation via the rank reduction criterion is proposed.In the proposed algorithm,with the generalized array manifold(GAM)model,the nominal DOAs are firstly decoupled from the original array manifold.Then the nominal DOAs are obtained when the rank of a trickily formulated matrix drops.Next,the angular spreads are estimated from the central moments of the angular distribution.The proposed algorithm can be applied to arbitrary arrays.When the array configuration satisfies certain conditions,a polynomial rooting based search-free method is also derived for nominal DOA estimation.The proposed algorithm can handle the case when multiple sources have different angular distributions.As compared to the popular ESPRIT-ID algorithm,the proposed algorithm can achieve higher accuracy,can handle more sources,and applies to a more general array structure.(2)The DOA estimation problem of point sources using a massive uniform linear array(ULA)is studied and a low-complexity based on the discrete Fourier transform(DFT)technique is proposed.The conventional direction of arrival(DOA)estimation algorithms,e.g.,MUSIC,root-MUSIC,and ESPRIT,may not be effective when applying to massive antenna array configuration because they not only require a large amount of receive snapshots but also suffer from forbiddingly high computational complexity.In this paper,we propose a low-complex DOA estimation algorithm for massive uniform linear array(ULA).We firstly obtain coarse initial DOA estimates via the fast Fourier transmission(FFT)and then search for the accurate estimates within a very small region.The proposed algorithm needs one snapshot only and could achieve very high accuracy that is close to the Cramér-Rao bound(CRB).Meanwhile,it is hardware friendly and can be easily implemented in practice.(3)The spatial spectrum estimation of non-circular point sources using an arbitrary array is investigated and a non-circular generalized propagator method(NC-GPM)is proposed.The non-circular signals(e.g.,the amplitude modulation(AM)and binary phase shift keying(BPSK)signals),whose elliptic covariance is not equal to zero,have been widely exploited in wireless communication systems.The non-circularity can be utilized to improve the estimation accuracy of DOA and extend the degree of freedom(DOF).The proposed NC-GPM imposes no constraint on the array configuration and can achieve decoupled DOA estimation by using one-dimensional(1-D)spectral search,with no need of estimating the non-circular phases.When the array geometry satisfies certain conditions,a proposed polynomial rooting-based method,which avoids spectral search and reduces the complexity further,can be applied.The proposed NC-GPM fully exploits the non-circularity property so that it outperforms some conventional algorithms in terms of estimation accuracy and the maximum number of detectable sources.(4)The two-dimensional(2-D)DOA estimation using an L-shaped array is discussed and a successive generalized ESPRIT(S-GESPRIT)algorithm is proposed.The L-shaped array,which consists of two line subarrays connected orthogonally at one end of each subarray,has advantages of simple structure and relatively high estimation accuracy.Thus,2D-DOA estimation using L-shaped array has drawn considerable attention.To achieve 2D-DOA estimation,the proposed S-GESPRIT algorithm needs 1-D spectral searches only and is much more computationally efficient than the conventional 2D-GESPRIT algorithm.Meanwhile,it can be applied to non-uniform L-shaped array and exhibits very close 2D-DOA estimation performance and requires no additional pair matching.(5)The joint DOA and Doppler frequency estimation problem using monostatic multiple input multiple output(MIMO)radar is investigated and two novel methods based on the parallel factor analysis(PARAFAC),i.e.,the compressed sensing PARAFAC(CS-PARAFAC)and the reduced-dimension PARAFAC(RD-PARAFAC),are proposed.The RD-PARAFAC firstly utilizes a RD transformation,which can remove the redundant entries of steering matrix.After transformation,by means of linking the transformation signal matrix to trilinear model,the estimation of DOA and Doppler frequency can be achieved by adopting the trilinear alternating least square(TALS)method and the least square principle,respectively.In the proposed CS-PARAFAC algorithm,the joint estimation problem is firstly linked to the compressed sensing trilinear model,then the estimated compressed matrix can be derived through TALS method and the angle and Doppler frequency are jointly estimated with sparsity from the compressed matrices.Compared with the conventional PARAFAC algorithm,both the RD-PARAFAC and CS-PARAFAC algorithms have very close estimation performance on both DOA and Doppler frequency and have much lower computational complexity and smaller memory capacity.Furthermore,both of them can achieve automatically paired DOA and Doppler frequency estimation results.