High-resolution Parameters Estimation Algorithms For Near-field Complex Sources

Traditional spatial spectrum estimation algorithms generally assume that all sources are in the far-field(FF) of the array. However, in many practical applications, the radiating sources may lie in the near-field(NF) region of the array. And FF direction-of-arrival(DOA) estimation algorithms may suffer from performance deterioration or even become invalid in the NF scenario. Therefore, NF source parameters estimation has received considerable attention in the area of array signal processing during the past decade. Even though various techniques have been developed for dealing with this issue, there remain some essential problems to be addressed, which include: i) fully utilization of spatial, temporal, frequency and polarimetric information of the sources; ii) improving the estimation accuracy of the source parameters; iii) alleviating computational complexity; iv) robustness against steering vector errors; v) avoiding the parameter match problem and so on. To solve these problems, this dissertation studies on the high-resolution parameters estimation algorithms for NF complex sources. The main works can be summarized as the following three parts.The first part studies the NF source localization algorithms for scalar sensor array. Firstly, the NF array signal model and two classic NF parameters estimation algorithms are introduced. According to our analysis, these methods suffer from array aperture loss. Then, two algorithms for the parameter estimation of NF non-circular sources have been proposed, which are called non-circular generalized ESPRIT(NCGESPRIT) and real-valued non-circular rank reduction(RVNCRARE). NCGESPRIT extends the generalized ESPRIT algorithm for DOA estimation and presents a novel spectral function for range estimation. While NCGESPRIT requires complex-valued computations, RVNCRARE is implemented in the real-valued domain, which is computationally more efficient. After decoupling the extended NF steering vector, RVNCRARE obtains the DOA and range estimates by two real-valued rank reduction(RARE) estimators. Both NCGESPRIT and RVNCRARE can improve the estimation accuracy and resolve more sources than the conventional second-order-statistics-based methods. Finally, a fast 4-dimensional parameters estimation algorithm is presented for a single NF source with uniform circular array. Two correlation functions are devised to obtain the azimuth and elevation angles, range and frequency parameters. The proposed algorithm is computationally efficient since it avoids spectral search and eigenvalue decomposition. Moreover, it has a flexible configuration and the array aperture can be extended, which is preferable in practical applications.The second part studies the NF source localization algorithms for vector sensor array. Firstly, an ESPRIT-like NF parameters estimation algorithm is introduced using an array of two orthogonally polarized sensors. Since this algorithm suffers from array aperture loss and parameters matching problems, an improved method is proposed using nonsymmetrical array configuration. The proposed method can fully utilize both spatial and polarimetric degree-of-freedom. Finally, a closed-form 6-dimentional NF parameters estimation algorithm is presented using spatially separated electromagnetic vector sensor(EVS). This algorithm only requires a six-component EVS, which mitigate the mutual coupling effect between dipoles and loops. In addition, it is computationally efficient, since it circumvents high-order cumulants and spectral search.The third part proposes four algorithms for the parameters estimation of mixed FF and NF sources. Firstly, based on the distributed coherent array, a cumulant domain dual-size shift invariance(CDSSI) algorithm is presented, which is composed of two stages. In the first stage, the DOA estimation is decoupled from the range estimation by constructing a specific fourth-order cumulant matrix, where unambiguous coarse DOA estimates and cyclically ambiguous fine DOA estimates are obtained. In the second stage, another cumulant matrix is derived and decoupled to generate unambiguous and cyclically ambiguous estimates of range parameter. After disambiguation, the mixed sources can be localized with high accuracy. Compared with some existing algorithms, the proposed algorithm enjoys extended array aperture and higher estimation accuracy. Secondly, as a matter of fact, the source number is usually unknown in practice. To circumvent this issue, a novel algorithm for the localization of mixed FF and NF sources without estimating the source number is developed. By constructing multiple fourth-order spatial-temporal cumulant matrices and exploiting the joint diagonalization structure among these matrices, the algorithm decouples the DOA estimation from the range estimation. After that, the range parameters are determined via Capon beamforming technique. This algorithm avoids the performance deterioration induced by erroneous source number estimation, and it outperforms other methods in signal-to-noise ratio(SNR) threshold since both the spatial and temporal information are fully utilized. Thirdly, another algorithm for the joint estimation of DOA, range and frequency of mixed FF and NF sources is advanced. By properly choosing the sensor combination, four second order correlation matrices are constructed to generate closed-form estimates of the DOA, range and frequency parameters. Therefore, the resultant algorithm has a low computational cost because it only requires second-order statistics and avoids multi-dimensional search. Finally, a two-stage RARE(TSRARE) alogrithm is developed to solve the problem of mixed source parameters estimation under unknown mutual coupling. The FF DOA estimates are firstly obtained by RARE estimator. Then, by the mutual coupling compensation and the FF components elimination, NF DOA parameters are found through another RARE estimator. After that, the range parameters are generated from 1-D MUSIC spectral search. The proposed algorithm is robust to mutual coupling effects and achieves a reasonable classification of the signals types. Moreover, it is efficient in that it only requires second-order statistics and one dimensional spectral search.