| In high resolution array signal processing, signal sources are usually assumed to be point sources. However, most of the time there is a multipath scattering phenomenon between sources and the array in practical application scenes, such as wireless communi-cation in multipath environment, the sonar detection when sources are close to the probe platform, and the low-elevation radar target tracking, etc. Under this situation, a dis-tributed source model can be more consistent with the description of the characterization of the signal source. The state-of-the-art methods of distributed sources may meet the requirement of the direction-of-arrival (DOA) estimation precision in general situations. However, with the low signal-to-noise ratio (SNR), the limited snapshots, engineering realization and other harsh requirements, the parameter estimation performance of these methods needs to be improved urgently.However, the actual transmitted signal often has some known characteristics, such as the noncircularity and the spatial sparsity and so on. For researchers it is a new s-tudying direction and challenge to explore these features to improve the performance of parameter estimators of distributed sources. Therefore, this dissertation mainly explores and studies the problem of parameter estimation of distributed sources from the two as-pects:the noncircularity and the spatial sparsity of the signal. The main contributions and innovations are summarized as follows:1. Based on the signal noncircularity, two new one-dimensional parameter estima-tion algorithms of distributed sources are proposed. For coherently distributed noncircu-lar signals, firstly the data model of the first parameter estimation algorithm is presented. Then, based on the beamforming framework, the cost function with respect to the cen-tral DOA and the angular spread is obtained. In the procedure of the minimization of the cost function, the problem of the three-dimensional spectrum search has been trans-formed into that of the two-dimensional spectrum search, which effectively reduces the computational cost. In addition, the proposed algorithm has excellent estimation per-formance of central DOAs and angular spreads, and is able to estimate the parameter when the number of sources is larger than the number of array sensors. Due to taking the angular spread into consideration, the proposed algorithm also has good parameter estimation performance in the case of the big angular spread. For incoherently distribut-ed noncircular sources, the second algorithm can obtain the central DOA by using the cross-correlation matrix. By using the signal noncircularity to extend the received data from two closely spaced uniform linear arrays (ULAs), the proposed algorithm employs the singular-value-decomposition of the extended cross-covariance matrix. Then the re-lationship that the subspaces spanned by the left and right singular vectors respectively are same as that spanned by the generalized steering vectors is utilized, and two groups of central DOA estimates are obtained. After associating the two groups of central DOA es-timates, a smoothing procedure is used to complete the final estimation of central DOAs. Compared to the conventional methods, the proposed algorithm has excellent estimation performance in low SNR case.2. For two-dimensional parameter estimation problems of coherently distributed noncircular sources, three low-complexity algorithms are proposed. In the case of the double parallel ULAs, the first algorithm utilizes the signal noncircularity to extend the received data and its conjugate, and obtains the extended covariance matrix, then uses the selection matrices to construct a rotational invariance relation between two extended steering vectors, which effectively avoids the influence of the noncircular phase. Finally, employing the estimation parameter via rotational invariance technique (ESPRIT), the proposed algorithm obtains the central azimuth DOAs and the elevation DOAs of coher-ently distributed noncircular sources. Compared with the existing ESPRIT-like methods, the proposed algorithm has excellent estimation performance in the case of low SNR and small number of sensors. Unlike the first algorithm, the second proposed algorith-m focuses on the L-shaped array, and the constructed second statistic is the extended cross-covariance matrix. But the second algorithm is still an ESPRIT-like algorithm. In comparison with the other ESPRIT-like methods, their computational complexities are almost same, but due to the introduction of the cross-covariance matrix, the proposed algorithm is less sensitive to the additive Gaussian white noise, so that the estimation accuracy is better. The third proposed algorithm is also based on the L-shaped array, and uses the modified propagator to independently obtain the estimation of central azimuth and elevation DOAs, respectively. After that, the pair-matching of them is accomplished by searching the minimums of a cost function of the estimated central azimuth and ele-vation DOAs. The proposed algorithm not only has exploited the signal noncircularity to improve the parameter estimation performance, but also does not require the estimation and the eigenvalue decomposition of the extended sample covariance matrix, namely, has low computational complexity.3. Based on the incoherently distributed source model and the spatial sparsity, t-wo parameter estimation algorithms are proposed by using the sparse Bayesian learning method. The first algorithm gives a new sparse representation model of incoherently dis-tributed source. Taking the temporal correlation between snapshots into consideration in this model, the sparse matrix is reconstructed by using a block Bayesian learning method, and central DOAs are estimated. But the proposed algorithm has relative high computa-tional complexity, and its parameter estimation performance is constrained by the inher-ent bias that is from the spatial angle sampling. The second algorithm presents a new s-parse representation model with the angle quantization error and the angular spread. Then the central DOA estimation of distributed sources is obtained by utilizing the Bayesian compressive sensing method to reconstruct the sparse matrix. In the case of low SNR and small number of snapshots, the proposed two algorithms performs better than the conventional estimation methods of distributed sources, and has better resolution.4. Combining the signal noncircularity and the spatial sparsity, two parameter esti-mation algorithms are proposed for coherently distributed sources and incoherently dis-tributed sources, respectively. The new central DOA estimation algorithm of coherent-ly distributed sources is proposed based on the Stage-wise orthogonal matching pursuit method. By using the signal noncircularity, the algorithm combines the covariance matrix and the elliptic covariance matrix into an extended covariance matrix. And an overcom-plete sparse representation model is built by vectorizing the extended covariance matrix. Then, the parameter estimation problem is transformed into a sparse reconstruction prob-lem, which can estimate the central DOA. The proposed algorithm is applicable for any noncircular rate, and has excellent performance on SNR. In addition, for incoherently distributed sources, the signal noncircularity is exploited to increase the dimension of the received data matrix. Then, a new sparse representation model with the angle quantiza-tion error and the angular spread is constructed. Based on this model and the generalized approximately message passing method, a new central DOA estimation algorithm is pro-posed without the requirement of grid refinement. The computational complexity of the proposed algorithm is low and shows better central DOA estimation performance.
|
PDF Full Text Request