Direction-of-arrival estimation in large sensor arrays

Array signal processing is widely used in many different applications such as radar, sonar, mobile communications, and seismology. One of the most popular tasks of sensor arrays is Direction-Of-Arrival (DOA) estimation. Most conventional DOA estimation techniques are based on the idealistic assumption of exact knowledge of the signal propagation model and antenna array characteristics. In practice, this assumption is rarely satisfied and the performance of these methods is severely degraded. This thesis contributes to the development of novel high-resolution DOA estimation algorithms with an improved robustness against model mismatches.; We study the case of sparse arrays composed of multiple subarrays; a scenario that is gaining popularity as it extends the array aperture and enjoys simple formulation of the DOA estimation problem. In this setup, it may occur that some of the subarrays have unknown orientation errors. We address this problem, and develop a robust DOA estimation algorithm, which is applicable to arrays of arbitrary geometry. A computationally efficient search-free variant of the algorithm is proposed for arrays of specific structure.; We also investigate the case when the inter-subarray unknown parameters are time varying. These parameters may include the inter-subarray displacement vectors and/or channel perturbations (signal fade) between subarrays. A DOA estimation algorithm (root-RARE) applicable for a specific class of arrays has been developed for this scenario. We study the performance of this algorithm and suggest an alternative technique to deal with this situation. Our proposed algorithm is applicable to arrays of arbitrary geometry and enjoys a significant performance improvement.; We also study the case when the sensor array exhibits unknown perturbations to sensor positions. This scenario has attracted a lot of research interest, and many DOA estimation algorithms have been proposed to solve this problem. In general, these algorithms either use calibration sources to estimate the unknown array parameters or use unrealistic assumptions about the array geometry, perturbations structure, and perturbations statistics to jointly estimate the array parameters and the DOAs. For most of the existing algorithms, a computationally expensive multi-dimensional optimization procedure is usually employed. We propose a new DOA estimation algorithm, which does not impose any restrictions on array geometry, signal structure or perturbations statistics. The algorithm does not require a calibration phase every time the array is employed, and has an acceptable computational complexity. A special search-free algorithm is also suggested for a specific class of arrays and perturbations.; The problem of beamspace matrix design is also studied. Conventional approaches to this problem take into account the sectors-of-interest only without considering the possible out-of-sector interference. Hence, the robustness of these methods is not sufficient and other techniques for beamspace design are of great interest. We develop robust formulations to the beamspace design problem based on a matrix filter design approach. We also propose adaptive beamspace techniques, where the beamspace matrix is designed and updated based on the array snapshots. These adaptive techniques enjoy substantial robustness against strong out-of-sector interference sources.; Numerical examples are provided for every problem we address to demonstrate the validity of our analysis.