Study On Adaptive Beamforming With Robustness And Sparsity In Sensor Arrays

Adaptive Beamforming techniques have attracted considerable interest and are wide-ly used in wireless communications, radar, sonar, navigation, radio astronomy, biomedical imaging, seismology, speech signal processing and so on. Minimum variance distortion-less response (MVDR) has a superior performance on interference-plus-noise suppression compared to data-independent beamformers as long as the desired array steering vector and the array covariance matrix are known or can be estimated accurately. In practice, steering vector mismatch is usually unavoidable and the number of snapshots available is relatively small, which can lead to dramatically degrade for array performance. On the other hand, recent researches show that utilizing the sparsity of systems can improve the performance in the engineering application, in which adaptive beamforming with sparse constraint has shown some advantages such as low sidelobe and robustness against mismatches. In this thesis, we focus on the robustness and sparsity for adaptive beamforming applications in sensor arrays. Concretely, the main points are as follows.1. The system model and primary principle for adaptive beamforming in array pro-cessing are introduced, and then the low-complexity on-line algorithms for adaptive beam-forming are present, such as the least mean squares (LMS) and recursive least squares (RLS) implementation for sample matrix inverse (SMI) beamforming, constant modulus algorithms (CMA) and least squares constant modulus algorithm (LSCMA) for blind adap-tive beamforming.2. Reviews systematically some typical approaches for robust adaptive beamforming, which are linearly constrained minimun variance (LCMV) beamformer, diagonal loading method, eigenspace based beamformer, Bayesian approach and set-based worse-case beam-former. Then according to the latest researches on robust beamforming, we mainly analysis two kinds of most important research results, which are iterative robust minimum variance beamforming and robust beamforming using multi-dimensional covariance fitting. Last-ly, we compare the performance of the algorithms related in above such as computational complexity, advantages and disadvantages and so on.3. In adaptive beamforming with sparse constraint, recent research on MVDR beamforming and spectral self-coherent restoral (SCORE) beamforming shows that an added sparse constraint on beampattern can suppress the sidelobe and improve array performance. As a matter of fact, the desired beampattern itself has the characteristic of a priori sparsity in the most conditions, which can be utilized to improve the array performance. With such the fact, we proposed a kind of least squares constant constant modulus blind beamforming with sparse constraint, which exploits the constant modulus property of the desired signal and the sparsity of the desired beampattern. The proposed algorithm exhibits a better performance than the well-known LSCMA. On the other hand, lots of researches show that MVDR beamforming is sensitive to even minor mismatches between the actual the presumed array steering vectors or inexact estimation of signal covariance matrix due to finite sample snapshots. In comparison, beamforming using constant modulus optimization criterion shows some advantages to the mismatches. In addition, in the presence of constant modulus interferences, the proposed algorithm in above may be not able to capture the desired signal of interest (SOI) exactly. Based on the consideration above, we proposed a kind of robust least squares constant modulus beamforming with sparse constraint, which exploits the DOA (direction of arrival) for SOI and a priori information for the sparsity of the desired beampattern. The algorithm shows that even in the presence of a large DOA mismatch, the approach can still capture SOI, and have a better convergence speed and SINR performance than linearly constrained LSCMA (LC-LSCMA). In addition, compared with MVDR beamforming based on sparse constraint, the algorithm shows a better robustness for steering vector mismatch.