Robust Beamforming Algorithms With Subspace Steering Vector Uncertainties

Robust adaptive beamforming is one of the most attractive fields in array signalprocessing, which has a wide application in communication, sonar, radar, seismology,medical imaging and so on. Due to the effect of errors such as steering vector error andcovariance matrix error, the performance of adaptive beamformer suffers significantdegradation. Hence, it is necessary to research the robustness of adaptive beamforming.In this thesis, the algorithms related to robust adaptive beamforming are discussedand two new algorithms are proposed. The first algorithm is to overcome the drawbacksin the MDB method based on known interference subspace (MDB with known IS) andMultirank MVDR method based on GSC. The second one is a new algorithm based onthe subspace uncertainty. Both the algorithms can improve the outputsignal-to-interference-plus-noise ratio, which is verified by simulations. The mainresearch content is summarized as follows:(1) Introduce the model and the basic knowledge of beamforming. The effect ofsteering vector errors on the performance is presented. Then several classical algorithmsrelated on robust adaptive beamforming are introduces, along with their advantages anddisadvantages.(2) Introduce three signal model based on subspace, and the application of the threesignal model in radar, sonar and wireless communication. Then the method to computethe steering vector of desired signal is presented.(3) Give the method to construct linear subspace and point out the fact that thesubspace needed in MDB with known IS should be known a prior and the influence ofinterference on the error covariance matrix should be neglected in Multirank MVDR. Toovercome the above demerits, a new algorithm is proposed in this thesis and a carefulanalysis is given. Through simulation results, the proposed algorithm has a similarperformance with MDB with known IS and has a better performance than MultirankMVDR when the DOA of interferences is close to that of desired signals.(4) Propose a novel algorithm based on subspace uncertainty set. When signalsubspace estimation is not perfect, based on the conventional method of signal subspace estimation, we add another constraint, i.e., the signal subspace intersect with a knownsubspace where the steering vector is located. This problem can be solved by alternativeiteration in two sub-problems which both have a closed-form solution. The numericalresults show that the proposed algorithm outperforms diagonal loading method andsignal-subspace projection in the case of low SNR.