Robust Adaptive Beamforming Based On Beamformer Sensitivity

Recently, adaptive beamforming has been applied in sonar, radar, communication and others. The standard Capon beamformer achieves the maximum output SINR in error-free case. However, estimation errors of the array covariance matrix and signal steering vector can result in severe performance deteriorations of the standard Capon beamformer, especially if the training data contains the SOI component. Hence, how to enhance beamforming robustness when estimation errors, become a research hotspot.In the early research, diagonal loading method is used to increase the robustness, but there are trade-off between robustness and suppression. Nearly twenty years, many methods for computing the loading factor by steering vector uncertainty set constraint has been raised, which may result in an insufficient suppression of interferers and noise. Until recent years, the algorithm using beamforming sensitivity is proposed, to enhance the robustness by reducing the sensitivity of adaptive beamforming method for covariance matrix perturbation. In this thesis, we proposed a new sensitivity method. The main work is as follows:(1) Introduce the basic theoretical knowledge of adaptive beamforming, introduces several robust adaptive beamforming algorithms and its derivation. This thesis also introduces the definition of beamformer sensitivity and its role in beamforming algorithm.(2) Propose a new algorithm based on beamformer sensitivity. Innovation points of the algorithm are: Firstly, introduce the definition of generalized sensitivity, put forward the method based on choosing the minimize sensitivity of the classic capon weight vector solution in the uncertainty set of spherical signal steering vector, which can improve the robustness of capon; Secondly, analysis the inadaptation of white noise sensitivity in colored noise environment which may cause disturbance of sampling covariance matrix, defined the correlated random error sensitivity based on matrix perturbation; Finally, presents a more intuitive graphical method to derived the spherical signal steering vector uncertainty set into a conical steering vector uncertainty set, and the weight vector is calculated by a simple Lagrange method, which has a rather same computational complexity.(3) Simulate the proposed sensitivity method and methods described above, which shows the sensitivity method has better performance in most situations.