Beamforming Design Based On Regularized Covariance Matrix Estimation

Beamforming,also known as spatial filtering,is an important aspect of array signal processing.The aim of beamforming is to enhance the desired signal and suppress interferences by filtering the signal in the spatial domain,and to change the weighting factor of each element adaptively according to the change of environment.The estimation of the covariance matrix is one of the challenges in beamforming.The main task of this thesis is to find a suitable algorithm to estimate the covariance matrix and construct a beamformer with good performance.The conventional approach to estimating the covariance matrix known as the sample covariance matrix(SCM)performs poorly when the sample size N is not sufficiently large.In recent years,several regularization-based covariance estimators have been proposed to enhance the accuracy of estimation.Most of these methods only consider how to accurately restore the real covariance matrix,but do not consider the practical application of covariance matrix in beamforming.In practice,these regularization algorithms may not optimize beamforming.In this thesis,two optimal beamformer design structures based on the maximum signal-to-noise plus interference ratio(SINR)criterion and the minimum mean square error(MSE)criterion are considered.A new cross-validation method for selecting the diagonal-loading factors is proposed to estimate the covariance matrix of MVDR beamformer.Through the simulation results,we compare the performance of several different regularization methods.We show that the proposed cross validation method achieves near-optimal selection of the diagonal-loading factors.