Two-Dimensional Rank-Deficient Capon Estimator And Its Applications

Spectral estimation is a fundamental topic in the signal processing field, and attracts many experts'attentions. The standard Capon spectral estimator is the one based on adaptive filter bank. To obtain the accurate estimates in case of finite sample available, the filter with longer tap is always selected in this kind of estimator. Because the number of snapshots is decreased, the sample covariance matrix is rank-deficient, which results in the conventional Capon estimator un-valuable.In this thesis, the two-dimensional (2-D) Capon spectral estimator, which is based on the rank-deficient covariance matrix, is studied. We extended the one-dimensional (1-D) modified version of the rank-deficient robust Capon filter-bank spectral estimator to the 2-D case, which is referred to as 2-D MRD-RCF spectral estimator. To improve the accuracy of the covariance matrix, a forward-backward method is employed for estimating the covariance matrix. With the method for covariance estimation, the 2-D MRD-RCF spectral estimator can be further renamed as 2-D FB MRD-RCF spectral estimator. The simulation results show the better performance of both methods.Fine synthetic aperture radar (SAR) images result in the super performance of automatic target recognition (ATR). Because both 2-D MRD-RCF spectral estimator and 2-D FB MRD-RCF spectral estimator are the spectral estimators with super resolution, they can be introduced into analyzing the fine characteristics of SAR images. With the qualitative and quantitative analyses of the processed SAR images, both estimators can improve the resolution of SAR image. Furthermore, both estimators can suppress the speckles and retain the shadow information in the SAR image.