Shrinkage-based capon and APES for spectral estimation

In this letter, we propose shrinkage-based Capon (S-Capon) and APES (S-APES) spectral estimators by minimizing the mean-square error (MSE) of standard Capon and APES in a linear regression framework. The proposed methods are shown to give more accurate spectral estimates but lower resolution than the methods they based on. We combine Capon with the proposed S-Capon and S-APES to overcome the resolution limit of shrinkage-based methods for estimation of both frequency and amplitude of spectral lines. The so-obtained Capon-SCapon and Capon-SAPES spectral estimators, which have about the same computational complexity as Capon, are compared with the Capon-APES (CAPES) by numerical examples. Simulations show that the Capon-SCapon performs similarly to CAPES in a wide range of signal-to-noise ratio, and the Capon-SAPES always gives more accurate spectral amplitude (less bias and lower MSE) than CAPES.