Structured covariance estimation via maximal representation of convex sets

Abstract
A method that utilizes the properties of the set of structured covariance matrices in order to perform covariance estimation is presented. The constraint space of allowable covariance matrices is characterized as a convex cone within the space of positive definite Hermitian matrices. The Caratheodory representation theorem is invoked to show that any member of this cone can be represented as a positive weighted sum of a small number of generating dyads. Simulation results confirm the computational efficiency of this approach and its significance in an adaptive beamforming application.

This publication has 7 references indexed in Scilit: