Advances in adaptive orthogonal filtering with applications to source localization

摘要

Adaptive orthogonal filters in Givens rotation format are developed for adaptively estimating the eigenstructure of autocorrelation matrices and spectral density matrices. For the former case a triangular array is developed by exploiting the connection between orthogonal adaptive unit norm filtering and the Schur eigenvalue deflation technique. For the latter case a class of adaptive paraunitary filters is developed with an efficient update algorithm that overcomes the computational burden of a recursive maximum-likelihood approach. Based on simulations, the new adaptive paraunitary filters appear to yield a single stationary point even in environments where a recursive maximum-likelihood approach gives many local minima.