Superfast Convergence Rate in Adaptive Arrays

This paper introduces a class of model-matched Toeplitz covariance matrix estimation (MM TCME) algorithms for adaptive arrays. Adaptive filters employing these algorithms are referred to as TMI filters. When the angular separation between the interference sources is not too close to a certain statistical resolution limit (SRL), the convergence rate for TMI filters is superior to that of loaded persymmetric covariance matrix inversion (LPMI) filters and to that of the well-known loaded SMI (LSMI) filters. In terms of the 3dB average SNR loss, for the TMI filters, the required training sample size is about ml2 (m is the number of interference sources), while that for the LPMI and LSMI filters is about $m$ and 2m, respectively. Since the TMI filters may suffer severe SNR degradation when the angular separation between the sources is near the SRL two remedies for dealing with this problem are discussed herein.