Subspace Estimation Along a Frequency Band Through Projection Matrix Approximation

In this paper, we present a method to estimate the signal subspace at all frequencies in given band, which is computed from usual set of frequency-bin sample covariance matrices wideband estimation. Fundamentally, exploits similarity between any two near-by produce an improved along band. Its key idea consists modeling by means projection matrix function approximated polynomial. The provides improvements: reduced-size representation frequency and quality improvement direction-of-arrival (DOA) estimators such as Incoherent Multiple Signal Classification (IC-MUSIC) Modified Test Orthogonality Projected Subspaces (MTOPS). paper includes derivation asymptotic bounds for bias root-mean-square (RMS) error estimate, numerical assessment its combination with previous DOA estimators.