Signal subspace decomposition of ideal covariance matrices

Abstract

Xu and Kailath have proposed (1992) a fast algorithm for signal subspace decomposition that exploits the special matrix structure associated with signal subspace algorithms. This work presents some modifications which eliminate the need to estimate the noise eigenvalue in order to estimate the orthonormal basis of an ideal covariance matrix. The algorithm yields the exact signal subspace and in so doing yields the exact subspace dimension. The modifications presented reduce the computational load by at least a factor of four.<>