EFFICIENT ESTIMATION OF THE SIGNAL SUBSPACE WITHOUT EIGENDECOMPOSITION

A method of obtaining estimates of a set of basis vectors spanning the signal subspace without eigendecomposition is described. Each basis v ector can be determined as the solution to a linear least-squares pred iction problem, thereby offering a reduction in computation of one ord er of magnitude compared with eigendecomposition-based methods. Experi ments suggest that the proposed method has performance equal to that o f MUSIC.