Efficient tracking of time-varying signal subspaces

Abstract
An algorithm for tracking d principal eigenvectors of an M-dimensional sample data covariance matrix is described. This algorithm requires O(Md/sup 2/) multiplications per iteration yet has performance comparable to algorithms having O(M/sup 2/d/sup 2/) complexity. A proof of the algorithm's convergence is given along with the results of several computer simulations.

This publication has 5 references indexed in Scilit: