On Convergence Properties of Subspace Trackers Based on Orthogonal Iteration
This paper studies convergence properties of subspace trackers using orthogonal iteration. In the context of blind estimation of a time-varying channel, orthogonal iteration and its variants have been widely considered for tracking the channel parameters by updating the eigen-decomposition of an exponentially weighted correlation matrix. While it is well known that orthogonal iteration converges exponentially with arbitrary initial conditions, orthogonal-iteration-based subspace trackers can only inherit these merits when the channels considered undergo extremely slow time-variations. In this paper, the authors generalize the traditional (i.e., fixed subspace) convergence analysis of the orthogonal iteration to include non-stationary situations as well.