Fast on-line algorithm for computing reduced-rank Wiener filters

Abstract

The reduced rank Wiener filter (RRWF) can be utilized wherever a desired signal needs to be extracted from random background noise or deterministic interference. Common applications are echo cancellation, equalization, neural network learning, and spectral line enhancement. This paper introduces a novel algorithm for the fast on-line computation of the RRWF. The algorithm is derived by making a certain approximation to the Alternating Power (AP) method to reduce its computational complexity from O(m2r) to O(max(m2,mn)) (or O(mn) if the input is white). Simulations show that somewhat surprisingly the computational saving does not come at the cost of estimation accuracy or convergence speed.