Noise cancellation using a split-path adaptive filter

A new structure for adaptive noise cancellation is suggested to improve the convergence characteristics of extracting a non-stationary correlated signal corrupted by random noise. The canceller composes of two linear phase filters which are connected in parallel and both filters are being adapted independently according to Widrow´s least-mean-square (LMS) algorithm. The merits of the new configuration are discussed and a comparison with the conventional structure that consists of a single transversal filter is also included. Experimental results show that the proposed canceller has a superior performance in the context of removing contaminated noise since gradient fluctuation during adaptation could be suppressed considerably which in turn speeds up the rate of convergence