Convergence acceleration of the LMS algorithm using successive data orthogonalization

We propose a new adaptive filtering algorithm whose convergence rate is very fast even for a highly correlated input signal. It is well-known that convergence rate gets worse when the input signal to an adaptive filter is correlated. Introducing an orthogonal constraint between successive input signal vectors makes us overcome the slow convergence caused by the correlated input signal. It is shown that the proposed algorithm yields highly improved convergence speed and tracking capability for both time invariant and time varying environments, while being very simple both in computation and implementation.