A novel kurtosis driven variable step-size adaptive algorithm (Q2732811)

The paper proposes a new least mean squares filter with time-varying step-size. The latter is selected on the basis of the kurtosis of the estimation error, and, as a consequence, the performance of the filter becomes less sensible to the presence of strong noise. A kurtosis estimator is presented, and the convergence properties of the filter are analyzed. Simulation results are also examined.