Improved data-selective LMS-Newton adaptation algorithms

Abstract

Improved versions of two known LMSN algorithms are proposed. In these algorithms, data-selective weight adaptation is performed and in this way reduced steady-state misalignment is achieved relative to that in the known LMSN algorithms while requiring a similar number of iterations to converge. On the other hand, for a constant misalignment a significant reduction in the convergence speed can be achieved. In addition, the modified algorithms require a reduced number of updates, which leads to a reduced amount of computation relative to that required by the known LMSN algorithms.