Delayed least-mean-square algorithm

Abstract

The conventional delayed least-mean-square (LMS) algorithm converges more slowly than the LMS algorithm due to the delay introduced in coefficient adaptation. The authors derive a new delayed LMS algorithm, which is a realisation of the orthogonality principle, by minimising the a posteriori squared error. Computer simulations show that it does not suffer from the introduced delay and has comparable convergence performance to the LMS algorithm.