On efficient implementations of the set-membership NLMS algorithm for real-time applications

This paper addresses two efficient implementations of the Set-Membership Normalized Least-Mean Squares (SM- NLMS) algorithm suited for real-time applications with strong limitation in computational complexity. It is shown how to take advantage of a particular property of set-membership algorithms, their data selective updating, to improve performance with no additional costs concerning computational complexity or hardware. Convergence speed and misadjustment of the proposed alternative implementations are compared to those of the conventional implementation of the SM-NLMS algorithm when the computational complexity per iteration is limited.