Efficient set-membership filtering algorithms for wireless sensor networks

Abstract

The paper discusses three main adaptive filtering algorithms with partial updates and low computational complexities that converge fast and have a significantly better mean square error (MSE) performance than their non selective-update versions when they are tuned well. The algorithms are set-membership normalized least mean squares (SM-NLMS), SM affine projection (SM-AP) and SM recursive least squares (SM-RLS, also known as BEACON). The lifetime of a wireless sensor network (WSN) is often governed by its power consumption. We show how the previous works for energy prediction, channel estimation, localization and data replication in WSNs can be improved in both accuracy and energy conservation by employing these algorithms. We derive two simplified versions of the SM-AP and BEACON algorithms to further minimize the computational load. The probable drawbacks of the algorithms and the alternative solutions are also investigated. To exhibit the improvements and compare the algorithms, computer simulations are conducted for different scenarios. The purpose is to show that many signal processing algorithms for WSNs can be replaced by one general low complexity algorithm which can perform different tasks by minor parameter adjustments.