Analysis of distributed parameter estimation in WSN with unreliable nodes

Abstract

In this article we analyze the diffusion normalized least mean square (NLMS) and its set-membership version (SM-NLMS) diffusion algorithms in a scenario where sensor nodes are subjected to different noise variances. We show through simulation that the SM-NLMS is a more robust algorithm in such condition, in addition to the provided reduced energy consumption. We also show that, in such context, the reduced feedback SM-NLMS (RF-SM-NLMS) presents a similar performance when compared to the SM-NLMS with an additional energy saving and lower channel occupancy. Moreover, we propose an adaptive way to choose the SM-NLMS and RF-SM-NLMS parameters in order to provide further performance enhancement in the presence of nodes subjected to different noise variances.