Diffusion Affine Projection Algorithm for Multitask Networks

Abstract

Distributed adaptive networks achieve better estimation performance by exploiting temporal as well as spatial diversity. In this paper, we consider the problem of estimating multiple optimal parameter vectors (also termed as tasks) under correlated input, over a sensor network, where the nodes within the same cluster are engaged in estimating a common optimum parameter vector in distributed manner. For this, we present an efficient multitask diffusion affine projection algorithm (APA). The proposed scheme uses a regularized term to promote similarity among the parameter vectors estimated by neighboring clusters. Usage of APA makes the algorithm robust against correlated input. We present important results on the mean and mean square convergence of the proposed strategy. Simulations are carried out to demonstrate the effectiveness of the proposed algorithm. Compared to the non-cooperative APA, the proposed multitask diffusion APA exhibits remarkably improved performance in terms of both convergence rate and steady-state MSD.