AN ADAPTIVE COMBINATION RULE FOR DIFFUSION LMS BASED ON CONSENSUS PROPAGATION

Diffusion least-mean-square (LMS) algorithm is a method that estimates an unknown global vector from its linear measurements obtained at multiple nodes in a network in a distributed manner. This paper proposes a novel combination rule in the algorithm used to integrate the local estimates at each node by using the idea of consensus propagation, which is known to be a fast algorithm to achieve the average consensus. Moreover, we optimize constants involved in the proposed combination rule in terms of the steady state mean-square-deviation (MSD) and show an adaptive combination rule, along with an adaptive implementation. Simulation results demonstrate that the proposed combination scheme achieves better MSD performance than conventional combination schemes.