Convex Combination of Diffusion Strategies over Distributed Networks

Diffusion adaptation is a useful strategy for distributed estimation over networks. Though several information fusion strategies for the diffusion adaptation have been proposed in the literature, it can be restrictive to use a single strategy especially for networks operating in non-stationary environments. Inspired by the convex combination of adaptive filters, in this paper we propose to benefit the performance of two distinct strategies by appropriately combining their fusion coefficients. The combination coefficient on each node is determined by minimizing the overall squared estimation error in a local and online manner. Simulation results highlight favorable properties of the proposed combination scheme, with both static and dynamic fusion components.