Randomized incremental protocols over adaptive networks

Abstract

We introduce an incremental cooperation mode into the framework of adaptive networks (AN). The method applies to generic topologies and avoids the need to establish a Hamiltonian cycle over the network, generalizing the original incremental mode, while keeping nearly the same mean-square performance, as illustrated by the simulations. We motivate the new mode by relying on an LMS rule at the nodes, and mean-square analysis is provided.