Partial-diffusion recursive least-squares estimation over adaptive networks

In the diffusion strategies for distributed estimation over adaptive networks, each node calculates a weighted average of the intermediate parameter estimates of its neighboring nodes. Thus, all the nodes should continuously share their intermediate estimates with their neighbors. In this paper, we consider exchanging a predetermined number of elements of each intermediate estimate vector at each iteration rather than the entire vectors. We examine two different schemes, i.e., stochastic and sequential partial-diffusion schemes, for selecting the to-be-diffused elements at each iteration. Accordingly, we propose a partial-diffusion recursive least-squares (PDRLS) algorithm that can alleviate internode communications at the expense of estimation performance. Simulation results show that the communication-performance trade-off offered by the proposed algorithm is indeed lucrative.