Partial-Diffusion Least Mean-Square Estimation Over Networks Under Noisy Information Exchange
For researchers in distributed adaptive filtering, this work extends analysis of PDLMS to realistic noisy links, but the contribution is incremental as it generalizes existing algorithms.
This paper analyzes the partial-diffusion least-mean-square (PDLMS) algorithm over networks with noisy information exchange, showing that noise disrupts the trade-off between communication cost and estimation performance compared to ideal links.
Partial diffusion scheme is an effective method for reducing computational load and power consumption in adaptive network implementation. The Information is exchanged among the nodes, usually over noisy links. In this paper, we consider a general version of partial-diffusion least-mean-square (PDLMS) algorithm in the presence of various sources of imperfect information exchanges. Like the established PDLMS, we consider two different schemes to select the entries, sequential and stochastic, for transmission at each iteration. Our objective is to analyze the aggregate effect of these perturbations on general PDLMS strategies. Simulation results demonstrate that considering noisy link assumption adds a new complexity to the related optimization problem and the trade-off between communication cost and estimation performance in comparison to ideal case becomes unbalanced.