Opportunistic routing under unknown stochastic models

Abstract

We consider opportunistic routing in wireless ad-hoc networks under an unknown probabilistic local broadcast model. The objective is to design online learning algorithms that govern the sequential selection of relaying nodes based on the realizations of the probabilistic wireless links. The performance measure of interest is regret, defined as the expected additional cost accumulated over time when compared with the optimal centralized opportunistic routing policy under a known model of the wireless links. We propose both centralized and distributed online learning algorithms that achieve the optimal logarithmic regret order.