Statistically assisted routing algorithms (SARA) for hop count based forwarding in wireless sensor networks

The main goal of this paper is to provide routing– table-free online algorithms for wireless sensor networks (WSNs) to select cost (e.g., node residual energies) and delay efficient paths. As basic information to drive the routing process, both node costs and hop count distances are considered. Particular emphasis is given to greedy routing schemes, due to their suitability for resource constrained and highly dynamic networks. For what concerns greedy forwarding, we present the Statistically Assisted Routing Algorithm (SARA), where forwarding decisions are driven by statistical information on the costs of the nodes within coverage and in the second order neighborhood. By analysis, we prove that an optimal online policy exists, we derive its form and we exploit it as the core of SARA. Besides greedy techniques, sub–optimal algorithms where node costs can be partially propagated through the network are also presented. These techniques are based on real time learning LRTA algorithms which, through an initial exploratory phase, converge to quasi globally optimal paths. All the proposed schemes are then compared by simulation against globally optimal solutions, discussing the involved trade–offs and possible performance gains. The results show that the exploitation of second order cost information in SARA substantially increases the goodness of the selected paths with respect to fully localized greedy routing. Finally, the path quality can be further increased by LRTA schemes, whose convergence M. Rossi ( ) · M. Zorzi Department of Information Engineering of the University of Padova, via Gradenigo 6/B, 35131 Padova, Italy e-mail: {michele.rossi, michele.zorzi}@dei.unipd.it R.R. Rao University of California San Diego (UCSD), 9500 Gilman Drive, La Jolla, CA, 92093-0436, USA e-mail: [email protected] can be considerably enhanced by properly setting real time search parameters. However, these solutions fail in highly dynamic scenarios as they are unable to adapt the search process to time varying costs.