Issues in energy optimization of reinforcement learning based routing algorithm applied to ad-hoc networks

Ad-hoc networks represent a class of networks which are highly unpredictable. The critical work of such networks is performed by the underlying routing protocols. Decision in such an unpredictable environment and with a greater degree of successes can be best modelled by a reinforcement learning algorithm. In this paper we consider SAMPLE, a collaborative reinforcement learning based routing algorithm, which performs competitively with other routing protocols of similar category. A major concern of SAMPLE is its energy consumption, as most of the wireless nodes are driven by finite battery power. Energy conservation has a direct bearing on the network survivability and also affects the underlying quality of services. Energy conservation is not just the problem of the network layer, but it must be considered at the data link layer. Thus we consider a cross-layer energy conservation algorithm SPAN which models a solution by aiding the routing protocol at the network layer with a backbone of stable energy nodes and conserves the energy of remaining nodes by extracting the best features of IEEE 802.11 power saving mode. Most of the network survivability issues should consider scalable scenarios and our work extends our applied energy optimization framework to scalable scenarios. A scalable scenario can be best visualized if we can gain insight into the performances of underlying mobility models. Thus we extend our work to the analysis of the underlying mobility model and its impact on energy conservation in the traffic-mobility dimensions. We also verify the simulation results statistically using hypothesis testing to prove the superiority of our energy conservation attempts for SAMPLE.