QLAR: A Q-learning based adaptive routing for MANETs

Abstract

Mobile Ad-hoc Networks are highly reconfigurable networks of mobile nodes which communicate by wireless links. The main issues in MANETs include the mobility of the network nodes, energy limitations and bandwidth. Thus, routing protocols should explicitly consider network changes into the algorithm design. In order to support service requirements of multimedia and real-time applications, the routing protocol must provide Quality of Service (QoS) in terms of packets loss and average End-to-End Delay (ETED). This work proposes a Q-Learning based Adaptive Routing model (QLAR), developed via Reinforcement Learning (RL) techniques, which has the ability to detect the level of mobility at different points of time so that each individual node can update routing metric accordingly. The proposed protocol introduces: (i) new model, developed via Q-Learning technique, to detect the level of mobility at each node in the network; (ii) a new metric, called Qmetric, which account for the static and dynamic routing metrics, and which are combined and updated to the changing network topologies. The proposed metric and routing model in this paper are deployed on the Optimized Link State Routing (OLSR) protocol. Extensive simulations validate the effectiveness of the proposed model, through comparisons with the standard OLSR protocols.