Effects of network characteristics on learning mechanism for routing in cognitive radio ad hoc networks

In cognitive radio (CR) networks, unlicensed users (or secondary users, SUs) can explore and exploit white spaces, which are the underutilized licensed channels, conditional on acceptable interference to the licensed users (or primary users, PUs). This paper investigates the effects of network characteristics on the network performance of a routing scheme called Cognitive Radio Q-routing (CRQ-routing), which applies an artificial intelligence approach called reinforcement learning (RL). CRQ-routing considers the dynamicity and unpredictability of the PUs´ activities, and finds least-cost routes to destination nodes in a CR network. Using RL, each SU node observes and learns about its operating environment as time goes by, and subsequently establishes least-cost routes, which help to achieve satisfactory SUs´ network performance and minimizes interference to PUs´ activities. Simulation results show that, network performance (i.e. SUs´ interference to PUs, SUs´ end-to-end delay, SUs´ packet loss rate, and SUs´ throughput) is slightly affected by network characteristics, although the overall network performance degrades as the number of nodes in a CR network increases and there is random placement of destination nodes which causes the length of routes to vary in a particular network. While this paper applies RL, similar trends and circumstances are believed to occur in other kinds of learning mechanisms applied to the CR networks.