Cognitive channel selection for opportunistic spectrum access

Abstract

Cognitive Radio has been proposed to address the problem of spectrum scarcity by exploiting under-utilized frequency bands (FBs). With opportunistic spectrum access, cognitive secondary users (SUs) may access vacant FBs originally assigned to Primary Users (PUs). Considering autonomous cognitive radio networks, where no central entity can share environment information, SUs have to operate as independent agents. Such agents are confronted with PU traffic which may follow either deterministic patterns, as in TV transmission, or stochastic patterns, as in packet-switched or circuit-switched networks. However, the traffic stochastic patterns cannot, in general, reflect the dynamic changes in the FBs, especially when these channels are accessed by not registered users as the SUs. Addressing this issue, a novel reinforcement-learning (R-L) scheme is proposed, based on Q-learning with two alternatives. The proposed scheme enables SUs to use information acquired solely by own sensing and evaluate the available FBs accordingly in order to set an optimal channel selection and sensing order. The evaluation of FBs is based on (i) the occupancy probability and (ii) the mean duration of the time that the PUs are idle. The effectiveness of both alternatives is examined in different environment characteristics. Simulations show that the proposed scheme succeeds in setting the FBs in the optimal sensing order in both static and dynamic cases providing SUs with self-awareness and efficient channel utilization.