A Distributed Learning Automata Scheme for Spectrum Management in Self-Organized Cognitive Radio Network

We propose a distributed Learning Automata (LA) for spectrum management problem in Cognitive Radio (CR) networks. The objective is to design intelligent Secondary Users (SUs) which can interact with the RF environment and learn from its different responses through the sensing. It is assumed there is no prior information about the Primary Users (PUs) and other SUs activities while there is no information exchange among SUs. Each SU is empowered with an LA which operates in the RF environment with different responses. That is, the SUs are considered as agents in a self-organized system which select one channel as an action and receive different responses from the environment based on how much their selected actions are favorable or unfavorable. Using these responses, SUs control their accesses to the channels for appropriate spectrum management with the objective to incur less communication delay, less interference with PUs, and less interference with other SUs. The proposed LA-based distributed algorithm is investigated in terms of asymptotic convergence and stability. Simulation results are provided to show the performance of the proposed scheme in terms of SUs' waiting times, interference with other SUs, the number of interruptions by PUs during their transmissions, and fairness.