Energy-efficient dynamic spectrum access using no-regret learning

In this paper, we consider a cross-layer design of dynamic spectrum access in distributive cognitive radio (CR) networks. We model the licensed channel as a finite-state Markov channel (FSMC) and the CR user selects one channel to access and decides transmission rate and power in order to maximize its energy efficiency. We propose a game theoretic framework to formulate this problem and apply a learning algorithm called modified regret-matching leading to correlated equilibrium which is more practical than the regret-matching learning algorithm applied in the related work. The only thing that each user needs to know is his own realized payoffs and actions. From the simulation results, the modified learning algorithm provides impressive performance.