Competitive Spectrum Access in Cognitive Radio Networks: Graphical Game and Learning

Competitive spectrum access is studied for cognitive radio networks. Based on the assumption of rational secondary users, the spectrum access is modeled as a graphical game, in which the payoff of a secondary user is dependent on only other secondary users that can cause significant interference. The Nash equilibrium in the graphical game is computed by minimizing the sum of regrets. To alleviate the local knowledge of payoffs (each secondary user knows only its own payoff for different channels), a subgradient based iterative algorithm is applied by exchanging information across different secondary users. When information exchange is not available, learning for spectrum access is carried out by employing stochastic approximation (more specifically, the Kiefer-Wolfowitz algorithm). The convergence of both situations is demonstrated by numerical simulations.