Spectrum Monitoring in Cognitive Radio Networks: a Game-Theoretical Approach

Ming Li

2015-01-01

Abstract:Opportunistic spectrum access is a great way to improve the spectrum utilization, but it also generates spectrum misuse problems. We employ limited number of monitors to dynamically monitor all channels concerned, then detect and punish the malicious secondary users. The problem is which channels to monitor, how long to monitor each channel, and in which order to monitor them. We model this problem as an adversarial multi-armed bandit problem with switching costs. To solve this problem, we propose two algorithms. The first algorithm selects strategies based on monitoring history and maintains same strategy for some time to reduce switching costs. We prove its expected weak regret to be O(T^{2/3}) where T is the time horizon. The second algorithm is modified to have more parameters and cover all channels more efficiently. We show there exists a confidence bound of the new algorithm's weak regret (still O(T^{2/3}). We also simulate the algorithm with different types of adversaries. Simulation results indicate the bounds hold and our algorithms sometimes outperform the best fixed strategy.