QoS-Based Blind Spectrum Selection with Multi-armed Bandit Problem in Cognitive Radio Networks

In the framework of cognitive radio, joint spectrum sensing and access strategies have been extensively studied recently. As a matter of fact, the sensing ability of cognitive radio is limited and the channel statistics may not be known as a priori. In this paper, we investigate the blind spectrum selection with the multi-armed bandit model, considering both primary user activities and channel quality to meet diverse QoS requirements, e.g. high transmission success rate for real-time applications and high throughput for best-effort applications. Firstly we propose a policy k th-UCB1 which is based on the UCB1 policy for multi-armed bandit problem but converges to the k th-best arm. Then we design a distributed order-optimal policy for multiple users accessing the rank-best channels according to their QoS requirements. The expected regret of proposed policy is proved to be logarithmic in the number of time slots and the simulation results implies it has better performance.