Observation vs statistics: Near optimal online channel access in cognitive radio networks

We investigate efficient channel learning and opportunity utilization problem in cognitive radio networks (CRN). We find that the sensing order of multiple channels and channel accessing policy play a critical role in designing effective and efficient scheme to maximize the throughput. Leveraging this important finding, we propose a near optimal online channel access policy. We prove that, our policy can converge to an optimal point in a guaranteed probability. Further, we design a computational efficient channel access policy, integrating optimal stopping theory and multi-armed bandit policy effectively. The computational complexity is reduced from O(K NK) to O(K), where N is the number of channels, and K is the maximum number of sensing/probing times in each procedure. Our simulation results validate our policy, showing at least 40% performance improvement over statistically optimal but fixed policy.