Optimal Action Point for Dynamic Spectrum Utilization Under Rayleigh Fading
LI Bowen Li, Panlong Yang, Jinlong Wang, Liang Shen, Yitao Xu, Yijing Liu and Xiang-Yang Li
In this study, we consider dynamic spectrum utilization in cognitive radio networks (CRNs) under Rayleigh fading, where the link-qualities and availabilities of all channels are time-varying and user could know the exact state of currently observed channel only. The main goal is to derive the optimal action point for determining when and which channel to access and when to switch channel, maximizing system throughput while avoiding causing harmful interference to primary users. We formally describe the continuous decision making process leveraging a two dimension optimal stopping analytical framework, and prove that the optimal channel accessing/switching action can be achieved by a simple one-threshold policy. Finding the optimal action point is then equivalent to attaining the optimal threshold. For the scenario that channel statistics is known a priori, we attain the optimal threshold by constructing a multi-absorbing states Markov chain analysis model. While for the more complex and practical scenario that channel statistics is unknown and non-stationary, a computational efficient online learning algorithm is built by appealing to multi-armed bandit formulation.
Keywords: cognitive radio networks, dynamic spectrum utilization, optimal action point, optimal stopping, multi-armed bandit.