Machine-Learning-Based Opportunistic Spectrum Access in Cognitive Radio Networks

Abstract

The explosive growth of wireless devices and data rate demands makes spectrum scarcity a serious problem. A promising solution is to employ OSA, which enables SUs to seek and opportunistically exploit the underutilized spectrum without interrupting the data transmission of PUs. However, the real-world implementation of OSA still faces several critical challenges including lack of global information, the dilemma of exploration and exploitation, and channel access competition. In this article, we propose a machine-learning-based OSA framework by integrating MAB and matching theory. First, we start from the single-SU scenario without global information while considering the volatility of channel availability. We propose an occurrence-aware OSA (OA-OSA) framework based on the UCB algorithm, which can achieve long-term optimal network throughput performance and a well-balanced trade-off between exploration and exploitation based on only local information. Then we extend OA-OSA to the multi- SU scenario with channel access competitions, and derive an OCA-OSA framework by integrating OA-OSA and the Gale-Shapley algorithm. Simulation results demonstrate that the proposed frameworks achieve superior performance in network throughput and less deviation from optimal performance with global information.