Licensed and Unlicensed Spectrum Management for Cognitive M2M: A Context-Aware Learning Approach

Abstract

Edge computing has emerged as a promising solution for relieving the tension between resource-limited machine type devices (MTDs) and computational-intensive tasks. To realize successful task offloading with limited spectrum, we focus on the cognitive machine-to-machine (CM2M) paradigm which enables a massive number of MTDs to either opportunistically use the licensed spectrum that is temporarily available, or to exploit the under-utilized unlicensed spectrum. We formulate the channel selection problem with both licensed and unlicensed spectrum as an adversarial multi-armed bandit (MAB) problem, and combine the exponential-weight algorithm for exploration and exploitation (EXP3) and Lyapunov optimization to develop a context-aware channel selection algorithm named $\mathrm {C}^{2}$ -EXP3. $\mathrm {C}^{2}$ -EXP3 can learn the long-term optimal channel selection strategy based on only local information, while dynamically achieving service reliability awareness, energy awareness, and backlog awareness. Specifically, we provide a rigorous theoretical analysis and prove that $\mathrm {C}^{2}$ -EXP3 can achieve a bounded deviation from the optimal performance with global state information. Four existing algorithms are compared with $\mathrm {C}^{2}$ -EXP3 to demonstrate its effectiveness and reliability under various simulation settings.