Machine Learning for Spectrum Access and Sharing

This chapter focuses on online learning algorithms that were developed for dynamic spectrum access (DSA), in which a cognitive user aims to learn the occupancy of the spectrum in the presence of external users to improve the spectral usage. The focus is thus on how quickly the cognitive user can learn the external process, and not on the interaction between cognitive users. The chapter discusses the more general model where multiple cognitive users share the spectrum, and the goal is to effectively allocate channels to cognitive users in a distributed manner in order to maximize a certain global objective. It then provides an overview of model-dependent solutions and discusses the very recent developments of artificial intelligence algorithms based on deep learning for DSA that can effectively self-adapt to complex real-world settings.