Dynamic Spectrum Management via Machine Learning: State of the Art, Taxonomy, Challenges, and Open Research Issues

Dynamic spectrum management (DSM) plays an increasingly important role in wireless communication networks for improving spectral efficiency. Conventionally, DSM is realized with the support of accurate information or with the dependence on assumptions about the network, which could be challenging and impractical in the Internet of Things where a large number of users need to be served. The application of machine learning into DSM is promising to address these issues, and many investigations have focused on this application. This article aims to survey the state-of-the-art research results along this direction. We devise a taxonomy to categorize the literature based on the operation modes, learning paradigms, enabling functions, and design objectives. Moreover, the key challenges are outlined to facilitate the application of machine learning for DSM. Finally, we present several open issues as the future research direction.