Hierarchical Meta-learning Models with Deep Neural Networks for Spectrum Assignment

Abstract

In this paper we consider a data-driven approach and apply machine learning methods to facilitate frequency assignment. Specifically, a hierarchical meta-learning architecture that harnesses the predictive capability of both statistical and deep learning approaches is proposed to predict a diverse range of spectrum usage patterns. Using spectrum measurements, network simulations are conducted to evaluate the effectiveness of the proposed architecture. It is shown that the hierarchical meta- learning models with deep recurrent neural networks have great potential for predicting spectrum usage patterns to facilitate multi-tier spectrum assignments.