Learning-based predictive resource allocation for machine type communications

The goal of this thesis is to design novel frameworks that will facilitate the development of predictive resource allocation mechanisms for machine type communications (MTC). This is done by exploiting a set of machine learning tools such as causal inference, multi-armed bandits (MABs) and deep learning. MTC networks often encompass a very large number of machine type devices (MTDs) that require wireless radio resources to send small data packets in the uplink of cellular networks. A predictive resource allocation framework based on the concept of the so-called fast uplink grant is proposed for MTC. In this scheme, the base station (BS) predicts the set of active MTDs at each time step and proactively allocates resources to them. Challenges such as source traffic prediction and optimal resource allocation for predictive resource allocation are addressed in this thesis.