A Machine Learning Assisted Cell Selection Method for Drones in Cellular Networks

We apply machine learning techniques to predict the cell quality for the aerial drones connecting with a standard cellular network on the ground. Stationary and strong spatial correlation of the aerial channels allow for exploiting predictive techniques for optimal cell selection based on few available neighboring observations. Yet, drastic cell quality changes due to the side lobes of base-station antenna patterns require advanced solutions for accurate prediction. In this paper, we propose a conditional random field based framework to predict a drone's best (or top few) candidates for the serving cell. Our results, assuming realistic antenna patterns as well as errors in the location estimates, show a high prediction accuracy, thereby illustrating the feasibility of exploiting learning approaches to predict the aerial channel environment.