A ML based empirical Model for next Cell ID Prediction

Abstract:
Improved mobility management in cellular networks has been a central research focus for several decades. For aiding handover (HO) decisions, several mechanisms for estimating the movement patterns and trajectories of users were proposed. Modeling and estimating the dynamic movement behavior of mobile users is quite challenging. Despite several existing approaches in the scientific literature, still some potential for improvement persists. For example, common UE localization techniques on which the user trajectory prediction and the HO decisions rely on are not accurate enough for indoor environments – they only provide 3-30 meter accuracy. Hence, in this work we follow a different approach by aiming not at a fine-grained user trajectory prediction but at a more coarse-grained prediction of the next cells an user will most likely move to. For that we propose a novel Machine Learning (ML) based next serving cells prediction method. It is able to predict the next cell associations of a user several time steps ahead. This provides an opportunity for optimizing HO decisions, thereby paving the way for further efficiency improvements in cellular networks, e.g. through mobility load balancing. The performance of our novel method is evaluated through simulations applying the ns3-Gym framework.