GraphHO: A Graph-based Handover Optimization System for Cellular Networks

Abstract

Handover optimization is an important task for the load balancing and mobility robustness in cellular networks. However, the cells in a cellular network often overlap and present strong interactions with nearby neighborhood. This renders the handover optimization a challenging problem. In this paper, we propose a novel graph convolutional neural network to capture the complex interaction between overlapping cells. With this graph model, we further develop a contextual bandit solution to optimize the handover efficiency of a cellular networks. Practical challenges derived from the real-world deployment, such as noisy environment and safety constraint, are also well-investigated and addressed. Extensive experiments in a simulation platform and a real-world cellular network demonstrate that our solution can significantly improve the network quality without a prejudice of network stability.