Neighbor Auto-Grouping Graph Neural Networks for Handover Parameter Configuration in Cellular Network

The mobile communication enabled by cellular networks is the one of main foundations our modern society. Optimizing performance and providing massive connectivity with improved coverage user experience has a considerable social economic impact on daily life. This relies heavily configuration network parameters. However, increase in both size complexity networks, management, especially parameter configuration, becoming complicated. current practice, which largely experts' prior knowledge, not adequate will require lots domain experts high maintenance costs. In this work, we propose learning-based framework for handover configuration. key challenge, case, to tackle complicated dependencies between neighboring cells jointly optimize whole network. Our addresses challenge two ways. First, introduce novel approach imitate how responds different states values, called auto-grouping graph convolutional (AG-GCN). During stage, instead solving global optimization problem, design local multi-objective strategy where each cell considers several metrics balance its own neighbors. We evaluate proposed algorithm via simulator constructed using real data. demonstrate that parameters model can find, achieve better average throughput compared those recommended as well alternative baselines, bring quality stability. It potential massively reduce costs arising from human expert intervention maintenance.