Predicting Interference Graphs from Data: A Machine Learning Approach

Future telecommunication networks, including 5G and 6G, envision the co-existence of public and private networks with access to a massive number of wireless devices. In practice, such a network leads to serious interference from concurrent transmission in closely separated frequency bands which leads to adjacent channel interference (ACI). The ill-effects of the interference can be mitigated by having a comprehensive understanding of the interference relation between the wireless devices. However, existing approaches are not equipped to identify the devices causing interference. In this work, we model the interference in the network as a graph with devices as nodes and edges representing interference relations. We propose neural network architectures to predict the interference graph. The proposed models are trained and tested for ACI graphs on co-existing private and public network data. The best-performing model accurately predicts the interference graph and generalizes well across different numbers of wireless devices with an average AUROC of 0.92.