Cell Tower Localisation using Graph Convolutional Networks and Positional Encoding

Capacity planning and monitoring solutions require accurate estimation of nearby cellular tower location. Currently available techniques use war-driving or crowdsourced datasets to estimate locations. We propose an approach for cell tower localisation that can potentially have higher utility and business value for end users. Using deep learning techniques, namely Graph Convolutional Networks (GCNs), along with positional encoding helps us in closely predicting how far the user is from the cell tower. In this work, we use data acquired from three or more User Equipments (UEs) to estimate the tower location. Cell towers emit radio-frequency signal with certain power and the UEs receive this signal where the received power is inversely proportional to the distance. The data used for the computation consists of power readings recorded by the UE and its location. Our approach employs positional encoding along with GCN on real world datasets and shows promising results.