Deep Learning based Localization of LTE eNodeBs from Large Crowdsourced Smartphone Datasets

In recent years, the rapid increase in location-based services has sparked a growing interest in localization. Typically, when GPS is unavailable, many of these services rely on the locations of nearby network infrastructure, notably cell towers, to locate smartphone devices. Although tower locations are known to each mobile operator, this information is not publicly available in many countries, and instead, is estimated via simple heuristic algorithms that rely on hand-crafted features. In this paper, we devise a novel image classification formulation that employs a Deep Convolutional Neural Network to locate cell towers using crowdsourced measurements from smartphones and operator-side tower licensing data. The proposed approach reinforces the ability to extract complex network characteristics from crowdsourced data which, in turn, serves as the foundation for further investigation.We obtain an effective location accuracy of 85% with a mean distance localization error of only 44 metres, effectively outperforming the baseline method used in the industry. Moreover, transfer learning is applied to demonstrate the effectiveness of this classifier in locating towers belonging to mobile operators or metropolitan areas beyond those used during the training phase of the algorithm.