Countrywide Basestation Localization with Timing Advance Measurements from Crowdsourcing

Abstract

Recently, there has been increased interest in using network traces from crowdsourcing for anomaly detection or operator benchmarking. Likewise, the inference of the network topology, particularly the localization of basestations, has been an active field of research. In this work, we conduct a large-scale analysis of a probabilistic basestation localization technique operating on crowdsourced Timing Advance measurements collected through an Android application. We evaluate the performance using ground-truth locations for 4,946 LTE cells, covering a representative subset of the infrastructure of an Austrian operator. Careful analysis confirms the well-known quality issues of crowdsourced data, with 5.8% of samples effectively reporting negative multipath components. Incorporating the notion of erroneous measurements into our system model ensures that the presence of such outliers is reflected in the posterior confidence of the estimator. Overall, this adapted approach significantly outperforms the baseline estimators from the literature, achieving an accuracy of above 97.5% in a cell-identification framework and mean distance errors of around 100 m. Deviations from multipath and erroneous samples remain the primary error source, given that for a low-multipath subset, the performance can be further improved to 99% and 50 m, respectively.