Localizing Cell Towers from Crowdsourced Measurements

Abstract

Today, several internet sites exist that aim to provide the locations and number of cellular network antennas worldwide. For example [1],[2] and [3]. What makes this task difficult to accomplish is the lack of information available about the whereabouts and number of antennas. Only in a few countries are correct locations for some cellular network antennas known. Otherwise, these sites base their knowledge about cellular network antenna locations on measurement data collected from crowdsourcing. OpenCellID uses a simple and primitive algorithm for estimating antenna locations based on such measurements. In this thesis we suggest an alternative approach to localize cellular network antennas based on data provided by OpenCellID. We start by giving an introduction to the problem, and give a brief overview of related work. This includes localization of mobile devices in addition to localization of cellular network antennas. We then present some background information for our algorithm development. Next we develop two similar algorithms for localizing cellular network antennas. One utilizes distance between measurements, the other utilizes Received Signal Strength (RSS) values among measurements. We experiment with the two algorithms on theoretical generated test data, and argue that utilizing RSS gives the most accurate estimated antenna locations. Next we present the OpenCellID data. We explore this data in detail before defining two subsets we will test our two algorithms on. One subset contains measurement data where correct antenna locations are known. The other contains measurement data for antennas in the Bergen City Center area. We then estimate cellular network antenna locations with our two algorithms for the two subsets. Our tests will show that utilizing RSS estimates more accurate antenna locations when correct antenna locations are known and can be compared to. We end the thesis by analyzing two measurement distribution patterns, and propose how the algorithms can be improved.