Parallel accurate localization from cellular network

Abstract

Localization from cellular network (LCN) is of great importance for mobile operator. Different from GPS, LCN utilizes operator's network infrastructure to identify the location of mobile devices. The advantage of LCN is that operator can calculate all mobile devices’ location according to the mobile big data, without each device reporting its location got from GPS. The challenges of LCN are localization accuracy and computing efficiency. In this paper, we propose three methods of localization by analyzing mobile signal strength data collected by cellular network per 480 milliseconds. Using an optimized propagation model which considers the penetration loss through buildings, we get fairly high localization accuracy. To improve the localization efficiency, a parallel method based on Spark is proposed to process and analyze mobile big data. Through the experiments of tracking people in real cellular network, our methods reach an accuracy of 100 meter (mean error), which exceeds the international standard (125 meter by FCC). The parallel method also gets a significant efficiency promotion, which reduce 97% time cost than serial methods.