Data Fusion of Multiple Spatio-Temporal Data Sources for Improved Localisation in Cellular Network

Abstract

An accurate and reliable estimation of subscribers' locations in a cellular network is becoming increasingly important for not only telco-related services but also commercial domains. The data collected in cellular network for locating subscribers could come from multiple sources with different characteristics such as accuracy, noise variance and spatial and temporal resolutions. Given various localisation techniques, it would be advantageous to utilize the multiple data sources to obtain an accurate location rather than relying on single type of measurement. Data fusion, which integrates multiple types of measurement, is an promising solution to provide location estimation with better accuracy, reliability and coverage. In this work, we proposed a data fusion framework using multiple spatio-temporal data sources. Existing solutions in the literature general rely on generative models based on attributes like Received Signal Strength (RSS), Angle of Arrival (AOA), and/or Round Trip Delay Time (RTT) that may not be available in practice due to various reasons. We address the problem from a pure data driven perspective. The challenges of practical implementation such as oscillation removal and noise estimation are discussed in depth. Moreover, the proposed framework is deployed into production and fully evaluated with data sources from a telco in Singapore.