Ph.D XXXIX

Supervisor: Angelo Montanari

Mail: marturano.alberto@spes.uniud.it

Research Project

Database-Driven Cellular Fingerprinting for Reliable and Scalable Geolocation

Cellular network fingerprinting is a data-driven approach to mobile-device geolocation based on comparing radio measurements observed by a device with previously collected network fingerprints. At scale, fingerprinting databases may contain millions of sparse, heterogeneous, and noisy observations acquired across different devices, network conditions, geographical areas, and time periods. Cellular geolocation therefore becomes not only a localization problem, but also a data-management problem involving data quality, reliability, indexing, similarity search, and scalable query processing.

This research investigates database-supported methods for improving the reliability, accuracy, and computational efficiency of large-scale cellular fingerprinting systems. A first research direction concerns data quality and robustness, including the detection of anomalous or geographically inconsistent observations and the analysis of factors that may affect localization reliability. A second direction focuses on efficient similarity search over large collections of sparse weighted observations, with particular attention to k-nearest-neighbor search and to techniques that reduce candidate evaluation, memory and storage requirements, and database execution costs without degrading geolocation quality.

The overall goal is to develop reliable and scalable methods for database-driven cellular geolocation by combining data-quality analysis, anomaly detection, and efficient exact similarity search. While cellular localization provides the primary application domain, the underlying methods are also relevant to more general similarity-search problems involving large collections of sparse weighted vectors and weighted-overlap measures.