From 5G Single-Cell Signals to Pervasive Smartphone Positioning: A Real-World Study

Abstract

Accurate and reliable positioning is a cornerstone of pervasive computing. However, GNSS is not always available for pervasive services while cellular-network-based alternatives have proven unsuccessful, due to coarse position accuracy or complex network setup. In this work, we investigate 5G singlecell positioning by leveraging Timing Advance and Angle of Arrival measurements, two key indicators already used in 5G NR for communication. Our approach does not require tight network synchronization among base stations, unlike other 5G positioning techniques. We present the first real-world study of single-cell positioning in a 5G network with a commercial gNB and an off-the-shelf smartphone moving up to 461 meters away from the gNB across four urban trajectories. Our analysis reveals two key challenges: coarse angular and timing resolution, and severe angle and range errors under Non-Line-of-Sight (NLOS) conditions. To overcome these challenges, we present a framework that (i) synthesizes higher-resolution Angle-ofArrival estimates from real 5G beam patterns, (ii) classifies LOS/NLOS conditions with a Convolutional Neural Network trained on beam signal-strength heatmaps, and (iii) refines angle and ranging using multipath-aware corrections. Our system reduces the median positioning error from 86.8 m to 16.5 m with a single gNB, without the aid of external sensors.