User equipment positioning in 5G NR and beyond

Precise positioning of User Equipment is (UE) a cornerstone for emerging Internet of Things (IoT) applications and intelligent industries, including smart factories, autonomous vehicles, robotics, and smart cities. To address these demands, 5G introduces substantial enhancements in bandwidth, latency, architecture, and dedicated signaling that enable advanced localization services. Despite this potential, accurate positioning in real-world deployments remains challenging. Conventional methods rely on signal's arrival time or angle measurements, which are severely affected by practical impairments such as multi-path propagation, Non-Line-of-Sight (NLoS) links between the user and Transmission Reception Points (TRPs), and synchronization errors arising from hardware imperfections. Moreover, much of the state-of-the-art in the literature relies on simulations or synthetic data, which overlook these real-world limitations and fail to capture the complexity of operational environments, particularly indoors or in dense urban scenarios. In this dissertation, we adopt an experimental perspective to close this gap. We design and implement a complete 5G positioning system that follows the 3rd Generation Partnership Project (3GPP) specifications, leveraging standard positioning signals, protocols, and architecture. Specifically, we present the first experimental evaluation of the uplink Time Difference of Arrival (TDoA) method integrated within the OpenAirInterface (OAI) platform, including the New Radio Positioning Protocol A (NRPPa) and the Location Management Function (LMF). Furthermore, we deploy this system in diverse indoor and outdoor testbeds using commercial Open RAN (O-RAN) Radio Units (RUs), enabling an unprecedented validation of standard-compliant 5G positioning in practice. Our results demonstrate both the effectiveness and the limitations of TDoA-based methods in real deployments. Motivated by these findings, we further explore advanced solutions based on data-driven and learning-based approaches. Leveraging the large-scale datasets collected from our experimental testbeds, we design supervised and unsupervised models with sensor fusion that can outperform conventional TDoA in scenarios with insufficient measurement quality. Finally, we benchmark these learning-based solutions against both TDoA and commercial high-accuracy alternatives such as Real-Time Kinematic (RTK) positioning, highlighting their potential to overcome constraints of standard methods.

These proposed open-source frameworks, spanning both 5G and beyond-5G standards, together with the publicly shared datasets from our testbeds, constitute a unique contribution to the positioning research community.