AoA Services in 5G Networks: A Framework for Real-World Implementation and Systematic Testing

TL;DR

This work tackles the need for low-latency, high-precision positioning in 5G by moving localization into the radio access network and leveraging uplink SRS-based AoA with a single-anchor, network-native approach. It introduces the first fully open-source end-to-end testbed that couples NVIDIA Sionna RT ray tracing with a Keysight PROPSIM channel emulator and a USRP N310-based hardware setup, enabling hardware-in-the-loop, repeatable experiments. Two subspace AoA methods, MUSIC and ESPRIT, are implemented on a four-element ULA and augmented with a calibration procedure and cylindrical correction to achieve sub-degree to a few-degree accuracy under realistic multipath and SNR conditions. The results demonstrate the feasibility of network-native, single-anchor localization in real 5G networks and provide a practical platform for reproducible evaluation and future xApp-based localization services.

Abstract

Accurate positioning is a key enabler for emerging 5G applications. While the standardized Location Management Function (LMF) operates centrally within the core network, its scalability and latency limitations hinder low-latency and fine-grained localization. A practical alternative is to shift positioning intelligence toward the radio access network (RAN), where uplink sounding reference signal (SRS)-based angle-of-arrival (AoA) estimation offers a lightweight, network-native solution. In this work, we present the first fully open-source 5G testbed for AoA estimation, enabling systematic and repeatable experimentation under realistic yet controllable channel conditions. The framework integrates the NVIDIA Sionna RT with a Keysight PROPSIM channel emulator and includes a novel phase calibration procedure for USRP N310 devices. Experimental results show sub-degree to few-degree accuracy, validating the feasibility of lightweight, single-anchor, network-native localization within next-generation 5G systems.

Figures (6)

• Figure 1: Calibration example, by using a 4-element .
• Figure 2: Geometric scenario of cylindrical correction. MUSIC naturally estimates the orange angle.
• Figure 3: Experimental testbed: at the top, the Sionna ray tracer, which fetch the channels to the PROPSIM; to the left, the UE connected through a douplex connection to the O-RAN gNB.
• Figure 4: An example of AoA estimation with and without multipath for the same urban scenario. In orange the estimated $\hat{\theta}$, while in green, the corrected $\hat{\theta}_{XY}$.
• Figure 5: MUSIC vs ESPRIT in three different multipath depth scenarios, after cylindrical correction.