5G indoor localization tested in real railway environment with AI methods

RAIL-5G: A Challenging Real-World Dataset and Benchmark for AI-Based 5G Indoor Localization

Machine Learning

Summary

Indoor positioning using radio signals is important for places like factories, but it's hard because signals bounce off metal and objects. The authors collected a big new dataset from inside a railway maintenance building with lots of obstacles and metal. They tested different AI and traditional methods to find devices inside this challenging environment. Their results show that while some AI methods help reduce errors and labeling work, achieving very accurate location under tough conditions is still quite difficult. They shared their dataset and code to help others improve 5G localization in real industrial settings.

What this means in practice

Authors

Jonas Pirkl, Andreas Eidloth, Maximilian Kasparek, Tobias Feigl, Christopher Mutschler

Abstract

Reliable radio-based indoor localization is essential for industrial automation, yet existing methods are often evaluated in environments that do not fully reflect the severe multipath and non-line-of-sight (NLOS) conditions encountered in real deployments. This limits our understanding of how AI, i.e., modern geometric, unsupervised, and semi-supervised localization methods generalize to industrial 5G settings. We present RAIL-5G, a public real-world 5G NR channel impulse response (CIR) benchmark recorded in an operational railway maintenance facility. RAIL-5G covers an 811m$^2$ indoor area with eight distributed radio units, robot-mounted user equipment (UE) trajectories, high-accuracy iGPS reference positions, and severe metal-induced multipath caused by train compartments and maintenance infrastructure. We benchmark geometric time-difference of arrival (TDoA), supervised fingerprinting, unsupervised channel charting, and self-supervised wireless pretraining methods. Our results highlight the difficulty of RAIL-5G: on the full dataset, TDoA reaches 7.02m mean average error (MAE), the best unsupervised method reaches 2.46m, and supervised fingerprinting achieves 0.52m. Self-supervised pretraining reduces the labeling effort by up to one order of magnitude, but the remaining gap to supervised learning shows that robust AI-based localization under severe multipath remains an open challenge. The dataset and code are available at https://iis.fraunhofer.de/rail-5g.