On the Limitations of Ray-Tracing for Learning-Based RF Tasks in Urban Environments

Antenna placement and orientation, not solver settings, determine whether ray-tracing simulations can faithfully replicate outdoor radio signals in dense cities, according to a study that tested the Sionna ray-tracer against real measurements across central Rome [1]. The work, posted on arXiv and last revised in June 2026, used a measurement set of 1,664 user-equipments and six base-station sites in the Italian capital [1]. Researchers systematically varied path depth, diffuse, specular and refraction flags, carrier frequency, and antenna altitude, radiation pattern and orientation [1]. Fidelity was scored via Spearman correlation between measured and simulated powers, and by a k-nearest-neighbor localization algorithm that relied on RSSI-based fingerprints [1]. Across all experiments, solver hyper-parameters had an immaterial effect on the chosen metrics [1]. Antenna locations and orientations, by contrast, proved decisive [1]. A simple greedy optimization improved the Spearman correlation by 5% to 130% for different base stations, while kNN-based localization error using only simulated data as reference points fell by one-third on real-world samples — though it remained twice as high as the error obtained with purely real data [1]. The authors conclude that precise geometry and credible antenna models are necessary but not sufficient; faithfully capturing residual urban noise remains an open challenge for transferable, high-fidelity outdoor RF simulation [1]. The study also acknowledges that it used homogeneous materials across the entire scene and that the sensitivity of RSSI signals to building materials was not explored [4]. The team could not obtain easy-to-use 3D models of Rome buildings supported by Sionna’s engine, and they note that complicated building shapes, balconies and other architectural details may significantly affect radio propagation in ways that simplistic models miss [4]. Rome’s built environment is marked by a complex urban structure: recent spatial analysis shows the city exhibits not only a central core but also well-defined intermediate zones shaped by its size and layered history, with distinct clusters of services and tourism sites such as Ostia Antica and Parco di Veio [9]. The simulation-to-reality gap documented in the study undermines the effectiveness of ray-tracing for downstream tasks such as signal-strength prediction, wireless localization and outdoor mapping [4].