In Ray Tracing We Trust? On the Usage of Ray Tracing for Vehicular Network Simulations

The evolution of GPUs has made the use of ray tracing techniques for network simulations feasible despite its computational demand. Ray tracing is an appealing technique for cooperative driving simulations as well because vehicular scenarios are complex by nature and the proper estimation of communication performance is crucial. One recent yet already well-known ray tracing tool is Sionna which, thanks to the support by NVIDIA, has been growing fast and is already being widely adopted by the community. The usage of ray tracing tools is, however, far from trivial as it is necessary to configure many parameters, create scenarios, and ensure that the results are meaningful. To address these issues, this paper performs a simulation study comparing Sionna with real-world measurements and existing empirical models, in particular the Veins obstacle shadowing model and the ETSI 3GPP Urban Micro model, looking at different parameters that include hardware used for computation, maximum number of reflections, frequencies, and corner cases. To this aim we develop a tool that couples Veins and Sionna, enabling users to easily generate Veins/SUMO and Sionna 3D scenarios from OpenStreetMap and enabling the faithful reproduction of vehicular mobility inside the ray tracer. The results of the study show that particular attention needs to be paid in the choice of the parameters depending on the type of scenario, that corner cases can lead to wrong outcomes, and that the issue is exacerbated by hardware floating point precision. Finally, we show that the ETSI 3GPP model can sometimes result in optimistic path loss values in non line-of-sight (NLOS) conditions.