Learning to Sample Ray Paths for Faster Point-to-Point Ray Tracing

Eertmans, Jérome [UCL] Di Cicco, Nicola [PoliMi] Oestges, Claude [UCL] Jacques, Laurent [UCL] Vitucci, Enrico Maria [UniBo] Degli-Esposti, Vittorio [UniBo] et al.

Ray Tracing has been extensively utilized in recent years to model radio propagation. Depending on the application, two main variants exist: Ray Launching and Point-to-Point Ray Tracing. In contrast to Ray Launching, Point-to-Point Ray Tracing aims to identify all possible paths between a pair of nodes. This typically entails a greater computational complexity, as the number of paths to be traversed increases exponentially in proportion to the size of the scene and the number of interactions permitted. Nevertheless, only a relatively small proportion of these paths ultimately prove to be valid. In this paper, we present a Machine Learning model that is capable of learning how to sample the valid paths from the set of all possible paths. The complexity of the Machine Learning model is linear with respect to the number of objects in the scene. Moreover, in contrast to recent proposals in the literature, our model is designed to accommodate input scenes of any size and does not depend on electromagnetic properties such as frequency.