Differentiable Ray Tracing for THz Radio Channel Characterization with Point Clouds

Niklas Vaara, Lafir N. Mohamed, Peize Zhang, Tuomas Maatta, Lam Huynh, Pekka Sangi, Miguel Bordallo Lopez, Pekka Kyösti, Janne Heikkila

Abstract: Differentiable ray tracing-based radio propagation modeling is a powerful tool for learning material properties from channel measurements. Prior approaches rely on idealized, manually crafted triangle mesh models, which often lack smaller geometric details that become important at higher frequencies. In this paper, we perform differentiable ray tracing directly on a noisy point cloud created from RGB-D images, eliminating the need for handcrafted geometry. We optimize channel parameters using THz channel measurements collected at a center frequency of 318 GHz with a 4 GHz bandwidth. Our results demonstrate that the channel characteristics can be extracted well with available measurements and ray tracing. With the optimized parameters, we achieve a mean relative rms delay spread error of less than 5% and a mean absolute error for received power of about 0.36 dB at novel receiver locations.