Discretization by rasterization is introduced into the method of images (MI) in the context of 3D deterministic radio propagation modeling as a way to exploit spatial coherence of electromagnetic propagation for fine-grained parallelism. Traditional algebraic treatment of bounding regions and surfaces is replaced by computer graphics rendering of 3D reflections and double refractions while building the image tree. The visibility of reception points and surfaces is also resolved by shader programs. The proposed rasterization is shown to be of comparable run time to that of the fundamentally parallel shooting and bouncing rays. The rasterization does not affect the signal evaluation backtracking step, thus preserving its advantage over the brute force ray-tracing methods in terms of accuracy. Moreover, the rendering resolution may be scaled back for a given level of scenario detail with only marginal impact on the image tree size. This allows selection of scene optimized execution parameters for faster execution, giving the method a competitive edge. The proposed variant of MI can be run on any GPU that supports real-time 3D graphics.
COBISS.SI-ID: 29665319
Method of images (MI) is one of the oldest methods for radio wave propagation prediction based on the ray-tracing principle. Although the MI was originally restricted to the radio environments with prevailing reflection phenomena, it is also used in indoor scenarios in which through-wall transmission make a significant contribution to the received signal power. Exact handling of propagation paths, either in the form of polyhedra bounding regions or in the form of some other equivalent geometrical description, is usually complemented with the use of visibility trees to contain excessive growth of source images. However, strict visibility trees and double refractions on parallel planes involved in through-wall transmissions are not well-suited to each other. Here we study visibility inaccuracy, which is usually ignored. We propose a source image translation heuristic based on the wall depth, material and field of view. We show that the proposed double refraction modelling improves accuracy of strict visibility trees, which gives a better fit of predicted signal to the theoretically correct solution.
COBISS.SI-ID: 30271015