Neural Network Based Ray Tracing on a Digital Twin for Performance Approximation Using Parameter Analysis

Accurately estimating received power in wireless networks is essential for optimizing signal quality, coverage, capacity, and managing interference. This paper introduces a method that combines machine learning with ray tracing simulations in a digital twin environment to approximate performance in wireless systems. Using MATLAB and OpenStreetMap, data on various ray types: line of sight, reflected, refracted, and diffracted is gathered across different transmitter-receiver setups. Key features, like propagation delay, path loss, propagation distance, phase shift, number of interaction and receiver position, are fed into a multilayer neural network to predict received power. By integrating clustering techniques, this approach enhances prediction accuracy and reduces signal estimate errors, outperforming traditional models. Experimental results show the neural network's superior accuracy and efficiency, supporting real-time optimization in complex urban environments.