Convolutional Neural Networks for Radio Frequency Ray Tracing

Ray tracing simulations are a common tool for commercial and military planning. In this work, we explore whether neural networks can be trained to accurately reproduce the results of these simulations. We create a large, procedurally generated urban dataset with which to train neural networks. We then utilize a tailored convolutional autoencoder to learn from the dataset. By iteratively predicting each set of reflected rays, we find the model is able to predict sharp, distinct ray propagation with low error for low numbers of reflections. This error increases significantly at two reflections, indicating further work is needed to capture the more complicated behavior that higher reflection numbers entail.