Augmented RF Propagation Modeling

Abstract

High-speed, accurate RF propagation models are essential for interference prediction between different spectrum users. We propose a simple and effective regression-based estimation method that can be used both for correcting errors in existing propagation models, such as the Terrain Integrated Rough Earth Model (TIREM), and also as a stand-alone empirical propagation model where no prior model is required. A neural network regression model (NNRM) that is informed by real-world channel and propagation characteristics is developed to obtain accurate path loss predictions throughout a 2.9 km by 2.6 km map of the University of Utah. The proposed NNRM is trained and tested on three path loss measurement campaign data sets collected across the campus, resulting in a 58% to 87% reduction in loss difference variance when used to correct TIREM, and a 59% to 76% reduction in measured signal strength variance when used as an empirical propagation model.