Map-Based Path Loss Prediction in Multiple Cities Using Convolutional Neural Networks
Ryan G. Dempsey, Jonathan Ethier, Halim Yanikomeroglu
TL;DR
The paper tackles the problem of predicting path loss in urban environments by leveraging a CNN that directly learns from high-resolution 2-D obstruction height maps. It uses dual data sources—UK Ofcom drive-test measurements and digital surface model clutter data—assembled into fixed-size, multi-channel path-profile inputs and trained with six-city geographic cross-validation. Results show RMSE around 7.3–7.8 dB across holds, with the CNN outperforming the ITU-R P.1812 baseline and remaining competitive with a feature-based FCN, demonstrating the value of high-dimensional DSM information for generalizable propagation prediction. Overall, the work validates a map-based deep learning approach that eliminates the need for engineered clutter metrics while providing robust path loss estimates suitable for spectrum planning across multiple cities.
Abstract
Radio deployments and spectrum planning benefit from path loss predictions. Obstructions along a communications link are often considered implicitly or through derived metrics such as representative clutter height or total obstruction depth. In this paper, we propose a path-specific path loss prediction method that uses convolutional neural networks to automatically perform feature extraction from 2-D obstruction height maps. Our methods result in low prediction error in a variety of environments without requiring derived metrics.