The First Indoor Pathloss Radio Map Prediction Challenge
Stefanos Bakirtzis, Çağkan Yapar, Kehai Qiu, Ian Wassell, Jie Zhang
TL;DR
This work introduces the ICASSP 2025 First Indoor Pathloss Radio Map Prediction Challenge to advance data-driven indoor radio propagation models for directional transmissions. It provides a Ranplan-based indoor ray-tracing dataset with 25 environments for training across three frequency bands and five antenna patterns, plus a separate test set with unseen geometries and patterns. The challenge evaluates three generalization tasks using RMSE-based metrics and a weighted score, highlighting state-of-the-art DL approaches such as U-Net variants, transformer-based encoders, and physics-informed features. Key findings include RMSEs in the low 0.05 range for top methods, a weighted RMSE of $9.41$ dB for the best entry, and systematic increases in error for larger buildings and wall-adjacent regions, underscoring the ongoing need for robust indoor propagation models and fair comparison frameworks.
Abstract
To encourage further research and to facilitate fair comparisons in the development of deep learning-based radio propagation models, in the less explored case of directional radio signal emissions in indoor propagation environments, we have launched the ICASSP 2025 First Indoor Pathloss Radio Map Prediction Challenge. This overview paper describes the indoor path loss prediction problem, the datasets used, the Challenge tasks, and the evaluation methodology. Finally, the results of the Challenge and a summary of the submitted methods are presented.