AIRMap: AI-Generated Radio Maps for Wireless Digital Twins

AIRMap is a deep-learning framework designed for rapid radio-map estimation, utilizing a U-Net autoencoder that processes 2D elevation maps and building heights. Trained on 1.2 million samples from the Boston area, it achieves a path gain prediction accuracy of under 4 dB RMSE in just 4 ms per inference, outperforming traditional ray tracing methods by over 100 times. This framework's integration with the Colosseum emulator and Sionna SYS platform indicates its potential for real-time applications in wireless digital twins, making it a significant advancement for practitioners in wireless network simulation.