M. Geis, S. Häger, C. Wietfeld, "Taming mmWave Connectivity Prediction with DRaGon: AI Propagation Modeling for Cluttered Industrial Environments," in IEEE Future Networks World Forum (FNWF), Bangalore, India, November 2025

Future 6G networks are expected to enable a wide range of industrial applications that demand ultra-high data rates. To meet these requirements, communication systems will increasingly rely on millimeter-wave (mmWave) frequencies, which offer large available bandwidths but exhibit complex and sensitive propagation behavior. Accurate network planning for such environments requires reliable Radio Environmental Maps (REMs), which in turn depend on precise channel modeling. Machine learning (ML) has shown promise in bridging the gap between computation time and prediction accuracy left by traditional modeling methods, but it has not yet been transferred to large-scale, complex Indoor Factory (InF) environments. In this paper, we present an ML-based propagation model for taming mmWave connectivity prediction in a densely cluttered real-world industrial scenario. The model utilizes features derived from privacy-preserving level-of-detail environmental representations, such as top and side-view image projections. We evaluate two ML model types– LGBM and ANN– on a dataset collected in a complex industrial environment. Both models accurately capture the underlying propagation characteristics and outperform traditional empirical models by 8.18dB in prediction RMSE and by being approximately 30-times faster than ray tracing. We demonstrate the potential of ML-based models to support fast and reliable connectivity map generation, making them suitable for efficient mmWave network planning for industrial scenarios.