Fast Indoor Radio Propagation Prediction using Deep Learning

We introduce efficient deep learning models that estimate maps of two indoor radio propagation parameters: i) received power and ii) cell association. These models, based on U-Net architectures, take floor plans and transmitter positions as inputs, and estimate the two maps. Training data was obtained from simulations using the IEEE 802.11ax path-loss model at 5GHz without co-channel interference. Validation shows errors below 1 dB (received power) and 1 % pixel error rate (cell association). Our models reduce simulation times by 30x to 40x compared to traditional methods (e.g., ray-tracing, dominant path), making them valuable for wireless network planning and deployment tasks that require iterative procedures.