Deep Learning Based Received Signal Strength Estimation
Mohammed Mallik, Guillaume Villemaud · 2025
As network densification becomes crucial for future mobile networks, deploying base stations (BS)s grows increasingly challenging due to the complex procedures in planning; moreover, poor layout can result in inadequate coverage. A coverage prediction model helps operators address these issues by identifying coverage gaps, optimizing BS placement, evaluating service quality, and creating radio maps for spectrum sharing, interference management, and localization. Current models rely on base station height, transmission power, or path loss models which are limited to oversimplified scenarios, and remain analytically intractable for more realistic network situations. In this work, we present a convolutional neural network based method to accurately predict received signal strength (RSS) and coverage maps of a network from its topology, thus overcoming the limitations of empirical path loss models. This approach uses the network topology as features in images to train and predict RSS and coverage maps. Experimental results show that our method can produce accurate RSS estimates with a Mean Absolute Error of -33 dBm and is able to generalize well to unseen environments.