IRGAN: cGAN-based Indoor Radio Map Prediction

Radio map or radio coverage prediction in indoor and outdoor remains a challenge of great interest due to the large number of applications it allows. Many techniques such as data-driven interpolation methods, model-based data fitting methods, model-based prediction methods exist. However, each of them has their limitations (computation time, accuracy, level of input information required, generalization to different environments). Also indoors seems less studied than outdoors. In this paper, a multi-material model and its enhanced version boosted by attention mechanism for indoor radio map prediction are introduced. The proposed models are based on conditional Generative Adversarial Networks (cGANs) and take floor plan images as input. We also propose a new method for floor plan images preprocessing by segmentation to characterize the materials in the environment of interest, which alleviates user's effort. The validity and efficiency of our method in generating high quality radio maps in different environments and its ability to consider the electromagnetic properties of materials are verified on two simulated datasets. Numerical results show that our approach outperforms state of the art methods.