DISC-Gen: DImension and Spectrum Conscious Generative Models for Universal Indoor Radio Design

A generative model of designing layout of indoor radio transmitters for a floor plan is introduced by formulating it as an image-to-image translation problem between a floor plan and its RSRP heatmap of the layout with reasonable coverage. The proposed model also considers physical dimension of a floor plan and spectrum band to learn how radio signals propagate and propose a desirable layout of radios from accumulated indoor radio designs by human experts. Comparative study between several generative models, including diffusion and consistency models, shows that high-resolution GAN with the joint embedding of dimension and spectrum band outperforms others. A few cases of the AI-generated designs show potentials of finding new and more efficient designs than traditional approaches.