Abstract
Radio maps reflect the spatial distribution of signal strength and areessential for applications like smart cities, IoT, and wireless networkplanning. However, reconstructing accurate radio maps from sparse measurementsremains challenging. Traditional interpolation and inpainting methods lackenvironmental awareness, while many deep learning approaches depend on detailedscene data, limiting generalization. To address this, we propose MARS, aMulti-scale Aware Radiomap Super-resolution method that combines CNNs andTransformers with multi-scale feature fusion and residual connections. MARSfocuses on both global and local feature extraction, enhancing featurerepresentation across different receptive fields and improving reconstructionaccuracy. Experiments across different scenes and antenna locations show thatMARS outperforms baseline models in both MSE and SSIM, while maintaining lowcomputational cost, demonstrating strong practical potential.