Delay-Doppler Radiomap for High Accuracy Channel Prediction

This paper proposes FEC+ (Fully connected and Convolutional layers Enhanced Channel predictor), a novel deep learning architecture designed to improve radiomap-based channel prediction by leveraging delay-Doppler (DD) domain characteristics, particularly suitable for high-mobility wireless communication scenarios. FEC+ processes spatial and motionrelated input features through an integration of fully connected and convolutional layers in a model-aware manner, enabling adaptive refinement based on positional and kinematic information. By jointly modeling spatial structures and dynamic user behavior, the proposed framework effectively captures spatiotemporal correlations in the DD domain, leading to more accurate and robust channel state prediction. Experimental results show that FEC+ significantly outperforms existing AIassisted methods, providing a scalable and reliable solution for next-generation wireless communication systems.