ODE-Former for Mobile Channel Prediction: A Novel Learning Structure Leveraging the Physics Continuity

Abstract

Obtaining accurate channel state information (CSI) is crucial and challenging for multiple-input multiple-output (MIMO) wireless communication systems. With the increasing antenna scale and user mobility, traditional channel estimation approaches suffer greatly from high signaling overhead and channel aging problems. By exploring the intrinsic correlation among a set of historical CSI instances, channel prediction is proven to increase the CSI accuracy while lowering the signaling overhead significantly. Existing works view this problem as a regular discrete sequence prediction task while ignoring the unique physics property of wireless channels. This letter proposes a novel former-like learning structure based on neural ordinary differential equations (NODEs) inclusively designed for accurate and flexible channel prediction. The proposed network aims to represent wireless channels’ implicit physics spatial-temporal continuity by integrating the Neural ODE into a former-like learning structure. Our proposed method impeccably fits channel matrices’ mathematics features and enjoys solid network interpretability. Experimental results show that the proposed learning approach outperforms existing methods from the perspective of accuracy, flexibility, and robustness.