Near-Field Tracking with Extremely Large-Scale RIS: A Sparse Learning Approach

In this paper, we investigate the employment of the extremely large-scale reconfigurable intelligent surface (XL-RIS) in dynamic near-field wireless communication systems. Given the spherical-wavefront nature in the near-field context, channel estimation becomes more challenging, in particular for time-varying channels due to the mobility of the transceivers. This implies that frequent feedback of high-dimensional cascaded channel state information (CSI) associated with XL- RIS entails substantial overhead. To tackle this challenge, we propose a channel tracking scheme for the near-field XL- RIS regime. Specifically, a Kalman filter (KF) based framework is proposed, in which two learning-based networks are employed for obtaining the channel information of the first time slot, thereby acquiring initial prior information for the KF technique. Subsequently, leveraging the known prior information and the temporal correlation of the time-varying channel, the cascaded channel matrix for the next time slot is continuously predicted. Following that, the prediction is updated and refined using the observation from the current time slot as a benchmark, enhancing the accuracy and reliability of the channel tracking process. Finally, simulation results are provided to verify the effectiveness of the proposed scheme.