Time-Varying XL-MIMO Channel Tracking by Image Keypoint Detection

In the near-field region of extremely large-scale multi-input multi-output (XL-MIMO) systems, efficient channel estimation is a critical challenge, often demanding substantial computational resources. This issue becomes even more pressing in dynamic, time-varying environments where both users and scatterers are in motion, necessitating lower computational complexity for real-time channel estimation. In this paper, we propose an innovative solution for near-field XL-MIMO systems with time-varying channels, introducing a fast and accurate channel estimation and tracking scheme inspired by keypoint detection techniques from computer vision. First, we design and train a high-precision, anchor-free channel keypoint detector (CKDet) using a fine-grained orthogonal matching pursuit (OMP) framework as an effective channel estimator. Building on this, we present a novel conditional cascaded OMP-based channel tracking scheme that exploits spatial correlations between consecutive time slots to significantly reduce computational complexity. After obtaining the keypoint locations at all time slots, we apply the Hungarian algorithm to match users and scatterers across all time slots, enabling the construction of motion trajectories for use in environmental sensing applications. Experimental results validate the proposed channel tracking algorithm, showcasing its superior performance, speed, and resilience across a range of signal-to-noise ratios (SNR).