Model-Driven Channel Extrapolation for Massive Fluid Antenna

Fluid antenna (FA), with its unique adaptability and precise signal control capabilities, have emerged as potential components for the future of wireless communication. However, the massive number of ports in FA poses challenges in channel state information (CSI) estimation. Existing methods, predominantly relying on deep learning (DL)-based channel extrapolation, face limitations in acquiring CSI for all antenna ports and suffer from generalization problem caused by sequence-dependent model. In this paper, we propose a model-driven channel extrapolation algorithm based on the theoretical analysis of Jake's channel model. Diverging from existing DL-based approaches, our method builds on theoretical and empirical analyses of channel correlations in Jake's model, allowing for robust training with minimal observed ports and ensuring robust channel extrapolation across all ports on FA. Numerical experiments validate the effectiveness and robustness of our proposed algorithm, achieving a low normalized mean square error (NMSE) down to -30 dB in channel extrapolation with just 30 observed ports during training, even as the total number of ports reaches ten thousand.