Channel State Information Extrapolation in Fluid Antenna Systems Based on Masked Language Model

Fluid antenna systems introduce higher degrees of freedom for multiple-input multiple-output but face challenges in port selection. The lack of channel state information (CSI) makes it difficult to compute signal-to-interference plus noise ratio, which serves as a benchmark for communication performance. In this paper, we propose a CSI extrapolation approach based on deep learning and a masked language model. The proposed approach (PA) utilizes incomplete CSI and innovatively incorporates position information encoding to extrapolate complete CSI. PA achieves low normalized mean squared error and outage probability under highly incomplete CSI constraints, demonstrating effective port selection in the scenario with 5 user equipments.