Deep Learning-Driven Time-Varying Fading Channel Estimation for Pinching Antenna Systems

The flexible deployment and dynamic characteristics of pinching antenna systems (PASS) make the channel estimation tricky. To address this issue, we propose a deep learning (DL)-based time-varying channel estimation method for PASS. More specifically, i) first, we present the pinching antenna (PA) position as a binary encoding to handle the data inconsistency; ii) second, we design a heterogeneous encoder, which maps the pilot signal and the PA position information into the high-dimensional space and performs feature extraction; iii) subsequently, we adopt the cross-attention mechanism to capture the dynamic connections between temporal and spatial information; iv) finally, we design a gated recurrent unit decoder to handle the channel spatio-temporal correlation and reconstructed mutability. Simulation results show that the proposed channel estimator can effectively track the channel state changes, achieve high adaptability for PASS with different numbers of antennas.