Fast Port Selection using Temporal and Spatial Correlation for Fluid Antenna Systems

Fluid antenna system (FAS) is a flexible antenna structure that obtains tremendous space diversity by allowing the antenna to change its position (or port) in a given space. The extraordinary performance requires FAS to always switch to the port with the largest signal-to-noise ratio (SNR) from the large number of ports. In practice, however, this means that a large number of channel observations are required and the overhead could outweigh the benefits. In this paper, we exploit the spatial and temporal correlation of the port channels using a machine learning approach. The proposed algorithm first estimates all the port channels in space from a small number of observations, then predicts the port channels in the subsequent time slots. Re-observations are used to reduce error propagation in long short-term memory (LSTM) rolling window regression. Simulation results demonstrate that the proposed algorithm can achieve promising performance with few re-observations in high-mobility scenarios.