Performance of Machine Learning Aided Fluid Antenna System with Improved Spatial Correlation Model

Fluid antenna has emerged as a new antenna technology that enables software-controllable position reconfigurability for great diversity and multiplexing benefits. The performance of fluid antenna systems has recently been studied for single and multiuser environments adopting a generalized spatial correlation model that accounts for the channel correlation between the ports of the fluid antenna. The recent work [1] further devised machine learning algorithms to select the best port of fluid antenna in a more practical setting in which only a small number of ports is observable in the selection process, and found that extraordinary outage probability performance can be obtained. However, there is a concern of how the spatial correlation parameters are set to reflect the actual correlation structure for accurately evaluating the system performance. In this paper, the method in [2] is used to set the correlation parameter so that the model can accurately characterize the correlation amongst the ports of a fluid antenna in a given space. This paper revisits the port selection problem for single-user fluid antenna system where learning-based algorithms are employed to select the best port when only a small subset of the channel ports are known. The new results demonstrate that the impact of spatial correlation on the performance becomes more pronounced but the machine learning aided fluid antenna system is still able to match the performance of maximum ratio combining (MRC) system with many uncorrelated antennas.