Low-Latency and Secure Port Selection for Fluid Antenna Systems: A Hardware-AI Co-Design

Abstract

Fluid Antenna Systems (FAS) enable spatial microdiversity in compact user equipment through dynamic antenna port selection; however, their performance is fundamentally constrained by the latency of the port-selection logic. This paper proposes a low-latency, hardware-aware FAS port-selection architecture that integrates a Non-Uniform Carry Look-Ahead Adder (NCLA)-based comparison engine with a digitally controlled RF switching network. Compared with conventional ripplecarry adder (RCA) and uniform carry look-ahead adder (CLA) designs, the proposed NCLA reduces the critical-path delay by approximately 70 % and 15 %, respectively. When embedded in a 32-port FAS control loop, this reduction translates into an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 8}-\mathbf{2 5} \boldsymbol{\%}$</tex> improvement in effective achievable rate. Moreover, the ultra-low-latency NCLA enables rapid secrecy-aware port switching, yielding up to <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$30-40 \%$</tex> secrecy-rate gains against a passive eavesdropper compared with random or latency-limited selection schemes. To further reduce control overhead, an O-RAN-compatible machine learning (ML) xApp is developed to predict the optimal FAS port from physical-layer features. Experimental results show that lightweight models, particularly logistic regression, achieve up to 97 % port-selection accuracy with over <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0 \%}$</tex> lower inference latency than kernel-based approaches. The proposed architecture offers a practical, scalable, and securityaware solution for real-time FAS control in future <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{5 G} / \mathbf{6 G}$</tex> user equipment.