Toward Practical Fluid Antenna Systems: Co-Optimizing Hardware and Software for Port Selection and Beamforming

TL;DR

This paper presents a hardware-software co-design approach for fluid antenna systems, integrating graph neural networks and FPGA acceleration to optimize beamforming and port selection with high efficiency and low latency.

Contribution

It introduces a novel GNN-RPS method for joint beamforming and port selection optimization, along with an FPGA-based accelerator for real-time inference in fluid antenna systems.

Findings

GNN-RPS achieves competitive communication performance.

FPGA accelerator maintains low latency for beamforming inference.

Scheduling algorithm reduces redundant computations.

Abstract

This paper proposes a hardware-software co-design approach to efficiently optimize beamforming and port selection in fluid antenna systems (FASs). To begin with, a fluid-antenna (FA)-enabled downlink multi-cell multiple-input multiple-output (MIMO) network is modeled, and a weighted sum-rate (WSR) maximization problem is formulated. Second, a method that integrates graph neural networks (GNNs) with random port selection (RPS) is proposed to jointly optimize beamforming and port selection, while also assessing the benefits and limitations of random selection. Third, an instruction-driven deep learning accelerator based on a field-programmable gate array (FPGA) is developed to minimize inference latency. To further enhance efficiency, a scheduling algorithm is introduced to reduce redundant computations and minimize the idle time of computing cores. Simulation results demonstrate that the proposed GNN-RPS approach achieves competitive communication performance. Furthermore, experimental evaluations indicate that the FPGA-based accelerator maintains low latency while simultaneously executing beamforming inference for multiple port selections.