A Graph-based Hybrid Beamforming Framework for MIMO Cell-Free ISAC Networks

This paper develops a graph-based hybrid beamforming framework for multiple-input multiple-output (MIMO) cell-free integrated sensing and communication (ISAC) networks. Specifically, we construct a novel MIMO cell-free ISAC network model. In this model, multiple dual-function base station (BS) transmitters employ distributed hybrid beamforming to enable simultaneous communication and sensing, while maintaining physical separation between the transmitters and the radar receiver. Building on this model, we formulate a multi-objective optimization problem under a power constraint to jointly improve communication and sensing performance. To solve it, the problem is first reformulated as a single-objective optimization problem. Then, a graph-based method composed of multiple graph neural networks (GNNs) is developed to realize hybrid beamforming with either perfect or imperfect channel state information. Once trained, the neural network model can be deployed distributively across BSs, enabling fast and efficient inference. To further reduce inference latency, a custom field-programmable gate array (FPGA)-based accelerator is developed. Numerical simulations validate the communication and sensing capabilities of the proposed optimization approach, while experimental evaluations demonstrate remarkable performance gains of FPGA-based acceleration in GNN inference.