Learning Beamforming in Cell-free Massive MIMO ISAC Systems

Abstract

Beamforming design is a critical enabler for integrated sensing and communication (ISAC) systems to achieve the full potential of the system. ISAC beamforming design in cell-free massive MIMO systems, compared to colocated MIMO systems, is further challenging due to the additional complexity of the problem with the large number of access points (APs). To tackle this problem, machine learning (ML) techniques can be utilized to design near-optimal beams with lower complexity, allowing instantaneous results with specialized ML hardware. In this paper, we first design a joint sensing and communication objective for cell-free massive MIMO, where the sum of the communication rates and logarithm of the sensing function is maximized. Then, we introduce machine learning for this problem. Specifically, we design a graph neural network (GNN) model to design joint sensing and communication beams in cell-free massive MIMO ISAC systems, where the model is a heterogenous GNN model for users, APs, and sensing target. Our results show that the proposed architecture can achieve near-optimal performance, and applies well to various network structures.

Topics & Concepts