AI/ML-driven beamforming with imperfect channel state information for multi-user MIMO transmission

This paper presents a novel machine learning (ML)-driven beamforming method for multi-user multiple-input multiple-output (MU-MIMO) transmission under imperfect channel state information (CSI). Traditional ML-based beamformers, typically employing fully connected layers, require extensive training across numerous user permutations, leading to scalability issues as the number of users increases. To overcome this limitation, we introduce a transformer neural network (TNN)-based approach, which demonstrates robust performance under imperfect CSI while scaling more effectively with user count. Comparative analysis against a message-passing graph neural network (MPGNN) architecture further highlights TNN’s superior scalability. These findings position TNNs as a promising solution for efficient and scalable beamforming in MU-MIMO systems with imperfect CSI.