CSI-based beamforming and localization for massive MIMO using deep learning

Abstract

In 6G technology, integrated sensing and communication (ISAC) is an emerging approach that enhances communication efficiency by enabling simultaneous sensing tasks. This paper leverages the multitasking capabilities of deep learning to optimize beamforming selection and user localization. A convolutional neural network (CNN) is employed to extract features from channel state information (CSI) data, which are then processed through a fully connected neural network to identify the optimal beamforming configuration and estimate user location. A weighted loss function is introduced to balance the importance of each task, ensuring that the model effectively prioritizes its objectives. Experimental results show that the proposed model achieves a Top-1 beamforming classification accuracy of up to 78.2% and a Top-3 accuracy of 99.21% with 64 antennas, while reducing localization error to as low as 2.11 meters. Compared to traditional single-task models, our approach improves classification accuracy by up to 7% and reduces localization error by up to 81%. This study highlights the potential of multitask learning in advancing ISAC capabilities and provides valuable insights for practical deployment in 6G systems.