Abstract
Singular value decomposition (SVD) is fundamental to channel decoupling and transmission optimization in massive multiple-input multiple-output (MIMO) systems. To reduce the computational complexity associated with conventional SVD, we propose an artificial intelligence (AI)-driven SVD method. The proposed method consists of three modules. A feature extraction module is first employed to learn channel features using different deep learning algorithms. Orthogonality refinement and implicit singular value calculation modules are then used for constraint enforcement and approximation, respectively. This modular design preserves mathematical fidelity compared with other baselines. The proposed convolutional neural network (CNN)-based algorithm together with the remaining modules offers the best trade-off between accuracy and efficiency.