A Row-Sparse Tensor Decomposition-Based Approach for Fading Channel Prediction

Accurate real-time channel state information (CSI) is essential for adaptive wireless systems. However CSI degrades rapidly in high-mobility scenarios. Conventional tensor decomposition-based channel prediction methods often suffer from significant errors due to unified subspace projections that ignore structural heterogeneity induced by channel sparsity. To address this issue, we propose a two-stage framework that explicitly leverages temporal row-sparse in multiple-input multiple-output (MIMO) channel tensors. First, a constrained Alternating Direction Method of Multipliers (ADMM) decouples the channel into sparse (dominant paths) and diffused (scattered multipath) components. Then, Tucker decomposition is used to predict the joint subspace of the sparse part, while a spatio-temporal autoregressive model handles the diffused part. Experiments on real-world data demonstrate that the proposed method significantly improves prediction accuracy and enhances transmission robustness in dynamic environments by effectively modeling channel heterogeneity.