Deep Unfolding-Based Near-Field Channel Estimation for 6G Communications

Extremely large-scale multiple-input-multiple-output (XL-MIMO) technology can significantly enhance array gain and spectral efficiency, and is considered a potential key technology for 6G communications. Additionally, artificial intelligence (AI), with its strong learning capabilities, supports sophisticated communication systems, particularly in large-scale antenna systems. To tackle the high complexity of traditional near-field XL-MIMO channel estimation, we propose an efficient deep learning-based method known asthe dictionary learning-based iterative shrinkage thresholding algorithm (DLISTA). The DLISTA integrates the near-field dictionary as a learnable parameter in the training process of the network, achieving a joint design of the dictionary and channel estimation algorithm. Furthermore, a new loss function is proposed for the DLISTA, which not only measures the error between the estimated and actual channels, but also reinforces the row-sparsity characteristics of the sparse channels generated by the network. Simulation results demonstrate that the DLISTA method achieves improved channel estimation performance.