Deep Learning-Based Near-Field Wideband Channel Estimation: A Joint LISTA-CP Approach

Extremely large-scale multiple-input-multipleoutput (XL-MIMO) is listed as one of the key 6G candidate technologies due to its ability to exponentially increase the communication spectrum efficiency. However, wideband XLMIMO channel estimation faces a new challenge known as near-field beam split effect. To cope with this challenge, we propose a deep learning based joint learned iterative shrinkage thresholding algorithm with partial weight-coupling (Joint LISTA-CP) to accurately recover the channel. Specifically, by studying the characteristics of the near-field wideband channel, we employ a frequency-dependent polar domain dictionary to ensure that the sparse polar domain channel vectors across different subcarriers share a common sparse support. Then, we transform the near-field wideband channel estimation problem into a multiple measurement vector recovery problem. Utilizing the common sparsity structure of different subcarriers, we design the Joint LISTA-CP network that integrates multiple subnetworks. The proposed network employs a multidimensional shrinkage thresholding operator to guarantee the common sparse property of different subcarrier channels. In addition, we rigorously demonstrate that the Joint LISTA-CP network achieves lower recovery error. Simulation results show that the Joint LISTA-CP method enhances the estimation accuracy of the near-field wideband channels.