Wideband Near-Field Channel Covariance Estimation for XL-MIMO Systems in the Face of Beam Split

Extremely large-scale multiple-input-multiple-output (XL-MIMO) techniques are capable of boosting the throughput of next 6G networks. However, due to the large number of antennas in the wideband XL-MIMO systems, they suffer from the so-called near-field beam split effect, where beams at different frequencies are focused on different locations. This effect imposes a severe performance loss on near-field channel estimation. To address this issue, in this paper, we investigate the wideband near-field channel covariance estimation problem of XL-MIMO systems. Explicitly, we first propose a sophisticated polar-domain dictionary for the wideband near-field channel at different frequencies. Specifically, the proposed polar-domain dictionary consists of several sub-matrices. For each sub-matrix, the angle is sampled from a unique range while the distance is sampled from the same range. Meanwhile, considering the beam split effect, we construct the corresponding dictionaries for different frequencies that can beneficially adjust the grid of the sampled angles and distances. Then we propose an algorithm termed as the correlation coefficient sparsity adaptive matching pursuit relying on parameter refinement (CCSAMP-PR) to reconstruct the near-field channel without any prior knowledge of the channel sparsity. More explicitly, an iterative parameter refinement step is conceived for gradually refining the angles and distances per iteration. Furthermore, the proposed algorithm is capable of adaptively adjusting its step size for ensuring rapid convergence. Finally, numerical results are provided for demonstrating that both our proposed polar-domain dictionary and CCSAMP-PR algorithm outperform the family of existing schemes.