Deep Unfolding-Based Non-Stationary Channel Estimation for Extremely Large-Scale Array

Extreme large-scale arrays (XL-arrays) is deemed as a promising technology to improve the spectrum efficiency for future wireless systems. However, for XL-arrays enabled wireless systems, users in general locate in the near-field regions, which results in a different channel model with the emergence of spatial non-stationarity (SnS), thus rendering the channel estimation more challenging as compared to that of the conventional far-field communications. This paper studies the channel estimation problem for millimeter-wave (mmWave) near-field communications. To address these challenges, this paper proposes a Dictionary Learning Iterative Shrinkage and Thresholding Algorithm (DLISTA), which formulates the sparse estimation problem of near-field non-stationary channels estimation problem as a sparse dictionary learning problem. The numerical results show that the proposed DLISTA outperforms various benchmark schemes.