Model-Driven Deep Learning-Aided Wideband Hybrid-Field THz UM-MIMO Channel Estimation (CCF C)

To efficiently implement Terahertz (THz) communications in the 6G era, the ultra-massive multiple-input multiple-output (UM-MIMO) technique is considered an essential building block. However, effective wideband THz UM-MIMO transmissions can never be achieved without pilot-inexpensive yet accurate channel estimation (CE) methods. In this article, we investigate the wideband THz UM-MIMO CE problem, accounting for the hybrid near- and far-field propagation characteristics, molecular absorption, and multi-path reflection. The CE problem is reformulated into a compressed sensing-aided counterpart, leveraging the inherent sparsity of THz UM-MIMO channels to reduce pilot overhead. We harness the power of model-driven deep learning and propose a deep unfolding (DU)-aided Bayesian learning (DUBL) CE algorithm. We tailor the structure of the deep neural network (DNN)-based unfolded expectation-maximization (EM) iteration, aiming to achieve efficient DUBL training performance. Simulation results demonstrate that the DUBL solution can offer substantial THz UM-MIMO CE gains over the considered representative benchmarks.