Multi-User Hybrid Far- and Near-Field THz UM-MIMO Channel Estimation via UNet-OAMP

his letter studies hybrid far- and near-field channel estimation for terahertz (THz)-band ultra-massive multiple-input multiple-output (UM-MIMO) systems in the presence of multi-user interference caused by non-orthogonal pilots. Specifically, we propose a two-stage model-driven estimator that modifies both the linear and nonlinear components of the traditional orthogonal approximate message passing (OAMP) algorithm by incorporating U-Net convolutional neural networks following the fixed-point principle for the nonlinear steps and by updating the decorrelated linear estimator matrix based on the channel correlation statistics. Simulation results verify that the proposed algorithm effectively mitigates interference from users with non-orthogonal pilots and still outperforms baseline algorithms by at least 1 dB, in terms of channel-estimation accuracy.