Physically Consistent High-Frequency Near-Field Modeling for Channel Estimation and Localization

Abstract

High-frequency wireless systems promise high communication data rates and precise sensing capabilities but face major challenges in channel modeling and signal propagation. For instance, to counter severe terahertz (THz)-band propagation losses from spreading loss, molecular absorption, and blockage, large and densely packed antenna arrays are used, where mutual coupling and near-field effects become significant. This work establishes a Maxwell-based analysis of the near-field–far-field transition and develops a mixture-gamma-based channel model for reconfigurable intelligent surface (RIS)-aided multiple-input multiple-output (MIMO) THz links that accounts for channel sparsity, absorption, mutual coupling, spatial correlation, and spherical-wave behavior in the near field. Building on these accurate models, and subject to computational constraints imposed by increasing array size and the promised terabit-per-second data rates, we propose low-complexity signal processing algorithms that substantially improve channel estimation and localization accuracy. In particular, reduced-subspace least-squares estimation is strengthened by incorporating mutual coupling information in a generalized RIS–MIMO near-field setting, and an array-of-subarrays architecture with a machine-learning-based refinement method achieves high-accuracy near-field localization while mitigating errors under coherent sources. These contributions provide a cohesive framework for physically consistent modeling and efficient signal processing in high-frequency near-field systems.