Deep Learning of Near Field Beam Focusing in Terahertz Wideband Massive MIMO Systems

Employing large antenna arrays and utilizing large bandwidth have the potential of bringing very high data rates to future wireless communication systems. To achieve that, however, new challenges associated with these systems need to be addressed. First, the large array aperture brings the communications to the near-field region, where the far-field assumptions no longer hold. Second, the analog-only (phase shifter based) beamforming architectures result in performance degradation in wideband systems due to their frequency unawareness. To address these problems, this paper proposes a low-complexity frequency-aware near-field beamforming framework for hybrid time-delay (TD) and phase-shifter (PS) based RF architectures. Specifically, a \textit{signal model inspired online learning} framework is proposed to learn the phase shifts of the quantized analog phase-shifters. Thanks to the model-inspired design, the proposed learning approach has fast convergence performance. Further, a low-complexity \textit{geometry-assisted} method is developed to configure the delay settings of the TD units. Simulation results highlight the efficacy of the proposed solution in achieving robust near-field beamforming performance for wideband large antenna array systems.