Effective Beam Training for Near-Field Wideband Terahertz Extremely Large-Scale Arrays

Abstract

Extremely large-scale array (XL-array) is envisioned as a key technology for future terahertz (THz) communication networks. However, the near-field beam split effect severely degrades beam training performance in wideband THz XL-array systems. In this letter, we reveal that the near-field array gain is considerably more sensitive to angle mismatch than to distance mismatch, which motivates an angle-priority two-stage beam training scheme. Specifically, angle beams generated by the controllable beam split effect are first exploited for angle search within partitioned angular regions, followed by distance refinement via controllable distance beams. Moreover, a central sub-array design is introduced to improve angle estimation robustness for nearby users. Numerical results demonstrate that the proposed scheme achieves near-optimal rate performance with significantly reduced training overhead.