A Hardware-Efficient Hybrid Beamforming Architecture for Near-Field Beam Training in 6G Wideband XL-MIMO Systems

Near-field wideband beam training is a key bottleneck in millimeter-wave (mmWave) and terahertz (THz) XL-MIMO systems, because channel acquisition depends jointly on angle and range, while the wide bandwidth introduces frequency-dependent beam split. Notably, such frequency-dependent beam splitting can also be exploited to enable fast near-field beam training. Existing distance-dependent beam-split (DDBS) training can exploit subcarriers to search multiple angles and distances simultaneously in a single time slot; however, it typically relies on employing true-time-delay (TTD) hardware with every antenna element, which is costly and power-consuming. This paper proposes a hardware-efficient array-of-sub-arrays (AoSA)-based true-time-delay/phase-shifter (TTD-PS) architecture, where one TTD is shared within each antenna sub-array (one TTD per sub-array). We characterize how delay sharing reshapes DDBS behavior by restricting the strong-gain region to a limited angular range/sector. We adopt practical 3-dB coverage rules that determine the required number of pilots to cover the entire angular range and distance domain, yielding an analytical expression for the training overhead. Building on these results, we develop a sector-shifted and distance-interleaved training procedure with least-squares (LS)-based focusing refinement. We also propose an off-grid match-filter method to improve robustness under noise and grid mismatch. It is analytically shown that under the AoSA-based TTD-PS architecture, a few pilots (e.g., three pilots, as shown in prior per-antenna TTD work for per-antenna TTD hardware) are not sufficient to cover the whole angle-distance space for wideband XL-MIMO. Simulations show that with a single TTD shared by two antennas, the proposed design achieves a favorable tradeoff, halving the number of TTD elements while approaching the performance of the ungrouped architecture with about 12 pilots.