Near-Field Beam Training with Analog Extremely Large True-Time-Delay Arrays

Abstract

As wireless systems have started to use millimeter-wave (mmWave) frequencies to exploit the abundant bandwidth, large antenna arrays with high beamforming gain have become necessary for compensating severe path loss. However, deploying large number of antennas extends the near-field (NF) region where effective beamforming requires the knowledge of not only the user equipment (UE)'s direction, but also distance. In this paper, we present a fast NF beam training method for analog extremely large (XL) true-time-delay (TTD) arrays using only a single orthogonal frequency-division multiplexing (OFDM) pilot symbol. In the proposed virtual sub-array sparse rainbow (VSSR) algorithm, we virtually partition the antenna array into smaller aperture sub-arrays so that a UE falls in the far-field region of each sub-array. By exploiting properties of the rainbow beamforming, we obtain near-orthogonal sub-array measurements such that UE directions with respect to each sub-array can be estimated with a sparse recovery algorithm. Then, these sub-array referenced UE directions are combined to estimate UE's location. Our simulation results demonstrate that the proposed method enables near-optimal beam training using only a single OFDM symbol, achieving an average of at least the 95% of the optimal array gain under various channel conditions.