Frequency-Dependent Beamforming for Hybrid Near-Far Field Communications Through Joint Phase-Time Arrays

Joint phase-time arrays (JPTA) provide a low-cost and energy-efficient solution for enabling flexible frequency-dependent beams in wideband scenarios by incorporating both true-time delays (TTDs) and phase shifters. This paper explores the potential of JPTA with a single radio frequency chain to serve multiple users simultaneously in hybrid near- and far-fields. We concentrate on optimizing subband allocation and JPTA-based hybrid beamforming to maximize the total concave utility function associated with user rates. To achieve this, we propose an unsupervised deep learning (DL) approach. Our DL framework includes a two-layer convolutional neural network for feature extraction, followed by a three-layer graph attention network (GAT) and a normalization module for optimizing resource allocation and beamforming. The GAT effectively captures the interactions between resource allocation and analog beamformers. Simulation results demonstrate the superiority of JPTA over conventional phased arrays in terms of user rate when serving multiple users concurrently. Furthermore, adopting a logarithmic function of user rates as the utility function results in greater fairness than simply maximizing sum rates.