Deep Learning Based Near-Field Positioning in True-Time-Delay Array Systems

Abstract

In millimeter-wave (mmW) networks, large antenna arrays can be deployed to combat high signal attenuation, yet creating also near-field (NF) effects in the relative proximity of the antenna system. While utilizing frequency-selective rainbow beams enabled by true-time-delay (TTD) analog beamformer, we study the mmW network localization capabilities in the NF domain via deep learning neural networks. By leveraging the unique properties of the rainbow beams, we show that the proposed deep learning model, referred to as RaiNet, is capable of accurately positioning the user using a single channel response measurement. The provided numerical results at different carrier frequencies show that the proposed deep learning approach enables significant improvements in localization accuracy, compared to the state-of-the-art benchmark methods. The study thus paves the way for advanced localization techniques in 6G systems, contributing to the development of more efficient and intelligent future networks.