Near-Field Sensing: A Low-Complexity Wavenumber-Domain Positioning Method

Abstract

A low-complexity wavenumber-domain positioning method is proposed for near-field sensing. Specifically, in the wavenumber domain, the power-concentrated region is sparse and closely related to the target's position. However, this relationship is complex and implicit. To address this, a bi-directional convolutional neural network (BiCNN) architecture is employed to capture the underlying relationship, enabling low-complexity, gridless target positioning. The simulation results reveal that the BiCNN method significantly reduces the computational complexity compared to the existing on-grid multiple signal classification (MUSIC) algorithm while achieving high accuracy.