Data-Driven Mixed-Field Sensing in ISAC Systems for Joint Classification and Estimation

Abstract

With the deployment of large-scale antenna arrays and high-frequency carriers, future integrated sensing and communication (ISAC) near-field and far-field users may coexist around the base station (BS), where different propagation fields lead to different channel structures and spatial parameters. If the propagation field is not properly identified, the BS may adopt a mismatched sensing model, resulting in degraded parameter estimation and unreliable link establishment. However, conventional mixed-field sensing methods usually rely on predefined steering-vector dictionaries and exhaustive angle or angle-distance search, which incurs high online complexity and may suffer from channel estimation mismatch. To address this issue, this paper proposes a data-driven mixed-field sensing framework for ISAC systems, where the estimated channel state information is used to jointly perform field classification and spatial parameter estimation. Specifically, a squeeze-and-excitation enhanced convolutional neural network (SE-CNN) is developed to extract field-dependent spatial features from the estimated channel. The proposed network simultaneously determines whether the user is located in the near-field or far-field region and estimates the corresponding spatial parameters, including angle and distance for near-field users and angle for far-field users. Simulation results show that the proposed SE-CNN achieves accurate field classification and effective angle/range estimation under different signal-to-noise ratio (SNR) levels, while avoiding exhaustive near-field/far-field dictionary search.