High-Resolution Sensing in Communication-Centric ISAC: Deep Learning and Parametric Methods

Abstract

This paper introduces two novel algorithms designed to address the challenge of super-resolution sensing parameter estimation in bistatic configurations within communication-centric integrated sensing and communication (ISAC) systems. Our approach leverages the estimated channel state information derived from reference symbols originally intended for communication to achieve super-resolution sensing parameter estimation. The first algorithm, IFFT-C2VNN, employs complex-valued convolutional neural networks to estimate the parameters of different targets, achieving significant reductions in computational complexity compared to traditional methods. The second algorithm, PARAMING, utilizes a parametric method that capitalizes on the knowledge of the system model, including the transmit and receive array geometries, to extract the sensing parameters accurately. Through a comprehensive performance analysis, we demonstrate the effectiveness and robustness of both algorithms across a range of signal-to-noise ratios, underscoring their applicability in realistic ISAC scenarios.