Deep Learning-Assisted Target Classification Using OTFS Signaling

Abstract

A new modulation method, orthogonal time frequency space (OTFS), can support reliable data transmission by representing the signal in the delay-Doppler (DD) domain for high-mobility applications. In particular, the parameters of in-environment reflectors can be obtained from the representation of wireless channels in the DD domain, making it possible to provide sensing capability. In this paper, we propose a deep learning (DL) based target classification method using OTFS signaling. In our approach, to enhance the network performance, a 2D correlation method is utilized to extract features for data preprocessing. Subsequently, inspired by the residual learning technique, a deep neural network incorporating the attention mechanism is designed to distinguish sensing targets from the coarse estimation results. Through simulation experiments, we demonstrate that our proposed network exhibits superior performance in terms of efficiency and accuracy for OTFS sensing applications.