Exploiting Structural Sparsity and Delay-Doppler Decoupling for Low-Complexity OTFS-ISAC Receivers

Abstract

In this work, the problems of channel estimation, radar sensing, and data detection are addressed for monostatic integrated sensing and communications (ISAC) applications within orthogonal time frequency space (OTFS) systems operating with a reduced cyclic prefix (RCP). Specifically, the delay-Doppler (DD) input-output relationship is formulated in a discrete representation that enables signal-independent disjoint parameter estimation by encapsulating fractional delay and Doppler effects through distinct, structurally sparse matrices. This exact algebraic separability is directly exploited to develop a low-complexity parameter estimation framework for the communication channel, which is seamlessly adapted for monostatic radar sensing on backscattered data frames. To enhance path detection robustly and safeguard estimation accuracy under low signal-to-noise ratio (SNR) regimes where traditional stopping criterionc(SC)-based methods fail, a deep learning (DL) architecture is integrated to perform model order selection via multi-class classification. Furthermore, a path-wise variant of the iterative Landweber method, designated as iterative matched filtering and combining (IMFC), is introduced for low-complexity data detection by leveraging the identical structural sparsity unlocked by the decoupled framework. Simulation results indicate the proposed estimation scheme achieves lower normalized mean squared error (NMSE) than conventional channel estimation algorithms and sensing performance close to the Cramer-Rao lower bound (CRLB). Finally, the IMFC equalizer is shown to deliver bit error rate (BER) performance comparable to the traditional linear minimum mean squared error (LMMSE) benchmark while dramatically reducing the computational load.