Model-Based Deep Unfolding for Efficient Zero-Padded OTFS Channel Estimation

Wireless communication systems in high-speed mobile environments remain a current research focus. Orthogonal Time Frequency and Space (OTFS) modulation, an innovative technology, meets the high-Doppler needs of future mobile systems. Zero-Padded OTFS (ZP-OTFS), a variant of OTFS, involves having multiple rows with zero values along the delay axis. Nonetheless, to achieve peak performance, the ZP-OTFS system necessitates precise channel information. This paper proposes an efficient model-based deep unfolding Channel Estimation (CE) scheme for zero-padded orthogonal time frequency space systems. Specifically, we formulate the ZP-OTFS CE task as a two-dimensional Compressed Sensing (CS) problem, leveraging the sparsity of the OTFS delay-Doppler domain channel. To solve this, we propose an enhanced CE method using deep unfolding Iterative Shrinkage Thresholding Algorithm (CE-DU-ISTA), which addresses the shortcomings of traditional CS methods that often require excessive iterations and computational time. We provide theoretical insights into CE-DU-ISTA, including its convergence rate. Additionally, we derive theoretical generalization error bounds for the CE-DU-ISTA network using Rademacher complexity. Numerical results demonstrate that our proposed algorithm significantly outperforms conventional benchmark methods while exhibiting superior generalization capabilities, a hallmark of model-based approaches.