频道(广播)
多普勒效应
算法
电子工程
计算机科学
信噪比(成像)
信号处理
物理
时频分析
噪音(视频)
估计理论
声学
控制理论(社会学)
数学
拓扑(电路)
探测理论
作者
Fei Duan,Xilong Liu,Jun Chang,Guan Gui
出处
期刊:IEEE Transactions on Vehicular Technology
[Institute of Electrical and Electronics Engineers]
日期:2026-01-01
卷期号:: 1-6
标识
DOI:10.1109/tvt.2026.3736276
摘要
This paper addresses the critical challenge of Orthogonal Time-Frequency Space (OTFS) channel estimation in high-mobility scenarios where fractional Doppler effects are prominent. This phenomenon degrades the channel's inherent sparsity into a complex structural sparsity, a challenge that highlights the dilemma between the limited accuracy of conventional methods and the lack of interpretability in pure data-driven models. To resolve this dilemma, we propose a physics-aware deep unfolding methodology based on the Fast Iterative Shrinkage-Thresholding Algorithm (FISTA), presented as a two-stage innovation. First, our baseline model, FUNet (FISTA with U-Net), embeds a powerful U-Net based denoiser designed to explicitly learn the patterns of structural sparsity. Second, our ultimate model, FUNet-PA (FUNet-Physics Aware), further enhances this denoiser by integrating the novel Physics-Aware Adaptive Fusion Module (PAAFM) to resolve the mismatch with the channel's multiscale anisotropic physics. The simulation results validate that our proposed framework provides substantial gains in terms of accuracy and robustness of the estimation. Specifically, at a Signal-to-Noise Ratio (SNR) of 30 dB, our baseline FUNet improves the Normalized Mean Square Error (NMSE) by approximately 8 dB over competitive deep learning approaches, while the integration of PAAFM in FUNet-PA yields a further approximately 2 dB enhancement.
科研通智能强力驱动
Strongly Powered by AbleSci AI