Deep Learning for Delay-Doppler Sensing and Echo Restoration in Urban Low-Altitude Wireless Networks
Year: 2025
Type: conference-paper
Abstract: This work tackles the issue of radar echo corruption caused by urban occlusions in low-altitude UAV surveillance scenarios. To address this issue, we propose a deep learning-based reconstruction framework built upon an improved UNet architecture. The network processes a Slow-Fast Time Echo Matrix (SFTEM), which captures both temporal structure and phase continuity. By incorporating Axial attention and residual connections, the model effectively restores radar echoes under severe pulse loss. A Partial-Edge Preserved Pulse Loss (PEPPL) model is introduced to simulate real-world occlusion patterns. Following signal restoration, range and velocity are jointly estimated through matched filtering and Doppler FFT. Experimental results demonstrate high robustness, achieving sub-meter range accuracy and a velocity error as low as 0.749,m/s, even under extreme occlusion conditions where over 90% of central pulses are missing.
Authors: Shuai Xu, Yunlong Cai
Institutions: Peng Cheng Laboratory, Zhejiang University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7167574523