Deep Delay-Parameter-Aware Radar Perception for Urban Low-Altitude Wireless Networks

Accurate delay parameter estimation is crucial for unmanned aerial vehicle (UAV) monitoring in urban low-altitude wireless networks (LAWN), where radar signals often suffer from building-induced occlusion and noise. We introduce DTFIRNet, a deep time-frequency inverse-reconstruction network designed to restore corrupted linear-frequency-modulated (LFM) radar echoes and precisely estimate target range based on delay parameters. DTFIRNet integrates an enhanced residual UNet with a lightweight spatio-temporal attention mechanism and a learnable inverse short-time Fourier transform (iSTFT) decoder, guided by time-frequency representations (TFRs) derived from STFT. Extensive experiments on synthetically corrupted LFM datasets demonstrate that DTFIRNet effectively recovers waveform details and achieves an average delay-based ranging error of only 0.61 meters, significantly outperforming existing TFR- or waveform-based methods.