Recognition-Guided UAV micro-Doppler Feature Extraction: A Real-World ISAC Validation

Abstract

Unmanned aerial vehicle (UAV) rotor micro-Doppler signatures are important for target sensing and recognition in integrated sensing and communication (ISAC) systems. However, in practical environments, these signatures are often severely corrupted by strong dynamic and static interference, resulting in observations with extremely low signal-to-interferenceplus-noise ratio (SINR). To address this issue, this paper proposes a recognition-guided denoising framework, termed RG-SOCA-DTNet, for UAV rotor micro-Doppler feature extraction. The proposed method operates directly on the 1D complex slow-time signal and combines nonlinear feature modeling with temporal modulation modeling in a two-stage training framework, where an auxiliary classifier provides recognition-guided supervision during denoiser training. Simulation results show that, with an average input SINR of −29.96 dB, the proposed method improves the average output SINR to 6.72 dB and outperforms the compared baselines. Real-world experiments on a sub-6 GHz ISAC hardware platform further validate the effectiveness of the proposed method under practical low-SINR conditions.