MFS: A Motion Feature Separation Model for UAV Detection Under Passive Radar

With unmanned aerial vehicle (UAV) applications in urban areas, reliable UAV detection becomes increasingly demanding to avoid air collisions and privacy disclosure. Passive radar methods could extract UAV echo features from the communication channel information when UAVs penetrate the wireless link and cause channel disturbance. However, weak UAV echoes and severe multi-path interference induce extremely low signal-to-interference-and-noise-ratio (SINR), and moving scatterers like vehicles also confuse UAV detection. This article proposes a motion feature separation (MFS) model to address these problems by analyzing the separability mechanism of UAV echoes. The MFS model uses long term evolution (LTE) as opportunity illuminators and requires no reference signal to suppress the interference and detect UAVs. In addition, by modeling delay-Doppler trajectories in the proposed model, UAVs can be distinguished from other moving scatterers. In the MFS model, a static interference canceller mitigates the main component of static interference via a U-Net based network. Then, a dynamic feature enhancer extracts motion features of both UAV echoes and moving scatterer echoes via a 3D-convolution and long-short-term memory (LSTM) network. Finally, a semantic delay-Doppler trajectory recognizer distinguishes UAV echoes from the other moving scatterer echoes by using an attention-aided LSTM network. The qualitative and quantitative experimental results verify that the MFS achieves a UAV detection accuracy of 92.31% at an SINR of -40 dB in urban, significantly higher than mainstream methods.