Deep Learning-Assisted ISAC for UAV Detection in Multipath Environments With Micro-Doppler

This letter addresses the challenge of high false alarms in integrated sensing and communications (ISAC) for detecting multiple Unmanned Aerial Vehicles (UAVs) in multipath environments. We propose a novel deep learning-based framework that leverages unique micro-Doppler signatures of UAVs within the same category, using OFDM as the ISAC waveform. By identifying unique signatures from micro-Doppler spectrograms using deep learning, sensing multipaths reflected from the same UAV can be associated via clustering. Simulation results demonstrate an 86% reduction in false alarm rates, with slight miss detection growth and comparable localization error performance to conventional solutions.