Radar Target Detection Based On OTFS Signaling: A Deep Learning Approach

Orthogonal time-frequency space (OTFS) offers significant advantages among various emerging modulation techniques by modulating the information in the delay-Doppler (DD) domain, making it promising for both communications and radar sensing applications. However, in multi-target detection scenarios, the masking effect poses challenges and diminishes the effectiveness of traditional constant false alarm rate (CFAR) detectors. Addressing this challenge and leveraging the capabilities of OTFS signaling, this paper proposes a novel deep-learning approach for radar target detection via OTFS. Our method involves preprocessing received symbols using two-dimensional cross-correlations, which are then adapted to the input of a neural network. Leveraging previously established learning weights, the deep neural network identifies potential target regions and assesses the likelihood of target presence, effectively controlling false alarm probabilities through adaptive thresholds. Through simulation experiments, we demonstrate the superiority of the method over traditional methods, particularly in scenarios with multiple targets and unknown signal-to-noise ratios (SNRs).