An Integrated Architecture for Robust End-to-End Radar Multitarget Detection in Jamming Environments

Abstract

When the radar suffers from the repeater deception jamming, the echoes of each target and the jamming signals will produce serious aliasing in time, frequency, and spatial domains; the real targets will be masked and the detection performance is extremely limited. To address this issue and ensure the target detection capability of the radar system in this complex jamming scenario, we propose an integrated end-to-end anti-jamming target detection architecture based on multiscale refinement and adaptive modulation (MRAM-net), which realizes the end-to-end processing from the input of the original radar echo to the output of multitarget position information when the type of jamming is unknown and the number of targets is uncertain. This integrated architecture departs from the conventional multistage serial processing paradigm, thereby mitigating error accumulation inherent in sequential processing and significantly reducing the complexity and implementation overhead of the radar signal processing system. First, we design the selective multiscale refinement block (SMRB), which effectively improves the target feature representation capability and suppresses redundant information. Furthermore, we propose a spectrogram-adaptive-modulated global aggregation (SAMGA) module, which utilizes time–frequency features as global prior information to guide and modulate the feature extraction process. Finally, we also propose a model training optimization strategy termed bidirectional spreading learning (BSL), which avoids the model falling into local optima and further improves the accuracy of the model by bidirectional spreading training. Simulation results demonstrate that MRAM-net achieves outstanding end-to-end detection performance and generalization ability across diverse jamming scenarios and varying numbers of targets.