RINN: A Radar-Informed Neural Network for Jamming Identification in Intrapulse Frequency Coding Radars Using Time-Frequency Discontinuity

Abstract

Interrupted sampling repeater jamming (ISRJ) and its variants pose a severe threat to intrapulse frequency coding radars, as effective countermeasures depend critically on accurate jamming signal identification. However, current jamming identification methods still face challenges, such as the insufficient generalization capability of handcrafted feature-based approaches and the fact that deep learning (DL) methods are predominantly applied to single-pulse systems and suffer from poor interpretability. To address these challenges, this article proposes a radar-informed neural network (RINN) that exploits the time–frequency discontinuity characteristics of ISRJ. First, RINN employs a hierarchical feature fusion architecture to extract the intrinsic time–frequency features of subpulses and capture the correlations among them, along with a binary mask mechanism to accommodate a variable number of subpulses. Second, an adaptive multi-information constrained loss function is designed, which incorporates the peak-to-sidelobe ratio (PSLR) similarity after pulse compression as a radar-informed constraint into the learning process, guiding the network to adhere to the principles of radar signal physics. Extensive experiments on a simulated dataset (five jamming types) and a hardware-in-the-loop test set demonstrate that RINN surpasses existing traditional and DL-based methods in accuracy, generalization, and robustness, while maintaining a lightweight design. Attention visualization analysis further confirms that the PSLR constraint effectively focuses the network’s attention on jamming-induced time–frequency regions rather than noise background, substantially improving the reliability and trustworthiness of the model.