Deep Learning Assisted Radar Jamming Detection from Target Returns for Joint Radar Communication Systems

Providing the coexistence of communications systems with spectrum sharing among multiple radio systems in the integrated sensing and communications (ISAC) systems, the radar is likely to be subject to self-interference with local and cooperative joint radar and communication (JRC) signals, bringing challenges in jamming detections among complicated electromagnetic environments. The correct awareness of jamming existence facilitates interference and spectrum management for maintaining the radar system’s performance against malicious attacks. This paper proposes a dual-channel deep learning (DL) assisted jamming detector method aligned with the existing pulse compression methods, where the target returns and the filtered feature representatives are both fed to the DL structure. The proposed method processes in the time domain and eliminates the need for extra frequency transforms, hereby showing high feasibility to most common radar systems. The presented DL structure consists of convolutional layers for acquiring spatial feature information followed by a recurrent neural network structure, i.e. long short-term memory (LSTM) to obtain time-sequential information related to time-varying noise changes. By demonstrating with a linear frequency modulated (LFM) pulse waveform radar simulator and smart jamming noise, the proposed method has excelled in detecting the smart jamming noise at a detection sensitivity of 1 dB. The simulation also suggests a 2 dB sensitivity improvement when the jamming amplitude doubles.