LSTM-based variational autoencoder for radar interference recognition

In radar jamming, Digital Radio Frequency Memory (DRFM) systems dynamically generate interferences according to the radar’s transmission waveform. The expansive parameter space of these interferences often results in poor identification performance by conventional recognition networks. To address the above challenge, this paper proposes a novel interference recognition approach that integrates the Long Short-Term Memory (LSTM) network with Variational Autoencoder (VAE). Firstly, LSTM-based VAE is used to encode both the transmitted waveforms and the received signals, including interferences, to extract feature vectors. These vectors are then merged to form a tensor and then, a convolutional neural network (CNN) is employed to perform the interference recognition based on the tensor. Since the CNN’s input is a fusion of feature vectors from both the transmitted waveform and interference, the proposed approach can effectively leverage the consistent relationship between these feature vectors. Simulation results show that our proposed model achieves a 6.56% improvement in recognition accuracy compared to the CNN-based model, under the parameters we have set.