LPI Radar Signal Recognition Using Deep-learning Neural Network

Currently, radar equipment uses Low Probability Intercepted (LPI) signals. Meanwhile, modulated radar signal is one of the important information in electronic reconnaissance, allowing the identification of the emission source. In order to improve the recognition of LPI radar signals, convolutional deep learning neural networks (CNN) are proposed in this paper. Specifically, the proposed CNN model is surveyed with different channel numbers and filter sizes. Survey results show that, the higher the parameter, the higher the identification accuracy; however, the slower the execution time. Therefore, it is necessary to select a network of a suitable size to achieve the required accuracy with the allowed execution time. In addition, preprocessing techniques also play an important role in enhancing the identity accuracy of the CNN network. Therefore, two techniques of STFT and WVD were explored. Although WVD offers higher recognition accuracy, the results show that it has a slower processing time than STFT.