Radar Signal Sorting Using Combined Residual and Recurrent Neural Network (CRRNN)

Abstract

Due to the density of the crowded electromagnetic environment nowadays and the complexity of modern radar signals, the performance of pulse repetition interval (PRI)-based sorting systems experience more deterioration than ever before. Such systems are considered unreliable when working in crowded circumstances, moreover, they require a long pulse stream and high signal-to-noise (SNR) ratio, which makes obtaining acceptable sorting accuracy a difficult task. In this paper, a new machine learning architecture, Combined Residual and Recurrent Neural Network (CRRNN), is proposed, where recurrent neural network (RNN) and residual neural network (ResNet) are incorporated to create an architecture which can be used to overcome the above-mentioned shortcomings of conventional sorting methods achieving more accuracy and stability. Separate ResNet and RNN models are investigated as well for comparison. Simulations are performed after discussion of the structure and the principle of work of each network architecture. Statistical results showing the high and reliable performance of the proposed method in different conditions are presented and discussed.