Gaussian dynamic recurrent unit for emitter classification

Abstract

Emitter classification plays a crucial role in electronic support measurement systems. A data-driven model named Gaussian dynamic recurrent unit is proposed to accomplish the end-to-end emitter classification tasks, which is simplified based on the long short-term memory unit by using a Gaussian function to characterize the relationship between the input and the network state, and a learnable exponentially weighted average to update the states. The Gaussian function and the dynamic exponentially weighted average amplify the difference between the input signals from different classes. The parameter size and computational time of many emitter classification methods are measured. The results show that the model proposed in this paper has the highest parameter efficiency, and low computational and storage costs. The results on a real-world dataset show that the proposed model has the highest accuracy and stability. These advantages have important significance for deploying the model in practical applications.