Radar Signal Recognition Based on Multi-Task Learning

Abstract

Radar signal recognition is an important topic in electronic countermeasures. However, with the growing complexity of the electromagnetic environment, accurately identifying radar signals faces great challenges, and insufficient feature extraction is a core factor leading to this issue. In this paper, we proposed a novel method for radar signal recognition based on multi-task learning (MTL) to tackle with the feature extraction problem. The method combines feature enhancement and signal recognition tasks for joint learning, improving model performance by comprehensive utilization of their correlation and feature sharing. Specifically, we adopt a feature enhancement network based on an autoencoder framework to enhance time-frequency features of radar signals. Then the learned representations are used to achieve signal classification with a deep residual network. Finally, this model, as a collaborative optimization algorithm, is end-to-end trained with interactive constrains using our designed loss function. Extensive experiments, including performance comparison, ablation experiments, recognition performance of multi-component radar signals, and hardware-in-the-loop simulation experiment are conducted to validate the effectiveness of the proposed method in different scenarios.

Keywords

• Deep learning
• multi-task learning (MTL)
• radar signal recognition
• feature enhancement
• loss function