HFRNet: An Efficient Architecture Based on Hierarchical Feature Retention for Radar Signal Modulation Recognition
Caiyu Liang, Weibo Huo, Yin Zhang, Ruobing Tang, Jifang Pei, Yulin Huang · IEEE Transactions on Aerospace and Electronic Systems · 2025
Automatic modulation recognition (AMR) of radar signals plays a crucial role in electronic reconnaissance. In the complex and rapidly changing electromagnetic environment, efficiently recognizing radar signals remains a challenging task. To achieve accurate and timely modulation recognition, we propose the Hierarchical Feature Retention Network (HFRNet)——an efficient recognition network architecture based on the time-frequency characteristics of radar signals. The core principle of HFRNet is to progressively retain signal features at intermediate network stages, enabling a more efficient mapping from raw signals to modulation types. To facilitate effective feature extraction, we design two fundamental modules: the time-frequency characteristics retention module (TFRM), which preserves signal characteristics through progressive filtering along the time and frequency dimensions; and the adaptive denoising operation (ADO), which adaptively filters out noise according to the signal-to-noise ratio (SNR). The integration of these modules ensures that essential signal characteristics are retained throughout the network's forward propagation, endowing HFRNet with robust recognition capabilities. Experiments demonstrate that HFRNet achieves superior recognition performance, with more than 75% fewer parameters and over 82% reduced computational complexity compared to state-of-the-art methods, significantly enhancing the efficiency of radar signal modulation recognition.