引文格式：郭鹏，牛知艺.基于时频交叉门控多尺度网络的雷达信号调制识别算法[J].西安邮电大学学报，2026,31(3):1-10.
GUO Peng, NIU Zhiyi. Radar signal modulation recognition algorithm based on the spectral-temporal cross-gated multi-scale network[J]. Journal of Xi'an University of Posts and Telecommunications, 2026, 31(3): 1-10.
基于时频交叉门控多尺度网络的雷达信号调制识别算法
郭 鹏1,2，牛知艺2,3
（1.中国电子科技集团公司第十研究所，四川 成都 610036；2.成都华力创通科技有限公司，四川 成都 610095；3.中国民用航空飞行学院，四川 成都 641419）
摘要：针对主流智能化雷达信号调制识别算法处理长序列雷达信号时，存在物理特征去耦合引发相关性丢失、降采样导致相位细节特征缺失的问题，提出一种基于时频交叉门控多尺度网络（Spectral-Temporal Cross-Gated Multi-Scale Network，STCG-MSNet）的雷达信号调制识别算法。该算法基于海森堡-加博尔测不准原理设计跨域门控机制，利用时频域互补性生成门控掩码，实现特征自适应去噪与增强；采用金字塔空洞卷积，在保持全时序分辨率的同时指数级扩大感受野，避免池化破坏相位突变；引入双统计量融合头，结合全局平均池化与全局最大池化提取完整信号特征。实验结果表明，该算法在低信噪比及复杂混合调制信号识别中显著优于MobileNetV4、CLDNN（Convolutiond，Long Short-Term Memory，Full Connected Deep Neural Networks）等主流算法，−10dB极端低信噪比下识别率达88.7%，0dB以上准确率接近100%，具备更强的理论完备性和抗噪鲁棒性。
关键词：雷达调制识别；低截获概率；时频交叉门控；测不准原理；多尺度网络
中图分类号：TP391 文献标志码：A
文章编号：2095-6533(2026)03-0001-10
Radar signal modulation recognition algorithm based on the spectral-temporal cross-gated multi-scale network
GUO Peng1,2, NIU Zhiyi2,3
(1. The Tenth Research Institute of China Electronics Technology Group Corporation, Chengdu 610036, China; 2. Cheng Du Hwa Create Co., Ltd, Chengdu 610095, China; 3. Civil Aviation Flight University of China; Chengdu 641419, China)
Abstract:Aiming at the problems of correlation loss caused by physical feature decoupling, and loss of phase detail features induced by downsampling when mainstream intelligent radar signal modulation recognition algorithms process long-sequence radar signals, a radar signal modulation recognition algorithm based on the spectral-temporal cross-gated multi-scale network (STCG-MSNet) is proposed. A cross-domain gating mechanism based on the Heisenberg-Gabor uncertainty principle is designed. The complementarity of time and frequency domains is adopted to generate gating masks, achieving adaptive feature denoising and enhancement. Pyramid dilated convolution is used to exponentially expand the receptive field while preserving full temporal resolution, avoiding phase mutation damage caused by pooling. A dual-statistic fusion head is introduced, which combines global average pooling and global max pooling to extract comprehensive signal features. Experimental results show that the algorithm significantly outperforms mainstream algorithms such as MobileNetV4 and CLDNN (Convolutiond, Long Short-Term Memory, Full Connected Deep Neural Networks) in recognizing low signal-to-noise ratio signals and complex mixed modulation signals. It achieves a recognition rate of 88.7% at an extremely low SNR of −10dB, and its accuracy approaches 100% when SNR is above 0dB. The algorithm has stronger theoretical completeness and noise robustness.
Keywords: radar modulation recognition; low probability of interception; spectral-temporal cross-gating; uncertainty principle; multi-scale network
基金项目：国家自然科学基金项目（2019YJ0455）
作者简介：郭鹏(1987—)，男，重庆江津人，硕士，成都华力创通科技有限公司高级工程师、信息系统项目管理师，中国电子学会高级会员，中国通信学会高级会员，主要研究方向为电子侦察、雷达与电子对抗。E-mail：guopeng_cetc@163.com
牛知艺(2001—)，男，山西太原人，中国民用航空飞行学院硕士研究生，主要研究方向为信号与信息处理。E-mail：861719782@qq.com