基于CLR和改进卷积神经网络的调制方式识别算法  

Modulation Recognition Algorithm Based on CLR and Enhanced Convolutional Neural Network

作　　者：王晓宇 张邦宁 杨宁 郭道省 WANG Xiaoyu;ZHANG Bangning;YANG Ning;GUO Daoxing(College of Communication Engineering,Army Engineering University of PLA,Nanjing 210007,China)

机构地区：[1]中国人民解放军陆军工程大学通信工程学院,江苏南京210007

出　　处：《无线电工程》2025年第1期67-75,共9页Radio Engineering

摘　　要：为提高非合作通信场景中调制方式识别任务在低信噪比(Signal to Noise Ratio,SNR)条件下的识别率及实时性,对卷积神经网络(Convolutional Neural Network,CNN)结构和学习率策略进行了改进,设计了一种基于循环周期学习率(Cyclic Learning Rate,CLR)策略和改进CNN的调制方式识别算法。为了突出信号特征,通过短时傅里叶变换(Short Time Fourier Transform,STFT)生成信号的时频图,引入注意力机制对CNN进行改进,用于抑制信号中的冗余信息,实现特征提取,增强在低SNR条件下算法的识别能力,通过设计CLR策略,对算法超参数进行设置,提高算法的收敛速度。实验结果表明,在-10 dB条件下,识别率可达92%,相较于其他经典算法,识别率得到显著提升,所提出的算法参数量小、计算复杂度低、收敛速度快。To enhance the recognition rate and real-time performance of modulation recognition tasks in non-cooperative communication scenarios under low Signal to Noise Ratio(SNR)conditions,an improved modulation recognition algorithm based on Cyclic Learning Rate(CLR)strategy and enhanced Convolutional Neural Network(CNN)is proposed.Initially,to highlight signal features,Short Time Fourier Transform(STFT)is employed to generate time-frequency images of the signals.Subsequently,an attention mechanism is introduced to modify CNN,suppressing redundant information in the signals,enabling efficient feature extraction,and strengthening recognition capabilities of the algorithm under low SNR conditions.Finally,the CLR strategy is designed to fine-tune the hyperparameters of the algorithm,enhancing its convergence speed.Experimental results demonstrate that the recognition rate can reach 92%under-10 dB conditions,significantly outperforming other classical algorithms.Additionally,the proposed algorithm exhibits a low parameter count,reduced computational complexity,and rapid convergence.

关 键 词：调制方式识别 短时傅里叶变换 卷积神经网络 注意力机制 循环周期学习率策略 

分 类 号：TN911.7[电子电信—通信与信息系统]