基于复数神经网络的通信调制信号降噪方法  

Noise reduction method of communication modulation signal based on complex neural network

作　　者：高明慧 唐晓刚 张斌权 潘协昭 彭钧 GAO Minghui;TANG Xiaogang;ZHANG Binquan;PAN Xiezhao;PENG Jun(School of Space Information,Space Engineering Univ.,Beijing 101416,China;Unit No.32039,Beijing 102300,China)

机构地区：[1]航天工程大学航天信息学院,北京101416 [2]32039部队,北京102300

出　　处：《海军工程大学学报》2025年第1期43-50,共8页Journal of Naval University of Engineering

基　　金：国家自然科学基金资助项目(62027801)。

摘　　要：针对复杂空间电磁环境下通信调制信号识别率低的难题,提出了一种基于复数长短期记忆(complex long and short-term memory,CLSTM)的自编码器降噪网络(complex noise reduction networks,CNRN)模型,设计了CLSTM的实部核、虚部核及复数交叉项,有效提取了带噪通信调制信号的复数域和时域融合特征,并构建了直接、间接和可视化三种评价方法。实验仿真表明:降噪后-20 dB信号信噪比提升15 dB,均方误差降低6.62,误比特率降低0.11;降噪后-20~18 dB信号三种典型调制识别网络的平均识别率分别提高了29.05%、38.76%、21.87%。To solve the problem of low recognition rate of communication modulated signals in complex spatial electromagnetic environments,a self-encoder noise reduction network(CNRN)model based on complex long and short-term memory(CLSTM)was proposed,where the CLSTM real kernel,imaginary kernel,and the complex cross terms were designed to effectively extract the fusion features of complex and time domains of the noise-carrying communication modulated signals,and the three methods,i.e.,direct,indirect,and visualization evaluation methods were constructed.The experimental simulation shows that the signal-to-noise ratio of the-20 dB signal after noise reduction is improved by 15 dB,the mean square error is reduced by 6.62,and the bit error rate is reduced by 0.11,and that the average recognition rates of the three typical modulation recognition networks of the-20~18 dB signal after noise reduction are improved by 29.05%,38.76%,and 21.87%,respectively.

关 键 词：通信调制信号 信号降噪 复数神经网络 

分 类 号：TN911.3[电子电信—通信与信息系统]