基于时隙IQ信号的瑞利衰落信道的调制信号识别  

Modulation Signal Recognition in Rayleigh Fading Channel Based on Time Slot IQ Signal

作　　者：马运壮 贺英[1] 郑鑫[1] MA Yunzhuang;HE Ying;ZHENG Xin(College of Electronics and Information,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学电子信息学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2023年第2期60-67,共8页Journal of Qingdao University(Engineering & Technology Edition)

基　　金：国家自然科学基金资助项目(61701271)。

摘　　要：针对瑞利信道调制信号存在的识别困难问题,本文提出一种基于时隙IQ信号的轻量级调制信号识别网络。将输入的IQ信号分不同的时隙,依次经过改进的GhostNet网络,提取信号的空间特征,并将多路时隙特征并行送入双向长短期记忆网络(bi-directional long short-term memory network,BiLSTM),提取信号的时域特征,最后由全连接网络(deep neural network,DNN)完成分类。同时,对仿真得到的多进制数字相位调制(multiple phase shift keying,MPSK)、多进制正交幅度调制(multiple quadrature amplitude Modulation,MQAM)等8种调制信号进行实验。实验结果表明,当慢衰落和快衰落信道在0 dB时,其识别准确率均能达到90%以上,说明GLDN模型具有良好的抗噪声和抗衰落性能,该模型对8种调制方式识别的F1-Score均在0.93以上,有效提升了高阶调制方式的识别性能,而且通过合理划分时隙长度,模型在更少的参数量下取得了较高的精度。该研究在非合作通信系统中具有重要意义。Automatic modulation recognition technology of wireless signal has important research significance in non-cooperative communication system.For the problem that it is difficult to identify modulated signals on the Rayleigh channel,a lightweight modulation signal recognition network named as GLDN is proposed,which is based on slot IQ signal.Firstly,the input IQ signals are divided into different time slots.The IQ signals of each time slot are sequentially processed by the improved GhostNet network,and the spatial features of IQ signals are extracted in the convolution layer.Then,the multi-path time slot features are parallelly input into the Bi-directional Long Short-Term Memory Network(BiLSTM)to extract the sequence features.Finally,the type of modulation mode is output by the deep neural network(DNN).The eight modulation signals of MPSK(multiple phase shift keying)and MQAM(Multiple Quadrature Amplitude Modulation)are obtained by simulation.The experimental results show that GLDN model has better anti-noise performance and anti-fading performance,and can achieve more than 90%recognition accuracy at 0dB in slow fading and fast fading channels.The F1-Score of all eight modulated signals is above 0.93,which effectively improves the recognition performance of higher-order modulation modes.By properly dividing the length of the time slot,the model can achieve higher accuracy with fewer parameters.

关 键 词：调制识别 瑞利信道 时隙IQ信号 GhostNet 双向长短期记忆网络 

分 类 号：TN911.3[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]