摘要
在时变色散信道和低信噪比下,基于时频图利用深度神经网络的短波信号识别的方法对并行多音信号的识别能够取得一定效果,但常用的相位调制的串行单音短波信号识别难以取得较好效果。由于短波信号帧结构中都含有同步帧和发射电平起控帧等特征数据,提出了一种基于同步帧注意力机制的卷积神经网络结构。测试结果表明,使用卷积神经网络(Convolutional Neural Networks,CNN)、双向长短时记忆(Bi-directional Long Short Term Memory,Bi-LSTM)和注意力机制搭建成的深度神经融合网络模型对短波基带信号直接进行检测识别能够取得较好效果。
Under the conditions of time-varying dispersive channel and low signal-to-noise ratio,the deep neural network method based on time-frequency diagram can achieve certain results for the recognition of parallel multi-tone signals.However,the commonly used phase-modulated serial monotone HF signal identification is difficult to achieve better results.Since the frame structure of shortwave signals contains feature data such as synchronization frames and emission level starting frames,a convolutional neural network structure based on the attention mechanism of synchronous frames is proposed.The test results indicate that the deep neural fusion network model built with CNN(Convolutional Neural Networks),Bi�LSTM(Bi-directional Long Short Term Memory)and attention mechanism can achieve better results in detecting and recognizing HF baseband signals directly.
作者
王鹏
黄伟强
WANG Peng;HUANG Weiqiang(Guangzhou Haige Communication Group Inc.,Co.,Guangzhou Guangdong 510663,China)
出处
《通信技术》
2023年第5期566-573,共8页
Communications Technology
