| 摘要: |
| 由于无线通信信道的开放性，通信信号在传输过程中容易受到各类自然或人为干扰影响，通信信号和干扰信号交织形成时频重叠信号,在低干信比条件下，传统信号识别方法性能不佳。针对这一问题，基于独立成分分析(Independent Component Analysis,ICA)算法和通道注意力机制(Channel Attention,CA)的卷积神经网络(Convolutional Neural Network,CNN)，提出了一种时频重叠信号识别方法(Overlapping Signals Recognition on ICA and CNN,OSR- IC)。该方法使用ICA算法将时频重叠信号分解为通信信号和干扰信号，通过快速傅里叶变换获得通信信号和干扰信号频谱图，以两类信号频谱图作为CNN网络的输入，引入通道注意力机制获取每个通道的权重进而改进网络特征表达能力，使用改进后的CNN网络对干扰信号进行识别。仿真实验表明，在干噪比为0 dB时，所提方法对干扰信号的识别率可达94%及以上。 |
| 关键词: 干扰信号识别 时频重叠信号 独立成分分析 通道注意力机制 卷积神经网络 |
| DOI：10.20079/j.issn.1001-893x.240808003 |
|
| 基金项目:国家重点研发计划(2021YFB2900404) |
|
| Recognition of Time-Frequency Overlapping Signals Based on Joint ICA and CNN |
| ZHOU Shangcong,ZHANG Yong,LI Dongfang,ZHANG Zhonghao,WANG Liu |
| (1.College of Communication Engineering,Chengdu University of Information Technology,Chengdu 610225,China;2.National Key Laboratory of Science and Technology on Communications,University of Electronic Science and Technology of China,Chengdu 611731,China;3.Key Laboratory of Meteorological Information and Signal Processing of Sichuan Provincial Universities,Chengdu 610225,China) |
| Abstract: |
| Due to the openness of wireless communication channels,communication signals are easily affected by various natural or artificial interferences during transmission.Communication signals and interference signals are intertwined to form time-frequency overlapping signals.Under low interference-to-signal ratio conditions,traditional signal recognition methods perform poorly.To address this problem,based on the independent component analysis(ICA) algorithm and the convolutional neural network(CNN) with the channel attention mechanism(CA),a method for time-frequency overlapping signal recognition on ICA and CNN(OSR-IC) is proposed.This method uses the ICA algorithm to decompose the time-frequency overlapping signal into communication signals and interference signals,obtains the spectrum of the communication signal and the interference signal through fast Fourier transform,uses the two types of signal spectrum as the input of the CNN network,introduces the channel attention mechanism to obtain the weight of each channel,and then improves the network feature expression ability,and uses the improved CNN network to identify the interference signal.Simulation experiments show that when the interference-to-noise ratio is 0 dB,the proposed method can achieve a recognition rate of 94% or more for interference signals. |
| Key words: interference signals identification time-frequency overlapping signal independent component analysis channel attention mechanism convolutional neural network |
《电讯技术》编辑部 