融合神经网络与瞬时自相关分区特征的自动调制分类方法研究

Research on automatic modulation classification method based on neural network and instantaneous autocorrelation partition feature

复杂电磁环境提升了认知电子战系统对外部电磁信号的感知、识别与分类能力的要求。针对认知电子战系统所需识别的多种复杂通信信号辐射源,基于瞬时自相关相位分布特征图进行分区特征提取,联合包络方差特征,并引入神经网络这一机器学习算法,构造出了针对多种通信信号自动分类的机器学习网络。提出采用信号瞬时自相关相位分区分布比率作为神经网络输人层信号,并逐层传递经过隐含层及输出层处理以实现信号之间的分类。采用计算机仿真实现机器学习分类网络的训练,并采用测试样本对算法性能进行了验证。仿真表明,信噪比大于13 dB时,分类网络可以获得大于90%的正确识别率。

The complex electromagnetic environment raises the requirement of the cognitive electronic warfare system on the ability to perceive,recognize and classify the external electromagnetic signals.According to the cognitive electronic warfare system needed to identify a variety of complex communication signal source,based on the instantaneous autocorrelation characteristics of phase distribution graph partitioning feature extraction,this paper combines envelope variance characteristics of neural network and introduces the structure of machine learning algorithm,constructes for a variety of automatic classification of machine learning network communication signals.The instantaneous auto-correlation phase partition distribution ratio is used as the input layer of neural network,and the signal is transmitted layer by layer and processed by hidden layer and output layer to realize the classification between signals.The training of machine learning classification network is realized by computer simulation,and the performance of the algorithm is verified by test samples.The simulation results show that when the SNR is greater than 13 dB,the classification network can obtain correct recognition rate of more than 90%.