基于谱相关特性和组合神经网络的数字调制信号识别

Modulation recognition of digital signals using spectral correlation and assembled neural networks

针对通信信号非稳定、信噪比(SNR)变化范围大的特性,利用调制信号的循环平稳特性,提取出五种对SNR和信号调制参数不敏感但对调制类型敏感的特征参量。为提高分类性能,设计了一种采用多个不同神经网络的组合分类器结构,采用输出向量加权表决的融合规则。仿真表明,低信噪比下组合神经网络分类器比单个神经网络分类器有更高的识别率。

Aiming at the feature of non-stabihty and largely changed character of SNR of communication signals, based on the spectral correlation feature of modulation signals, this paper extracted five characteristic parameters which were sensitive to modulation types but inactive to SNR and modulation parameter. And proposed a assembled classifier using some different neural networks to improve the performance, it adopted the fused method whose output vectors were added. The simulation results show that the assembled classifier has much higher recognition rate compared with the single neural network classifier.