分离通道联合卷积神经网络的自动调制识别

针对通信信号的自动调制识别需要大量特征提取的问题，提出了一种分离通道卷积神经网络自动调制识别算法。该算法通过结合深度学习中卷积神经网络(CNN)，分别提取时域信号的多通道和分离通道调制特征，再利用融合特征实现不同信号的分类。仿真结果表明，相比基于CNN的算法，所提算法在高信噪比下针对两个数据集的识别率分别提升7%和18%；此外，相比于基于特征提取的传统识别算法，其高阶调制识别性能平均提升3 dB。

Automatic modulation recognition based on separate channel combined convolutional neural networks

An automatic modulation recognition algorithm based on Separate Channel Combined Convolutional Neural Network(SCC-CNN) is proposed to solve the mass computation of feature extraction in conventional automatic modulation recognition methods.The algorithm can extract features from multi-channel and separate channels of time domain data respectively by combining the convolutional neural network of deep learning,and then different signals can be classified by combined features.The simulation results reveal that compared with the methods based on CNN the proposed algorithm can improve the accuracy under high signal-to-noise ratio(SNR) more than 7% and 18% on two different datasets;furthermore,high order modulation recognition performance is improved more than 3 dB compared with the methods based on feature extraction.