基于卷积神经网络的通信信号调制识别研究

Research on communication signal modulation recognition based on convolution neural network

摘要:

针对传统人工提取专家特征来进行通信信号识别的方法存在局限性大、低信噪比下准确率低的问题，提出一种复基带信号与卷积神经网络自动调制识别相结合的新方法。该方法将接收到的信号进行预处理，得到包含同相分量和正交分量的复基带信号，该信号作为输入卷积神经网络模型的数据集，通过多次训练调整模型结构以及卷积核、步长、特征图和激活函数等超参数，利用训练好的模型对通信信号进行特征提取和识别。实现了对2FSK、4FSK、BPSK、8PSK、QPSK、QAM16和QAM64 七种数字通信信号类型的识别分类。实验结果表明，当信噪比为0dB时，七种信号的平均识别准确率已达94.61%，验证了算法是有效的且在低信噪比条件下有较高的准确率。

Abstract:

In the task of communication signal recognition, to improve the limitation and low accuracy under the condition of low signal-to-noise ratio (SNR) which used the traditional manual extraction of expert feature, a new method of automatic modulation recognition based on complex baseband signals and convolutional neural network is proposed. In this method, the received signals are preprocessed to obtain the complex baseband signal containing In-Phase components and Quadrature components. The complex baseband signal is input to convolutional neural network model as data set, which trains the convolutional neural network model for many times and adjusts the model structure, filter size, stride, feature map, activation function and other super parameters. The trained convolutional neural network model is used to extract the features and classify signals. The classification and recognition of seven kinds of digital communication signals including 2FSK, 4FSK, BPSK, 8PSK, QPSK, QAM16 and QAM64 are realized. The experimental results show that when the SNR is 0db, the average recognition accuracy of seven types of signals can reach 94.61%, which proves that the algorithm is effective and has high accuracy under the condition of low SNR.