基于深度学习的通信辐射源调制样式识别方法

Deep-learning-based modulation pattern recognition method for communication emitter

针对自动调制分类中通信辐射源调制方式识别率低问题,提出了一种基于短时傅里叶变换(STFT)和卷积神经网络(CNN)结合的方法.该方法首先对通信辐射源信号进行小波阈值降噪,去除混在信号中的高斯白噪声;然后经过短时傅里叶变换,将一维时域信号变换成二维时频域图像,利用临近插值法降维;将时频图输入卷积神经网络进行训练,通过对超参数的选取,得到优化的卷积神经网络;最后采用softmax函数给出识别结果.仿真结果表明,当信噪比(SNR)为0 dB时,利用本文识别方法的宏平均值达到0.874以上,其性能显著优于传统方法.

Aiming at the problem of low recognition rate of communication emitter modulation pattern in automatic modulation classification, this paper proposes a method based on the combination of short-time Fourier transform(STFT) and convolutional neural network(CNN). This method first performs wavelet threshold noise reduction on the communication radiation source signal to remove white Gaussian noise mixed in the signal. And then by using STFT, the one-dimensional time-domain signal is converted into a two-dimensional time-frequency domain image, using neighbor interpolation to reduce dimensionality. The time-frequency map is also input to the CNN for training, and the optimal CNN is obtained by selecting the hyper-parameters. Finally, the softmax function is used to give the recognition results. Simulation results show that when the signal-to-noise ratio(SNR) is 0 dB,the average value of macros identified by this recognition method is above 0.874, and its performance is significantly better than that of the traditional method.