A signal modulation classification algorithm based on convolutional neural network

Abstract

As a hot research direction in the field of communication reconnaissance, signal modulation classification plays an increasingly important role in the field of national defense. Traditional signal modulation style classification methods are mostly based on the combination of feature engineering and pattern recognition. First, the expert features are extracted by manual design, and then the signal modulation is recognized by the pattern recognition algorithm. The limitation of this method is that an expert feature can only effectively identify a few specific modulation signals. Besides, the deficiency of the number of expert features will lead to a low classification accuracy. In order to improve the accuracy of signal modulation classification, a signal modulation classification algorithm based on convolutional neural network is proposed. Convolutional neural networks can achieve end-to-end classification without manually designing and extracting features. Convolutional neural networks can automatically extract various levels of abundant features through learning, which can improve classification accuracy. The convolutional neural network architecture designed in this paper includes: three convolutional layers, three pooling layers, and the last layer is the softmax classification layer, which outputs the classification results. Experimental results show that on a data set containing 32 I/O signals, when the signal-to-noise ratio is 6dB, the algorithm has a training accuracy rate of 92.7% and a test accuracy rate of 90.2%.