Spectrogram-Based Automatic Modulation Recognition Using Convolutional Neural Network

Abstract

We study a system for classifying modulation types with spectrograms obtained through short-time Fourier transform. AWGN-based carrier modulated signals and their spectrograms are generated. In order to extract features from spectrogram automatically, we learned our convolutional neural network model with the generated data. Even at low SNRs, the performance is fairly good, but additional modulation type applications and comparisons with others in various environments are necessary.