Automatic Modulation Recognition with Deep Learning Algorithms

Abstract

In this study, an automatic modulation classifier based on Convolutional Neural Network (CNN) was developed using deep learning algorithms. A synthetic dataset generated with GNU Radio consisting of eleven modulations at varying signal-tonoise ratios was used for classification. The time-frequency domain images of the complex signals are generated. An efficient approach, Synchrosqueezing Transform (SST), is used to provide time-frequency representations of complex signals. A CNN model has been developed for the classification of time-frequency analysis images produced by the Fourier-Based Synchrosqueezig Transform (FSST) method. The new dataset, consisting of images taken with FSST, has been turned into a three-dimensional input for the designed CNN model. A modulation classifier was prepared by training these inputs with the CNN model created using seven convolution layers and five dense layers. The proposed classifier provided a high performance of eighty-seven percent.