Classification of modulated radiosignals based on convolutional neural networks

Abstract

The work is devoted to the study of convolutional neural networks for use in the classification of modulated radio signals. Automatic classification of modulation signals has a wide variety of wireless applications. Also shown are modulated radio signals at different SNR levels. In this work, we used data from the DeepSig base of radio modulated signals, created using GNU Radio. Based on this, the classification of modulation using the latest generation convolutional neural networks is considered. The work shows graphs of dependences showing the accuracy of training the network, as well as matrices of inaccuracies with various types of modulated radio signals. It is shown that convolutional neural networks of the latest generation are the most suitable for solving this problem, since they have the ability to quickly learn and accurately determine.