IDENTIFICATION AND CLASSIFICATION OF SIGNALS USING GENERATIVE ADVERSARIAL NETWORKS

Abstract

Research has shown that machine learning holds promise as a technique to improve the identification and classification of signals of interest. This study proposes the use of machine learning, specifically generative adversarial networks, to classify received signals based on their down-converted, but not demodulated, in-phase and quadrature signals and evaluate their probability of being of interest. The approach used a generative adversarial network to train a classifier convolutional neural network to determine the likelihood that a received signal is of interest. We tested the ability of a semi-supervised generative adversarial network to classify signals of interest by modulation scheme. We then tested the ability of the semi-supervised generative adversarial network to identify unique signals of interest within a dataset of a single modulation scheme. We evaluated the performance of the network on accuracy, training time, and the amount of data needed to train the network. The results proved that a semi-supervised generative adversarial network could classify a signal by modulation scheme and identify signals within a single modulation scheme.