Deep Learning Based Sub-Nyquist Modulation Recognition

Abstract

In this paper, we designed a Convolutional Neural Network (CNN) for Sub-Nyquist modulation recognition and compare the performance Long Short-Term Memory (LSTM) network and Convolutional Long Short-term Deep Neural Network (CLDNN) respectively. Unlike conventional modulation recognition task that operates with Nyquist sampled rate, the network architectures for Sub-Nyquist modulation recognition were specifically designed with a certain number of neurons, layers, and other hyperparameters to effectively extract key features from Sub-Nyquist sampled signals and process larger volumes of data. The simulation results demonstrate that the CNN network has the best recognition accuracy of 98.01% on the GBsense dataset, followed by the CLDNN of 96.81% and LSTM of 87.51% respectively.