Unsupervised Neural Network for Modulation Format Discrimination and Identification

Abstract

We propose a new method to discriminate and identify the modulation format of signals based on an unsupervised neural network named convolutional Gaussian-Bernoulli restricted Boltzmann machine (CGBRBM). Tests are performed to demonstrate how the proposed method works and to evaluate the discrimination/identification accuracy for different input combinations. Signals of five modulation formats are used to test the CGBRBM-based algorithm including QPSK, 8QAM, 16QAM, 32QAM, and 64QAM. The results indicate the performance of the proposed method when dealing with various application scenarios and reveal the relation between discrimination accuracy and identification accuracy.