Intelligent Cyclic Spectrum Features Based Modulation Recognition Design

Abstract

This paper proposes an intelligent deep learning aided modulation recognition system. In this design, we utilize the deep residual shrinkage network (DRSN) to identify the modulation types with the cyclic spectrum (CS) features as the data set. With the aim to reduce the computational complexity, we first use a half part of the XY-plane of the 3-dimensional CS, which is transformed into a gray-scale image to compose the dataset. Thanks to the statistical characteristics evaluation with the CS, the data set is noise-resilient. Then we develop the DRSN with soft thresholding and attention mechanism to combat the noise and interference, and to reserve key features of received modulated signals. Simulation results demonstrate that the proposed system can achieve a higher classification accuracy than counterpart methods with lower computational complexity.