Automatic Modulation Classification Using Fractional Low Order Cyclic Spectrum and Deep Residual Networks in Impulsive Noise

Abstract

Automatic modulation classification (AMC) plays an important role in cognitive radio, surveillance and electronic warfare systems. However, due to the presence of impulsive noise in wireless communication systems, conventional AMC methods which are based on Gaussian noise assumption usually exhibit poor performance in impulsive noise environment. Aiming at realizing efficient AMC with impulsive noise, a novel AMC method based on fractional low order cyclic spectrum (FLOCS) and deep residual networks (ResNets) is proposed in this paper. In this method, FLOCS can effectively suppress impulsive noise, and meanwhile the discriminating features can also be extracted from FLOCS for AMC. Then, on the basis of these features, deep residual networks are employed to classify different modulation types. Monte Carlo simulation results demonstrate that the proposed AMC method can achieve superior classification performance in impulsive noise environment.