Automatic modulation classification, cyclostationary analysis, cyclic correntropy spectral density, impulsive environment, alpha-stable distribution. 

The continued growth in the use of wireless communication systems has been contributing to the search for new ways to exploit the maximum capacity of spectrum. In this context, cognitive radios appear as an appropriate option, able to offer an efficient use of the channel, allowing the dynamic access of the spectrum and guaranteeing large bandwidth to the users. An important cognitive radio feature is the ability to automatically recognize the signal type used in the channel. This attribute, known as automatic modulation classification (AMC), has several applications in communication systems, such as electronic surveillance, improved interoperability of networks and diverse communication systems. AMC techniques based on the feature detection obtained by the second order cyclostationary analysis are very common, however are not able to discriminate some types of digital modulations, such as an M-QAM modulation. On the other hand, the higher-order cyclostationary techniques used to extract singular descriptors of these modulations has a very high computational cost and are only suitable for AWGN channels. Although the AWGN model is widely used to characterize communication channels, there are several practical scenarios that are better modeled by non-Gaussian distributions, such as HF communication, where the environment is strongly contaminated by impulsive noise. The typically solution adopted for the signals cyclostationary analysis in impulsive environments consists of fractional lower-order statistics analysis. However, these techniques in the context of communication sistems are only used for spectral sensing. Thus, in view of the lacuna in the state of art, referring to AMC techniques in impulsive environments applied in high-order digital modulations, in this work we propose the investigation of new ciclostationary feature extraction methods for the M-QAM digital modulations, in impulsive noise environments, using the cyclic correntropy spectral density function (CCSD). The CCSD was able to extract the cyclostationary singular features M-QAM modulations, even when they were contaminated by alpha-stable impulsive noise. This demonstrated that CCSD is a promising tool to be employed in a digital modulation classification system.