Robust Automatic Modulation Recognition Using Convolutional Deep Neural Networks

This work looks into Automated Modulation Categorization (AMC), essential in military and civil projects. The Convolutional neural networks (CNN) form the foundation of a powerful AMC technique that is described in this paper to solve the existing drawback of many FB AMC techniques where they are designed to detect a limited number of modulation types and cannot easily be adapted to work with other types of modulation. It also saves power and increases accuracy. Eleven forms of modulation are analyzed in total across all categories. The two forms of conventional AMCs are feature-based and maximum likelihood (ML). Here, an efficient AMC method that makes use of CNN is proposed and implemented for modulation classification. The described approach may partition the incoming signals and decide systematically about their properties with no need for feature extraction. It also determines the most effective optimizer between the two approaches over the three different signal-to-noise ratios (high, medium, and low). The two methods are Adaptive Moment Estimation (Adam) and Stochastic Gradient Descent with Momentum (SGdm). A comparative analysis between the proposed CNN-based AMCs was performed.