Dense Layer Dropout Based CNN Architecture for Automatic Modulation Classification
Pratibha Dileep, Dibyaiyoti Das, Prabin Kumar Bora
Abstract
Automatic modulation classification (AMC) is an important part of signal identification for cognitive radio as well as military communication. The problem has been approached traditionally using either likelihood-based or feature-based methods. Since the problem is a classification task, a deep learning (DL) based approach can be an attractive solution. A number of convolutional neural network (CNN) based DL algorithms were introduced for AMC recently. The complex baseband signals that are represented as In-phase and Quadrature (IQ) samples are applied to train the CNN. We propose a new CNN architecture that significantly improves the classification accuracy over existing results in the literature while keeping the number of trainable parameters low. In this architecture, dropouts are applied only in the dense layers.