Automatic Modulation Classification using DenseNet

Wireless Communication over long distances has revolutionised the way world works, functions and interacts.Modulation of a signal containing information has made all this possible.The identification of modulation type of a signal forms an integral part in several military and civilian applications.While the classical approaches of modulation classification provides optimal accuracy, they are less robust, requires human interference in the form of engineering expertise and involves computational complexity.The ability to automatically identify the modulation type using deep learning is termed as Automatic Modulation Classification (AMC).This mitigates the need for hand-crafted features and provides state of the art classification accuracy.AMC using neural network (NN) architectures such as Convolutional Neural Networks (CNN), Residual Neural Networks (ResNets) provide excellent modulation classification.While CNN suffers from vanishing gradient,ResNet architecture is used to facilitate smooth gradient flow over the layers so that learning happens more effectively.In this study the main focus is to reduce the training time taken by ResNets which is due to huge number of parameters to be trained.Towards this, Densely Connected Convolutional Neural Network (DenseNet) architecture is trained with few parameters which takes less time and is found to achieve optimal results.