HARDWARE IMPLEMENTATION OF CONVOLUTIONAL NEURAL NETWORKS FOR MODULATION CLASSIFICATION

Abstract

In modern wireless communication, adaptive devices are needed to overcome different challenges. Adding artificial intelligence to these adaptive devices can further increase their capabilities. One of its aspects is that the receiver should know which modulation schemes to handle. So, to achieve this objective, different deep learning algorithms are used at the receiver side. In this work, we will be using deep learning techniques for modulation classification. For this purpose, high accuracy coupled with fast and efficient processing is required. Moreover, most of the traditional methods are insufficient to classify while providing high accuracy at high end processing. To fulfil the classification task, a convolutional neural network (CNN) is used. The modulated signals are converted into a constellation diagram image, following that, the data is put into a convolutional neural network for training. On the basis of training, thesis objective is achieved with 95.69% accuracy. To achieve an above-stated objective, there is a need to design a hardware architecture for a deep learning algorithm CNN for modulation classification. This thesis is aimed to design hardware architecture for modulation classifiers based on the convolutional neural network along with field programmable gate array (FPGA) prototyping solutions.