Automatic Modulation Classification using CNN and SVM

Abstract

AMC assumes paramount significance in modern wireless communication networks, facilitating rapid signal processing, interference reduction, and resource allocation with utmost efficiency. In this research paper, the incorporation of CNN and SVM techniques for AMC is examined by employing a variety of cutting-edge deep learning frameworks, including ResNet, Inception-ResNet, MobileNet, DenseNet, and AlexNet. For training and evaluations, HISARMOD2019 is employed which is a highly regarded benchmark dataset across the field. The method proposed involves a multi-phased process, from preprocessing and feature extraction to training the CNN and SVM classifiers with adapted raw data. Through the application of innovative deep learning strategies, the key features and nuances present in diverse modulation approaches can be automatically identified. Harnessing the unique capacity of SVM algorithms to detect minute differences within the data allows them to correctly categorize even the most complex instances with remarkable precision. Exhaustive testing is conducted to gauge the performance of the suggested methodology. The performance of each architecture across diverse SNR levels is assessed, evaluating its accuracy, resilience to noise, and speed. A side-by-side assessment of the models is undertaken to pinpoint their suitability in distinct scenarios. The study indicates that fusing CNN and SVM leads to more accurate modulation identification. Integrating deep learning features with SVM-based classification leads to impressive performance across spectrum of modulation schemes. Insights from the research shed light on the viability of diverse deep learning frameworks within practical wireless communication environments. Findings of this research can help steer the selection of suitable models to heighten the accuracy and efficiency of automatic modulation classification.