Indian Journal of Science and Technology
DOI: 10.17485/ijst/2020/v13i02/148648
Year: 2020, Volume: 13, Issue: 2, Pages: 200 – 212
Original Article
Jean Baptiste Bi Gouho1,*, Sidibé Karim1, Boko Aka2 and Michel Babri3
1Laboratoire de Mathématiques Informatique, Université Nangui Abrogoua, Abidjan, Côte d’Ivoire
2Institut de Recherche sur les Energies Nouvelles, Université Nangui Abrogoua, Abidjan, Côte d’Ivoire
3Institut National Polytechnique Félix Houphouët Boigny, Yamoussoukro, Côte d’Ivoire
*Author for correspondence:
Jean Baptiste Bi Gouho
E-mail ID: gouhobi@yahoo.fr
Objectives: This study aims to present a framework for Automatic Modulation recognition using Deep learning without feature extraction.
Methods: We study seven modulations using the In-Phase Quadrature constellation polluted by Additive White Gaussian Noise. We apply the K-means algorithm to normalize data transmitted and polluted by noise; the new diagram obtained is considered as an image and coded in pixel before entering in a Deep Neural Network where we apply 20% dropout on hidden layers to avoid overfitting. The simulation is carried out in Matlab.
Findings: Experiment performed on selected modulations following the proposed framework gives a good percentage of recognition equal to 96.12%. Our algorithm Deep Neural Network imaGe gives the best performance results at epoch equal to 2,000,000.
Applications: The outcome will be beneficial for researchers in Software-Defined Radio for civilian and military applications like electronic attacks and electronic protection.
Keywords: Modulations, I–Q Diagram Constellation, Clustering, Deep Neural Network, Dropout.
Subscribe now for latest articles and news.