DI-UMONS : Dépôt institutionnel de l’université de Mons

Recherche transversale
Rechercher
(titres de publication, de périodique et noms de colloque inclus)
2022-03-30 - Colloque/Présentation - poster - Anglais - 1 page(s)

Gros Alexander , Moeyaert Véronique , Mégret Patrice , "Analysis of CNN Data Input Shape for Automatic Modulation Classification" in Infortech’ Day, Mons, Belgique, 2022

  • Codes CREF : Technologie des télécommunications [transmission] (DI2556)
  • Unités de recherche UMONS : Electromagnétisme et Télécommunications (F108)
  • Instituts UMONS : Institut de Recherche en Technologies de l’Information et Sciences de l’Informatique (InforTech)
  • Centres UMONS : Centre de Recherche en Technologie de l’Information (CRTI)

Abstract(s) :

(Anglais) The AMC (Automatic Modulation Classification) domain has recently shown an increase of interest, particularly as an application of cognitive radios or physical layer security of wireless transmissions. Among all possible classification techniques, EMD (Empirical Mode Decomposition) and CNN (Convolutional Neural Networks) are interesting tools already used with success for classification. On the one hand, EMD is able to decompose single or bivariate signals into a finite number of IMFs (Intrinsic Mode Functions) that correspond to the decomposition algorithm stop criterium. On the other hand, CNN are already used in the scientific literature to classify modulations, starting from IQ values of modulated signals. In this work, it is proposed to study the type and the shape of the CNN inputs in order to improve the modulated signal recognition rates. Indeed, starting from a publicly available data base, the study of different combinations of input matrices is realised: 2D IQ signal only, 2D BEMD (Bivariate Empirical Mode Decomposition) and 3D BEMD. Those are applied to different modulated signals. It is shown that using BEMD instead of the simple IQ signal improves the overall accuracy of the recognition.