Open AccessOn classifiers for blind feature‐based automatic modulation classification over multiple‐input–multiple‐output channels
Author(s) -
Kharbech Sofiane,
Dayoub Iyad,
ZwingelsteinColin Marie,
Simon Eric Pierre
Publication year - 2016
Publication title -
iet communications
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.355
H-Index - 62
eISSN - 1751-8636
pISSN - 1751-8628
DOI - 10.1049/iet-com.2015.1124
Subject(s) - computer science , random subspace method , artificial intelligence , support vector machine , pattern recognition (psychology) , classifier (uml) , feature selection , decision tree , artificial neural network , machine learning , cross validation , context (archaeology) , paleontology , biology
Modulation recognition is crucial for a good environmental awareness required by cognitive radio systems. In this study, the authors design and compare models of four among the most commonly used classifiers for feature‐based automatic modulation classification (FB‐AMC) algorithms. Classifiers whose models will be designed are classification tree, K ‐nearest neighbours, artificial neural networks (ANNs), and support vector machines. In this study, they apply some statistical pattern recognition techniques in the context of blind FB‐AMC over multiple‐input–multiple‐output channels. Comparison criteria are classification accuracy and computational complexity. To improve the impartiality of this comparison, each classifier is optimally deployed by selecting its optimal model with respect to their context. Model selection for the classifiers is done using the ‘ k ‐fold cross‐validation’ model validation technique. The comparison study, within the considered context, shows that ANN classifiers have the best performance/complexity tradeoff.